
Citation 
 Permanent Link:
 http://ufdc.ufl.edu/AA00024910/00001
Material Information
 Title:
 Model enhancements and parameter extraction for the MMSPICE/QBBJT model
 Creator:
 Lee, TzungYin, 1969
 Publication Date:
 1997
 Language:
 English
 Physical Description:
 vii, 133 leaves : ill. ; 29 cm.
Subjects
 Subjects / Keywords:
 Bipolar transistors ( jstor )
Crowding ( jstor ) Doping ( jstor ) Electric current ( jstor ) Electric potential ( jstor ) Equivalent circuits ( jstor ) Modeling ( jstor ) Parametric models ( jstor ) Simulations ( jstor ) Transistors ( jstor ) Dissertations, Academic  Electrical and Computer Engineering  UF ( lcsh ) Electrical and Computer Engineering thesis, Ph. D ( lcsh ) City of Gainesville ( local )
 Genre:
 bibliography ( marcgt )
theses ( marcgt ) nonfiction ( marcgt )
Notes
 Thesis:
 Thesis (Ph. D.)University of Florida, 1997.
 Bibliography:
 Includes bibliographical references (leaves 128132).
 Additional Physical Form:
 Also available online.
 General Note:
 Typescript.
 General Note:
 Vita.
 Statement of Responsibility:
 by TzungYin Lee.
Record Information
 Source Institution:
 University of Florida
 Rights Management:
 All applicable rights reserved by the source institution and holding location.
 Resource Identifier:
 0027600329 ( ALEPH )
37232314 ( OCLC )

Downloads 
This item has the following downloads:

Full Text 
MODEL ENHANCEMENTS AND PARAMETER EXTRACTION FOR THE MMSPICE/QBBJT MODEL
BY
TZUNGYIN LEE
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE
UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA 1997
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude and appreciation to my advisor, Dr. Robert M. Fox, for his devoted guidance, patient encouragement, and support throughout my Ph.D study. It is a great pleasure to work with him. I would also like to give thanks to my supervisory committee, Dr. Gijs Bosman, Dr. Jerry G. Fossum, Dr. Mark Law, and Dr. Mang Tia for their willingness to provide me their time and guidance.
I am grateful to the Semiconductor Resource Company (SRC), Motorola, and Texas Instruments, Inc. for their technical and financial support throughout my research. I feel especially obligated to express my deep appreciation to Dr. Mark Foisy at Motorola for his invaluable comments and discussions. I am also grateful to Mr. Tom Vrotsos, Dr. Keith Green and Dr. Zhiliang Chen, for their guidance during my internship in Texas Instrument Inc. Without their support, the work could never have been finished. I would also like to thank Dr. Ranjit Gharpurey for the invaluable discussion on BJT substrate resistance calculation.
I would like to recognize in this achievement Dr. MingChang Liang and Messrs. David Zwidinger, Jonathan Brodsky, MingYeh Chuang, and Amed Nadeem, who helped me in many ways with sincere and profound discussions on related topics, and my best friend, Everett Yang, who helped me in editing the drafts of this dissertation.
My gratitude is also extended to many of my friends, who have made my life most cheerful. I cannot mention them all, but I would like to mention Dr. Paul Chen, Dr. Simon
ii
Wang, Dr. Bruce Liu, Jenfeng Yue, Jenyi Yue, and Chihou Vong. Without them, my life would have been difficult and barren.
My deepest gratitude goes to my parents ChenI and Kuolan Lee. They presented themselves as an example in several ways that I can be humble and responsible. In addition, without their unfailing love and financial support, I could never have come to the United States and begun my graduate studies.
I am pleased that my wife Fiona has been able to share the achievement with me. Being a wife of a student for many years is not easy at all. Finally I would like to give my praises to my Lord, Jesus Christ, for He is the beginning and He is the end.
iii
TABLE OF CONTENTS
page
ACKNOWLEDGEMENTS ................................................................................................ ii
A B S T R A C T ....................................................................................................................... vi
CHAPTERS
I IN T R O D U C T IO N ................................................................................................... 1
2 BACKGROUND, PHYSICS, AND MODEL FORMULATION OF THE
MMSPICE CHARGEBASED BIPOLAR TRANSISTOR MODEL QBBJT ... 10
2.1 B ackground R eview ............................................................................ 10
2. 1.1 EbersM oll M odel ...................................................................... 10
2.1.2 GummelPoon Model ................................................................ 11
2.2 Novel Compact Models for Bipolar Transistors ................................. 12
2.2.1 The Extended GummelPoon Model ......................................... 12
2.2.2 M EX TR A M ............................................................................... 13
2.2.3 MMSPICE/QBBJT Model ......................................................... 14
2.2.4 Vertical Bipolar Inter Company Model (VBIC95) .................... 16
2.3 Model Formulation of QBBJT ............................................................ 17
2.3.1 BJT Operation Modes ................................................................ 17
2.3.2 Formulation of DC CurrentVoltage Equations ......................... 22
2.4 QBBJT's SmallSignal Equivalent Circuit ......................................... 24
3 MODEL ENHANCEMENTS OF THE QBBJT MODEL .................................... 30
3.1 Selfheating M odeling ........................................................................ 30
3.2 DC Current Crowding Modeling ........................................................ 35
3.3 AC Current Crowding Modeling ........................................................ 44
3.3.1 Simple Model Formulation ........................................................ 44
3.3.2 Accounting for DC Current Crowding in AC Current Crowding A n aly sis ..................................................................................... 4 9
3.3.3 Model Implementation and Verification .................................... 52
3.3. 4 Conclusion and Discussion .............................................................. 55
iv
4 PARAMETER EXTRACTION FOR M[M[SPICE/QBBJT .................................... 58
4 .1 Introduction ......................................................................................... 58
4.2 Extraction of GurnmelPoonLike Parameters ................................... 68
4.3 Extraction of Doping Profiles ............................................................. 70
4.3.1 Collector Doping Profile ............................................................ 70
4.3.2 Base Doping Profile ................................................................... 75
4.3.3 Integration of Doping Profile Extraction Procedures ................ 76
4.3.4 Extraction Results and Discussion ............................................. 77
4.4 Extraction of QBBJT's SmallSignal Equivalent Circuit ................... 84
4.4.1 M ethodology .............................................................................. 84
4.4.2 Calibration of QBBJT's Extrinsic Resistance Parameters ......... 88
4.4.3 Evaluation of QBBJT's Charge Partition Factors ...................... 92
4 .5 C onclusion .......................................................................................... 95
5 SPICE MODELING OF BJT SUBSTRATE RESISTANCE .............................. 101
5.1 Introduction ....................................................................................... 10 1
5.2 Algorithm to Calculate BJT Substrate Resistance ............................ 104
5.3 Limitations of the Algorithm ............................................................ III
5.4 AC Verification of the Algorithm ..................................................... 113
5.5 Including BJT Substrate Resistance in Parameter Extraction .......... 116
5.6 C onclusions ....................................................................................... 118
6 SUMMARY AND SUGGESTED FUTURE WORK ......................................... 119
6 .1 S um m ary ........................................................................................... 119
6.2 Suggested Future Work ..................................................................... 122
APPENDIX T14E SUBSTRATE RESISTANCE CALCULATION PROGRAM
SU B T E ST ......................................................................................... 125
R E FE R E N C E S ................................................................................................................ 128
BIOGRAPHICAL SKETCH ........................................................................................... 133
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
MODEL ENHANCEMENTS AND PARAMETER EXTRACTION FOR THE MMSPICE/QBBJT MODEL
By
TzungYin Lee
May 1997
Chairman: Robert M. Fox
Major Department: Electrical and Computer Engineering
This dissertation addresses several enhancements for the MMSPICE/QBBJT model which are critical to contemporary bipolar transistors. First, selfheating is included by adding an extra thermal node into QBBJT's admittance matrix. Through the extra thermal node, the transistor's local temperature is simulated along with its terminal currents and voltages. Second, QBBJT's dc current crowding analysis is modified to better model the transistor's high injection behavior. Third, a compact and physical ac current crowding model is developed. AC current crowding is predominantly a nonquasistatic (NQS) behavior caused by the majority carriers transporting laterally along the base. By adapting the RC transmission line model, an equivalent circuit is developed to account for the NQS carrier transport. The equivalent circuit also accounts for the base resistance reduction due to dc current crowding. With the effective base width calculated from dc
Vi
current crowding analysis, the model can also be extended to account for the nonuniformity in the baseemitter admittance along the base.
This dissertation presents a complete parameter extraction methodology for the MMSPICE/QBBJT model. QBBJT is designed to take the model parameters derived from transistor's structural information such as layout and doping profile. In such a way, QBBJT enables simulation even when the process is not yet finalized. This attribute makes QBBJT especially useful in concurrent engineering applications. However, QBBJT still needs a methodology to calibrate or tune its parameters when representative measurements are available. This dissertation demonstrates a systematic methodology to extract QBBJT's model parameters from measurements. The methodology can be implemented in automated extraction tools. 'With this methodology, QBBJT now can be used in either the predictive mode for concurrent engineering applications or the precision mode for routine circuit verification.
Finally, the dissertation presents a systematic methodology to include substrate resistance in BJT parameter extraction and simulation. Substrate resistance is often neglected in compact BJT models, since it is layout dependent. Even for the most sophisticated parasitics extraction programs, substrate resistance is ignored because its calculation involves computationintensive threedimensional simulations. Based on a fast calculation algorithm which was initially developed to characterize substrate noise coupling in mixedsignal applications, the dissertation demonstrates an efficient way to eliminate the uncertainties resulting from the layout dependence of substrate resistance in BJT parameter extraction and circuit simulation.
Vii
CHAPTER 1
INTRODUCTION
Since the bipolar junction transistor (BIT) was invented in 1947, it has been employed in applications of analog amplification and digital switching. Complementarymetaloxidesemiconductor (CMOS) technology, because of its characteristics of lowpower consumption and highpacking density, is preferred in fabrication of ultra large scale integrated (ULSI) circuits. However, because of its advantages in highspeed operations, bipolar technology is still employed in a1 wide variety of applications, ranging from high performance mainframe computers and data links, to analog and digital telecommunications and wireless communications [1]. As aggressive scaling and advanced process technology continue to improve the speedpower performance of CMOS, CMOS threatens to overwhelm all the advantages of conventional bipolar transistors and become the ubiquitous technology. Nevertheless, the process and lithography breakthroughs can also improve the speed performance of bipolar transistors. Because bipolar transistors have advantages in operation speed over CMOS technology, they are often used in highperformance digital and lowcost highspeed RE applications.
As the result of its high power dissipation compared with CMOS technology, bipolar technology is rarely used in very highlevel integration. To reduce power dissipation with high levels of integration, contemporary bipolar technologies have been developed to maintain BIT's speed advantages while achieving higher integration level.
2
One approach is to merge highperformance bipolar devices into CMOS processes, and form Bipolar/CMOS (BiCMOS) processes. In such processes, bipolar transistors are employed only where the leverage is high enough to justify extra power consumption. By retaining the benefits of bipolar and CMOS technology, BiCMOS technology allows overall speedpowerdensity performance previously unattainable with either technology. BiCMOS gates now can reach a speed about twice that of CMOS gates with approximately equal power dissipation [2]. Availability of bipolar transistors along with CMOS devices also adds design flexibility in analog or mixedsignal applications. However, the manufacturing cost of BiCMOS ICs is considerably higher than that of CMOS ICs. In addition, insertion of bipolar processes into baseline CMOS processes usually constraints optimization of either bipolar or CMOS transistor performance [3].
Another direction for contemporary bipolar technology is to fabricate highperformance vertical pnp as well as npn bipolar transistors on the same chip. Complementary bipolar technology has long held the promise of achieving highspeed with lower power dissipation. Until recently, the bottleneck of complementary bipolar technology is to realize highperformance vertical pnp transistors with baseline npn bipolar processes [1]. Therefore, complementary bipolar technology completely is unlikely to replace vertical npn technology in the near future.
Currently the most advanced technology for realization of highspeed bipolar transistors is based on doublepoly selfaligned technology. Doublepoly selfaligned bipolar processes were developed over a decade ago, and rapidly became the favored technology for manufacturing highperformance bipolar transistors. In this process, the firstlayer poly forms the base contact. The base contact poly serves as local interconnect
3
and as a diffusion source to form lowresistive extrinsic base, Like the base contact poly, the secondlayer poly forms the emitter contact and serves as a diffusion source to form a shallow emitter junction. A typical structure for a doublepoly selfaligned bipolar transistors is shown in Fig. I 1.
.........  
............ ......
..........
WR
..........
.. ..........
Figure 11 Crosssection of common advanced doublepoly selfaligned bipolar transistors
Using highly doped polySi as an emitter contact reduces process complexity and maintains the separation between the metal contact and the very sensitive monocrystalline silicon region. The separation prevents sintering effects and metal spikes [4]. The polyemittercontact BJT also exhibits characteristics superior to its metalcontact counterpart. First, a shallow base junction can be formed with sufficient control on shallow emitter junction formation. The shallow base reduces base transit time and improves speed performance of the transistor. Secondly, bipolar transistors with poly emitter contacts
4
usually have better emitter efficiency than bipolar transistors with metal contacts, because of the lower carrier mobility in poly silicon and the interfacial oxide layer interposed between active emitter and poly emitter. As the thickness of emitter continues to decrease for shallower base junctions, conventional metalcontact transistors would suffer a serious currentgain reduction. This is because the emitter becomes transparent to the injected minority carriers from the base, when its thickness is less than the characteristic diffusion length of minority carriers. Finally, the poly emitter serves as a lettering source for metal impurities [4].
With doublepoly selfaligned technology, BJT's lateral dimensions are also significantly reduced. This improves BJT's speed performance in several aspects. First, emitter width is narrower than the minimum linewidth given by lithography. The reduction in emitter width reduces the intrinsic base resistance, thus easing the dc current crowding effect for BJT operating at high current densities. Secondly, the distance between the edges of emitter and base contacts, which is usually determined by the width of a spacer, is reduced so that BJT's extrinsic base resistance is reduced. Finally, as the extrinsic base area is reduced, the overall basecollector junction capacitance is reduced. The reduction in both extrinsic base resistance and basecollector junction capacitance serve to improve speed performance and power gain.
Reducing base thickness, which decreases base transit time, can further enhance speed performance. However, reducing base thickness always conflicts with the need to minimize intrinsic base resistance, which is essential in optimizing the BJT's power gain. In addition, reducing base thickness also reduces the voltage at which transistors break down because of base punchthrough. A possible way to compromise speed, power gain,
5
and breakdown voltage in designing a BJT is to increase the transistor's base doping. However, increasing base doping sacrifices transistor's current gain.
Using SiGe epitaxial film for the transistor base in siliconbased process is a costeffective solution to the above dilemma. Because the ratio between transistor's base and emitter current injections is exponentially amplified by the bandgap difference at the emitterbase junction, a SiGe heteroj unction bipolar transistor (HBT) can have 10 to 100 times the current gain of a homoj unction silicon bipolar transistor. This extra current gain can be traded for the speed performance by optimizing base doping and thickness. In addition, as the bandgap of SiGe can be manipulated by varying the Ge concentration in the Si base, a builtin electric field can be created to assist minority transport in base, which further enhance transistor's speed performance. Recently, SiGe HBTs with 75 GHz unity gain frequency have been reported.
Even for purely silicon technology, aggressively scaled doublepoly selfaligned bipolar transistors now can reach unitygain frequencies as high as 30 GHz [5], thus being suitable for very highspeed digital and analog applications. A decade ago these highspeed applications were reserved for only compound semiconductor devices. However, the device models, which are essential to simulation, design and optimization of bipolar devices and circuits, have not kept pace with the tremendous progress achieved in process technology.
In designing Bipolar/BiCMOS devices and circuits, technology CAD (TCAD), which explores and evaluates various design tradeoff without timeconsuming and costly device fabrication, is widely used in today's stateoftheart design process. In the past, technology design was mostly based on experimentbased (i.e. trialand error)
6
optimizations. However, as the costs and duration of development increase with the complexities of today's BipolarIBiCMOS technology, such a process is slow and expensive. TCAD predicts and optimizes the electrical performance of VLSI circuits without using expensive fabrications. TCAD reduces development costs, shortens development times, and helps maintain process control in design and production environments.
In conventional TCAD strategy, the output of a process simulator is applied as input to a device simulator, whose output in turn is used in extracting device model parameters for circuit simulations. The computational cost of the device simulation in this approach is very high, especially considering the inevitable uncertainty in doping profiles, even when they are tuned using SIMS other measurements.
MMSPICE developed in University of Florida is a mixedmode device/circuit simulator to facilitate TCAD design processes. MMSPICE merges device physics into circuit simulations, therefore reducing the computation cost. Its chargebased bipolar transistor model QBBJT is based on regional analysis of ambipolar transport equations, subject to moving boundary conditions. The model is thus consistent with device physics implied by the driftdiffusion relations used in most device simulators, but runs much faster. QBBJT's model parameters are mostly derived from device's layout and doping profile. Therefore, correlations between model parameters and process information are automatically maintained. However, the use of MMSPICE/QBBJT model has been rather limited. This is because that the MMSPICE/QBBJT model lacks a welldefined parameter extraction methodology. In practice, the MMSPICE/QBBJT model is used in a variety of modes. In concurrent engineering applications, QBBJT takes parameter derived from
7
simulated doping profiles and runs at predictive mode. In this way, QBBJT enables circuit design before the process is finalized; the design cycle is significantly shortened. The ability to predict performance from device structure is very useful in the early stages of product design. However, once the technology is mature and test devices are available, simulation accuracy and efficiency become the key issues.
The dissertation presents a parameter extraction methodology that allows QBBJT model parameters to be calibrated in accordance with measured data, enabling accurate simulations even when the process information is incomplete or inaccurate. This ability to use model parameters derived either from process/structural information or from electrical measurements makes the QBBJT model uniquely flexible. Now a single compact model can be used in either a ICAD or bestfittodata mode. As the technology matures, only simple revisions to the processderived parameters are need for transparent transitions between modes.
The dissertation also addresses several refinements to the MMSPICE/QBBJT model. First, selfheating is implemented in the QBBJT model by adding an extra thermal node to QBBJT's admittance matrix, which is used in SPICE's modified nodal analysis. Selfheating is inherent to all semiconductor devices. Selfheating is a significant effect even for bulksilicon bipolar transistors operating at modest power dissipation. Ignoring selfheating in the BJT model can cause problems in simulating analog circuits because BiT's output conductance is either overestimated for homojunction silicon bipolar transistors or underestimated for some compound heteroj unction bipolar transistors.
DC current crowding is an important twodimensional effect. The original QBBJT model account for current crowding around the emitter periphery by using the effective
8
baseemitter voltage, which can significantly underestimate BJT's high injection effects. In this dissertation, we present an alternative way to model current crowding by using an effective emitter area. The approach is consistent with the original dc current crowding analysis, but maintains correct high injection analysis.
AC current crowding effect is a relatively new topic in compact bipolar transistor modeling. While dc current crowding has been studied for over 30 years, ac current crowding has received relatively little attention. AC current crowding is a nonquasistatic (NQS) effect, and is often overlooked in the BJT's smallsignal equivalent circuit. This is because BJT's smallsignal equivalent circuits are usually derived using quasistatic (QS) approximations. As the total base charge storage is modeled by a lumped capacitor, the time required for the majority charges in the base to redistribute laterally along the base is overlooked. The QS approximations is equivalent to assuming that the carriers can travel with infinite velocity in certain regions. This dissertation presents a new compact model to account the NQS ac current crowding. This model is consistent with QBBJT's dc current crowding analysis, and is easily realized in most compact bipolar transistor model.
As bipolar transistors continues to be scaled down for the demand of faster operation and higher integration, substrate resistance becomes more and more influential to the overall performance of bipolar transistors. For most compact BJT models, substrate resistance is not included by default. A common approach to model substrate resistance is to optimize the collectortosubstrate capacitance to partially account for its effect. However, the approach can cause numerous difficulties in statistical modeling and circuit design, because BJT's substrate resistance is layout dependent. In this dissertation, a systematic methodology to include substrate resistance in SPICE BJT simulation is
9
presented. With the methodology, BJT's substrate resistance is consistently included in BJT's parameter extraction and circuit simulations.
The dissertation is organized as follows: Chapter 2 reviews several classic and novel compact models. At the end of Chapter 2, QBBJT's model formulation is briefly discussed to introduce several concepts in the following chapters. Chapter 3 addresses several additions and refinements of QBBJT. Chapter 4 presents a complete parameter methodology for QBBJT. Chapter 5 presents a systematic methodology to include substrate resistance in BJT parameter extraction and circuit simulation. Chapter 6 concludes the dissertation with a summary and suggested future work.
CHAPTER 2
BACKGROUND, PHYSICS, AND MODEL FORMULATION
OF THE MMSPICE CHARGEBASED BIPOLAR TRANSISTOR MODEL QBBJT
This chapter discusses several classic bipolar transistor models and a number of novel models designed to accommodate contemporary technology breakthroughs. This chapter provides concise comparisons between these models. The model formulation of the MMSPICE/QBBJT model is briefly reviewed at the end of this chapter to introduce several concepts elaborated in following chapters.
2.1 Background Review
2.1.1 EbersMoll Model
The EbersMoll model was an empirical model invented in early 50s [6]. The original EbersMoll model is only a dc large signal model which contains two diodes connected backtoback with a controlled current source in parallel with each of them. The controlled current sources represent current diffusing from emitter to collector in forward operation and from collector to emitter in reverse operation. The expanded EbersMoll model has extra elements added to its equivalent circuit to account for resistive and capacitive parasitics. The addition of fixed terminal resistors allows the EbersMoll model the ability to account for nonideal ohmic voltage drop across resistive parasitics. By including several lumped capacitors, the expanded EbersMoll model can account for
10
11
charge storage in the quasineutral base and depletion regions. This enables ac simulations of bipolar transistors. With the addition of several empirical parameters, the expanded EbersMoll model accounts for several secondary effects. For instance, the expanded EbersMoll model uses forward and reverse Early voltages to account for basewidth modulation and includes temperature coefficients for several parameters [7]. However, the EbersMoll model cannot sufficiently account for high current effects.
2.1.2 GummelPoon Model
As a physical supplement to EbersMoll model, the GummelPoon model was formulated based on the onedimensional integral chargecontrol relation (ICCR) [8]. Many of the GummelPoon model's parameters are bias dependent. The GummelPoon model includes several empirical parameters, such as the knee current IK, to account for high injection effects. Even though the GummelPoon model is the most popular model employed in a wide variety of circuit simulators, it has remained mostly unchanged for 20 years and is insufficient to model the advanced bipolar transistors.
The conventional GummelPoon model has the following major deficiencies: 1) It models basewidth modulation (Early) effects using constant Early voltages (VAF and VAR), which results in unacceptable errors in simulating output conductances for BJTs with thin bases. 2) It uses constant collector resistance, thereby overlooking collector modulation and quasisaturation effects. 4) Its smallsignal equivalent circuit is derived based on quasistatic (QS) principles. It thus fails to account for nonquasistatic (NQS) effects in transistors operating at very high frequencies. 5) It fails to account for some
12
extrinsic parasitics, which are negligible in large devices, but important for scaled devices, e.g., the baseemitter sidewall capacitances. 6) It is not scalable with geometry.
2.2 Novel Compact Models for Bipolar Transistors
A predictive model for scaled bipolar transistors must account for (1) minority carrier transport in quasineutral base region for typical base doping profile and all injection levels; (2) majority carrier transport in quasineutral collector accounting for nonlinear voltage drops across the epicollector and for quasisaturation; (3) forward and reverse basewidth modulation (Early) effects; (4) base widening (Kirk) effects; (5) impact ionization with consideration of nonlocal carrier energy transport relations; (6) twodimensional geometric effects, such as sidewall current injection and currentcrowding effects; (7) physical and continuously differentiable ac differential resistances and capacitances [9]; and (8) global and local thermal effects, i.e., selfheating. Many compact physical models based on these requirements have been proposed in the past ten years. The extended GummelPoon model [10], VBIC95 [11], MEXTRAM [12] and the MMSPICE chargebased bipolar transistor model (QBBJT) [14] are reviewed in the following sections.
2.2.1 The Extended GummelPoon Model
Using Kull's majority carrier transport equation in the quasineutral collector [10], the extended GummelPoon model accounts for quasisaturation effect by calculating nonlinear voltage drop across the quasineutral collector. The extended GummelPoon model is based on the formulation of the original GummelPoon model and inherits most
13
of its parameters. The extended GummelPoon model is easily implemented and is familiar to most circuit designers. Consequently, the extended GummelPoon model has been included in a wide variety of circuit simulators.
To include Kull's collector transport equation in the extended GummelPoon model, one artificial node is introduced at the basecollector junction. While the original integral chargecontrolled relation (ICCR) is employed to simulate the minority carrier transport in the quasineutral base, Kull's collector current equation is applied in the quasineutral collector. The collector current is then solved by forcing current continuity at the artificial node. The extended GummelPoon model models quasisaturation as collector resistance modulation, thus reducing the empiricism of using constant collector resistance. However, the extended GummelPoon model cannot properly simulate the IV characteristics of all possible operating regimes, because it assumes that the entire epicollector is quasineutral. For example, the extended GummelPoon model cannot physically simulate currentvoltage characteristics of BJT operating in the nonohmic quasisaturation regime where an excesscarrier induced SCR forms near the highly doped buried layer due to carrier velocity saturation at high collector current density.
2.2.2 MEXTRAM
MEXTRAM model was developed by de Graaff and Kloosterman in Phillips Research Lab. MEXTRAM differs from the original GummelPoon model in a number of aspects [12]: (1) MEXTRAM does not use the integral charge control relation (ICCR). It models minority carrier transport and charge storage for an exponentially doped base, which is common in advanced bipolar transistors. (2) The terminal currents and charge
14
storages of MEXTRAM model are formulated as functions of the doping profile and bias condition. (3) Basewidth modulation (i.e., Early effect) is not modeled with the constant Early voltages, but with the baseemitter and basecollector junction depletion charges.
The latest version of the MEXTRAM model also adds a unified majority carrier transport equation in the collector accounting for various conditions of quasisaturation [ 13]. The MEXTRAM model, used in the circuit simulator Pstar (Phillips) and Microwave Design System (HewlettPackard), has many advantages over the GummelPoon model for simulating BJT's quasisaturation and high current characteristics.
2.2.3 MMSPICE/QBBJT Model
The MMSPICE/QBBJT model was developed by Jeong and Fossum [14] at University of Florida. While QBBJT was being first implemented in SPICE2G.6, QBBJT is now available in at least one commercial SPICE simulator. QBBJT is similar to MEXTRAM in many aspects. Both QBBJT and MEXTRAM use de Graaff's equation to model minority carrier transport in the quasineutral base [15] and Kull's equation to model majority carrier transport in the quasineutral collector [10]. QBBJT, however, has several unique features. QBBJT uses moving boundary conditions for different bias conditions, thus including many physical effects such as the basewidth modulation (Early) effects and the base widening Early (Kirk) effect. QBBJT also allows for possible SCR formation in epicollector, thus eliminating the nonphysical assumption used in the extended GummelPoon model. QBBJT physically simulates BJT's currentvoltage characteristics for all possible operating regions.
15
Most of QBBJT's parameters are derived from device structure (layouts) and process information (doping profiles). This makes QBBJT suitable for TCAD and concurrent engineering applications, where a simple preprocessor (SUMM) is used to generate QBBJT's parameters directly from the process simulator output or from measured doping profiles. QBBJT predicts circuit performance even in the absence of fabricated devices. Therefore, QBBJT can be used in concurrent engineering applications, which efficiently shortens production cycles. With parameters derived from the estimation of process information, circuit designs can begin before representative test chips are available.
QBBJT is also suited for statistical circuit design applications. Since most of QBBJT's model parameters come directly from the device structure and process information, statistical correlations between circuit performance and process variations are automatically maintained. Such correlations are not usually maintained in empirical models such as GummelPoon. MMSPICE is much more computationally efficient than device simulators. This makes accurate statistical analyses, that require many simulations (such as Monte Carlo), practical.
QBBJT is chargebased. Instead of using lumped capacitive elements and transit time constants, regional charge storages are formulated in accordance with bias conditions and device structural information. The regional charge storage provides nonreciprocal transcapacitances.
The QBBJT model was made more realistic and predictive for simulating advanced scaled BJTs by including nonlocal modeling of impact ionization and nonideal twodimensional effects [ 16]. High electric fields and field gradients are common in BJTs
16
with aggressively scaled vertical dimensions. As electric fields change rapidly over distances comparable to the energyrelaxation meanfreepath, nonlocal effects such as carrier velocity overshoot have been shown to influence characteristics of BJTs significantly. Current crowding effects, caused by laterally distributed base resistance, is a prominent twodimensional effect. Current crowding degrades BJT current gain and unitygain frequency.
The QBBJT model was extended for SiGe HBT simulations by including bandgap variation parameters [17]. With the operation defined physically for the entire operating range in a unified and systematic way, the numerical robustness of QBBJT has been greatly enhanced. However, since QBBJT uses different model equations for different operating regions, first or higher order derivatives of currents and regional charges can still be discontinuous with improper model parameters. These discontinuities can cause numerical problems and errors in simulating analog circuits.
2.2.4 Vertical Bipolar Inter Company Model (VBLC95)
VBIC95 was proposed in 1995 by a group of IC and CAD industrial representatives as a replacement for the conventional GummelPoon model. The group evaluated compact BJT models developed since GummelPoon model. Their objective was to pinpoint the best model and bring it to the mainstream of bipolar IC designs. However, no models they evaluated satisfied all the requirements from the committee, so they decided to develop their own.
VBIC's formulation is similar to the extended GummelPoon model, although, with several unique features: 1) VBIC uses baseemitter and basecollector junction
17
depletion charges to model basewidth modulation effect, thereby eliminating the empiricism of using constant Early voltages. 2) VBIC adds a parasitic pnp modeled with a simplified GummelPoon model. 3) To avoid numerical problems, VBJC uses singlepiece, smooth functions whenever possible to model BJT's currents and smallsignal capacitances. 4) VBIC includes selfheating.
Table 21 summarizes the above discussion contrasting different models. Notice that although some attributes can be included in almost all models, some attributes are pertinent to only specific types of models, i.e., empirical or physical models. This is because that they are just formulated differently. For example, an empirical model is usually continuously differentiable. However, its parameters are usually hardly scalable. Also, a physical model can include as many physical effects as possible. However, it is usually more computationally intensive than an empirical model.
2.3 Model Formulation of OBBJT
2.3.1 BIT Operation Modes
Before QBBJT's model formulation is reviewed, all possible operation modes of a BJT must be defined. Depending on whether a spacechargeregion (SCR) is formed at basecollector metallurgical junction, BJT operation is divided into two distinct regions: saturation and forward regions. Different saturation conditions can be defined based on the velocity of majority carriers in epicollector. When the majority carriers in epicollector move with a velocity proportional to the local electrical field, the BJT is operating in ohmic quasi saturation region and the epicollector acts like a resistor. When the majority carriers reach their saturation velocity, excess majority carriers must be injected into the
18
Table 21 Comparison between compact BJT models
Extended MEXQBBJT VBIC
GP TREM
Base Use GummelPoon ICCR X
analysis Use de Graaff's base X X X
transport equation
Use Knee current to model X X
injection
Account for base pushout X X
Account for bandgap X
variation in base. SiGe HBT
simulations.
Early Use Early voltages X
effect Use depletion charges X X
Use moving boundaries X
Collector Use Kull's equation to model X X X X
analysis quasisaturation
Account for various SCR X X
formations in epicollector
Account for field dependent X X
mobility
AC Smallsignal equivalent X X X X
analysis circuit derived with quasistatic (QS) approximation
Bias dependent smallsignal X X X X
differential resistances and
capacitances
Account for vertical X X
nonquasistatic (NQS)
effects
Account for lateral nonquasi X
static (NQS) effects
19
Table 21  continued
Other DC current crowding X
effects Transient current crowding ___X
Selfheating __ X X
Collector spreading effect X
Two dimensional geometric X
(layout) effects
General Continuous smooth terminal X X
attributes currents
Continuous smooth X X
differential resistances and
resistances
Scalable parameters X
Numerical robustness X X
20
quasineutral collector, which induce a spacecharged region (SCR) near the buried layer. In this situation, BJT is operating in nonohmic quasisaturation region.
Different operating modes can also be defined for different bias conditions on an ICVBC plane as shown in Fig. 21, where all the boundary lines are defined as [17]. Different regions in Fig. 21 are defined as following: (1) saturation; (II) ohmic quasisaturation, where entire base and collector are quasineutral; (111) nonohmic quasisaturation, where excessive charge is injected from base region and an SCR forms at the end of epicollector; (IV) forwardactive, where the basecollector junction is reverse biased and an SCR forms at basecollector metallurgical junction; (V) forwardactive with the entire epicollector depleted due to adequately high VCB.
In Fig. 21, the thick solid boundary lines labeled ISCC and ISCCWEPI define the onset of SCR formation at basecollector junction. Above these two lines, no SCR is formed at the basecollector junction and BJT is operating at either ohmic or nonohmic quasi saturation. Below these two lines, an SCR formed at basecollector junction, and the BJT enters forwardactive operation. The thick dashed boundary line, =0,: q *A C NEWI' Vsat I is usually defined as the onset of base pushout. However, for collector currents greater than 1o, the quasineutral base does not necessarily extend into the epicollector. Actually, base pushout depends on the collector current level as well as the basecollector bias voltage. 10 represents the maximum current that the epicollector can carry before its majority carriers reach their saturation velocity, and above which extra carriers must be injected into quasineutral collector to sustain excess current.
21
IC
CRWEPI
I0
(V)'
(IV)
(II)
SCCWEPI C
VCRIT PC
(I): saturation region VBC
(II): ohmic quasisaturation region (III): nonohmic quasisaturation region
(IV): forwardactive
(V): forwardactive with the entire epicollector depleted
Figure 21 Operation mode boundaries on IcVBc plane.
22
2.3.2 Formulation of DC CurrentVoltage Equations
The formulation of QBBJT's currentvoltage equations and charge storages is based on de Graaff's equation to model minority carrier transport in the quasineutral base [12] and Kull's equation to model majority carrier transport in the quasineutral collector [10]. In deriving the minority carrier transport and charge storage equations, de Graaff assumed quasineutrality for an exponential base doping profile as
NA = NAO exp(lx/WBM), (21)
where NAO is the extrapolated peak base doping, 1 is the base doping grading factor, and WBM is the base metallurgical width. To account for intended or unintended bandgap variations which result in an aiding or retarding field in the quasineutral base, de Graaff's current equations are reformulated as
JnL JnLF JnLR for low injection conditions (22)
J,H JnHF JnHR for high injection conditions (23)
where
,exp(p"'Wb)
JnLF qDnL exp  n(xE) (24)
exp(Tp'Wb)
n(xc)
JnLR =qDnL x c) (25)
exp(l'Wb) I
nHF 2qD, exp(il"Wb) (26)
inHF 2 2qDHann" ex q"W In(xE) (26
exp(fl"Wb) 1n~E
n (xc)
nHR = 2qDnr n(xc) (27)
and the base doin grading factor and are defined as
and the base doping grading factor if' and i" are defined as
23
ri' = ( +AE/kT)/WBM (28)
4" = AE/2kTWBM (29)
to account for bandgap variations. To ensure a smooth and continuous transition from low injection to high injection, linear extrapolation is used to get a single analytic current equation for all injection levels:
= JnLF NA(XE) + JnHF n(XE) nLR NA(Xc) + JnHR n(XC)
Jn = (210)
NA(xE) + n(xE) NA(XC) + n(xc)
In the quasineutral collector, QBBJT uses an equation similar to Kull's formulation for the collector majority carrier transport equation:
qAgVTNEPI I1+K(BCO) BCO BCIT (211)
= W K(Bco) K( c) In 1 +K(Yact) V, (211)
where
F (;{.2 ni ( ?"
K(T) 1 + N expv) (212)
and WQNR is the width of quasineutral collector.
QBBJT uses baseedge and collectoredge quasiFermi level separations at the basecollector junction, ,BCJ and BCI, to link the base and collector transport analyses. QBBJT then solves the collector current by forcing current continuity at the basecollector junction. When WBCI is greater than basecollector junction barrier, QC, the BJT operates in saturation or quasisaturation modes and TBCI=T BCJ. No SCR exists between the base and collector. On the other hand, BJT operates in the forwardactive mode, where YBCI is modeled to be equal to %c and YBCJ is modeled as
24
qUBCJ q~' qc) VBc)]
exp(TT) = exp( Uc) + [exp ( U )_ exp(U IC]Ic (213)
where ICRIT denotes the critical boundary currents (IscC or ISCRWEPI) separating the quasisaturation and forwardactive operations as shown in Fig. 21.
QBBJT solves for the widths of baseemitter and basecollector SCRs to provide moving boundary conditions and to account for basewidth modulation (i.e. Early effect) in base analysis. The width of basecollector SCR is calculated based on the potential drop across the basecollector SCR (i.e., TBcITBcJ). While taking into consideration the background charge modulation induced by collector current injection, QBBJT calculates the width of basecollector SCR using the depletion approximation. The baseemitter SCR, which controls the reverse Early effect, is modeled for both forward and reverse bias. With the boundary conditions defined physically, the QBBJT model properly links base collector transport equations and ensures current continuity throughout all operating regimes.
2.4 OBBJT's SmallSignal Equivalent Circuit
Before constructing the smallsignal equivalent circuit, charge storages must be formulated. The formulation of excess charge stored in the quasineutral base is based on de Graaff's work. With a linear interpolation for intermediate injection, excess charge storage in the quasineutral base is formulated as: NA(XE) Q L + n(xE) Q QH (214)
QnB = NA(XE)+n(xE) 1hBNA(XE) + n(xE) QnB
where Q B and Q H are excess charge storages for low and high injection cases.
25
Excess charge stored in the emitter is formulated using the emitter minority transit time TE, an empirical parameter defined as:
QE = TE. AE. JEO" [exp(VBE/VT) 1]. (215)
Excess charge stored in the baseemitter depletion region is modeled as:
XE
QJE = q. A. fNA(x)dx (216)
0
where NA(x) is the exponentially graded base doping profile and xE is the baseside depletion width of baseemitter junction, modeled for both forward and reverse bias.
Excess charge storage in the quasineutral collector is modeled according to the electric field distribution shown as Fig. 22. In saturation and quasisaturation (case A of Fig. 22), charge stored in the unmodulated region is zero because the electric field gradient is zero. The excess charge stored in the collector is only the charge in the modulated region, which is expressed as 22 2 2
QQNR = IC [exp(VBCO) exp (VBC] qAnI WQNR (217)
In nonohmic quasisaturation (case B of Fig. 22) region, the excess charge in the collector is formulated as the sum of charge stored in the modulated region as (217) and in SCR at the end of the epicollector as
QEPI(SCR) = WSCR (qNEP1 Ic/'s), (218)
where WSCR is calculated with consideration of the background charge modulation caused by collector current injection.
26
Electrical field
WQNR I WUNMOD Depth
S~IJ
ES . .. 
(A)
Electrical field
WQNR W SCRD Depth
mo
(C)
Electrical field
WSCC=WSCR No Depth
When IC > 10
~(D)
Figure 22 Electrical field distribution of each operation
mode. A: saturation & ohmic quasisaturation. B: nonohmic
quasisaturation. C: forwardactive with epicollector not fully depleted. D: forwardactive with entire epicollector
depleted.
27
In forwardactive mode (case C of Fig. 22), excess charge storage in the collector is formulated as charge stored in basecollector depletion region
QEPI(scc) = Wscc" (qNEPI I/)s) (219)
where WscC is also calculated with consideration of background charge modulation caused by collector current injection.
In forwardactive mode with the entire epicollector depleted (case D of Fig. 22), in addition to charge stored in the epicollector, which is modeled as (219) with WsC replaced by WEPI, excess charge in the collector must include the charge stored in the buried layer as:
Q L A WEPI.
QBL FEP(Dc VBc) I 1 IC) eAES (220)
While all regional charge storages are defined and formulated physically, next step to formulate the smallsignal equivalent circuit is to partition these charge storages into different terminals. Charge partition is a typical technique to allocate a charge storage in a certain region into two different terminals so that the charge transport kinetics in the region can be partially accounted for. Through two empirical parameters B and c, QnB and QQNR, which represent the excess charge storage in the quasineutral base and in the extended base, are partitioned into the emitter and collector terminals.
The excess charge stored in the epicollector QEPI is modulated by the collector current, which is a strong function of collector current. QEPI is allocated into the emitter and collector terminals as QEPI = (W/ s) Ic and QEI (W/ls) 1, where W is the SCR width in the epicollector.
28
The resulting charge storages in baseemitter and basecollector terminals are formulated as:
E
QBE = B "QnB + C" QQNR + QE + QEPI + QJE (221)
QBC = (1B QnB + (1C) QQNR + QCPI+QBL
where QQNR is set to zero for case (C) and (D). QBL is set to zero for case (A), (B) and (C).
With all terminal charges defined for all bias conditions, the smallsignal equivalent circuit is formulated as shown in Fig. 23. This smallsignal equivalent circuit has several unique features: (1) Implementation of the smallsignal equivalent circuit is easy, since all the regional charge storages are derived from the dc solution; (2) The differential conductance (i.e., dI/dV terms) and capacitances (i.e., dQ/dV terms) are biasdependent; (3) The transcapacitances are included explicitly; (4) The extra delay (phase shift) caused by base widening is included implicitly with QQNR modeled in quasineutral collector; (5) The NQS carrier transport inside the quasineutral base is taken into consideration through charge partitioning.
29
C
RC
CBC(ext) PA'a Q CSUB
W IBCV BC QBC VBE
RB(ext) RB(int (aVBC + VBE
A A,
B19' C9
CL) CL.)
IF
+ rn +
I +
Ij CLI CL)
rrj
lz:
r!I ct rZI
R E
E
Figure 23 QBBJT's smallsignal equivalent circuit
CHAPTER 3
MODEL ENHANCEMENTS OF THE QBBJT MODEL
3.1 Selfheating Modeling
This chapter discusses several additions and refinements to the MMSPICE/QBBJT model. Selfheating occurs in all semiconductor devices and circuits. Selfheating is an increase of device local temperature caused by device's own power dissipation. As packing density and power dissipation of VLSI circuits continue to increase with aggressive scaling, selfheating can become a severe problem and must be accounted for in simulations and circuit designs. Selfheating is often observed in MOSFETs as a drain current reduction, because the channel carrier mobility is degraded by the increased lattice scattering [18]. While decreasing the drain current in MOSFETs, selfheating increases the base and collector currents in bipolar transistors by enhancing the minority carrier injection at the baseemitter junction, Selfheating increases BJT power dissipation, thus causing positive feedback in BJT operation.
Since the mobility decreases slowly with temperature, the drain current reduction in MOSFETs is not easily detected unless there is a large temperature change of tens of degrees. Selfheating is especially important for MOSFETs built on SOI wafers, because the thermal resistance of silicon dioxide is about two orders of magnitude greater than that of silicon substrate. However, since the intrinsic carrier concentration increases
30
31
exponentially with absolute temperature, significant base and collector current changes in bipolar devices can be observed a small change in temperature.
Selfheating affects analog circuits significantly by increasing BJT's output conductance. This can lead to a significant reduction in the voltage gain in highgain amplifiers [191. Circuits that depend on close matching of baseemitter voltages can be very sensitive to selfheating, because small differences in junction temperatures can cause significant mismatch in currents. Recent research demonstrates that neglecting selfheating can cause significant errors in parameter extraction [20]. Selfheating can cause more than 50% difference in extracted values of BJT knee current. Therefore, proper models should include selfheating.
While the current gain of homojunction BJTs has a positive temperature coefficient, the current gain of heterojunction BJTs (HBT) can have either positive or negative coefficient. The negative coefficient is a result of the bandgap difference at the baseemitter junction [21]. Physical modeling of selfheating depends on accurate calculation of the device local temperature. To simulate the local temperature, one thermal node representing the temperature inside the bipolar transistor must be added into the QBBJT model, making it a 5 node model as shown in Fig. 31. Thermal impedance is the response of device temperature to variations in device power. With the thermal impedance calculated theoretically as in [22] or evaluated as discussed in [23], the equivalent circuit of Fig. 31 can be used for dc, ac and transient analyses.
In Fig. 31, the device local temperature is treated as an electrical voltage, the power dissipation as a controlled current source, and then the thermal impedance as an electrical impedance. The controlled current source IPWR is calculated as
32
C
S T
Rc
ISUB Q)
ICP I ,
B Cb
3T7 CQBC ICT
SVTaE
iCQBE R E
E
Figure 31 Schematic of thermal implementation
33
IPWR = IC~ VCE + 'B*' VBE. The local temperature is then evaluated through multiplying the power dissipation by the thermal impedance. In this way, selfheating simulation can be merged into the electrical simulation. In the implementation of selfheating, an extra node T was added into QBBJT by simply modifying QBBJT's admittance matrix in SPICE's modified nodal analysis. The implementation involves minor modifications in 18 subroutines and addition of a new subroutine to update temperature dependent parameters in QBBJT model. Temperature dependent parameters are updated based on the temperature obtained from the thermal node T in every NewtonRaphson iteration. The resulting operating point is, thus, consistent with device's local temperature.
In this implementation, the ambient temerature is simulated as a voltage source Vtamb. In this way, we can simulate a BJT transistor at different ambient temperatures by simply varying the voltage source VTamb. This implementation allows easier extraction of the temperature dependence of BJT electrical characteristics than using the TEMP control card. For instance, we can vary the ambient temperature from 73 'C to 27 'C with a step of 5 'C using the DC sweep control as
*DC VTAMB 73 27 5
.PRINT DC I(VC)
In this way, we obtain directly the temperature dependence of the collector current. This also adds the flexibility to vary ambient temperature in frequency domain or time domain analyses. However, this implementation disables the original TEMP control card.
34
15 VBE=0.95V
VBE=0.85V VBE=O.80V
0
5
0.0 1.0 2.0 3.0
VCE (V)
Figure 32 Output characteristic plot: Symbol lines denote measurement data for BJT of 1.6gmx20gm emitter from Motorola 0.5mm BiCMOS process. Solid lines are the MMSPICE simulation results with selfheating. Dashed lines are simulation without selfheating.
35
Fig. 32 shows a simulated output characteristic demonstrating selfheating in a BJT. The thermal resistance is calculated with the method in [22]. The thermal resistance can also be extracted using temperature dependent measurements [23].
3.2 DC Current Crowding Modeling
Current crowding caused by laterally distributed base sheet resistance is an important 2D effect [24][25][26]. A half crosssection is shown in Fig. 33. As an ohmic drop results from dc base current flowing across intrinsic base, baseemitter voltage VBE can be nonuniformly distributed. This thus causes the emitter current injection in the periphery of the baseemitter junction to be greater than that in the center. DC current crowding is especially severe at high current levels, because the transverse ohmic voltage drop is significantly higher at high current levels. Current crowding can be observed by measuring the lowfrequency ac base resistance rb, which decreases with dc emitter current lE [24]. DC current crowding is usually associated with BJTs with large emitter widths. However, base thicknesses are scaled down to reduce base transit time for better speed performance, serious endtter current crowding occurs even for BJT's with moderate emitter widths.
Current crowding was first investigated by Pritchard in late 50's [24]. By cascading an infinite number of bipolar transistors with series base resistances, Pritchard derived the equivalent base resistance as one third of total base resistance. However, this result is invalid for the high basecurrent levels, which is typically involved in current crowding. Another analysis of emitter current crowding was done by Hauser [26]. Hauser manipulated the differential currentvoltage relations across the intrinsic base region. With
36
Base Emitter
IX I
N
Iy= Y=WEI2'
Emitter Base
iB(Y)
Figure 33 Half crosssection of a vertical bipolar transistors.
37
appropriate boundary conditions for the periphery and at the center, Hauser derived an analytical solution for total base current. Hauser's method is useful, but inadequate for advanced bipolar transistors, because it assumes that the base sheet resistance is constant, thus overlooking the base majority conductivity modulation, which occurs in high injection or base widening. QBBJT's dc current crowding model is mostly based on Hauser's approach. However, QBBJT includes consideration of base widening and majority carrier modulation. QBBJT calculates quasineutral base width and majority carrier concentration for each dc bias point. The biasdependent base sheet resistance thus eliminates the inadequacy of Hauser's model.
Current crowding often enhances high injection effects [31], because the peak current density in the periphery is enhanced by current crowding. Emitter current crowding is usually modeled with an effective base resistance in SPICE and effective baseemitter voltage in MMSPICE. These approaches are equivalent to averaging the nonuniforndy distributed emitter current injection along the base, thus underestimating the peak current injection in the periphery, as shown in Fig. 34.
To avoid the underestimation of the high injection effects, effective emitter area is used to model dc current crowding instead of effective VBE or effective base resistance. Effective emitter width was first derived by Hauser [26]. With the consideration of conductivity modulation, the effective base width is derived similarly to [16] and [26]. This model is consistent with previous QBBJT dc current crowding analysis, but better simulates the currentcrowding induced highinjection effects.
38
2.0
~1.5
;~1.0
Hauser's equation
Effective VBF
0. Effective emitter area
0.51
0.0 0.2 0.40.08
Position y (gm)
Figure 34 Two different models to approximate nonuniform current distribution along intrinsic base: (1) effective baseemitter voltage VBE(effO model, (2) effective emitter area AE(effD model.
39
In our current crowding analysis, the baseemitter junction is treated as an infinite number of diodes and resistors cascaded in series as shown in Fig. 33. Therefore, the baseemitter junction voltage as a function of position y, v(y), can be expressed as v(y) = VBE iB() Pd (31)
0
where p is the specific base resistivity and is expressed as p = WE/(2 p (QBB + QQNR)), (32)
QBB is the total charge of majority carrier in quasineutral base, and QQNR is the total charge of majority carrier in extended quasineutral base. QBB can QQNR are calculated from dc operating point, therefore making base resistivity biasdependent.
From Fig. 33, the base current as a function of position y can be expressed as
iB(Y) = iB(O)f 2"LE'JEO(eff)' expV)l ]d (33)
where LE is the length of the emitter area and JEO(eff) is the effective saturation current density of diode, which is the steadystate diode saturation current plus the transient term obtained from previous timestep. Substituting (31) into (33) and differentiating the resulting equation twice, we obtain
2
d iB O. diB
di + i (34)
dy2 V T
where we assume that exp(v(y)/VT) >> 1. The general solution to (34) can be expressed as
Atan B (35)
40
where A and B are two unknowns. At y=WE/2, iB(WE/2) should decrease to zero. Therefore,
iB(WE/2) = 0 (36)
Substituting (36) into (35), we can obtain the unknown B, and (35) becomes iB(Y) = A. tan 1 ] (37)
where z = (A p WE) VT. Hence, the total base current at y=O can be expressed as iB(0) = A tan(z) (38)
To solve the last independent unknown A, we can substitute (37) into the first two coupled integral equations (31) and (33), and obtain VBE sin(2z)
iB(O) = LEWEJEoexp( T)_ "2z (39)
By equating (38) and (39), we can solve A and the dependent variable z, thereby obtaining the total base current for a given baseemitter voltage VBE.
Notice from (39) that current crowding makes the total base current smaller than that expected from bipolar transistors with zero base spreading resistance. In the previous version of QBBJT, dc current crowding is modeled with an effective baseemitter voltage VBE(eff, that is
SV BE( efft)'
i(0) = LEWEJEOexp ( VBTf)' (310)
where
exp(VBE(eff)) = exp(VBE) sin2z)
41
The result from the VBE(etf) approach is similar to the result using base resistance in SPICE. Both underestimate the peak current injection level. A more physical model of current crowding should evaluate effective emitter area, thereby maintaining current density increase in the periphery due to the reduction in effective emitter area as Fig. 34.
To account for current crowding effect using effective emitter area, we inherit most of the original model formulations. However, we use an effective emitter area AE(eft) instead of effective baseemitter voltage VBE(effD in (310). The resulting effective emitter area AE(eff) is
AE(eff) _= AE~ si 2 WELE sin(2z) (2
2z E2z (2
With AE replaced by AE(eftD in the subroutine QBCT, QBBJT's current crowding model is modified for both DC and transient analysis. Fig. 35 and Fig. 36 show simulation results with the new current crowding modeling. The new model better simulates the currentcrowding induced highinjection effect, i.e. highinjection rolloff, in current gain P3 and unitycurrentgain frequency fT
In the original development of MMSPJCE, the use of effective emitter area was thought of to cause instability. For this reason, transient simulations were performed on different devices to check numerical stability of the new current crowding model. The results show that the numerical stability of the new current crowding model is at least as good as that of the old one. For large devices (WE > 1.0 gim), the new current crowding model actually improves convergence stability.
42
1 0 0 .0 0 V c p b ,0 1M e a s u r e m e n t
0 w/ crowding
90.0 0 w/o crowding
0
80.0
C
70.0
00
0
50.0 I
0.5 0.7 0.9 1.1
VBE (V)
Figure 35 Current gain j3vs. baseemitter voltage VBE.
From this figure we can see currentcrowding induced highinjection effect is better manifested using the new current
crowding model.
43
15.0 .
0 with
10.0 )crowding
:EO without
5.0 00
O) Measurement
0.01
0.4 0.6 0.8 1.0 1.2 1.4
VBE (V)
Figure 36 Unitycurrentgain frequency fT vS. baseemitter voltage VBE. From this figure we can see currentcrowding induced highinjection effect is better manifested
using the new current crowding model.
44
3.3 AC Current Crowding Modeling
3.3.1 Simple Model Formulation
QBBJT's ac analysis does not include smallsignal intrinsic base resistance, when dc current crowding analysis is active, which underestimates the total smallsignal base impedance. To model smallsignal base resistance, we must account for distributed base resistance as well as baseemitter capacitance (including both depletion and diffusion capacitances), which can lead to significant ac current crowding. Consider the distributed baseemitter RC network shown in Fig. 37. Such a network can be derived directly by linearizing the cascaded dioderesistor network as shown in Fig. 33. As the smallsignal admittance of the baseemitter capacitance increases with frequency, smallsignal base current tends to gather near the periphery as shown in Fig. 38. The phenomenon of nonuniform distribution of ac base current is called ac current crowding. As dc current also tends to crowd around the periphery of baseemitter junction (e.g. edge current crowding), the diffusion capacitance in the periphery is proportionally higher than that in the center. Therefore, ac current crowding is more severe with significant dc current crowding [27].
The phenomenon of dc current crowding was reported more than 30 years ago [26] and has been widely studied [271. AC current crowding, however, has received much less attention than its dc counterpart. AC current crowding is a nonquasistatic (NQS) effect. The NQS effect is the timedependent behavior cannot be obtained by extrapolating from steadystate analysis. In quasistatic (QS) chargebased modeling, different charging and discharging dynamics are modeled with transit time constants, which are calculated from a dc solution for excess charge storage. For instance, the base transit time can be calculated
45
rbdy y~cdy ZbI
Figure 37 Smallsignal equivalent circuit of the base resistance plus baseemitter capacitance.
1.2 1.0 C 0.8 .m
0.6 f 100MHz
.m
0.4 f 1GHz
0.2 0.0
position y (jim) Figure 38 Normalized smallsignal ac base current distribution along the intrinsic base. Calculated for typical BJT without significant dc current crowding.
46
as dQB/dIC, where QB is the excess charge storage in the quasineutral base and IC is the collector current. Actually, this base transit time is just the time required for the collector current to deliver charge to the quasineutral base. This approach thus neglects the time required for those charges to redistribute in the quasineural base.
The NQS charge redistribution time is especially important as the operating frequency approaches the reciprocal of carrier transit times. Since the lateral dimension of the base can be greater than its vertical dimension, the time required for the charge to redistribute laterally can be larger than the base transit time, which is predominantly controlled by the vertical dimension of the base. Therefore, the NQS effect caused by the lateral redistribution time can be significant, even when the BJT is operating at frequencies far below its unitygain frequency fT. As transistors are used at frequencies approaching their fTs, the lateral NQS effect must be carefully accounted for.
A common way to account for NQS effects is to use charge partitioning, which accounts for charge redistribution in a region by allocating the excess charge storage in a region to two different terminals. Charge partitioning is often used to account for NQS minority carrier transport in the quasineutral base, i.e. for vertical NQS effect [30]. Charge partitioning technique can also be adapted to account for majority charge redistribution in the quasineutral base, i.e. for the lateral NQS effects. In most compact BJT models, basecollector capacitance is partitioned into an intrinsic part and an extrinsic part, where the charge partition factor, i.e. XCJC in the GummelPoon model, is obtained from optimization of smallsignal sparameters. Actually, this is merely a way of using charge partitioning to account for the lateral NQS effect occurring in the intrinsic base to better fit the input smallsignal parameters, i.e. s I I or s 12.
47
To exactly represent the NQS charge redistribution time along the lateral base in ac current modeling would require cascading an infinite number of RC networks in series [28] [29]. Based on the series expansion of the input impedance of an RC transmission line, Versleijen derived a simple RC network, which consists of a resistor RJ/3 and capacitor C2.15 in parallel, to simulate the ac current crowding effect [28]. This approach is simple and useful. However, this model is not sufficient at high base current, because it does not account for the differential base resistance reduction caused by dc current crowding. Also, it does not account for the nonuniformity in the baseemitter admittance caused by dc current crowding.
The new approach described below is based on [28]. However, by careful manipulation of the series expansion of the RC transmission line input impedance, the new model considers the differential base resistance reduction. Also, with the effective base length obtained in the dc current crowding analysis, the new model accounts for the nonuniform distribution of baseemitter admittance.
Under linear and smallsignal conditions, the input impedance of a uniformly distributed transmission line can be expressed as Rb7 ot VYR b (313)
where Y 7 = Il/R9 + jwOC 7 is the total base admittance, and Rb is the intrinsic base resistance. (313) is equivalent to the ac base impedance derived by Pritchard [24]. Using the series expansion of coth(x) = I+ X + *~,(313) can be expressed as x 3 45
48
R R2 .(R1 + joC)
Zb 1 Rb Rb"R+jC,
Zb = jc + 3 (314)
Through simple algebraic manipulation, (314) can be further transformed as
1 1
Zb I + 1 + (315)
R + jmC~c (Rb/3) + (5R) + jm(C/5) Then the simplified equivalent circuit of the distributed RC base impedance is shown in Fig. 39. The equivalent circuit of Fig. 39 is similar to that in [28], but contains an extra component of 5R. At low current, the baseemitter resistance Rgt is usually much larger than Rb so that this 5Rr term can be neglected. However, since R.t decreases with base current, the magnitude of R. can become much smaller than Rb at high current. Therefore, 5R,, should be kept in the equivalent circuit.
Figu 39. Equivalent.circt of(3.
Figure 39 Equivalent circuit of (315).
49
The 5R,, term is to account for baseresistance reduction due to dc current crowding at high currents. In the equivalent circuit of Fig. 39, RW/3 is the lowcurrent dc intrinsic base resistance, which was first derived by Pritchard in 1958 [24]. However, at high currents, dc equivalent base resistance should be lowered, because the effective base width is reduced. By including 5R,,, the simple equivalent circuit of Fig. 39 accounts for the dc baseresistance reduction due to dc current crowding at high currents.
The C,,/5 term is used to model the NQS ac current crowding. At low frequency, the base impedance of the equivalent circuit of Fig. 39 is merely the resistance of the two resistors (5R. and RW/3) in parallel, because C7,/5 can be thought of as an open element. This implies that no significant ac current crowding occurs at low frequency. However, the effective base impedance becomes smaller at high frequency, because the ac current tends to crowd around the periphery, thus narrowing the effective base. With C,/5 which shunts the effective dc base resistance (5R., and Rb!3 in parallel), the equivalent circuit of Fig. 39 properly simulates ac current crowding. However, the C.15 term should not be thought of as a physical component, but as an empirical component to approximately simulate the nonquasistatic (NQS) delay or phase shift.
3.3.2 Accounting for DC Current Crowding in AC Current Crowding Analysis
In the above discussion, the impedance of the baseemitter pn junction was assumed to be uniformly distributed; that is, the smallsignal baseemitter resistance and capacitance, R7, and C7T, are constant along the lateral base dimension. This assumption is valid only for low currents, where VBE is almost a constant along the lateral base; i.e.,
50
there is no significant dc current crowding. Since R. and C. are exponentially biasdependent, they are nominiformly distributed along the lateral base dimension, even for very small variations in VBE due to dc current crowding. The net effect of the nonuniformly distributed baseemitter impedance on ac current crowding can be described as follows. Since the dc baseemitter voltage VBE at the periphery is higher than that at the center, the smallsignal baseemitter impedance is smaller than that in center, thereby causing more of the ac base current tends to flow through the periphery instead of the center. Hence, dc current crowding enhances ac crowding.
It is not easy to model the nominiformly distributed baseemitter impedance, since, unlike the uniform transmission, no analytic solution exists. Nevertheless, little modification in the solution of uniform transmission line is needed to approximate the nonuniform transmission line for the purpose of compact modeling. For a physical approximation, we first examine the net effect of dc crowding on ac current conduction. The net effect of dc current crowding is to push ac current toward the periphery. This is equivalent to compress the area where most of the ac base current flows. Hence, we employ the concept of effective area, formulated for dc current crowding, to approximate the nonuniform transmission line. To accommodate the effective area factor in the previous ac current crowding analysis, we adjust the effective length of the RC transmission line used in our previous ac current crowding analysis.
To verify the ac current model, we compare the solutions of a nonuniform transmission line and a uniform transmission line with an effective length. The reason we use this comparison as a preliminary verification is that the solutions of transmission lines can better illustrate the concept of effective length than numerical device simulation, the
51
results of which are not always easily interpreted with regard to the intrinsic physics. We will use numerical device simulation as a final check to the model implementation.
To derive the differential equation which controls currentvoltage relation of the nonuniform transmission line, the smallsignal baseemitter admittance is modeled as yr(y) = yoexp( BEY) (316)
where y. = l/R, + jwCn, y,0 is a constant, and n is the injection ideality factor. n equals to one at medium injection level. At high injection, n is between one and two. However, for simplicity, we use n equal to one for all the injection levels.
To obtain the baseemitter voltage as a function of position, we can substitute (37) into (31) and obtain
VBE(Y) = VBE(O)2VTln{ cos[z(l 2y/WE)]} (317)
where z is defined in (37). Then using (317) in (316), the smallsignal baseemitter impedance as a function of position is expressed as y,(y) = ylc' { sec [z(l 2y/WE)] }2 (318)
2
where yo' = y,,oexp [VBE(O)/VT] [cos (z)]
From Fig. 37, we can derive a pair of firstorder differential equations for the smallsignal currentvoltage relation along the lateral base as
dib(y) y(y)v(y) = Yo" {sec [z( I 2y/WE)] }2. V(Y) (319)
dy
dVb(y) rb ib(Y) (320)
52
Substituting (319) into (320), we can obtain
d2
dy2Vb(Y) = rb"Yno" {sec[z(1 2y/WE)] } v2b(y). (321)
dy
The above second order differential equation is similar to that of a uniform transmission line, but does not have an analytical solution because of the secant function. (321) can be solved numerically using the boundary value method. The boundary conditions of (321) are described as following: At y=0, vb(Y)=l, if a unity ac voltage source is applied. At Y=WE/2, ib(Y)=0.
Fig. 310 shows that the smallsignal voltage distribution along a nonuniform transmission line can be modeled approximately by a uniform transmission line with an effective length calculated from the dc current crowding analysis. Also, as shown in Fig. 311, the input admittance of a nonuniform transmission line as a function of frequency can be simulated properly by using a uniform transmission line with an effective length.
3.3.3 Model Implementation and Verification
To include the ac current crowding model into MMSPICE/QBBJT model, we implemented the equivalent circuit of Fig. 39 in QBBJT's ac admittance matrix in SPICE's modified nodal analysis.
MEDICI was used to verify the ac current crowding model and its implementation in MMSPICE/QBBJT model. In spite of being, strictly speaking, a quasistatic (QS) simulator, MEDICI can be used to simulate the nonquasistatic (NQS) extra phaseshift due to the charge redistribution in the device. The reasons are described as following: 1) MEDICI's mesh is usually small enough that the time required for the charge to be
53
1.0 ,..
0. f=100MHz
" 0.6 f=4GHz
0.4 <
"'' f= 10GHz
0.2 1"
0.0 0.2 0.4 0.6 0.8 1.0
position y (gtm)
0.0 f=100MHz
50.0 M C3 '3
, ""..f=4GHz
100.0 ",.
150.0 __ _ _ _
0.0 0.2 0.4 0.6 0.8 1.0
position y (gim)
Figure 310 Comparison between the numerical solution of (321) and the analytical solution of uniform transmission line. Symbol lines represent the exact solution. Thick lines represent the analytic solution of the transmission line with effective length. Thin lines represent the analytic solution of the transmission line with normal length.
54
0.050
0.040 Cy
0
0
0
0 0.030
Cz
E
0.020
0.010
0.000
108 109 1010
frequency (Hz)
80.0
70.0
60.0
50.0
06
40.0 CO
Cz
30.0
20.0
10.0
0.0
1o8 109 1010
frequency (Hz)
Figure 311 The input admittance of the nonuniform transmission line (symbol lines), the uniform transmission line with effective length (thick lines), and the uniform transmission line with normal length (thin lines).
55
redistributed in it can be assumed much smaller than any of device's characteristic transit time constants; therefore, we can assume that charges are instantaneously redistributed inside a mesh. 2) MEDICI's ac admittance matrix actually maintains all the interactions between adjacent meshes. Therefore, MEDICI enables simulations of NQS phaseshift due to the limited carrier transport velocity.
The MEDICIsimulated and MMSPICEsimulated base input impedances are shown in Fig. 312. While the magnitude of the base input admittance is only slightly influenced by ac current crowding, overlooking it causes significant errors in phase predictions at high frequency.
3.3.4 Conclusion and Discussion
This chapter describes addition of selfheating to the MMSPICE/QBBJT model. DC current crowding was modified to better simulate high injection effects. A compact equivalent circuit model was developed to model NQS ac current crowding effect. The model is consistent with the dc current crowding analysis and is easily implemented. Using the effective base width calculated from the dc current crowding analysis, the ac current crowding model can be extended to account for nominiformly distributed baseemitter differential admittance along the lateral base. The resulting ac current crowding model is verified against numerical solutions of a nominiformly distributed transmission line and MEDICI's simulations,
Notice that the NQS ac current crowding analysis may not be consistent with the existing transient crowding analysis. The ac and transient current crowding models emphasize different device physics for different applications. In fast switching (digital) applications, even when dc current crowding is not significant, the transient base current
56
0.040
0.030
> 0.020
E
0.010
0.0 10.0 20.0 30.0 40.0
Frequency (GHz) 100.0
80.0
60.0!
40.0 000
20.0
0.0 0.0 10.0 20.0 30 .0 40.0 Frequency (GHz)
Figure 312 Comparison between the MEDICI simulated base input admittance and the MMSPICE simulated base input admittance with and without ac current crowding. The symbol line represents the MIEDICI simulation. The thick solid line represents the MMSPICE simulation with ac current crowding. The thin line represents the MMSPICE simulation without ac current crowding.
57
can be very large causing severe transient crowding either in the periphery during tumon or in the center during turnoff. Hence, in fast largesignal switching applications, the key need is to properly simulate the transient crowding. For simplicity, QBBJT calculates transient base current by differentiating majority charge storage based on the solution from the previous time step. This is equivalent to treating the whole quasineutral base as a single capacitor, ignoring the physics of charge redistribution along the lateral base dimension. In ac analysis, the ac current is assumed small enough that the excess charge distribution does not significantly deviate from its steadystate solution, so the primary issue is to account for the small amount of charge redistribution along the lateral base dimension. The focus of the NQS ac: current crowding analysis discussed here is different from that of the transient crowding analysis. Therefore, the NQS ac current crowding analysis is not necessary to be consistent with the transient crowding analysis.
CHAPTER 4
PARAMETER EXTRACTION FOR MMSPICE/QBBJT
4.1 Introduction
QBBJT is implemented in at least one commercial SPICE version, and MMSPICE is available from its authors. QBBJT contains many features not provided by other available models, including accurate physical accounting for all quasi saturation mechanisms, emitter current crowding, base pushout, and selfheating. In spite of these advantages, use of QBBJT has been rather limited. One probable reason is that QBBJT lacks a welldefined parameterextraction methodology. In practice, even if QBBJT can be used with the model parameters derived from the device structures and process information, a circuit designer would still prefer a model with its model parameters calibrated for a given process to maximize the accuracy of circuit simulations. This chapter presents a systematic parameter extraction methodology for the QBBJT model parameters to be calibrated for a given process and used in its precision mode, even when the process information is not complete. The parameter extraction methodology, thus, successfully merges the QBBJT model into contemporary TCAD process and eliminates the need of involving another compact model with welldefined parameter methodology for precision purpose.
QBBJT's input parameter are derived from the collector and basedopant profiles. The model parameter extraction can be used to investigate the device structure. Device's 58
59
structures are usually obtained with reversebiased differential CV method or SIMS. Accuracy of the CV method is generally influenced by the junction abruptness and Debye length [32]. The applicable range of the CV method is limited by junction breakdown voltages. SIMS is reasonably accurate. However, SIMS generally demands very large test structures; the results thus represent predominant ID diffused dopant profile which may not be applicable directly to practicalsize devices. For instance, because of the 2D diffusion effects occurred in manufacturing processes, the emitterbase junction depth of the BJT's from the same process can be significant deviated for different emitter feature sizes, thereby influencing the peak base doping at the baseemitter junction. This effect can be also observed by comparing the electrical characteristics of devices with different emitter feature sizes. As shown in Fig. 41 and Fig. 42, the highinjection rolloff of the current gain P and unitygain frequency fT occur at higher current density for narrower emitter devices, because they have shallower emitterbase junctions due to the 2D diffusion effects, and thus higher peak base doping. This is also another reason why we need to calibrate the QBBJT's model parameters from electrical measurements for better precision, although it is also possible to evaluate them from process information.
We have found that the profilerelated QBBJT parameters we extracted correlate well with SIMS measurements. This suggests that when MMPSICE is used in this extraction mode, it provides an inexpensive way to infer profiles from nondestructive electrical measurements. Such an approach could be useful for statistical process control, or for investigating an unfamiliar transistor's fabrication process.
60
120.0
0 .8p.m x 20prm
a l.2pm x 20gm 01. g 6m x 20gm 100.0
: .. ........
S80.0
C)
60.0
40.0 .le+00 le+02 le+04 le+06
Collector current density, Ic/AE (A/cm2)
Figure 41 Current gain P vs. collector current density Jc
for different emitter sizes.
61
15.0 GE) AE = 0.8pgm x 20pm 3El AE = 1.2gm x 20gm
10.0
N
5.0
0 .0 W , I ni
0.0 2.0e+04 4.0e+04 6.0e+04 8.0e+04 1.0e+05 Ic/AE (A/cm2)
Figure 42 Unitycurrentgain frequency fT vs. collector
current density J for different emitter sizes.
62
The QBBJT model has 4 device line input parameters and 41 model card parameters. Their physical meanings and typical values are listed in Table 41 and Table 42. Basically, the QBBJT input parameters can be grouped into the following categories:
1. geometric layout parameters,
2. material dependent carrier transport parameters,
3. GummelPoonlike parameters,
4. collector doping profile parameters,
5. base doping profile parameters,
6. charge partition factors.
Table 41 Device line input parameters Name Description Units Default Typical
AE Effective emitter area m21.0e1 1 1.0e1 1
WE Effective emitter width m 1 .0e6 1 .0e6
AC Effective collector area AE AE
ABL Buried layer area M2jAC 
63
Table 42 Model card parameters Name Description Units Default Typical
UNEPI Electron mobility in epi cm2*V*ls1 1.0e3 1.0e3
collector
NEPI Epi doping density cm3 1.0e16 1.0e16
WEPI Epi collector width m 5.0e7 5.0e7
WBM Metallugical base width m 2.0e7 2.0e7
NAO Extrapolated peak base cm3 1.Oe18 1.0e 18
doping
ETA Base doping gradient 3.0 3.0
Al Preexponential coefficient cm1 0.0 7.03e5
for impact ionization rate
BI Exponential coefficient for V*cm1 0.0 1.23e6
impact ionization rate
VS Electron saturated drift cmes1 1.0e7 1.0e7
velocity
JEO Emitter saturation current A*m2 1.0e8 1.0e8
density
JSEO BE SCR saturation A*m2 1.0e4 1.0e4
current density
NEB BE SCR emission 2.0 2.0
coefficient
WSEO Zerobias BE SCR width m 5.0e8 5.0e8
PE BE junction potential V 1.0 1.0
barrier
ME BE junction exponential 0.4 0.4
factor
PC BC junction potential V 0.8 0.8
barrier
DNB Electron diffusivity in base cm2 10.0 10.0
PS CS junction barrier V 0.6 0.6
MS CS junction exponential 0.4 0.4
factor
CJS Zerobias CS junction Fcm2 0.0 1.0e4
capacitance
CRBI Intrinsic base resistance V.s.cm2 2.0e3 2.0e4
coefficient
TB Carrier lifetime in base s 1.0e7 1.0e7
64
Table 42 Model card parameters (Continued)
TC Carrier lifetime in collector s 1.0e7 1.0e7
CIF Forward current coupling 1.0 1.0
coefficient
CIR Reverse current coupling 1.0 1.0
coefficient
RC Extrinsic collector W 0.0 50
resistance
RB Extrinsic base resistance W 0.0 100
RE Extrinsic emitter resistance W 0.0 10
FB Basecharge partition 1.0 0.5
factor
FC Collectorcharge partition 1.0 0.5
factor
TE Emitter hole transit time s 1.0e9 1.0e9
JEOP Peripheral BE SCR Am1 0.0 2.0e1 1
saturation current density
NEBP Peripheral BE SCR 2.0 2.0
emission coefficient
UPBASE Hole mobility in intrinsic cm2oV1os 230 230
base
LBE Width of intrinsic base m 0.0 1.5e6
region
JCOP Substrate saturation Amq 0.0 1.5e15
current density
DNBH Electron diffusivity in base cm2*s1 10.0 10.0
for high injection
DEGE Bandgap reduction in base eV 0.0 0.0
emitter junction
DEGC Bandgap reduction in base eV 0.0 0.0
collector junction
WEM Emitter junction depth m 0.0 0.0
DEGA Average bandgap reduction eV 0.0 0.0
in emitter
65
Our extraction strategy is described as follows. First, we assume that geometric layout parameters such as the emitter and buried layer dimensions are known. The materialdependent carrier transport parameters, such as majority carrier mobility in the epicollector, are initially assumed to have their typical values. These values are later refined in the extraction. Then, we extract the GummelPoonlike parameters using existing techniques developed for the GummelPoon model. Although some of the GummelPoonlike parameters are not accurate enough for ac simulation, they are sufficient for the subsequent extraction procedures. With these GummelPoonlike parameters, we then extract the collector and base doping profiles through optimization of output characteristics and fT plot. Finally, we evaluate the charge partition factors and calibrate the extrinsic resistance parameters in accordance with measured smallsignal parameters.
To evaluate the charge partition factors and calibrate the extrinsic resistance parameters, we need to extract QBBJT's smallsignal equivalent circuit. Generally, there are two different approaches to extract BJT's smallsignal circuit. (1) Numerical optimization, which obtains parameters through fitting the equivalent circuit to measured sparameters. (2) Direct extraction, which obtains parameters step by step through manipulating the smallsignal parameters (z, y, or h parameters) derived from the equivalent circuit.
Optimization is a numerical technique used to determine one or more model parameters from measurement. Optimization is especially useful in the fine tuning of multiple parameters to obtain the best model fit to measurements. However, the optimization of multiple parameters is usually not desirable for extracting parameters for a
66
physical model, because it often results in nonunique or nonphysical parameter sets when given different initial guesses for parameters. There are several methods that prevent optimization from generating nonphysical parameter sets: 1) providing closer initial guesses based on dc parameter extraction; 2) imposing physical constraints; 3) using stepbystep deembedding techniques with special testing structures [33]; and 4) using advanced optimization algorithms, such as the stimulated diffusion technique [34].
In contrast, direct extractions avoid generating nonunique parameter sets. Therefore, a direct extract methodology is preferred in obtaining a unique set of parameters for a physical model. A direct extraction method is also more computationally efficient than global optimization. Based on the hparameters and zparameters derived from BJT's hybridT equivalent circuit, some direct extraction methods have been reported [35][36]. However, these methods are not useful to determine QBBJT's smallsignal parameters, because they use the oversimplified hybridT model. The extraction of smallsignal parameters is very sensitive to the topology of the equivalent circuit.
QBBJT's smallsignal equivalent circuit is based on the hybridit equivalent circuit shown in Fig. 43. Recently, a methodology to extract hybridit equivalent circuit was reported [37]. The method was also employed to investigate smallsignal base impedance [27]. The method is based on the fact that BJT's substrate parasitics are not significant at low frequencies. Using this fact, a BJT's lowfrequency zparameters can be analytically derived from the simplified lowfrequency equivalent circuit shown in the dashed box in Fig. 43. Through manipulation of the analytic form of BJT's lowfrequency zparameters, smallsignal resistances and capacitances such as rbb, ree, rcc and Ccc can be expressed explicitly. However, [37] is not completly direct extraction method, because it does not
67
ccsu
22Z22"
Figure 43 The QBBJT's smallsignal equivalent circuit
68
provide a direct way to extract or calculate BJT's baseemitter capacitance C,,. In addition, this method requires a final optimization to obtain the optimal value for the basecollector capacitance partition factor, which accounts for the NQS effects caused by the distributed base impedance.
This chapter presents a complete parameter extraction methodology for the MMSPICE/QBBJT model. With the extraction methodology, QBBJT's parameters can now be evaluated either from process information or from measured data.
The extraction methodology also includes the extraction of BJT's smallsignal hybridit equivalent circuit from measured sparameters. With the extracted smallsignal equivalent circuit, we can further calibrate the resistance parameters and evaluate the charge partition parameters. The extraction of BJT smallsignal equivalent circuit parameters uses no optimization of multiple parameters. Therefore, it ensures the uniqueness of extracted parameters and maintains their physical meanings.
This chapter is organized as follows. Section 4.2 Section discusses the parameters which can be extracted with the techniques developed for the GumnmelPoon model. Section 4.3 discusses the extraction of QBBJT's physical parameters, such as the collector and base doping profiles. Section discusses the extraction of QBBJT's smallsignal equivalent circuit. Section 4.5 discusses remaining parameters and concludes this chapter with a summary.
4.2 Extraction of GummelPoonLike Parameters
Some of the QBBJT parameters are analogous to similar parameters in the GummelPoon model and can be extracted from a Guminel plot [7]. Saturation collector
69
current densities JCS, JEO and JSEO are extracted from a linear fitting to the Gummel plots in the medium and low current range. JCS is not a model parameter. JC will be used to determine DNB during the optimization of the collector doping parameters. Junction capacitance parameters, such as barrier potential and nonideality factor, can be extracted from reversebias CV measurements on largearea devices. These junction capacitances will be used in the optimization of base doping parameters.
Parasitic resistances RC, RB and RE (collector, base and emitter resistances) are required in subsequent extraction and optimizations of collector and base doping profiles. Because QBBJT calculates the voltage drops across the epicollector and intrinsic base (base region underneath the emitter), thereby including the biasdependent intrinsic collector resistance and base resistance, RC and RB represent the extrinsic part of the total collector and base resistances. The intrinsic collector resistance usually dominates the total collector resistance in the forwardactive region, but it becomes negligible in hard saturation. The extrinsic collector resistance can therefore be extracted from the slopes of IC vs. VCE curves in hard saturation (i. e., near IC= 0) [7].
Emitter and base resistances can be extracted simultaneously using Ning and Tang's method [38]. Assuming IB degrades only due to the VBE drop across the base and emitter resistances, the potential drop can be expressed as VT' lnIOI)I = RE +(RE+RB)/jB. (41)
Therefore, the intercept of the (VT'* ln(IBO/IB))/IC VS. l/P curve gives RE and the slope gives (RE+RB). A typical base resistance vs. bias plot extracted using this method is shown in Fig. 44. The intrinsic base resistance predominates at low injection, but it becomes negligible at high currents because of base majority charge modulation and
70
1.6gmx2Ogm
35.0
E
VCE =2.0 V
o 30.0
4)25.0 ,20.0
15.0 1
0.80 0.90 1.00 1.10
VBE MV
Figure 44 Extracted base resistance vs. VBE Plot
shrinking of the effective emitter width due to current crowding. Therefore, the extrinsic base resistance can be estimated from the minimum (highinjection) value of the total base resistance,
4.3 Extraction of Doping Profiles
4.3.1 Collector Doping Profile
Collector parameters WEPN (epilayer thickness) and NEPI (epilayer doping) determine the epilayer resistance, which in turn controls the quasisaturation IN characteristics. NEPI is directly related to basecollector breakdown voltage as well as the threshold voltage of PMOS devices in n well BiCMOS technologies [27]. NEPI also controls the onset of base pushout and the width of the basecollector SCR, and thus
71
strongly affects collector transit time. As base widths are progressively scaled down, collector transit time becomes an increasingly significant part of the total transit time and hence strongly affects highfrequency ac characteristics.
Since the electrical performance of a bipolar transistor operating in quasisaturation is primarily controlled by its collector structure, we can extract collector parameters by optimizing them to fit the QBBJT model prediction to the quasisaturation IV characteristics. Unlike most compact BJT models (for example, the extended GummelPoon model) model the quasisaturation operation by assuming that the entire epicollector is quasineutral, QBBJT accounts for all possible SCR formation in the epicollector. QBBJT solves de Graaff's base transport equation [12] and Kull's collector transport equation [10] subject to moving boundary conditions [17]. Therefore, QBBJT properly simulates currentvoltage characteristics accurately in quasisaturation.
To improve the stability and convergence rate of the optimization, PC (BC junction potential barrier) and REp! (epilayer resistance) are used as the target variables for the optimization instead of optimizing NEPI and WEPI directly. This preserves correlations among the model parameters and helps to eliminate nonphysical parameter combinations. Because the optimization of the collector parameters is fairly insensitive to the base doping gradient, the base grading factor ETA can temporarily be assigned an typical value. Assuming the base doping at the basecollector metallurgical junction equals the collector doping, NEPI, WEPI, and NAO (extrapolated peak base doping) can be related to PC and REp! as
PC = 2. 1n(NEPI/ni) (42)
REPI = WEPI/(q UNEPI. Ac NEPI) (43)
72
NAO = NEPI. exp(ETA) (44)
Calculating WBM (metallurgical base width) and DNB (average minority diffusivity in base) requires a value for the total base doping, QBO. This parameter can be extracted from the forward Early voltage if selfheating is eliminated [21]. The Early voltage VA, defined as VA = (1/Ic)Ilc/@VcB, can be expressed as [39]
II =~ QBO QJE jc(45) =S/(WscB WScc)'
where Fs is the silicon permittivity, WSCB and Wscc are the baseside and collectorside depletion widths of the basecollector junction, and QJE and QJC are the depletion charges of the emitter and collector junctions. Under lowinjection conditions (i. e., neglecting background charge modulation induced by collector current) WSCc and WSCB can be calculated using the depletion approximation. Qjc can be calculated as q NEPI. Wscc. With the baseemitter depletion width estimated assuming a oneside abrupt junction, QJE can be obtained by integrating the base dopant across the baseemitter depletion region. Since VA is welldefined for every iteration value of QBO, QBO can be obtained by optimizing it to fit the VAVCB curve.
With QBO obtained from the previous step, WBM and DNB can be calculated as WBM QBO ETA
NAO 1 exp(ETA) (46)
DNB Jcs" NG (47)
2
q hie
73
where NG is the Gumnmel number, defined as NG = (QBOQJE Q1c)/q, and JCS is extracted from the Gummel plot as discussed in Section 2. With all the correlated parameters recalculated for every iteration, the collector parameters can be obtained from optimizing them to fit the model prediction to the measurement data. However, From (47), it is clear that variation of the bandgap can critically affect the calculated value of DNB. For a contemporary bipolar transistor with a very thin and exponentially doped base, the base doping at the emitter side can be as high as 1018 cm3, causing significant bandgap reduction. Overlooking such bandgap variation would lead to significant errors in the calculation of DNB due to the exponential dependence of n je on the bandgap variation. Extraction of the bandgap variation parameter is possible using temperaturedependent measurement [40]. However, for simplicity we used the empirical model in [41] to estimate the bandgap variation at the emitter edge of the base using the value of NAO for every iteration.
To ensure the reasonable values of the bandgap variation parameters, the intermediate values of DNB can be monitored to check the consistency. Based on the estimated basedopant profile and the minority mobility model in [42], a reasonable range of DNB (typically from 6 to 12 cm2/sec.) can be obtained using Einstein's relation. If the intermediate values of DNB go out of the valid range, the bandgap variation parameters can be finetuned in accordance with the calculation of [41]. The flowchart of the optimization procedures of collector parameters is shown in Fig. 45.
74
Assign ETA=3.0
Initial guess of PC and REPI, S Calculate NEPI, WEPI, NAO
Extract OBO by fitting VAVcB curve Tune bandgap With Gummel number known,
variation Calculate WBM and DNB
parameters
Calculate next
NO Is DNB I PC and REP,
NO in valid range? n Ep
using LevenburgYES Marquart algorithm
Invoke MMSPICE/QBBJT simulation Calculate the sum of square errors, P
Does g NO
Fi e4 F reach mdoim um? r Done
Figure 45 Flowchart of collector doping extraction
75
4.3.2 Base Doping Profile
Although total base doping QBO can be obtained by fitting the forward Early voltage VA as a function of VBC, the actual basedopant grading factor ETA cannot be obtained precisely from dc measurements because dc currentvoltage characteristics depend only weakly on basedopant grading. However, the base transit time 'rB depends strongly on the basedopant grading because of the electric force it causes. QBBJT physically accounts for the dependence of the base transit time on the builtin electrical field, so ETA can be extracted by optimizing it to fitfT data.
The total forward transit time, rT = ('CB + "C + TE) can be estimated from the unity currentgain frequency using cT = 1/(27t fT). For narrow base widths, 'tc and "E can become comparable to "TB and must be carefully accounted for in order to correctly predictfTp
In its regional charge analysis, QBBJT models rc as QCIC, where Qc = Wscc" (qNEp Ic/Us). Using NEPI obtained from the collectorprofile
extraction, WsCc can be calculated based on the depletion approximation, modified to account for the background charge modulation induced by collector current, which allows an accurate estimate of 'rc.
QBBJT uses the empirical parameter TE (emitter minoritycarrier transit time) to account for TE. TE can be related to TE as TrE = QEEI'C = (TE. I)Ic TE/j3, where QEE is the extra charge storage in the emitter. If the emitter junction is very shallow (< 0.1 gm) and has a metal contact, the extra charge storage in emitter is generally very small and
76
"E can often be neglected. However, if the emitter is thick and uses a polysilicon contact which can accumulate a significant charge, rE can be comparable to "TB and must be calculated independently; if the detailed emitter structure is known, 'rE can be estimated using the formula in [44].
ETA can be extracted by optimizing ETA to fit fT vs. VBE curves using the previously extracted values of NEPI and WEPI. Since the quasineutral base can extend into epicollector through either ohmic or nonohmic quasisaturation mechanisms, it is important to select the right bias range for the fT measurements used in extracting ETA. VYBE should be restricted to ensure that Ic is smaller than the threshold for base pushout at Io = qA'usNEPI [39], and VCB must be kept high enough to avoid quasisaturation.
4.3.3 Integration of Doping Profile Extraction Procedures
The optimization procedures described above were implemented using the Optimization Toolbox in MATLAB. A translation program was written in C that converts MATLAB outputs to MMSPICE/QBBJT input parameters, invokes MMSPICE simulations, and returns the simulation results to MATLAB. The MMSPICE code was modified slightly to reduce the number of required file accesses. Also, a provision was added to calculate fT directly from the regional charges in dc analysis rather than by extrapolation using ac analyses.
The empirical parameters Jcs, JEO, JSEO, and the parasitic RE, RC and RB are extracted from Gummel plots as described in Section II. Then, with a default value assigned to ETA, the collector parameters are extracted by fitting to quasisaturation IV
77
curves. Next the emitter transit time "rE is estimated using the formula in [44], and the basedopant grading factor ETA is extracted from optimization of fT vs. VBE curves as in Section III. As a final step, the zerobias baseemitter SCR width WSEO is tuned to fit f3 VS. VBE curves.
4.3.4 Extraction Results and Discussion
The parameter extraction procedures are tested on doublepoly selfaligned BJT devices from a Motorola 0.5gm BiCMOS process. Three sets of data are taken. They are 1) Gummel plots, e.g. log(Ic) and log(IB) vs. VBE with VCE as parameter, 2) IC vs. VCE plots with VBE as parameter, 3) fT vs. VBE plots. The dc measurements are done with HP4145A semiconductor parameter analyzer. To avoid significant selfheating, the forward Early voltage data, which are used to extract QBO in optimization of collector structural parameters, are evaluated at lower VBE. In our present case, the forward Early voltages are extracted from Ic vs. VCE curve at VBE=0.8V. Because of their easycalibration nature, sparameter measurements, which are used most widely for characterizing smallsignal behaviors of devices, are employed to extract fT data. With the Cascade air coplanner microwave probes, the sparameter measurements are done using the HP8510A network analyzer set in conjunction with the HP4140 dc power supply source. Since smallsignal current gain 3 can be evaluated from sparameter data as 13 = s21/[(I s11) (1 +s22) +s12s21] [45], fT vs. VBE curve can be obtained from extrapolating 13 vs. frequency plot to unity gain at each VBE.
78
In spite of taking 2D effects into consideration[ 16], e.g. sidewall currents and extrinsic parasitic capacitances, the QBBJT model is virtually a IlD model. To minimize the influence of 2D effects on parameters extraction, measurements are done on a considerably large (1 .6jim x 20gmi) emitter device. Large devices also help to reduce the importance of deembedding work in sparameter measurements. However, large devices aggravate emitter current crowding. The current crowding effect causes devices to enter high injection earlier by increasing current densities corresponding to reduced effective emitter area. Therefore, the current crowding makes the P3 and fT vs. VBE curves rolloff faster at high current density. Physically accounting for the current crowding as discussed in Chapter 2, the QBBJT model eliminates the drawbacks of parameter extraction on large devices.
The extracted collector parameters and basedopant grading factor were found to agree closely with corresponding parameters extracted directly from SIMS profiles as shown in Table 43. However, the extracted base width was 16% larger than that indicated by SIMS, which led to about a 2:1 error in the peak doping NAO. This discrepancy can be explained by noting that the SIMS measurements were performed on very large area devices. Because of twodimensional diffusion effects, the emitter junction of small devices is expected to be shallower than that of large devices. Therefore, the base of the small device should be wider and contain more dopants than predicted by SIMS.
79
Table 43 Comparison between SIMS and extraction results NEPI WEPI NAO ETA/WBM WBM
SIMS 4e16 0.50gm 1.00e18 20.10gm"1 0.140 gm
1e17
Extraction 7.4e16 0.47[jm 2.12e18 20.36 jarm1 0.163gm
Difference%  6% 112% 1% 16.4%
The simulations with extracted parameters are shown in Fig. 46 through Fig. 49. A new version of the QBBJT model which includes selfheating and modified current crowding model is used to simulate device characteristics with the extracted parameters. The thermal impedance is calculated theoretically as in [22] or evaluated as discussed in [23]. Notice that the extracted physical parameters such as base and collector doping profiles are also applicable to the BJT's of the same structure but with different emitter feature sizes. For instance, NAO, WBM, ETA, NEPI, and WEPI extracted from the I.6gmx20gm device can be used for the 0.8gmx20gm device without repeating the complete extraction procedures. This suggests that the model is truly scalable and thus can reduce the parameter extraction efforts significantly.
One bonus feature of the parameter extraction methodology is that with the parameters calibrated with measurements, the firstoder derivative discontinuities in QBBJT's collector current can be removed. Depending on the existence of basecollector SCR, QBBJT uses two different sets of equations to calculate the collector current. This can causes the firstorder derivative discontinuity in QBBJT's collector current, when operation transits between quasisaturation and forwardactive modes, if given
80
(a) 1.6pmx20gm 100
VCE = 2.0 V
102
1041
o 106
108
1010
0.4 0.6 0.8 1.0 1.2
VBE (V)
(b) 0.8mx20m 101
VCE = 2.0 V
103 0.0,
S10"
10"
1011
0.4 0.6 0.8 1.0 1.2
VBE (V) Figure 46 Gummel plot, log(IC) & log(IB) vs. VBE. Symbol lines are the measurement data. Solid lines are simulation results.
81
(a) 1.6,mx20gm 120.0
VCE = 2.0 V = 100.0
S80.0
60.0
40.0 ,
1010 108 106 104 102 100 IC (A)
(b) 0.8pmx20gm 120.0
VCE = 2.0 V 100.0
80.0
60.0
40.0 ,.
1010 108 106 104 102 100 IC (A) Figure 47 Current gain 3 vs. VBE. Symbol line is the measurement data. Solid line is simulation with current crowding effect. Dashedline is simulation without current crowding.
82
(a) 1.6gmx2Ogm
15.02
10.0
5.0 0.0
5.02
0.0 1.0 2.0 3.0
VCE (V)
(b) 0.8gmx2Ogm 10.0
5.0 ....0.0
5.0
0.0 1.0 2.0 3.0
VCE (V)
Figure 48 IC vs. VCE plot. Symbol lines are the measurement data. Solid lines are simulation results with selfheating. Dashedline are simulation results without selfheating.
83
15.0
0
o
10.0 a
0
0
0
5.0 0
0.0
0.4 0.6 0.8 1.0 1.2 1.4
VBE (V)
Figure 49 fT VS. VBE plot. Symbol line is the measurement data. Solid line is simulation with current crowding effect. Dashedline is simulation without current crowding.
84
inappropriate parameters. Because the collector parameters are obtained from optimization, which in turn conform the model prediction to fit the measurement data, the discontinuity problem can be avoided as shown in Fig. 4 10.
4.4 Extraction of OBBJT's SmallSignal Equivalent Circuit
4.4.1 Methodology
In previous sections, most of QBBJT's parameters were extracted from dc currentvoltage characteristics, lowfrequency CV measurements and fT plots. The resulting model parameters give very accurate predictions of BJT's dc characteristics. However, some of these parameters might not be optimal for ac simulations, because isolation of their individual effects from dc measurement is difficult. For instance, the base and emitter resistances can not be accurately determined through dc measurements, because their effects are coupled together at dc. Also, for accurate ac simulation, we still need to know the charge partition factors, which are used in QBBJT to empirically account for the nonquasistatic (NQS) carrier transport in the quasineutral base.
This section presents a new methodology to directly extract QBBJT's smallsignal equivalent circuit, from which we can evaluate the optimal value of RE, RB, and charge partition factors for accurate ac simulation. The extraction methodology begins with the calculation of the differential current gain and baseemitter resistance from the de currentvoltage characteristics. At medium to highinjection levels, the base current is dominated by the minority carrier diffusion current in the emitter, given as 1B = IBOexp(VBE/IVT), (48)
where VBE' is internal bias voltage. Therefore, r.~ can expressed as
85
30.0
25.0 simulated Ic
simulated dIc/dV 20.0 S15.0 10.0
5.0
0.0
5.0
0.0 1.0 2.0 3.0
VCE (V) Figure 410 Simulated Ic vs. VCE and its first derivative.
86
r = 1/(dIB/dVBE') = VT/IB (49)
Because the ideality (exponential) factor of collector current is not necessarily equal to unity at high injection, we cannot directly obtain gm by direct differentiation of the collector current IC with respect to the baseemitter voltage VBE'. Instead, we should use dIc dIc/dIB fPac
gM ac = (410)
n dVBE' dVBE'/dlB r, (410)
where fac is lowfrequency commonemitter current gain, which can be obtained from BJT's dc collector and base currents. In [37], gm is expressed as gn= gnoexp(jOrd) (411)
where the exponential term is used to account for the phaseshift caused by the nonquasistatic (NQS) minority transport in the quasineutral base. Since such a NQS effect has been accounted for in QBBJT with the charge partition technique, we use (410) instead of (411).
Since contemporary silicon bipolar transistors are usually built on a very lightly doped substrate or epilayer, the resulting substrate parasitics are insignificant at low frequencies. With these substrate parasitics neglected at low frequency, BJT zparameters without consideration of probe pad parasitics are expressed as z11 = rbb+ ree +Z1/(l +gmZt) (412)
Z12 = ree + ZC/(1 + gaZ") (413)
Z21 ree + [Z/( 1 + g1Z7)] ( 1 g,11/joCcc) (414)
z22 = r'cc +ree + (Z7 + 1/j0Cec) /(1 + ginZ) (415)
where Z7 = rn/( I + jorC). Subtracting zl2 from z11 yields the base resistance rbb as rbb = Re(zI,z12). (416)
87
Subtracting z21 from z22 yields the basecollector capacitance Ccc as
C = l/[Co"Im(z22z21)]. (417)
To extract emitter resistance ree, we multiply the numerator and denominator of the second term of (413) by gm and obtain
ree = Re(z12 a/g,) (418)
where Oxac Pac(Pac + 1) is the lowfrequency commonbase current gain.
To compute the collector resistance rcc, we cannot directly subtract z21 from z22, because z21 converted from sparameters is usually unreasonably large at low frequencies [37]. To compute collector resistance rcc, we subtract z12 from z22 and obtain rcc = Re[z22 z12 I + (419)
The above equation contains another unknown Cr. To calculate C,,, we use BJT's first dominant pole extracted from the unitygain frequency fT as Pdominant=PaIc(2 tfT). From the hybridit equivalent circuit of Fig. 43, BJT's first dominant pole is expressed as
Pdoininant = C~r7 + Ccc[r= + (1 + gmrrt)(ree + rcc)]. (420)
Notice that (419) and (420) are independent and contain only two unknowns rcc and C.. Therefore, we can solve rcc and C7, from (419) and (420).
With r7T, gm, rbb, ree, rcc, Ccc, and C7, calculated with (49), (410), and (416)(420), BJT's lowfrequency equivalent circuit (i.e. the circuit in the dashedbox in Fig. 43) is directly extracted from lowfrequency zparameters. This equivalent circuit fits closely to measured smallsignal parameters at low frequency. However, it does not accurately simulate BJT's smallsignal parameters at high frequencies, because this equivalent circuit does not take substrate parasitics into considerations. Therefore, we extract substrate
88
parasitics from the discrepancy between the simulated and measured data at high frequencies.
The extraction of substrate parasitics begins with the calculation of z22 of the lowfrequency equivalent circuit. Then the calculated z22 is compared with the measured z22. From the difference, the substrate parasitics are extracted as Z22 = Z22' II (rsub+.I). (421)
where z22' is the simulated z22 of the lowfrequency equivalent circuit and z22" is the measured z22.
In practical situations, only Rsub is extracted from this step. We use the Ccs measured with lowfrequency CV measurements, and extract Rsub from fitting (421) to the measured z22 from 100MHz to 7.5GHz. Since the optimization contains only one variable, we ensure the uniqueness of Rsub, and maintain its physical meaning. The extraction methodology of BJT substrate resistance will also be used in the next chapter to verify the calculation of BJT substrate resistance.
4.4.2 Calibration of QBBJT's Extrinsic Resistance Parameters
The extraction methodology was tested on bipolar transistors of the 0.6gm BiCMOS process of Texas Instruments, Inc. Sparameters are used because they are easier to measure at high frequencies than other smallsignal parameters such as z, y, or hparameters. Many calibration and deembedding techniques are also welldeveloped for sparameter measurements. A HP85 IOC network analyzer system was used to measure BJT sparameters. These sparameters were taken from 100 MHz to 7.5 GHz for IB swept from
89
1 gA to I mA and VCE from 1 V to 3 V. The resulting sparameters were deembedded with measurements of open structures. Then, BJT's zparameters were converted from the deembeded sparameters.
The baseemitter resistance r., and transconductance g. are calculated directly from BJT's dc bias currents using (49) and (410). Therefore, the r. and gm are consistent with the prediction of QBBJT using the parameters extracted from Chapter 4. Other components are extracted with the procedures described previously.
Figure 411 shows the extraction of ree. QBBJT models the emitter resistance with a constant parameter RE. RE can be extracted from NingTang's method or the floating collector technique as discussed previously. However, the accuracy of RE extracted with these dc methods is usually limited, because the effects of emitter resistance and base resistance are coupled in dc measurements. Here, we provide an alternative way to determine RE from ac measurements. Notice from Fig. 411 that the ac extracted RE reaches a constant value when IB is greater than a threshold value. Therefore, we can extract RE from the dashed line in Fig. 411.
Figure 412 illustrates the extraction of rce. QBBJT calculates the intrinsic collector resistance internally and leaves the extrinsic collector resistance RC as an input parameter. RC can be extracted from the output characteristics of a BJT operating at saturation region [7]. However, the extrinsic collector resistance RC extracted using this method is usually dependent on IB, because the intrinsic collector resistance is not totally negligible even if the BJT operates at saturation region. Here, we provide an alternative way to extract RC for a BJT operating at forward active region. Notice from Fig. 412 that the total collector resistance extracted from BJT's highfrequency sparameters is almost a
90
30.0
25.0 VCE=1.0V
VCE=1.5V
SVCE=2.0V 20.0 VCE=2.5V
E VCE=3.0V
o 15.0
10.0
5.0
5 .. .m.... ... ........ i.... .
0.0
1 e06 1 e05 1 e04 1 e03 Ib (A)
Figure 411 The extracted smallsignal emitter resistance ree for different bias conditions.
500.0
VCE=1.0V 400.0 VCE=1.5V
 VCE=2.0V
E 3VCE=2.5V 0.0 VCE=3.0V
200.0
100.0
0.0
1 e06 1 e05 1 e04 1 e03 Ib (A)
Figure 412 The extracted smallsignal collector resistance rc for different bias conditions.
91
constant when IB is greater than a threshold value. When a BJT is operating at forwardactive region, its intrinsic collector resistance can be approximated as the epicollector resistance. Therefore, RC can be extracted as RC = RC(totai)REPI, (422)
where REPI = WEPI/(q UNEPI Ac. NEPI).
Figure 413 illustrates the extraction of total base resistance rbb. Notice that rbb decreases with BJT's base current 'B, and eventually becomes a constant at very high current. This phenomenon is primarily due to the majority charge modulation and dc current crowding. Because the effective base width decreases with base current and the intrinsic resistance eventually becomes zero, the total base resistance approaches the value of the extrinsic base resistance at high current. Therefore, we can extract the extrinsic base resistance RB when the ac extracted total base resistance approaches a constant at high current.
QBBJT calculates the intrinsic base resistance in two different ways. If the current crowding analysis is on, the intrinsic base resistance is internally accounted for in the current crowding analysis as discussed in Chapter 3. If the current crowding analysis is off, the intrinsic base resistance is calculated as RB(intrinsic) = CRBIWEWE (423)
QBB (3
where WE is the emitter width, QBB is the total base majority charge, and CRBI is an input parameter. In this case, the intrinsic base resistance is a constant for all bias conditions. CRBI can be theoretically calculated as CRBI = 1 (424)
n. UPBASE'
92
200.0
VCE=1.0V
 VCE=1.5V
150.0 VCE=2.0V
VCE=2.5V
5 VCE=3.OV
.S 100.0
50.0 .
0.0 le06 1e05 1"04 le03
Ib (A)
Figure 413 The extracted smallsignal base resistance rbb
for different bias conditions.
where UPBASE is the majority mobility in the base and n is equal to 3 for single base contact configurations and 12 for double base contact configurations. With QBB calculated from the total base doping for low and medium injection conditions, CRBI can also be tuned to fit the ac extracted total base resistance.
4.4.3 Evaluation of QBBJT's Charge Partition Factors
Figure 414 shows the extraction of the baseemitter and basecollector capacitances C, and Ccc. Ccc is roughly a constant at low currents, but becomes proportional to C,, for base currents greater than IOOjiA. This phenomenon is caused by nonquasistatic (NQS) minority transport in the quasineutral base. The derivation of the equivalent circuit of Fig. 43 is mainly based on the quasistatic (QS) approximation.
93
10o1
VCE=1.OV
1010 VCE=1.5V
' VCE=2.OV
101 VCE=2.5V
VCE=3.OV
102
11
1 e06 1 e05 1 e704 1e03 Ib (A)
1010
VCE=1.OV
1011 VCE=1.5V
VCE=2.OV U~. VCE=2.5V
8 1012 VCE=3.OV
10141e66 l5 1754 1 e03
Ib (A)
Figure 414 The extracted smallsignal baseemitter and basecollector capacitances C,, and CCC for different bias conditions.

Full Text 
MODEL ENHANCEMENTS AND PARAMETER EXTRACTION FOR
THE MMSPICE/QBBJT MODEL
BY
TZUNGYIN LEE
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE
UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1997
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude and appreciation to my advisor, Dr.
Robert M. Fox, for his devoted guidance, patient encouragement, and support throughout
my Ph.D study. It is a great pleasure to work with him. I would also like to give thanks to
my supervisory committee, Dr. Gijs Bosman, Dr. Jerry G. Fossum, Dr. Mark Law, and Dr.
Mang Tia for their willingness to provide me their time and guidance.
I am grateful to the Semiconductor Resource Company (SRC), Motorola, and
Texas Instruments, Inc. for their technical and financial support throughout my research. I
feel especially obligated to express my deep appreciation to Dr. Mark Foisy at Motorola
for his invaluable comments and discussions. I am also grateful to Mr. Tom Vrotsos, Dr.
Keith Green and Dr. Zhiliang Chen, for their guidance during my internship in Texas
Instrument Inc. Without their support, the work could never have been finished. I would
also like to thank Dr. Ranjit Gharpurey for the invaluable discussion on BJT substrate
resistance calculation.
I would like to recognize in this achievement Dr. MingChang Liang and Messrs.
David Zwidinger, Jonathan Brodsky, MingYeh Chuang, and Amed Nadeem, who helped
me in many ways with sincere and profound discussions on related topics, and my best
friend, Everett Yang, who helped me in editing the drafts of this dissertation.
My gratitude is also extended to many of my friends, who have made my life most
cheerful. I cannot mention them all, but I would like to mention Dr. Paul Chen, Dr. Simon
u
Wang, Dr. Bruce Liu, Jenfeng Yue, Jenyi Yue, and Chihou Vong. Without them, my life
would have been difficult and barren.
My deepest gratitude goes to my parents ChenI and Kuolan Lee. They presented
themselves as an example in several ways that I can be humble and responsible. In
addition, without their unfailing love and financial support, I could never have come to the
United States and begun my graduate studies.
I am pleased that my wife Fiona has been able to share the achievement with me.
Being a wife of a student for many years is not easy at all. Finally I would like to give my
praises to my Lord, Jesus Christ, for He is the beginning and He is the end.
TABLE OF CONTENTS
page
ACKNOWLEDGEMENTS ii
ABSTRACT vi
CHAPTERS
1 INTRODUCTION 1
2 BACKGROUND, PHYSICS, AND MODEL FORMULATION OF THE
MMSPICE CHARGEBASED BIPOLAR TRANSISTOR MODEL  QBBJT ...10
2.1 Background Review 10
2.1.1 EbersMoll Model 10
2.1.2 GummelPoon Model 11
2.2 Novel Compact Models for Bipolar Transistors 12
2.2.1 The Extended GummelPoon Model 12
2.2.2 MEXTRAM 13
2.2.3 MMSPICE/QBBJT Model 14
2.2.4 Vertical Bipolar Inter Company Model (VBIC95) 16
2.3 Model Formulation of QBBJT 17
2.3.1 BJT Operation Modes 17
2.3.2 Formulation of DC CurrentVoltage Equations 22
2.4 QBBJTâ€™s SmallSignal Equivalent Circuit 24
3 MODEL ENHANCEMENTS OF THE QBBJT MODEL 30
3.1 Selfheating Modeling 30
3.2 DC Current Crowding Modeling 35
3.3 AC Current Crowding Modeling 44
3.3.1 Simple Model Formulation 44
3.3.2 Accounting for DC Current Crowding in AC Current Crowding
Analysis 49
3.3.3 Model Implementation and Verification 52
3.3. 4 Conclusion and Discussion 55
ÃV
4PARAMETER EXTRACTION FOR MMSPICE/QBBJT
58
4.1 Introduction 58
4.2 Extraction of GummelPoonLike Parameters 68
4.3 Extraction of Doping Profiles 70
4.3.1 Collector Doping Profile 70
4.3.2 Base Doping Profile 75
4.3.3 Integration of Doping Profile Extraction Procedures 76
4.3.4 Extraction Results and Discussion 77
4.4 Extraction of QBBJTâ€™s SmallSignal Equivalent Circuit 84
4.4.1 Methodology 84
4.4.2 Calibration of QBBJTâ€™s Extrinsic Resistance Parameters 88
4.4.3 Evaluation of QBBJTâ€™s Charge Partition Factors 92
4.5 Conclusion 95
5 SPICE MODELING OF BJT SUBSTRATE RESISTANCE 101
5.1 Introduction 101
5.2 Algorithm to Calculate BJT Substrate Resistance 104
5.3 Limitations of the Algorithm 111
5.4 AC Verification of the Algorithm 113
5.5 Including BJT Substrate Resistance in Parameter Extraction 116
5.6 Conclusions 118
6 SUMMARY AND SUGGESTED FUTURE WORK 119
6.1 Summary 119
6.2 Suggested Future Work 122
APPENDIX THE SUBSTRATE RESISTANCE CALCULATION PROGRAM
SUBTEST 125
REFERENCES 128
BIOGRAPHICAL SKETCH
133
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
MODEL ENHANCEMENTS AND PARAMETER EXTRACTION FOR
THE MMSPICE/QBBJT MODEL
By
TzungYin Lee
May 1997
Chairman: Robert M. Fox
Major Department: Electrical and Computer Engineering
This dissertation addresses several enhancements for the MMSPICE/QBBJT
model which are critical to contemporary bipolar transistors. First, selfheating is included
by adding an extra thermal node into QBBJTâ€™s admittance matrix. Through the extra
thermal node, the transistorâ€™s local temperature is simulated along with its terminal
currents and voltages. Second, QBBJTâ€™s dc current crowding analysis is modified to better
model the transistorâ€™s high injection behavior. Third, a compact and physical ac current
crowding model is developed. AC current crowding is predominantly a nonquasistatic
(NQS) behavior caused by the majority carriers transporting laterally along the base. By
adapting the RC transmission line model, an equivalent circuit is developed to account for
the NQS carrier transport. The equivalent circuit also accounts for the base resistance
reduction due to dc current crowding. With the effective base width calculated from dc
VI
current crowding analysis, the model can also be extended to account for the
nonuniformity in the baseemitter admittance along the base.
This dissertation presents a complete parameter extraction methodology for the
MMSPICE/QBBJT model. QBBJT is designed to take the model parameters derived from
transistorâ€™s structural information such as layout and doping profile. In such a way,
QBBJT enables simulation even when the process is not yet finalized. This attribute makes
QBBJT especially useful in concurrent engineering applications. However, QBBJT still
needs a methodology to calibrate or tune its parameters when representative
measurements are available. This dissertation demonstrates a systematic methodology to
extract QBBJTâ€™s model parameters from measurements. The methodology can be
implemented in automated extraction tools. With this methodology, QBBJT now can be
used in either the predictive mode for concurrent engineering applications or the precision
mode for routine circuit verification.
Finally, the dissertation presents a systematic methodology to include substrate
resistance in BJT parameter extraction and simulation. Substrate resistance is often
neglected in compact BJT models, since it is layout dependent. Even for the most
sophisticated parasitics extraction programs, substrate resistance is ignored because its
calculation involves computationintensive threedimensional simulations. Based on a fast
calculation algorithm which was initially developed to characterize substrate noise
coupling in mixedsignal applications, the dissertation demonstrates an efficient way to
eliminate the uncertainties resulting from the layout dependence of substrate resistance in
BJT parameter extraction and circuit simulation.
CHAPTER 1
INTRODUCTION
Since the bipolar junction transistor (BJT) was invented in 1947, it has been
employed in applications of analog amplification and digital switching. Complementary
metaloxidesemiconductor (CMOS) technology, because of its characteristics of low
power consumption and highpacking density, is preferred in fabrication of ultra large
scale integrated (ULSI) circuits. However, because of its advantages in highspeed
operations, bipolar technology is still employed in a wide variety of applications, ranging
from high performance mainframe computers and data links, to analog and digital
telecommunications and wireless communications [1]. As aggressive scaling and
advanced process technology continue to improve the speedpower performance of
CMOS, CMOS threatens to overwhelm all the advantages of conventional bipolar
transistors and become the ubiquitous technology. Nevertheless, the process and
lithography breakthroughs can also improve the speed performance of bipolar transistors.
Because bipolar transistors have advantages in operation speed over CMOS technology,
they are often used in highperformance digital and lowcost highspeed RF applications.
As the result of its high power dissipation compared with CMOS technology,
bipolar technology is rarely used in very highlevel integration. To reduce power
dissipation with high levels of integration, contemporary bipolar technologies have been
developed to maintain BJTâ€™s speed advantages while achieving higher integration level.
l
2
One approach is to merge highperformance bipolar devices into CMOS processes, and
form Bipolar/CMOS (BiCMOS) processes. In such processes, bipolar transistors are
employed only where the leverage is high enough to justify extra power consumption. By
retaining the benefits of bipolar and CMOS technology, BiCMOS technology allows
overall speedpowerdensity performance previously unattainable with either technology.
BiCMOS gates now can reach a speed about twice that of CMOS gates with
approximately equal power dissipation [2], Availability of bipolar transistors along with
CMOS devices also adds design flexibility in analog or mixedsignal applications.
However, the manufacturing cost of BiCMOS ICs is considerably higher than that of
CMOS ICs. In addition, insertion of bipolar processes into baseline CMOS processes
usually constraints optimization of either bipolar or CMOS transistor performance [3],
Another direction for contemporary bipolar technology is to fabricate high
performance vertical pnp as well as npn bipolar transistors on the same chip.
Complementary bipolar technology has long held the promise of achieving highspeed
with lower power dissipation. Until recently, the bottleneck of complementary bipolar
technology is to realize highperformance vertical pnp transistors with baseline npn
bipolar processes [1]. Therefore, complementary bipolar technology completely is
unlikely to replace vertical npn technology in the near future.
Currently the most advanced technology for realization of highspeed bipolar
transistors is based on doublepoly selfaligned technology. Doublepoly selfaligned
bipolar processes were developed over a decade ago, and rapidly became the favored
technology for manufacturing highperformance bipolar transistors. In this process, the
firstlayer poly forms the base contact. The base contact poly serves as local interconnect
3
and as a diffusion source to form lowresistive extrinsic base. Like the base contact poly,
the secondlayer poly forms the emitter contact and serves as a diffusion source to form a
shallow emitter junction. A typical structure for a doublepoly selfaligned bipolar
transistors is shown in Fig. 11.
Figure 11 Crosssection of common advanced doubleÂ¬
poly selfaligned bipolar transistors
Using highly doped polySi as an emitter contact reduces process complexity and
maintains the separation between the metal contact and the very sensitive monocrystalline
silicon region. The separation prevents sintering effects and metal spikes [4], The polyÂ¬
emittercontact BJT also exhibits characteristics superior to its metalcontact counterpart.
First, a shallow base junction can be formed with sufficient control on shallow emitter
junction formation. The shallow base reduces base transit time and improves speed
performance of the transistor. Secondly, bipolar transistors with poly emitter contacts
4
usually have better emitter efficiency than bipolar transistors with metal contacts, because
of the lower carrier mobility in poly silicon and the interfacial oxide layer interposed
between active emitter and poly emitter. As the thickness of emitter continues to decrease
for shallower base junctions, conventional metalcontact transistors would suffer a serious
currentgain reduction. This is because the emitter becomes transparent to the injected
minority carriers from the base, when its thickness is less than the characteristic diffusion
length of minority carriers. Finally, the poly emitter serves as a gettering source for metal
impurities [4],
With doublepoly selfaligned technology, BJTâ€™s lateral dimensions are also
significantly reduced. This improves BJTâ€™s speed performance in several aspects. First,
emitter width is narrower than the minimum linewidth given by lithography. The
reduction in emitter width reduces the intrinsic base resistance, thus easing the dc current
crowding effect for BJT operating at high current densities. Secondly, the distance
between the edges of emitter and base contacts, which is usually determined by the width
of a spacer, is reduced so that BJTâ€™s extrinsic base resistance is reduced. Finally, as the
extrinsic base area is reduced, the overall basecollector junction capacitance is reduced.
The reduction in both extrinsic base resistance and basecollector junction capacitance
serve to improve speed performance and power gain.
Reducing base thickness, which decreases base transit time, can further enhance
speed performance. However, reducing base thickness always conflicts with the need to
minimize intrinsic base resistance, which is essential in optimizing the BJTâ€™s power gain.
In addition, reducing base thickness also reduces the voltage at which transistors break
down because of base punchthrough. A possible way to compromise speed, power gain,
5
and breakdown voltage in designing a BJT is to increase the transistorâ€™s base doping.
However, increasing base doping sacrifices transistorâ€™s current gain.
Using SiGe epitaxial film for the transistor base in siliconbased process is a cost
effective solution to the above dilemma. Because the ratio between transistorâ€™s base and
emitter current injections is exponentially amplified by the bandgap difference at the
emitterbase junction, a SiGe heterojunction bipolar transistor (HBT) can have 10 to 100
times the current gain of a homojunction silicon bipolar transistor. This extra current gain
can be traded for the speed performance by optimizing base doping and thickness. In
addition, as the bandgap of SiGe can be manipulated by varying the Ge concentration in
the Si base, a builtin electric field can be created to assist minority transport in base,
which further enhance transistorâ€™s speed performance. Recently, SiGe HBTs with 75 GHz
unity gain frequency have been reported.
Even for purely silicon technology, aggressively scaled doublepoly selfaligned
bipolar transistors now can reach unitygain frequencies as high as 30 GHz [5], thus being
suitable for very highspeed digital and analog applications. A decade ago these highÂ¬
speed applications were reserved for only compound semiconductor devices. However, the
device models, which are essential to simulation, design and optimization of bipolar
devices and circuits, have not kept pace with the tremendous progress achieved in process
technology.
In designing Bipolar/BiCMOS devices and circuits, technology CAD (TCAD),
which explores and evaluates various design tradeoff without timeconsuming and costly
device fabrication, is widely used in todayâ€™s stateoftheart design process. In the past,
technology design was mostly based on experimentbased (i.e. trialanderror)
6
optimizations. However, as the costs and duration of development increase with the
complexities of todayâ€™s Bipolar/BiCMOS technology, such a process is slow and
expensive. TCAD predicts and optimizes the electrical performance of VLSI circuits
without using expensive fabrications. TCAD reduces development costs, shortens
development times, and helps maintain process control in design and production
environments.
In conventional TCAD strategy, the output of a process simulator is applied as
input to a device simulator, whose output in turn is used in extracting device model
parameters for circuit simulations. The computational cost of the device simulation in this
approach is very high, especially considering the inevitable uncertainty in doping profiles,
even when they are tuned using SIMS other measurements.
MMSPICE developed in University of Florida is a mixedmode device/circuit
simulator to facilitate TCAD design processes. MMSPICE merges device physics into
circuit simulations, therefore reducing the computation cost. Its chargebased bipolar
transistor model QBBJT is based on regional analysis of ambipolar transport equations,
subject to moving boundary conditions. The model is thus consistent with device physics
implied by the driftdiffusion relations used in most device simulators, but runs much
faster. QBBJTâ€™s model parameters are mostly derived from deviceâ€™s layout and doping
profile. Therefore, correlations between model parameters and process information are
automatically maintained. However, the use of MMSPICE/QBBJT model has been rather
limited. This is because that the MMSPICE/QBBJT model lacks a welldefined parameter
extraction methodology. In practice, the MMSPICE/QBBJT model is used in a variety of
modes. In concurrent engineering applications, QBBJT takes parameter derived from
7
simulated doping profiles and runs at predictive mode. In this way, QBBJT enables circuit
design before the process is finalized; the design cycle is significantly shortened. The
ability to predict performance from device structure is very useful in the early stages of
product design. However, once the technology is mature and test devices are available,
simulation accuracy and efficiency become the key issues.
The dissertation presents a parameter extraction methodology that allows QBBJT
model parameters to be calibrated in accordance with measured data, enabling accurate
simulations even when the process information is incomplete or inaccurate. This ability to
use model parameters derived either from process/structural information or from electrical
measurements makes the QBBJT model uniquely flexible. Now a single compact model
can be used in either a TCAD or bestfittodata mode. As the technology matures, only
simple revisions to the processderived parameters are need for transparent transitions
between modes.
The dissertation also addresses several refinements to the MMSPICE/QBBJT
model. First, selfheating is implemented in the QBBJT model by adding an extra thermal
node to QBBJTâ€™s admittance matrix, which is used in SPICEâ€™s modified nodal analysis.
Selfheating is inherent to all semiconductor devices. Selfheating is a significant effect
even for bulksilicon bipolar transistors operating at modest power dissipation. Ignoring
selfheating in the BJT model can cause problems in simulating analog circuits because
BJTâ€™s output conductance is either overestimated for homojunction silicon bipolar
transistors or underestimated for some compound heterojunction bipolar transistors.
DC current crowding is an important twodimensional effect. The original QBBJT
model account for current crowding around the emitter periphery by using the effective
baseemitter voltage, which can significantly underestimate BJTâ€™s high injection effects.
In this dissertation, we present an alternative way to model current crowding by using an
effective emitter area. The approach is consistent with the original dc current crowding
analysis, but maintains correct high injection analysis.
AC current crowding effect is a relatively new topic in compact bipolar transistor
modeling. While dc current crowding has been studied for over 30 years, ac current
crowding has received relatively little attention. AC current crowding is a nonquasistatic
(NQS) effect, and is often overlooked in the BJTâ€™s smallsignal equivalent circuit. This is
because BJTâ€™s smallsignal equivalent circuits are usually derived using quasistatic (QS)
approximations. As the total base charge storage is modeled by a lumped capacitor, the
time required for the majority charges in the base to redistribute laterally along the base is
overlooked. The QS approximations is equivalent to assuming that the carriers can travel
with infinite velocity in certain regions. This dissertation presents a new compact model to
account the NQS ac current crowding. This model is consistent with QBBJTâ€™s dc current
crowding analysis, and is easily realized in most compact bipolar transistor model.
As bipolar transistors continues to be scaled down for the demand of faster
operation and higher integration, substrate resistance becomes more and more influential
to the overall performance of bipolar transistors. For most compact BJT models, substrate
resistance is not included by default. A common approach to model substrate resistance is
to optimize the collectortosubstrate capacitance to partially account for its effect.
However, the approach can cause numerous difficulties in statistical modeling and circuit
design, because BJTâ€™s substrate resistance is layout dependent. In this dissertation, a
systematic methodology to include substrate resistance in SPICE BJT simulation is
9
presented. With the methodology, BJTâ€™s substrate resistance is consistently included in
BJTâ€™s parameter extraction and circuit simulations.
The dissertation is organized as follows: Chapter 2 reviews several classic and
novel compact models. At the end of Chapter 2, QBBJTâ€™s model formulation is briefly
discussed to introduce several concepts in the following chapters. Chapter 3 addresses
several additions and refinements of QBBJT. Chapter 4 presents a complete parameter
methodology for QBBJT. Chapter 5 presents a systematic methodology to include
substrate resistance in BJT parameter extraction and circuit simulation. Chapter 6
concludes the dissertation with a summary and suggested future work.
CHAPTER 2
BACKGROUND, PHYSICS, AND MODEL FORMULATION
OF THE MMSPICE CHARGEBASED BIPOLAR
TRANSISTOR MODEL  QBBJT
This chapter discusses several classic bipolar transistor models and a number of
novel models designed to accommodate contemporary technology breakthroughs. This
chapter provides concise comparisons between these models. The model formulation of
the MMSPICE/QBBJT model is briefly reviewed at the end of this chapter to introduce
several concepts elaborated in following chapters.
2.1 Background Review
2.1,1 EbersMoll Model
The EbersMoll model was an empirical model invented in early 50s [6]. The
original EbersMoll model is only a dc large signal model which contains two diodes
connected backtoback with a controlled current source in parallel with each of them. The
controlled current sources represent current diffusing from emitter to collector in forward
operation and from collector to emitter in reverse operation. The expanded EbersMoll
model has extra elements added to its equivalent circuit to account for resistive and
capacitive parasitics. The addition of fixed terminal resistors allows the EbersMoll model
the ability to account for nonideal ohmic voltage drop across resistive parasitics. By
including several lumped capacitors, the expanded EbersMoll model can account for
10
11
charge storage in the quasineutral base and depletion regions. This enables ac simulations
of bipolar transistors. With the addition of several empirical parameters, the expanded
EbersMoll model accounts for several secondary effects. For instance, the expanded
EbersMoll model uses forward and reverse Early voltages to account for basewidth
modulation and includes temperature coefficients for several parameters [7]. However, the
EbersMoll model cannot sufficiently account for high current effects.
2.1.2 GummelPoon Model
As a physical supplement to EbersMoll model, the GummelPoon model was
formulated based on the onedimensional integral chargecontrol relation (ICCR) [8].
Many of the GummelPoon modelâ€™s parameters are bias dependent. The GummelPoon
model includes several empirical parameters, such as the knee current IK, to account for
high injection effects. Even though the GummelPoon model is the most popular model
employed in a wide variety of circuit simulators, it has remained mostly unchanged for 20
years and is insufficient to model the advanced bipolar transistors.
The conventional GummelPoon model has the following major deficiencies: 1) It
models basewidth modulation (Early) effects using constant Early voltages (VAF and
VAR), which results in unacceptable errors in simulating output conductances for BJTs
with thin bases. 2) It uses constant collector resistance, thereby overlooking collector
modulation and quasisaturation effects. 4) Its smallsignal equivalent circuit is derived
based on quasistatic (QS) principles. It thus fails to account for nonquasistatic (NQS)
effects in transistors operating at very high frequencies. 5) It fails to account for some
12
extrinsic parasitics, which are negligible in large devices, but important for scaled devices,
e.g., the baseemitter sidewall capacitances. 6) It is not scalable with geometry.
2.2 Novel Compact Models for Bipolar Transistors
A predictive model for scaled bipolar transistors must account for (1) minority
carrier transport in quasineutral base region for typical base doping profile and all
injection levels; (2) majority carrier transport in quasineutral collector accounting for
nonlinear voltage drops across the epicollector and for quasisaturation; (3) forward and
reverse basewidth modulation (Early) effects; (4) base widening (Kirk) effects; (5) impact
ionization with consideration of nonlocal carrier energy transport relations; (6) two
dimensional geometric effects, such as sidewall current injection and currentcrowding
effects; (7) physical and continuously differentiable ac differential resistances and
capacitances [9]; and (8) global and local thermal effects, i.e., selfheating. Many compact
physical models based on these requirements have been proposed in the past ten years.
The extended GummelPoon model [10], VBIC95 [11], MEXTRAM [12] and the
MMSP1CE chargebased bipolar transistor model (QBBJT) [14] are reviewed in the
following sections.
2.2.1 The Extended GummelPoon Model
Using Kullâ€™s majority carrier transport equation in the quasineutral collector [10],
the extended GummelPoon model accounts for quasisaturation effect by calculating
nonlinear voltage drop across the quasineutral collector. The extended GummelPoon
model is based on the formulation of the original GummelPoon model and inherits most
13
of its parameters. The extended GummelPoon model is easily implemented and is
familiar to most circuit designers. Consequently, the extended GummelPoon model has
been included in a wide variety of circuit simulators.
To include Kullâ€™s collector transport equation in the extended GummelPoon
model, one artificial node is introduced at the basecollector junction. While the original
integral chargecontrolled relation (ICCR) is employed to simulate the minority carrier
transport in the quasineutral base, Kullâ€™s collector current equation is applied in the quasiÂ¬
neutral collector. The collector current is then solved by forcing current continuity at the
artificial node. The extended GummelPoon model models quasisaturation as collector
resistance modulation, thus reducing the empiricism of using constant collector resistance.
However, the extended GummelPoon model cannot properly simulate the IV
characteristics of all possible operating regimes, because it assumes that the entire epi
collector is quasineutral. For example, the extended GummelPoon model cannot
physically simulate currentvoltage characteristics of BJT operating in the nonohmic
quasisaturation regime where an excesscarrier induced SCR forms near the highly doped
buried layer due to carrier velocity saturation at high collector current density.
2.2.2 MEXTRAM
MEXTRAM model was developed by de Graaff and Kloosterman in Phillips
Research Lab. MEXTRAM differs from the original GummelPoon model in a number of
aspects [12]: (1) MEXTRAM does not use the integral charge control relation (ICCR). It
models minority carrier transport and charge storage for an exponentially doped base,
which is common in advanced bipolar transistors. (2) The terminal currents and charge
14
storages of MEXTRAM model are formulated as functions of the doping profile and bias
condition. (3) Basewidth modulation (i.e., Early effect) is not modeled with the constant
Early voltages, but with the baseemitter and basecollector junction depletion charges.
The latest version of the MEXTRAM model also adds a unified majority carrier
transport equation in the collector accounting for various conditions of quasisaturation
[13]. The MEXTRAM model, used in the circuit simulator Pstar (Phillips) and Microwave
Design System (HewlettPackard), has many advantages over the GummelPoon model
for simulating BJTâ€™s quasisaturation and high current characteristics.
2.2.3 MMSPICE/OBBJT Model
The MMSP1CE/QBBJT model was developed by Jeong and Fossum [14] at
University of Florida. While QBBJT was being first implemented in SPICE2G.6, QBBJT
is now available in at least one commercial SPICE simulator. QBBJT is similar to
MEXTRAM in many aspects. Both QBBJT and MEXTRAM use de Graaffâ€™s equation to
model minority carrier transport in the quasineutral base [15] and Kullâ€™s equation to
model majority earner transport in the quasineutral collector [10]. QBBJT, however, has
several unique features. QBBJT uses moving boundary conditions for different bias
conditions, thus including many physical effects such as the basewidth modulation
(Early) effects and the base widening Early (Kirk) effect. QBBJT also allows for possible
SCR formation in epicollector, thus eliminating the nonphysical assumption used in the
extended GummelPoon model. QBBJT physically simulates BJTâ€™s currentvoltage
characteristics for all possible operating regions.
15
Most of QBBJTâ€™s parameters are derived from device structure (layouts) and
process information (doping profiles). This makes QBBJT suitable for TCAD and
concurrent engineering applications, where a simple preprocessor (SUMM) is used to
generate QBBJTâ€™s parameters directly from the process simulator output or from
measured doping profiles. QBBJT predicts circuit performance even in the absence of
fabricated devices. Therefore, QBBJT can be used in concurrent engineering applications,
which efficiently shortens production cycles. With parameters derived from the estimation
of process information, circuit designs can begin before representative test chips are
available.
QBBJT is also suited for statistical circuit design applications. Since most of
QBBJTâ€™s model parameters come directly from the device structure and process
information, statistical correlations between circuit performance and process variations
are automatically maintained. Such correlations are not usually maintained in empirical
models such as GummelPoon. MMSP1CE is much more computationally efficient than
device simulators. This makes accurate statistical analyses, that require many simulations
(such as Monte Carlo), practical.
QBBJT is chargebased. Instead of using lumped capacitive elements and transit
time constants, regional charge storages are formulated in accordance with bias conditions
and device structural information. The regional charge storage provides nonreciprocal
transcapacitances.
The QBBJT model was made more realistic and predictive for simulating
advanced scaled BJTs by including nonlocal modeling of impact ionization and nonideal
twodimensional effects [16]. High electric fields and field gradients are common in BJTs
16
with aggressively scaled vertical dimensions. As electric fields change rapidly over
distances comparable to the energyrelaxation meanfreepath, nonlocal effects such as
carrier velocity overshoot have been shown to influence characteristics of BJTs
significantly. Current crowding effects, caused by laterally distributed base resistance, is a
prominent twodimensional effect. Current crowding degrades BJT current gain and unity
gain frequency.
The QBBJT model was extended for SiGe HBT simulations by including bandgap
variation parameters [17]. With the operation defined physically for the entire operating
range in a unified and systematic way, the numerical robustness of QBBJT has been
greatly enhanced. However, since QBBJT uses different model equations for different
operating regions, first or higher order derivatives of currents and regional charges can still
be discontinuous with improper model parameters. These discontinuities can cause
numerical problems and errors in simulating analog circuits.
2.2.4 Vertical Bipolar Inter Company Model ÃVBIC95)
VBIC95 was proposed in 1995 by a group of IC and CAD industrial
representatives as a replacement for the conventional GummelPoon model. The group
evaluated compact BJT models developed since GummelPoon model. Their objective
was to pinpoint the best model and bring it to the mainstream of bipolar IC designs.
However, no models they evaluated satisfied all the requirements from the committee, so
they decided to develop their own.
VBICâ€™s formulation is similar to the extended GummelPoon model, although,
with several unique features: 1) VBIC uses baseemitter and basecollector junction
17
depletion charges to model basewidth modulation effect, thereby eliminating the
empiricism of using constant Early voltages. 2) VBIC adds a parasitic pnp modeled with a
simplified GummelPoon model. 3) To avoid numerical problems, VBIC uses singleÂ¬
piece, smooth functions whenever possible to model BJTâ€™s currents and smallsignal
capacitances. 4) VBIC includes selfheating.
Table 21 summarizes the above discussion contrasting different models. Notice
that although some attributes can be included in almost all models, some attributes are
pertinent to only specific types of models, i.e., empirical or physical models. This is
because that they are just formulated differently. For example, an empirical model is
usually continuously differentiable. However, its parameters are usually hardly scalable.
Also, a physical model can include as many physical effects as possible. However, it is
usually more computationally intensive than an empirical model.
2.3 Model Formulation of OBBJT
2.3.1 BJT Operation Modes
Before QBBJTâ€™s model formulation is reviewed, all possible operation modes of a
BJT must be defined. Depending on whether a spacechargeregion (SCR) is formed at
basecollector metallurgical junction, BJT operation is divided into two distinct regions:
saturation and forward regions. Different saturation conditions can be defined based on the
velocity of majority carriers in epicollector. When the majority carriers in epicollector
move with a velocity proportional to the local electrical field, the BJT is operating in
ohmic quasisaturation region and the epicollector acts like a resistor. When the majority
carriers reach their saturation velocity, excess majority earners must be injected into the
18
Table 21 Comparison between compact BJT models
Extended
GP
MEX
TREM
QBBJT
VBIC
Base
analysis
Use GummelPoon ICCR
X
Use de Graaffs base
transport equation
X
X
X
Use Knee current to model
injection
X
X
Account for base pushout
X
X
Account for bandgap
variation in base. SiGe HBT
simulations.
X
Early
effect
Use Early voltages
X
Use depletion charges
X
X
Use moving boundaries
X
Collector
analysis
Use Kullâ€™s equation to model
quasisaturation
X
X
X
X
Account for various SCR
formations in epicollector
X
X
Account for field dependent
mobility
X
X
AC
analysis
Smallsignal equivalent
circuit derived with quasiÂ¬
static (QS) approximation
X
X
X
X
Bias dependent smallsignal
differential resistances and
capacitances
X
X
X
X
Account for vertical
nonquasistatic (NQS)
effects
X
X
Account for lateral nonquasi
static (NQS) effects
X
19
Table 21 â€” continued
Other
effects
DC current crowding
X
Transient current crowding
X
Selfheating
X
X
Collector spreading effect
X
Two dimensional geometric
(layout) effects
X
General
attributes
Continuous smooth terminal
currents
X
X
Continuous smooth
differential resistances and
resistances
X
X
Scalable parameters
X
Numerical robustness
X
X
20
quasineutral collector, which induce a spacecharged region (SCR) near the buried layer.
In this situation, BJT is operating in nonohmic quasisaturation region.
Different operating modes can also be defined for different bias conditions on an
Ic'VBc plane as shown in Fig. 21, where all the boundary lines are defined as [17].
Different regions in Fig. 21 are defined as following: (I) saturation; (II) ohmic quasiÂ¬
saturation, where entire base and collector are quasineutral; (III) nonohmic quasiÂ¬
saturation, where excessive charge is injected from base region and an SCR forms at the
end of epicollector; (IV) forwardactive, where the basecollector junction is reverse
biased and an SCR forms at basecollector metallurgical junction; (V) forwardactive with
the entire epicollector depleted due to adequately high VCB.
In Fig. 21, the thick solid boundary lines labeled ISCC and ISCCWEPI define the
onset of SCR formation at basecollector junction. Above these two lines, no SCR is
formed at the basecollector junction and BJT is operating at either ohmic or nonohmic
quasisaturation. Below these two lines, an SCR formed at basecollector junction, and the
BJT enters forwardactive operation. The thick dashed boundary line,
lQ  q â– Ac â€¢ NEPI â– vsat, is usually defined as the onset of base pushout. However, for
collector currents greater than I0, the quasineutral base does not necessarily extend into
the epicollector. Actually, base pushout depends on the collector current level as well as
the basecollector bias voltage. I0 represents the maximum current that the epicollector
can carry before its majority carriers reach their saturation velocity, and above which extra
carders must be injected into quasineutral collector to sustain excess current.
21
Figure 21 Operation mode boundaries on IcVgc plane.
22
2.3.2 Formulation of DC CurrentVoltage Equations
The formulation of QBBJTâ€™s currentvoltage equations and charge storages is
based on de Graaffâ€™s equation to model minority carrier transport in the quasineutral base
[12] and KulTs equation to model majority carrier transport in the quasineutral collector
[10]. In deriving the minority carrier transport and charge storage equations, de Graaff
assumed quasineutrality for an exponential base doping profile as
na = nao' exp(qx/W5M), (21)
where Naq is the extrapolated peak base doping, q is the base doping grading factor, and
WBM is the base metallurgical width. To account for intended or unintended bandgap variÂ¬
ations which result in an aiding or retarding field in the quasineutral base, de Graaffâ€™s curÂ¬
rent equations are reformulated as
Jni = Jâ€žlf ~ 2nLR fÂ°r lÂ°w injectiÂ°n conditions (22)
JnH = J nHF  JnHR for high injection conditions (23)
where
, _ â€ž exP(n'wb) , N
JnLF4DnlT\ eXp(T]'Wb)
n(xc)
J"LR ^ qDnLV[ exp(y\'Wh)~ 1
(24)
(25)
exp(r\"Wh)
JnHF s 2(iDnH^"p{^w ) {n{XE)
(26)
KHR 
n(xr)
exp(r\ Wh)  1
(27)
and the base doping grading factor qâ€™ and qâ€ are defined as
23
tT = (11 +A E/kT)/WBM
(28)
tT = &E/2kTWBM
(29)
to account for bandgap variations. To ensure a smooth and continuous transition from low
injection to high injection, linear extrapolation is used to get a single analytic current
equation for all injection levels:
J _ JnLF ' Na(xe) + JnHF ' n(xE) JnLR ' NA^XC> + JnHR ' n(xC) ^ f)\
n NA(xE) + n{xE) Na{xc) + n{xc)
In the quasineutral collector, QBBJT uses an equation similar to Kullâ€™s
formulation for the collector majority carrier transport equation:
lc ~
qA\ynV tNep i
W
QNR
K^BC0)K^BCJ)ln
1 +KCÂ¥bco>
Li +kcvbci).
+
BCO
BCI
(211)
where
KmsfiSH5 â– (2i2>
and WqNR is the width of quasineutral collector.
QBBJT uses baseedge and collectoredge quasiFermi level separations at the
basecollector junction, and H'gQ, to link the base and collector transport analyses.
QBBJT then solves the collector current by forcing current continuity at the basecollector
junction. When VFBCI is greater than basecollector junction bander, Oc, the BJT operates
in saturation or quasisaturation modes and VFBCI=T'BCJ. No SCR exists between the base
and collector. On the other hand, BJT operates in the forwardactive mode, where NKgQ is
modeled to be equal to c and T'sq is modeled as
24
exp
Ã^bcA
(QVbc\
K kT
j = explyrrv
exp
q$>
c
kT
exp
f^BC
{ kT
[c
(213)
'CRIT
where Icrit denotes the critical boundary currents (Iscc or IsCRWF.Pl) separating the
quasisaturation and forwardactive operations as shown in Fig. 21.
QBBJT solves for the widths of baseemitter and basecollector SCRs to provide
moving boundary conditions and to account for basewidth modulation (i.e. Early effect)
in base analysis. The width of basecollector SCR is calculated based on the potential drop
across the basecollector SCR (i.e., 'Rbcf'^BCj) While taking into consideration the
background charge modulation induced by collector current injection, QBBJT calculates
the width of basecollector SCR using the depletion approximation. The baseemitter
SCR, which controls the reverse Early effect, is modeled for both forward and reverse
bias. With the boundary conditions defined physically, the QBBJT model properly links
base collector transport equations and ensures current continuity throughout all operating
regimes.
2.4 OBBJTâ€™s SmallSignal Equivalent Circuit
Before constructing the smallsignal equivalent circuit, charge storages must be
formulated. The formulation of excess charge stored in the quasineutral base is based on
de Graaffâ€™s work. With a linear interpolation for intermediate injection, excess charge
storage in the quasineutral base is formulated as:
Qnl
Na(xe)
QnB +
n(xE)
Na(xe) + n(xE)
â– Q
H
nB
(214)
where <2
L
nB
and Qâ€žb
are excess charge storages for low and high injection cases.
25
Excess charge stored in the emitter is formulated using the emitter minority transit
time TE, an empirical parameter defined as:
Qe = TEAEJE0[exp(VBE/VT)l]. (215)
Excess charge stored in the baseemitter depletion region is modeled as:
XE
Qje = qAA(x)dx (216)
0
where NA(x) is the exponentially graded base doping profile and xE is the baseside depleÂ¬
tion width of baseemitter junction, modeled for both forward and reverse bias.
Excess charge storage in the quasineutral collector is modeled according to the
electric field distribution shown as Fig. 22. In saturation and quasisaturation (case A of
Fig. 22), charge stored in the unmodulated region is zero because the electric field
gradient is zero. The excess charge stored in the collector is only the charge in the
modulated region, which is expressed as
2.2 2., 2 .2
Cj A V rrYl: C^Afl: /r\  'j\
Qqnr  7 [exp(l/sco)  exp(VBC/)]  â€” WqNR
1C yv EPI
In nonohmic quasisaturation (case B of Fig. 22) region, the excess charge in the
collector is formulated as the sum of charge stored in the modulated region as (217) and
in SCR at the end of the epicollector as
Qepi(SCR) = wscr'^nepi~^c^x>s^' (218)
where WSCR is calculated with consideration of the background charge modulation caused
by collector current injection.
26
Electrical field
Figure 22 Electrical field distribution of each operation
mode. A: saturation & ohmic quasisaturation. B: nonohmic
quasisaturation. C: forwardactive with epicollector not
fully depleted. D: forwardactive with entire epicollector
depleted.
27
In forwardactive mode (case C of Fig. 22), excess charge storage in the collector
is formulated as charge stored in basecollector depletion region
Qepi(SCC) = ^scc' (qNEpi Ic/^s) â€¢ (219)
where Wscc is also calculated with consideration of background charge modulation
caused by collector current injection.
In forwardactive mode with the entire epicollector depleted (case D of Fig. 22),
in addition to charge stored in the epicollector, which is modeled as (219) with WSCE
replaced by WEPI, excess charge in the collector must include the charge stored in the
buried layer as:
c A W fpr
Qbl =  VBC)   >c)  <22Â°)
While all regional charge storages are defined and formulated physically, next step
to formulate the smallsignal equivalent circuit is to partition these charge storages into
different terminals. Charge partition is a typical technique to allocate a charge storage in a
certain region into two different terminals so that the charge transport kinetics in the
region can be partially accounted for. Through two empirical parameters C,B and QnB
and Qqnr, which represent the excess charge storage in the quasineutral base and in the
extended base, are partitioned into the emitter and collector terminals.
The excess charge stored in the epicollector QEPI is modulated by the collector
current, which is a strong function of collector current. QEpÂ¡ is allocated into the emitter
E C
and collector terminals as QEPI = (W/x>s) â€¢ Ic and QEPI = (W/x>s) â– I0, where W is
the SCR width in the epicollector.
28
The resulting charge storages in baseemitter and basecollector terminals are
formulated as:
Qbe 
^B ' QnB + CC â€™ QqNR + Qe+ QePI + QjE
(221)
Qbc ~
(1  â€¢ QnB + ( 1 _ Cc) ' QqNR + QePI + QbL
(222)
where Qq^r is set to zero for case (C) and (D). QBL is set to zero for case (A), (B) and (C).
With all terminal charges defined for all bias conditions, the smallsignal
equivalent circuit is formulated as shown in Fig. 23. This smallsignal equivalent circuit
has several unique features: (1) Implementation of the smallsignal equivalent circuit is
easy, since all the regional charge storages are derived from the dc solution; (2) The
differential conductance (i.e., dl/dV terms) and capacitances (i.e., dQ/dV terms) are bias
dependent; (3) The transcapacitances are included explicitly; (4) The extra delay (phase
shift) caused by base widening is included implicitly with QqrÂ¡r modeled in quasineutral
collector; (5) The NQS carrier transport inside the quasineutral base is taken into
consideration through charge partitioning.
29
Figure 23 QBBJTâ€™s smallsignal equivalent circuit
CHAPTER 3
MODEL ENHANCEMENTS OF THE QBBJT MODEL
3.1 Selfheating Modeling
This chapter discusses several additions and refinements to the MMSPICE/QBBJT
model. Selfheating occurs in all semiconductor devices and circuits. Selfheating is an
increase of device local temperature caused by deviceâ€™s own power dissipation. As
packing density and power dissipation of VLSI circuits continue to increase with
aggressive scaling, selfheating can become a severe problem and must be accounted for
in simulations and circuit designs. Selfheating is often observed in MOSFETs as a drain
current reduction, because the channel carrier mobility is degraded by the increased lattice
scattering [18]. While decreasing the drain current in MOSFETs, selfheating increases
the base and collector currents in bipolar transistors by enhancing the minority carrier
injection at the baseemitter junction. Selfheating increases BJT power dissipation, thus
causing positive feedback in BJT operation.
Since the mobility decreases slowly with temperature, the drain current reduction
in MOSFETs is not easily detected unless there is a large temperature change of tens of
degrees. Selfheating is especially important for MOSFETs built on SOI wafers, because
the thermal resistance of silicon dioxide is about two orders of magnitude greater than that
of silicon substrate. However, since the intrinsic earner concentration increases
30
31
exponentially with absolute temperature, significant base and collector current changes in
bipolar devices can be observed a small change in temperature.
Selfheating affects analog circuits significantly by increasing BJTâ€™s output
conductance. This can lead to a significant reduction in the voltage gain in highgain
amplifiers [19]. Circuits that depend on close matching of baseemitter voltages can be
very sensitive to selfheating, because small differences in junction temperatures can
cause significant mismatch in currents. Recent research demonstrates that neglecting selfÂ¬
heating can cause significant errors in parameter extraction [20]. Selfheating can cause
more than 50% difference in extracted values of BJT knee current. Therefore, proper
models should include selfheating.
While the current gain of homojunction BJTs has a positive temperature
coefficient, the current gain of heterojunction BJTs (HBT) can have either positive or
negative coefficient. The negative coefficient is a result of the bandgap difference at the
baseemitter junction [21]. Physical modeling of selfheating depends on accurate
calculation of the device local temperature. To simulate the local temperature, one thermal
node representing the temperature inside the bipolar transistor must be added into the
QBBJT model, making it a 5 node model as shown in Fig. 31. Thermal impedance is the
response of device temperature to variations in device power. With the thermal impedance
calculated theoretically as in [22] or evaluated as discussed in [23], the equivalent circuit
of Fig. 31 can be used for dc, ac and transient analyses.
In Fig. 31, the device local temperature is treated as an electrical voltage, the
power dissipation as a controlled current source, and then the thermal impedance as an
electrical impedance. The controlled current source Ipwr is calculated as
Figure 31 Schematic of thermal implementation
33
IpwR ~ Ic'Vce +IÂ¡3'Vbe The local temperature is then evaluated through
multiplying the power dissipation by the thermal impedance. In this way, selfheating
simulation can be merged into the electrical simulation. In the implementation of selfÂ¬
heating, an extra node T was added into QBBJT by simply modifying QBBJTâ€™s
admittance matrix in SPICEâ€™s modified nodal analysis. The implementation involves
minor modifications in 18 subroutines and addition of a new subroutine to update
temperaturedependent parameters in QBBJT model. Temperature dependent parameters
are updated based on the temperature obtained from the thermal node T in every Newton
Raphson iteration. The resulting operating point is, thus, consistent with deviceâ€™s local
temperature.
In this implementation, the ambient temerature is simulated as a voltage source
Vtamb 1Â° ^is way, we can simulate a BJT transistor at different ambient temperatures by
simply varying the voltage source VTamb. This implementation allows easier extraction of
the temperature dependence of BJT electrical characteristics than using the TEMP control
card. For instance, we can vary the ambient temperature from 73 Â°C to 27 Â°C with a step
of 5 Â°C using the DC sweep control as
.DC VTAMB 73 27 5
.PRINT DC I(VC)
In this way, we obtain directly the temperature dependence of the collector current. This
also adds the flexibility to vary ambient temperature in frequency domain or time domain
analyses. However, this implementation disables the original TEMP control card.
34
Figure 32 Output characteristic plot: Symbol lines denote
measurement data for BJT of 1.6pmx20im emitter from
Motorola 0.5mm BiCMOS process. Solid lines are the
MMSP1CE simulation results with selfheating. Dashed
lines are simulation without selfheating.
35
Fig. 32 shows a simulated output characteristic demonstrating selfheating in a
BJT. The thermal resistance is calculated with the method in [22]. The thermal resistance
can also be extracted using temperature dependent measurements [23].
3.2 DC Current Crowding Modeling
Current crowding caused by laterally distributed base sheet resistance is an
important 2D effect [24] [25] [26]. A half crosssection is shown in Fig. 33. As an ohmic
drop results from dc base current flowing across intrinsic base, baseemitter voltage VBE
can be nonuniformly distributed. This thus causes the emitter current injection in the
periphery of the baseemitter junction to be greater than that in the center. DC current
crowding is especially severe at high current levels, because the transverse ohmic voltage
drop is significantly higher at high current levels. Current crowding can be observed by
measuring the lowfrequency ac base resistance rb, which decreases with dc emitter
current IE [24], DC current crowding is usually associated with BJTs with large emitter
widths. However, base thicknesses are scaled down to reduce base transit time for better
speed performance, serious emitter current crowding occurs even for BJTâ€™s with moderate
emitter widths.
Current crowding was first investigated by Pritchard in late 50â€™s [24], By cascading
an infinite number of bipolar transistors with series base resistances, Pritchard derived the
equivalent base resistance as one third of total base resistance. However, this result is
invalid for the high basecurrent levels, which is typically involved in current crowding.
Another analysis of emitter current crowding was done by Hauser [26]. Hauser
manipulated the differential currentvoltage relations across the intrinsic base region. With
36
Emitter
Figure 33 Half crosssection of a vertical bipolar
transistors.
37
appropriate boundary conditions for the periphery and at the center, Hauser derived an
analytical solution for total base current. Hauserâ€™s method is useful, but inadequate for
advanced bipolar transistors, because it assumes that the base sheet resistance is constant,
thus overlooking the base majority conductivity modulation, which occurs in high
injection or base widening. QBBJTâ€™s dc current crowding model is mostly based on
Hauserâ€™s approach. However, QBBJT includes consideration of base widening and
majority carrier modulation. QBBJT calculates quasineutral base width and majority
carrier concentration for each dc bias point. The biasdependent base sheet resistance thus
eliminates the inadequacy of Hauserâ€™s model.
Current crowding often enhances high injection effects [31], because the peak
current density in the periphery is enhanced by current crowding. Emitter current
crowding is usually modeled with an effective base resistance in SPICE and effective
baseemitter voltage in MMSPICE. These approaches are equivalent to averaging the
nonuniformly distributed emitter current injection along the base, thus underestimating the
peak current injection in the periphery, as shown in Fig. 34.
To avoid the underestimation of the high injection effects, effective emitter area is
used to model dc current crowding instead of effective VBE or effective base resistance.
Effective emitter width was first derived by Hauser [26]. With the consideration of
conductivity modulation, the effective base width is derived similarly to [16] and [26],
This model is consistent with previous QBBJT dc current crowding analysis, but better
simulates the currentcrowding induced highinjection effects.
38
Position y (Â¡im)
Figure 34 Two different models to approximate
nonuniform current distribution along intrinsic base: (1)
effective baseemitter voltage VBF(efF) model, (2) effective
emitter area AE(etFl model.
39
In our current crowding analysis, the baseemitter junction is treated as an infinite
number of diodes and resistors cascaded in series as shown in Fig. 33. Therefore, the
baseemitter junction voltage as a function of position y, v(y), can be expressed as
v(y) = vbe f iB(0 ' P,
" O
where p is the specific base resistivity and is expressed as
P = WE/(2 â– \\,p â€¢ (Qbb + Qqnr)),
(31)
(32)
Qbb is the total charge of majority carrier in quasineutral base, and QqNR is the total
charge of majority carrier in extended quasineutral base. QBB can QqNR are calculated
from dc operating point, therefore making base resistivity biasdependent.
From Fig. 33, the base current as a function of position y can be expressed as
zb(t) ~ zfi(())'"J 2 â€¢ Le J EO(eff)' exp[â€œ^] 1
dÂ£,
(33)
where LE is the length of the emitter area and JEo(eff) is the effective saturation current
density of diode, which is the steadystate diode saturation current plus the transient term
obtained from previous timestep. Substituting (31) into (33) and differentiating the
resulting equation twice, we obtain
di
2_
dig p .
d7 + vrlB"dyâ€™
B
(34)
where we assume that exp(v(y)/Vr) Â» 1 . The general solution to (34) can be expressed
as
iB(y)  A â– tan
Ap_
2Vt
B
(35)
40
where A and B are two unknowns. At y=WE/2, iB(WE/2) should decrease to zero.
Therefore,
iB(WE/2) = 0 (36)
Substituting (36) into (35), we can obtain the unknown B, and (35) becomes
i'nOO = A  tan
z\\2y.
1
(37)
where z = (ApWE)/VT. Hence, the total base current at y=0 can be expressed as
iB(0) = A â€¢ tan(z) (38)
To solve the last independent unknown A, we can substitute (37) into the first two
coupled integral equations (31) and (33), and obtain
Â¡3(0) = LeWEJE0txp. (39)
By equating (38) and (39), we can solve A and the dependent variable z, thereby
obtaining the total base current for a given baseemitter voltage VBE.
Notice from (39) that current crowding makes the total base current smaller than
that expected from bipolar transistors with zero base spreading resistance. In the previous
version of QBBJT, dc current crowding is modeled with an effective baseemitter voltage
vBE(eff)â€™ that is
â– B(0) = lewejeozxâ– (310)
where
exP(VBE(eff)}  exP(VBÂ£) 2z ^ '
(311)
41
The result from the VBE(efÂ« approach is similar to the result using base resistance in
SPICE. Both underestimate the peak current injection level. A more physical model of
current crowding should evaluate effective emitter area, thereby maintaining current
density increase in the periphery due to the reduction in effective emitter area as Fig. 34.
To account for current crowding effect using effective emitter area, we inherit most
of the original model formulations. However, we use an effective emitter area AE/eff)
instead of effective baseemitter voltage VBE(et^ in (310). The resulting eifective emitter
area AE(eff) is
AE{eff)
a sin(2z) _
Â£ 2z ~
wele
sin(2z)
2 z
(312)
With Ae replaced by AE(efE) in the subroutine QBCT, QBBJTâ€™s current crowding
model is modified for both DC and transient analysis. Fig. 35 and Fig. 36 show
simulation results with the new current crowding modeling. The new model better
simulates the currentcrowding induced highinjection effect, i.e. highinjection rolloff, in
current gain (3 and unitycurrentgain frequency fT.
In the original development of MMSPICE, the use of effective emitter area was
thought of to cause instability. For this reason, transient simulations were performed on
different devices to check numerical stability of the new current crowding model. The
results show that the numerical stability of the new current crowding model is at least as
good as that of the old one. For large devices (WE > 1.0 pm), the new current crowding
model actually improves convergence stability.
42
VBE (V)
Figure 35 Current gain (3 vs. baseemitter voltage VBE.
From this figure we can see currentcrowding induced high
injection effect is better manifested using the new current
crowding model.
43
Figure 36 Unitycurrentgain frequency fT vs. base
emitter voltage VBE. From this figure we can see currentÂ¬
crowding induced highinjection effect is better manifested
using the new current crowding model.
44
3.3 AC Current Crowding Modeling
3.3.1 Simple Model Formulation
QBBJTâ€™s ac analysis does not include smallsignal intrinsic base resistance, when
dc current crowding analysis is active, which underestimates the total smallsignal base
impedance. To model smallsignal base resistance, we must account for distributed base
resistance as well as baseemitter capacitance (including both depletion and diffusion
capacitances), which can lead to significant ac current crowding. Consider the distributed
baseemitter RC network shown in Fig. 37. Such a network can be derived directly by
linearizing the cascaded dioderesistor network as shown in Fig. 33. As the smallsignal
admittance of the baseemitter capacitance increases with frequency, smallsignal base
current tends to gather near the periphery as shown in Fig. 38. The phenomenon of
nonuniform distribution of ac base current is called ac current crowding. As dc current
also tends to crowd around the periphery of baseemitter junction (e.g. edge current
crowding), the diffusion capacitance in the periphery is proportionally higher than that in
the center. Therefore, ac current crowding is more severe with significant dc current
crowding [27].
The phenomenon of dc current crowding was reported more than 30 years ago [26]
and has been widely studied [27], AC current crowding, however, has received much less
attention than its dc counterpart. AC current crowding is a nonquasistatic (NQS) effect.
The NQS effect is the timedependent behavior cannot be obtained by extrapolating from
steadystate analysis. In quasistatic (QS) chargebased modeling, different charging and
discharging dynamics are modeled with transit time constants, which are calculated from a
dc solution for excess charge storage. For instance, the base transit time can be calculated
45
Figure 37 Smallsignal equivalent circuit of the base
resistance plus baseemitter capacitance.
position y (pm)
Figure 38 Normalized smallsignal ac base current
distribution along the intrinsic base. Calculated for typical
BJT without significant dc current crowding.
46
as dQB/dIc, where QB is the excess charge storage in the quasineutral base and Ic is the
collector current. Actually, this base transit time is just the time required for the collector
current to deliver charge to the quasineutral base. This approach thus neglects the time
required for those charges to redistribute in the quasineural base.
The NQS charge redistribution time is especially important as the operating
frequency approaches the reciprocal of carrier transit times. Since the lateral dimension of
the base can be greater than its vertical dimension, the time required for the charge to
redistribute laterally can be larger than the base transit time, which is predominantly
controlled by the vertical dimension of the base. Therefore, the NQS effect caused by the
lateral redistribution time can be significant, even when the BJT is operating at frequencies
far below its unitygain frequency fT. As transistors are used at frequencies approaching
their fTs, the lateral NQS effect must be carefully accounted for.
A common way to account for NQS effects is to use charge partitioning, which
accounts for charge redistribution in a region by allocating the excess charge storage in a
region to two different terminals. Charge partitioning is often used to account for NQS
minority carrier transport in the quasineutral base, i.e. for vertical NQS effect [30],
Charge partitioning technique can also be adapted to account for majority charge
redistribution in the quasineutral base, i.e. for the lateral NQS effects. In most compact
BJT models, basecollector capacitance is partitioned into an intrinsic part and an extrinsic
part, where the charge partition factor, i.e. XCJC in the GummelPoon model, is obtained
from optimization of smallsignal sparameters. Actually, this is merely a way of using
charge partitioning to account for the lateral NQS effect occurring in the intrinsic base to
better fit the input smallsignal parameters, i.e. si 1 or si2.
47
To exactly represent the NQS charge redistribution time along the lateral base in ac
current modeling would require cascading an infinite number of RC networks in series
[28] [29]. Based on the series expansion of the input impedance of an RC transmission
line, Versleijen derived a simple RC network, which consists of a resistor R^ and
capacitor Cn/5 in parallel, to simulate the ac current crowding effect [28]. This approach is
simple and useful. However, this model is not sufficient at high base current, because it
does not account for the differential base resistance reduction caused by dc current
crowding. Also, it does not account for the nonuniformity in the baseemitter admittance
caused by dc current crowding.
The new approach described below is based on [28]. However, by careful
manipulation of the series expansion of the RC transmission line input impedance, the
new model considers the differential base resistance reduction. Also, with the effective
base length obtained in the dc current crowding analysis, the new model accounts for the
nonuniform distribution of baseemitter admittance.
Under linear and smallsignal conditions, the input impedance of a uniformly
distributed transmission line can be expressed as
cothVVÂ¡â€™
(313)
where Y% = l/RK + j(X>Cn is the total base admittance, and R/, is the intrinsic base
resistance. (313) is equivalent to the ac base impedance derived by Pritchard [24], Using
the series expansion of coth(x) =  +
i 3
1 x x
+ ..., (313) can be expressed as
x 3 45
48
tRâ€ž Ry (R~l + jÂ®CK)
(314)
3
45
Through simple algebraic manipulation, (314) can be further transformed as
(315)
Then the simplified equivalent circuit of the distributed RC base impedance is shown in
Fig. 39. The equivalent circuit of Fig. 39 is similar to that in [28], but contains an extra
component of SR^. At low current, the baseemitter resistance RK is usually much larger
than Rb so that this SR^ term can be neglected. However, since Rrt decreases with base
current, the magnitude of Rn can become much smaller than Rb at high current. Therefore,
5Rn should be kept in the equivalent circuit.
C*/5
If
Figure 39 Equivalent circuit of (315).
49
The 5RK term is to account for baseresistance reduction due to dc current
crowding at high currents. In the equivalent circuit of Fig. 39, Rb/3 is the lowcurrent dc
intrinsic base resistance, which was first derived by Pritchard in 1958 [24], However, at
high currents, dc equivalent base resistance should be lowered, because the effective base
width is reduced. By including 5Rn, the simple equivalent circuit of Fig. 39 accounts for
the dc baseresistance reduction due to dc current crowding at high currents.
The Cn/5 term is used to model the NQS ac current crowding. At low frequency,
the base impedance of the equivalent circuit of Fig. 39 is merely the resistance of the two
resistors (5Rn and R^/3) in parallel, because Cn/5 can be thought of as an open element.
This implies that no significant ac current crowding occurs at low frequency. However, the
effective base impedance becomes smaller at high frequency, because the ac current tends
to crowd around the periphery, thus narrowing the effective base. With CrJ5 which shunts
the effective dc base resistance (SRj,. and Rb/3 in parallel), the equivalent circuit of Fig. 3
9 properly simulates ac current crowding. However, the Cn/5 term should not be thought
of as a physical component, but as an empirical component to approximately simulate the
nonquasistatic (NQS) delay or phase shift.
3.3.2 Accounting for DC Current Crowding in AC Current Crowding
Analysis
In the above discussion, the impedance of the baseemitter pn junction was
assumed to be uniformly distributed; that is, the smallsignal baseemitter resistance and
capacitance, Rn and Cn, are constant along the lateral base dimension. This assumption is
valid only for low currents, where VBE is almost a constant along the lateral base; i.e.,
50
there is no significant dc current crowding. Since Rn and Cj,. are exponentially bias
dependent, they are nonuniformly distributed along the lateral base dimension, even for
very small variations in VBE due to dc current crowding. The net effect of the
nonuniformly distributed baseemitter impedance on ac current crowding can be described
as follows. Since the dc baseemitter voltage VBE at the periphery is higher than that at the
center, the smallsignal baseemitter impedance is smaller than that in center, thereby
causing more of the ac base current tends to flow through the periphery instead of the
center. Hence, dc current crowding enhances ac crowding.
It is not easy to model the nonuniformly distributed baseemitter impedance, since,
unlike the uniform transmission, no analytic solution exists. Nevertheless, little
modification in the solution of uniform transmission line is needed to approximate the
nonuniform transmission line for the purpose of compact modeling. For a physical
approximation, we first examine the net effect of dc crowding on ac current conduction.
The net effect of dc current crowding is to push ac current toward the periphery. This is
equivalent to compress the area where most of the ac base current flows. Hence, we
employ the concept of effective area, formulated for dc current crowding, to approximate
the nonuniform transmission line. To accommodate the effective area factor in the
previous ac current crowding analysis, we adjust the effective length of the RC
transmission line used in our previous ac current crowding analysis.
To verify the ac current model, we compare the solutions of a nonuniform
transmission line and a uniform transmission line with an effective length. The reason we
use this comparison as a preliminary verification is that the solutions of transmission lines
can better illustrate the concept of effective length than numerical device simulation, the
51
results of which are not always easily interpreted with regard to the intrinsic physics. We
will use numerical device simulation as a final check to the model implementation.
To derive the differential equation which controls currentvoltage relation of the
nonuniform transmission line, the smallsignal baseemitter admittance is modeled as
?*(y) = ynoexP[â€”^) > (316)
where y%  1 /Rn + j(i>CK, yno is a constant, and n is the injection ideality factor, n
equals to one at medium injection level. At high injection, n is between one and two.
However, for simplicity, we use n equal to one for all the injection levels.
To obtain the baseemitter voltage as a function of position, we can substitute (37)
into (31) and obtain
yBE(y) = VBE(0)2VT\n
cos [z( 1  2y/ VrÂ£)] 1
cos(z)
7
(317)
where z is defined in (37). Then using (317) in (316), the smallsignal baseemitter
impedance as a function of position is expressed as
y^y) = yj â– (sec[z(i2y/wE)]Y
(318)
where y izo = yreoexp[yBÂ£(0)/V7][cos(z)].
From Fig. 37, we can derive a pair of firstorder differential equations for the
smallsignal currentvoltage relation along the lateral base as
didy) 2
â€”= TjcCvMy) = yno' â– (sec[z(l 2y/WE)]} â– vh(y)
(319)
dvdy)
= ~rh â– hÃy)
dy
(320)
52
Substituting (319) into (320), we can obtain
,2
â€”vft(y) = rh â– yj â– {$ec[z.(l  2y/W E)]f â– vh(y). (321)
dy
The above second order differential equation is similar to that of a uniform
transmission line, but does not have an analytical solution because of the secant function.
(321) can be solved numerically using the boundary value method. The boundary
conditions of (321) are described as following: At y=0, vÂ¿(y)=l, if a unity ac voltage
source is applied. At y=WE/2, ih(y)=0.
Fig. 310 shows that the smallsignal voltage distribution along a nonuniform
transmission line can be modeled approximately by a uniform transmission line with an
effective length calculated from the dc current crowding analysis. Also, as shown in Fig.
311, the input admittance of a nonuniform transmission line as a function of frequency
can be simulated properly by using a uniform transmission line with an effective length.
3.3.3 Model Implementation and Verification
To include the ac current crowding model into MMSPICE/QBBJT model, we
implemented the equivalent circuit of Fig. 39 in QBBJTâ€™s ac admittance matrix in
SPICEâ€™s modified nodal analysis.
MEDICI was used to verify the ac current crowding model and its implementation
in MMSPICE/QBBJT model. In spite of being, strictly speaking, a quasistatic (QS)
simulator, MEDICI can be used to simulate the nonquasistatic (NQS) extra phaseshift
due to the charge redistribution in the device. The reasons are described as following: 1)
MEDICFs mesh is usually small enough that the time required for the charge to be
phase(vb(y))
53
Figure 310 Comparison between the numerical solution of
(321) and the analytical solution of uniform transmission
line. Symbol lines represent the exact solution. Thick lines
represent the analytic solution of the transmission line with
effective length. Thin lines represent the analytic solution of
the transmission line with normal length.
54
frequency (Hz)
Figure 311 The input admittance of the nonuniform
transmission line (symbol lines), the uniform transmission
line with effective length (thick lines), and the uniform
transmission line with normal length (thin lines).
55
redistributed in it can be assumed much smaller than any of deviceâ€™s characteristic transit
time constants; therefore, we can assume that charges are instantaneously redistributed
inside a mesh. 2) MEDICIâ€™s ac admittance matrix actually maintains all the interactions
between adjacent meshes. Therefore, MEDICI enables simulations of NQS phaseshift
due to the limited carrier transport velocity.
The MEDICIsimulated and MMSPICEsimulated base input impedances are
shown in Fig. 312. While the magnitude of the base input admittance is only slightly
influenced by ac current crowding, overlooking it causes significant errors in phase
predictions at high frequency.
3.3.4 Conclusion and Discussion
This chapter describes addition of selfheating to the MMSPICE/QBBJT model.
DC current crowding was modified to better simulate high injection effects. A compact
equivalent circuit model was developed to model NQS ac current crowding effect. The
model is consistent with the dc current crowding analysis and is easily implemented.
Using the effective base width calculated from the dc current crowding analysis, the ac
current crowding model can be extended to account for nonuniformly distributed base
emitter differential admittance along the lateral base. The resulting ac current crowding
model is verified against numerical solutions of a nonuniformly distributed transmission
line and MEDICIâ€™s simulations.
Notice that the NQS ac current crowding analysis may not be consistent with the
existing transient crowding analysis. The ac and transient current crowding models
emphasize different device physics for different applications. In fast switching (digital)
applications, even when dc current crowding is not significant, the transient base current
56
Frequency (GHz)
Figure 312 Comparison between the MEDICI simulated
base input admittance and the MMSPICE simulated base
input admittance with and without ac current crowding. The
symbol line represents the MEDICI simulation. The thick
solid line represents the MMSPICE simulation with ac
current crowding. The thin line represents the MMSPICE
simulation without ac current crowding.
57
can be very large causing severe transient crowding either in the periphery during turnon
or in the center during turnoff. Hence, in fast largesignal switching applications, the key
need is to properly simulate the transient crowding. For simplicity, QBBJT calculates
transient base current by differentiating majority charge storage based on the solution
from the previous time step. This is equivalent to treating the whole quasineutral base as a
single capacitor, ignoring the physics of charge redistribution along the lateral base
dimension. In ac analysis, the ac current is assumed small enough that the excess charge
distribution does not significantly deviate from its steadystate solution, so the primary
issue is to account for the small amount of charge redistribution along the lateral base
dimension. The focus of the NQS ac current crowding analysis discussed here is different
from that of the transient crowding analysis. Therefore, the NQS ac current crowding
analysis is not necessary to be consistent with the transient crowding analysis.
CHAPTER 4
PARAMETER EXTRACTION FOR MMSPICE/QBBJT
4.1 Introduction
QBBJT is implemented in at least one commercial SPICE version, and MMSPICE
is available from its authors. QBBJT contains many features not provided by other
available models, including accurate physical accounting for all quasisaturation
mechanisms, emitter current crowding, base pushout, and selfheating. In spite of these
advantages, use of QBBJT has been rather limited. One probable reason is that QBBJT
lacks a welldefined parameterextraction methodology. In practice, even if QBBJT can be
used with the model parameters derived from the device structures and process
information, a circuit designer would still prefer a model with its model parameters
calibrated for a given process to maximize the accuracy of circuit simulations. This
chapter presents a systematic parameter extraction methodology for the QBBJT model
parameters to be calibrated for a given process and used in its precision mode, even when
the process information is not complete. The parameter extraction methodology, thus,
successfully merges the QBBJT model into contemporary TCAD process and eliminates
the need of involving another compact model with welldefined parameter methodology
for precision purpose.
QBBJTâ€™s input parameter are derived from the collector and basedopant profiles.
The model parameter extraction can be used to investigate the device structure. Deviceâ€™s
58
59
structures are usually obtained with reversebiased differential CV method or SIMS.
Accuracy of the CV method is generally influenced by the junction abruptness and Debye
length [32], The applicable range of the CV method is limited by junction breakdown
voltages. SIMS is reasonably accurate. However, SIMS generally demands very large test
structures; the results thus represent predominant ID diffused dopant profile which may
not be applicable directly to practicalsize devices. For instance, because of the 2D
diffusion effects occurred in manufacturing processes, the emitterbase junction depth of
the BJTâ€™s from the same process can be significant deviated for different emitter feature
sizes, thereby influencing the peak base doping at the baseemitter junction. This effect
can be also observed by comparing the electrical characteristics of devices with different
emitter feature sizes. As shown in Fig. 41 and Fig. 42, the highinjection rolloff of the
current gain 3 and unitygain frequency fT occur at higher current density for narrower
emitter devices, because they have shallower emitterbase junctions due to the 2D
diffusion effects, and thus higher peak base doping. This is also another reason why we
need to calibrate the QBBJTâ€™s model parameters from electrical measurements for better
precision, although it is also possible to evaluate them from process information.
We have found that the profilerelated QBBJT parameters we extracted correlate
well with SIMS measurements. This suggests that when MMPSICE is used in this
extraction mode, it provides an inexpensive way to infer profiles from nondestructive
electrical measurements. Such an approach could be useful for statistical process control,
or for investigating an unfamiliar transistorâ€™s fabrication process.
Current gain (5
60
Figure 41 Current gain P vs. collector current density Jc
for different emitter sizes.
fT (GHz)
61
Figure 42 Unitycurrentgain frequency fT vs. collector
current density Jc for different emitter sizes.
62
The QBBJT model has 4 device line input parameters and 41 model card
parameters. Their physical meanings and typical values are listed in Table 41 and Table 4
2.Basically, the QBBJT input parameters can be grouped into the following categories:
1. geometric layout parameters,
2. material dependent carrier transport parameters,
3. GummelPoonlike parameters,
4. collector doping profile parameters,
5. base doping profile parameters,
6. charge partition factors.
Table 41 Device line input parameters
Name
Description
Units
Default
Typical
AE
Effective emitter area
m2
1.0e11
1.0e11
WE
Effective emitter width
m
1 .Oe6
1 .Oe6
AC
Effective collector area
m2
AE

ABL
Buried layer area
m2
AC
â€”
63
Table 42 Model card parameters
Name
Description
Units
Default
Typical
UNEPI
Electron mobility in epi
collector
cm2Â«V1Â«s'1
1.0e3
1.0e3
NEPI
Epi doping density
cm'3
1.0e16
1.0e16
WEPI
Epi collector width
m
5.0e7
5.0e7
WBM
Metallugical base width
m
2.0e7
2.0e7
NAO
Extrapolated peak base
doping
cm'3
1.0e18
1.0e18
ETA
Base doping gradient

3.0
3.0
Al
Preexponential coefficient
for impact ionization rate
cm'1
0.0
7.03e5
Bl
Exponential coefficient for
impact ionization rate
VÂ»cm"1
0.0
1.23e6
VS
Electron saturated drift
velocity
cmÂ»s'1
1.0e7
1.0e7
JEO
Emitter saturation current
density
A*m'2
1 .Oe8
1 .Oe8
JSEO
BE SCR saturation
current density
AÂ»m'2
1.0e4
1.0e4
NEB
BE SCR emission
coefficient

2.0
2.0
WSEO
Zerobias BE SCR width
m
5.Oe8
5.Oe8
PE
BE junction potential
barrier
V
1.0
1.0
ME
BE junction exponential
factor
â€
0.4
0.4
PC
BC junction potential
barrier
V
0.8
0.8
DNB
Electron diffusivity in base
cm2
10.0
10.0
PS
CS junction barrier
V
0.6
0.6
MS
CS junction exponential
factor
â€œ
0.4
0.4
CJS
Zerobias CS junction
capacitance
FÂ»cm'2
0.0
1 .Oe4
CRBI
Intrinsic base resistance
coefficient
V*s*cm'2
2.0e3
2.0e4
TB
Carrier lifetime in base
s
1 .Oe7
1,0e7
64
Table 42 Model card parameters (Continued)
TC
Carrier lifetime in collector
s
1.0e7
1.0e7
CIF
Forward current coupling
coefficient

1.0
1.0
CIR
Reverse current coupling
coefficient
â€œ
1.0
1.0
RC
Extrinsic collector
resistance
W
0.0
50
RB
Extrinsic base resistance
W
0.0
100
RE
Extrinsic emitter resistance
w
0.0
10
FB
Basecharge partition
factor

1.0
0.5
FC
Collectorcharge partition
factor
â€œ
1.0
0.5
TE
Emitter hole transit time
s
1 .Oe9
1.0e9
JEOP
Peripheral BE SCR
saturation current density
AÂ»m"1
0.0
2.0e11
NEBP
Peripheral BE SCR
emission coefficient

2.0
2.0
UPBASE
Hole mobility in intrinsic
base
cm2Â»Vâ€œ1Â»s'1
230
230
LBE
Width of intrinsic base
region
m
0.0
1.5e6
JCOP
Substrate saturation
current density
A*m'1
0.0
1.5e15
DNBH
Electron diffusivity in base
for high injection
cm2*s"1
10.0
10.0
DEGE
Bandgap reduction in base
emitter junction
eV
0.0
0.0
DEGC
Bandgap reduction in base
collector junction
eV
0.0
0.0
WEM
Emitter junction depth
m
0.0
0.0
DEGA
Average bandgap reduction
in emitter
eV
0.0
0.0
65
Our extraction strategy is described as follows. First, we assume that geometric
layout parameters such as the emitter and buried layer dimensions are known. The
materialdependent carrier transport parameters, such as majority carrier mobility in the
epicollector, are initially assumed to have their typical values. These values are later
refined in the extraction. Then, we extract the GummelPoonlike parameters using
existing techniques developed for the GummelPoon model. Although some of the
GummelPoonlike parameters are not accurate enough for ac simulation, they are
sufficient for the subsequent extraction procedures. With these GummelPoonlike
parameters, we then extract the collector and base doping profiles through optimization of
output characteristics and fT plot. Finally, we evaluate the charge partition factors and
calibrate the extrinsic resistance parameters in accordance with measured smallsignal
parameters.
To evaluate the charge partition factors and calibrate the extrinsic resistance
parameters, we need to extract QBBJTâ€™s smallsignal equivalent circuit. Generally, there
are two different approaches to extract BJTâ€™s smallsignal circuit. (1) Numerical
optimization, which obtains parameters through fitting the equivalent circuit to measured
sparameters. (2) Direct extraction, which obtains parameters step by step through
manipulating the smallsignal parameters (z, y, or h parameters) derived from the
equivalent circuit.
Optimization is a numerical technique used to determine one or more model
parameters from measurement. Optimization is especially useful in the fine tuning of
multiple parameters to obtain the best model fit to measurements. However, the
optimization of multiple parameters is usually not desirable for extracting parameters for a
66
physical model, because it often results in nonunique or nonphysical parameter sets when
given different initial guesses for parameters. There are several methods that prevent
optimization from generating nonphysical parameter sets: 1) providing closer initial
guesses based on dc parameter extraction; 2) imposing physical constraints; 3) using step
bystep deembedding techniques with special testing structures [33]; and 4) using
advanced optimization algorithms, such as the stimulated diffusion technique [34],
In contrast, direct extractions avoid generating nonunique parameter sets.
Therefore, a direct extract methodology is preferred in obtaining a unique set of
parameters for a physical model. A direct extraction method is also more computationally
efficient than global optimization. Based on the hparameters and zparameters derived
from BJTâ€™s hybridT equivalent circuit, some direct extraction methods have been
reported [35] [36], However, these methods are not useful to determine QBBJTâ€™s small
signal parameters, because they use the oversimplified hybridT model. The extraction of
smallsignal parameters is very sensitive to the topology of the equivalent circuit.
QBBJTâ€™s smallsignal equivalent circuit is based on the hybrid7T equivalent circuit
shown in Fig. 43. Recently, a methodology to extract hybrid7i equivalent circuit was
reported [37]. The method was also employed to investigate smallsignal base impedance
[27], The method is based on the fact that BJTâ€™s substrate parasitics are not significant at
low frequencies. Using this fact, a BJTâ€™s lowfrequency zparameters can be analytically
derived from the simplified lowfrequency equivalent circuit shown in the dashed box in
Fig. 43. Through manipulation of the analytic form of BJTâ€™s lowfrequency zparameters,
smallsignal resistances and capacitances such as rbb, ree, rcc and Ccc can be expressed
explicitly. However, [37] is not completly direct extraction method, because it does not
67
Figure 43 The QBBJTâ€™s smallsignal equivalent circuit
68
provide a direct way to extract or calculate BJTâ€™s baseemitter capacitance Cn. In addition,
this method requires a final optimization to obtain the optimal value for the basecollector
capacitance partition factor, which accounts for the NQS effects caused by the distributed
base impedance.
This chapter presents a complete parameter extraction methodology for the
MMSPICE/QBBJT model. With the extraction methodology, QBBJTâ€™s parameters can
now be evaluated either from process information or from measured data.
The extraction methodology also includes the extraction of BJTâ€™s smallsignal
hybridJi equivalent circuit from measured sparameters. With the extracted smallsignal
equivalent circuit, we can further calibrate the resistance parameters and evaluate the
charge partition parameters. The extraction of BJT smallsignal equivalent circuit
parameters uses no optimization of multiple parameters. Therefore, it ensures the
uniqueness of extracted parameters and maintains their physical meanings.
This chapter is organized as follows. Section 4.2 Section discusses the parameters
which can be extracted with the techniques developed for the GummelPoon model.
Section 4.3 discusses the extraction of QBBJTâ€™s physical parameters, such as the collector
and base doping profiles. Section discusses the extraction of QBBJTâ€™s smallsignal
equivalent circuit. Section 4.5 discusses remaining parameters and concludes this chapter
with a summary.
4.2 Extraction of GummelPoonLike Parameters
Some of the QBBJT parameters are analogous to similar parameters in the
GummelPoon model and can be extracted from a Gummel plot [7]. Saturation collector
69
current densities Jcs, JEO and JSEO are extracted from a linear fitting to the Gummel
plots in the medium and low current range. Jcs is not a model parameter. Jcs will be used
to determine DNB during the optimization of the collector doping parameters. Junction
capacitance parameters, such as barrier potential and nonideality factor, can be extracted
from reversebias CV measurements on largearea devices. These junction capacitances
will be used in the optimization of base doping parameters.
Parasitic resistances RC, RB and RE (collector, base and emitter resistances) are
required in subsequent extraction and optimizations of collector and base doping profiles.
Because QBBJT calculates the voltage drops across the epicollector and intrinsic base
(base region underneath the emitter), thereby including the biasdependent intrinsic
collector resistance and base resistance, RC and RB represent the extrinsic part of the total
collector and base resistances. The intrinsic collector resistance usually dominates the
total collector resistance in the forwardactive region, but it becomes negligible in hard
saturation. The extrinsic collector resistance can therefore be extracted from the slopes of
Iq vs. Vce curves in hard saturation (i. e., near Ic = 0) [7].
Emitter and base resistances can be extracted simultaneously using Ning and
Tangâ€™s method [38]. Assuming IB degrades only due to the VBE drop across the base and
emitter resistances, the potential drop can be expressed as
VT â– In{IB0/IB)/IC  RE + (RE + RB)/$ . (41)
Therefore, the intercept of the (VT â– \n(IB0/IB))/Ic vs. 1/(3 curve gives RE and the
slope gives (RE+RB). A typical base resistance vs. bias plot extracted using this method is
shown in Fig. 44. The intrinsic base resistance predominates at low injection, but it
becomes negligible at high currents because of base majority charge modulation and
70
Figure 44 Extracted base resistance vs. VBE plot
shrinking of the effective emitter width due to current crowding. Therefore, the extrinsic
base resistance can be estimated from the minimum (highinjection) value of the total base
resistance.
4.3 Extraction of Doping Profiles
4.3.1 Collector Doping Profile
Collector parameters WEPI (epilayer thickness) and NEPI (epilayer doping)
determine the epilayer resistance, which in turn controls the quasisaturation IV
characteristics. NEPI is directly related to basecollector breakdown voltage as well as the
threshold voltage of PMOS devices in n well BiCMOS technologies [27]. NEPI also
controls the onset of base pushout and the width of the basecollector SCR, and thus
71
strongly affects collector transit time. As base widths are progressively scaled down,
collector transit time becomes an increasingly significant part of the total transit time and
hence strongly affects highfrequency ac characteristics.
Since the electrical performance of a bipolar transistor operating in quasiÂ¬
saturation is primarily controlled by its collector structure, we can extract collector
parameters by optimizing them to fit the QBBJT model prediction to the quasisaturation
IV characteristics. Unlike most compact BJT models (for example, the extended
GummelPoon model) model the quasisaturation operation by assuming that the entire
epicollector is quasineutral, QBBJT accounts for all possible SCR formation in the epi
collector. QBBJT solves de Graaffâ€™s base transport equation [12] and Kullâ€™s collector
transport equation [10] subject to moving boundary conditions [17]. Therefore, QBBJT
properly simulates currentvoltage characteristics accurately in quasisaturation.
To improve the stability and convergence rate of the optimization, PC (BC
junction potential barrier) and REPÂ¡ (epilayer resistance) are used as the target variables
for the optimization instead of optimizing NEPI and WEPI directly. This preserves
correlations among the model parameters and helps to eliminate nonphysical parameter
combinations. Because the optimization of the collector parameters is fairly insensitive to
the base doping gradient, the base grading factor ETA can temporarily be assigned an
typical value. Assuming the base doping at the basecollector metallurgical junction
equals the collector doping, NEPI, WEPI, and NAO (extrapolated peak base doping) can
be related to PC and REPI as
(42)
PC = 2 â€¢ In (NEPI/nÂ¡)
RepÂ¡ = WEPI/(q â– UNEP I â– Ac â– NEPI)
(43)
72
NAO = NEPI â– exp(ETA) (44)
Calculating WBM (metallurgical base width) and DNB (average minority
diffusivity in base) requires a value for the total base doping, QB0. This parameter can be
extracted from the forward Early voltage if selfheating is eliminated [21]. The Early
voltage V4, defined as VA = (l/Ic)dIc/dVCB, can be expressed as [39]
Qbo Qje ~ Qjc
(45)
A eS/'^SCB +
where es is the silicon permittivity, WSCB and Wscc are the baseside and collectorside
depletion widths of the basecollector junction, and QJE and QJC are the depletion charges
of the emitter and collector junctions. Under lowinjection conditions (i. e., neglecting
background charge modulation induced by collector current) Wscc and WSCB can be
calculated using the depletion approximation. QJC can be calculated as q â– NEPI â– W scc â–
With the baseemitter depletion width estimated assuming a oneside abrupt junction, QJE
can be obtained by integrating the base dopant across the baseemitter depletion region.
Since VA is welldefined for every iteration value of QB0, Qbq can be obtained by
optimizing it to fit the VAVCB curve.
With Qb0 obtained from the previous step, WBM and DNB can be calculated as
WBM
Qbo ETA
NAO 1  exp {ETA)
JcS ' Ng
q â€¢ nie
(46)
DNB =
(47)
73
where NG is the Gummel number, defined as iVG = (Qbo~Qje~ Qjc^V Â» an<^ ^CS is
extracted from the Gummel plot as discussed in Section 2. With all the correlated
parameters recalculated for every iteration, the collector parameters can be obtained from
optimizing them to fit the model prediction to the measurement data. However, From (4
7), it is clear that variation of the bandgap can critically affect the calculated value of
DNB. For a contemporary bipolar transistor with a very thin and exponentially doped base,
the base doping at the emitter side can be as high as 1018 cm'3, causing significant
bandgap reduction. Overlooking such bandgap variation would lead to significant errors in
the calculation of DNB due to the exponential dependence of nie on the bandgap variation.
Extraction of the bandgap variation parameter is possible using temperaturedependent
measurement [40]. However, for simplicity we used the empirical model in [41] to
estimate the bandgap variation at the emitter edge of the base using the value of NAO for
every iteration.
To ensure the reasonable values of the bandgap variation parameters, the
intermediate values of DNB can be monitored to check the consistency. Based on the
estimated basedopant profile and the minority mobility model in [42], a reasonable range
of DNB (typically from 6 to 12 cm /sec.) can be obtained using Einsteinâ€™s relation. If the
intermediate values of DNB go out of the valid range, the bandgap variation parameters
can be finetuned in accordance with the calculation of [41]. The flowchart of the
optimization procedures of collector parameters is shown in Fig. 45.
74
Tune bandgap
variation
parameters
Extract QBO by fitting VAVCB curve
With Gummel number known,
Calculate WBM and DNB
Calculate next
PC and REPi
using Levenburg
Marquart algorithm
Invoke MMSPICE/QBBJT simulation
Calculate the sum of square errors, e
Figure 45 Flowchart of collector doping extraction
75
4.3.2 Base Doping Profile
Although total base doping QB0 can be obtained by fitting the forward Early
voltage VA as a function of VBC, the actual basedopant grading factor ETA cannot be
obtained precisely from dc measurements because dc currentvoltage characteristics
depend only weakly on basedopant grading. However, the base transit time xB depends
strongly on the basedopant grading because of the electric force it causes. QBBJT
physically accounts for the dependence of the base transit time on the builtin electrical
field, so ETA can be extracted by optimizing it to fit fT data.
The total forward transit time, xr = (xB + xc + xE) can be estimated from the
unity currentgain frequency using xT  1 /(2k â– fT). For narrow base widths, xc and xE
can become comparable to xB and must be carefully accounted for in order to correctly
predict fT.
In its regional charge analysis, QBBJT models xc as Qc/Ic, where
Qc  â€¢ (g./VÂ£p//(./uQ. Using NEPI obtained from the collectorprofile
extraction, Wscc can be calculated based on the depletion approximation, modified to
account for the background charge modulation induced by collector current, which allows
an accurate estimate of xc.
QBBJT uses the empirical parameter TE (emitter minoritycarrier transit time) to
account for xE. xE can be related to TE as xÂ£ = QEE/lc = (TE â– IB)/Ic~TE/$, where
Qee is the extra charge storage in the emitter. If the emitter junction is very shallow (< 0.1
pm) and has a metal contact, the extra charge storage in emitter is generally very small and
76
xE can often be neglected. However, if the emitter is thick and uses a poly silicon contact
which can accumulate a significant charge, xE can be comparable to xB and must be
calculated independently; if the detailed emitter structure is known, xE can be estimated
using the formula in [44].
ETA can be extracted by optimizing ETA to fit fT vs. VBE curves using the
previously extracted values of NEPI and WEPI. Since the quasineutral base can extend
into epicollector through either ohmic or nonohmic quasisaturation mechanisms, it is
important to select the right bias range for thtfT measurements used in extracting ETA.
VBE should be restricted to ensure that Ic is smaller than the threshold for base pushout at
Iq = qAx>sNEPI [39], and VCB must be kept high enough to avoid quasisaturation.
4.3.3 Integration of Doping Profile Extraction Procedures
The optimization procedures described above were implemented using the
Optimization Toolbox in MATLAB. A translation program was written in C that converts
MATLAB outputs to MMSPICE/QBBJT input parameters, invokes MMSPICE
simulations, and returns the simulation results to MATLAB. The MMSPICE code was
modified slightly to reduce the number of required file accesses. Also, a provision was
added to calculate fT directly from the regional charges in dc analysis rather than by
extrapolation using ac analyses.
The empirical parameters Jcs, JEO, JSEO, and the parasitic RE, RC and RB are
extracted from Gummel plots as described in Section II. Then, with a default value
assigned to ETA, the collector parameters are extracted by fitting to quasisaturation IV
77
curves. Next the emitter transit time xE is estimated using the formula in [44], and the
basedopant grading factor ETA is extracted from optimization offT vs. VBE curves as in
Section III. As a final step, the zerobias baseemitter SCR width WSEO is tuned to fit [3
vs. VBE curves.
4.3.4 Extraction Results and Discussion
The parameter extraction procedures are tested on doublepoly selfaligned BJT
devices from a Motorola 0.5jam BiCMOS process. Three sets of data are taken. They are
1) Gummel plots, e.g. log(Ic) and log(IB) vs. VBE with VCE as parameter, 2) Ic vs. VEE
plots with VBE as parameter, 3) fT vs. VBE plots. The dc measurements are done with
HP4145A semiconductor parameter analyzer. To avoid significant selfheating, the
forward Early voltage data, which are used to extract QB0 in optimization of collector
structural parameters, are evaluated at lower VBE. In our present case, the forward Early
voltages are extracted from Ic vs. VCE curve at VBE=0.8V. Because of their easy
calibration nature, sparameter measurements, which are used most widely for
characterizing smallsignal behaviors of devices, are employed to extract fT data. With the
Cascade air coplanner microwave probes, the sparameter measurements are done using
the HP8510A network analyzer set in conjunction with the HP4140 dc power supply
source. Since smallsignal current gain (3 can be evaluated from sparameter data as
P = ~s2t/[(l sn) â€¢ (1 + s22) + ^12%] [45], fT vs. VBE curve can be obtained from
extrapolating 3 vs. frequency plot to unity gain at each VBE.
78
In spite of taking 2D effects into consideration[16], e.g. sidewall currents and
extrinsic parasitic capacitances, the QBBJT model is virtually a 1D model. To minimize
the influence of 2D effects on parameters extraction, measurements are done on a
considerably large (1.6pm x 20pm) emitter device. Large devices also help to reduce the
importance of deembedding work in sparameter measurements. However, large devices
aggravate emitter current crowding. The current crowding effect causes devices to enter
high injection earlier by increasing current densities corresponding to reduced effective
emitter area. Therefore, the current crowding makes the (3 and fT vs. VBE curves rolloff
faster at high current density. Physically accounting for the current crowding as discussed
in Chapter 2, the QBBJT model eliminates the drawbacks of parameter extraction on large
devices.
The extracted collector parameters and basedopant grading factor were found to
agree closely with corresponding parameters extracted directly from SIMS profiles as
shown in Table 43. However, the extracted base width was 16% larger than that indicated
by SIMS, which led to about a 2:1 error in the peak doping NAO. This discrepancy can be
explained by noting that the SIMS measurements were performed on very large area
devices. Because of twodimensional diffusion effects, the emitter junction of small
devices is expected to be shallower than that of large devices. Therefore, the base of the
small device should be wider and contain more dopants than predicted by SIMS.
79
Table 43 Comparison between SIMS and extraction results
NEPI
WEPI
NAO
ETA/WBM
WBM
SIMS
4e16 
1e17
0.50pm
1.00e18
20.10 pm'1
0.140 pm
Extraction
7.4e16
0.47pm
2.12e18
20.36 pm'1
0.163pm
Difference%
â€”
6%
112%
1%
16.4%
The simulations with extracted parameters are shown in Fig. 46 through Fig. 49.
A new version of the QBBJT model which includes selfheating and modified current
crowding model is used to simulate device characteristics with the extracted parameters.
The thermal impedance is calculated theoretically as in [22] or evaluated as discussed in
[23]. Notice that the extracted physical parameters such as base and collector doping
profiles are also applicable to the BJTâ€™s of the same structure but with different emitter
feature sizes. For instance, NAO, WBM, ETA, NEPI, and WEPI extracted from the
1.6pmx20pm device can be used for the 0.8pmx20pm device without repeating the
complete extraction procedures. This suggests that the model is truly scalable and thus can
reduce the parameter extraction efforts significantly.
One bonus feature of the parameter extraction methodology is that with the
parameters calibrated with measurements, the firstoder derivative discontinuities in
QBBJTâ€™s collector current can be removed. Depending on the existence of basecollector
SCR, QBBJT uses two different sets of equations to calculate the collector current. This
can causes the firstorder derivative discontinuity in QBBJTâ€™s collector current, when
operation transits between quasisaturation and forwardactive modes, if given
80
(a) 1.6pmx20im
(b) 0.8pmx20gm
Figure 46 Gummel plot, log(lc) & log(lB) vs. VBE. SymÂ¬
bol lines are the measurement data. Solid lines are simulaÂ¬
tion results.
81
(a) 1.6jj.mx20im
(b) 0.8gmx20im
IC (A)
Figure 47 Current gain P vs. VBE. Symbol line is the meaÂ¬
surement data. Solid line is simulation with current crowdÂ¬
ing effect. Dashedline is simulation without current
crowding.
Ic (mA) lc (mA>
82
(a) 1.6(imx20jxm
(b) 0.8amx20pm
VCE (V)
Figure 48 Ic vs. VCE plot. Symbol lines are the measureÂ¬
ment data. Solid lines are simulation results with selfheatÂ¬
ing. Dashedline are simulation results without selfheating.
83
Figure 49 fT vs. VBE plot. Symbol line is the measurement
data. Solid line is simulation with current crowding effect.
Dashedline is simulation without current crowding.
84
inappropriate parameters. Because the collector parameters are obtained from
optimization, which in turn conform the model prediction to fit the measurement data, the
discontinuity problem can be avoided as shown in Fig. 410.
4.4 Extraction of OBBJTâ€™s SmallSignal Equivalent Circuit
4.4.1 Methodology
In previous sections, most of QBBJTâ€™s parameters were extracted from dc current
voltage characteristics, lowfrequency CV measurements and fT plots. The resulting
model parameters give very accurate predictions of BJTâ€™s dc characteristics. However,
some of these parameters might not be optimal for ac simulations, because isolation of
their individual effects from dc measurement is difficult. For instance, the base and emitter
resistances can not be accurately determined through dc measurements, because their
effects are coupled together at dc. Also, for accurate ac simulation, we still need to know
the charge partition factors, which are used in QBBJT to empirically account for the
nonquasistatic (NQS) carrier transport in the quasineutral base.
This section presents a new methodology to directly extract QBBJTâ€™s smallsignal
equivalent circuit, from which we can evaluate the optimal value of RE, RB, and charge
partition factors for accurate ac simulation. The extraction methodology begins with the
calculation of the differential current gain and baseemitter resistance from the dc current
voltage characteristics. At medium to highinjection levels, the base current is dominated
by the minority carrier diffusion current in the emitter, given as
^b ~ ^soexP( Vbe 4^r)â€™ (48)
where VBEâ€™ is internal bias voltage. Therefore, rrt can expressed as
Ic (mA), dIc/dV (mA/V)
85
Figure 410 Simulated Ic vs. VCE and its first derivative.
rK = 1 /{dIB/dVBE') = VT/IB (49)
Because the ideality (exponential) factor of collector current is not necessarily equal to
unity at high injection, we cannot directly obtain gm by direct differentiation of the
collector current Ic with respect to the baseemitter voltage VBEâ€™. Instead, we should use
(410)
where (3ac is lowfrequency commonemitter current gain, which can be obtained from
BJTâ€™s dc collector and base currents. In [37], gm is expressed as
Sm = Snro^H^d)
(411)
where the exponential term is used to account for the phaseshift caused by the nonquasiÂ¬
static (NQS) minority transport in the quasineutral base. Since such a NQS effect has
been accounted for in QBBJT with the charge partition technique, we use (410) instead of
(411).
Since contemporary silicon bipolar transistors are usually built on a very lightly
doped substrate or epilayer, the resulting substrate parasitics are insignificant at low
frequencies. With these substrate parasitics neglected at low frequency, BJT zparameters
without consideration of probe pad parasitics are expressed as
^11 1 bb ree ^
^12 ~ ree + + gmZn)
Z21 = ''Â«+[V(1+tÃ³)Kl4/)(ÃºCÂ«)
z22 = rcc + ree + (ZK+ 7Â®Ccc)/( 1 + gmZ%)
(415)
(413)
(414)
(412)
where Z% = rK/( 1 + ja>rnCK). Subtracting z12 from zÂ¡ Â¡ yields the base resistance rbb as
87
Subtracting z21 from z22 yields the basecollector capacitance Ccc as
Ccc = l/[co Im(z22z2l)]. (417)
To extract emitter resistance ree, we multiply the numerator and denominator of the
second term of (413) by gm and obtain
ree = R
where aac = fiac/(fiac + 1) is the lowfrequency commonbase current gain.
To compute the collector resistance rcc, we cannot directly subtract z21 from z22,
because z21 converted from sparameters is usually unreasonably large at low frequencies
[37]. To compute collector resistance rcc, we subtract z12 from z22 and obtain
rcc = Re
Zl2 Z{1 j
The above equation contains another unknown Cn. To calculate Cn, we use BJTâ€™s
first dominant pole extracted from the unitygain frequency fT as Pdominant^aA^Kfr)
1
1
(419)
From the hybrid7t equivalent circuit of Fig. 43, BJTâ€™s first dominant pole is expressed as
Pdominant = Cnrn + CccK + ( 1 + 8mr,zXree + rcc^ â€¢ (4â€˜20)
Notice that (419) and (420) are independent and contain only two unknowns rcc
and Cjj. Therefore, we can solve rcc and Cn from (419) and (420).
With rn, gm, rbb, ree, rcc, Ccc, and Cn calculated with (49), (410), and (416)(4
20), BJTâ€™s lowfrequency equivalent circuit (i.e. the circuit in the dashedbox in Fig. 43)
is directly extracted from lowfrequency zparameters. This equivalent circuit fits closely
to measured smallsignal parameters at low frequency. However, it does not accurately
simulate BJTâ€™s smallsignal parameters at high frequencies, because this equivalent circuit
does not take substrate parasitics into considerations. Therefore, we extract substrate
parasitics from the discrepancy between the simulated and measured data at high
frequencies.
The extraction of substrate parasitics begins with the calculation of z22 of the low
frequency equivalent circuit. Then the calculated z22 is compared with the measured z22.
From the difference, the substrate parasitics are extracted as
(421)
where z22â€™ is the simulated z22 of the lowfrequency equivalent circuit and z22â€ is the
measured z22.
In practical situations, only Rsub is extracted from this step. We use the Ccs
measured with lowfrequency CV measurements, and extract Rsub from fitting (421) to
the measured z22 from 100MHz to 7.5GHz. Since the optimization contains only one
variable, we ensure the uniqueness of Rsub, and maintain its physical meaning. The
extraction methodology of BJT substrate resistance will also be used in the next chapter to
verify the calculation of BJT substrate resistance.
4.4,2 Calibration of OBBJTâ€™s Extrinsic Resistance Parameters
The extraction methodology was tested on bipolar transistors of the 0.6im
BiCMOS process of Texas Instruments, Inc. Sparameters are used because they are easier
to measure at high frequencies than other smallsignal parameters such as z, y, or h
parameters. Many calibration and deembedding techniques are also welldeveloped for s
parameter measurements. A HP85 IOC network analyzer system was used to measure BJT
sparameters. These sparameters were taken from 100 MHz to 7.5 GHz for IB swept from
89
1 0.A to 1 mA and VCE from 1 V to 3 V. The resulting sparameters were deembedded
with measurements of open structures. Then, BJTâ€™s zparameters were converted from the
deembeded sparameters.
The baseemitter resistance rK and transconductance gm are calculated directly
from BJTâ€™s dc bias currents using (49) and (410). Therefore, the rK and gm are consistent
with the prediction of QBBJT using the parameters extracted from Chapter 4. Other
components are extracted with the procedures described previously.
Figure 411 shows the extraction of ree. QBBJT models the emitter resistance with
a constant parameter RE. RE can be extracted from NingTangâ€™s method or the floating
collector technique as discussed previously. However, the accuracy of RE extracted with
these dc methods is usually limited, because the effects of emitter resistance and base
resistance are coupled in dc measurements. Here, we provide an alternative way to
determine RE from ac measurements. Notice from Fig. 411 that the ac extracted RE
reaches a constant value when IB is greater than a threshold value. Therefore, we can
extract RE from the dashed line in Fig. 411.
Figure 412 illustrates the extraction of rcc. QBBJT calculates the intrinsic
collector resistance internally and leaves the extrinsic collector resistance RC as an input
parameter. RC can be extracted from the output characteristics of a BJT operating at
saturation region [7]. However, the extrinsic collector resistance RC extracted using this
method is usually dependent on IB, because the intrinsic collector resistance is not totally
negligible even if the BJT operates at saturation region. Here, we provide an alternative
way to extract RC for a BJT operating at forward active region. Notice from Fig. 412 that
the total collector resistance extracted from BJTâ€™s highfrequency sparameters is almost a
90
Figure 411 The extracted smallsignal emitter resistance
ree for different bias conditions.
Figure 412 The extracted smallsignal collector resistance
rcc for different bias conditions.
91
constant when IB is greater than a threshold value. When a BJT is operating at forward
active region, its intrinsic collector resistance can be approximated as the epicollector
resistance. Therefore, RC can be extracted as
RC = RC(tota[j Repi, (422)
where REPI = WEPI/(q â– UNEP I â– Ac â€¢ NEP1).
Figure 413 illustrates the extraction of total base resistance rbb. Notice that rbb
decreases with BJTâ€™s base current IB, and eventually becomes a constant at very high
current. This phenomenon is primarily due to the majority charge modulation and dc
current crowding. Because the effective base width decreases with base current and the
intrinsic resistance eventually becomes zero, the total base resistance approaches the value
of the extrinsic base resistance at high current. Therefore, we can extract the extrinsic base
resistance RB when the ac extracted total base resistance approaches a constant at high
current.
QBBJT calculates the intrinsic base resistance in two different ways. If the current
crowding analysis is on, the intrinsic base resistance is internally accounted for in the
current crowding analysis as discussed in Chapter 3. If the current crowding analysis is
off, the intrinsic base resistance is calculated as
n n
iXU (intrinsic)
CRBI WE WE
Q
(423)
BB
where WE is the emitter width, QBB is the total base majority charge, and CRBI is an
input parameter. In this case, the intrinsic base resistance is a constant for all bias
conditions. CRBI can be theoretically calculated as
CRBI =
n â– UPBASEâ€™
(424)
92
Figure 413 The extracted smallsignal base resistance rbb
for different bias conditions.
where UPBASE is the majority mobility in the base and n is equal to 3 for single base
contact configurations and 12 for double base contact configurations. With QBB calculated
from the total base doping for low and medium injection conditions, CRBI can also be
tuned to fit the ac extracted total base resistance.
4.4.3 Evaluation of OBBJTâ€™s Charge Partition Factors
Figure 414 shows the extraction of the baseemitter and basecollector
capacitances Cn and Ccc. Ccc is roughly a constant at low currents, but becomes
proportional to Cn for base currents greater than 100(iA. This phenomenon is caused by
nonquasistatic (NQS) minority transport in the quasineutral base. The derivation of the
equivalent circuit of Fig. 43 is mainly based on the quasistatic (QS) approximation.
93
Ib (A)
Figure 414 The extracted smallsignal baseemitter and
basecollector capacitances Cn and Ccc for different bias
conditions.
94
Therefore, the resulting components of the equivalent circuit will be influenced by NQS
effects, when the QS equivalent circuit is used to fit the highfrequency measurement. In
the present case, the basecollector capacitance Ccc is actually comprised of two terms.
The first term is the physical basecollector capacitance which can be derived from the
depletion charge storage at the basecollector junction. The second term is the NQS term,
which is modeled in QBBJT as part of the minority charge storage in the quasineutral
base. At low current, because the minority charge storage in the quasineutral base is very
small, the basecollector capacitance is primarily determined by its physical term.
However, at high current, the NQS term dominates the value of the extracted base
collector capacitance, because the minority charge storage in the quasineutral base
becomes much larger than that at low current. Therefore, the extracted basecollector
capacitance increases with the base current, when the base current is greater than a
threshold value.
QBBJT uses the base charge partition factor FB to account for the NQS effects in
the quasineutral base. At high current, Ccc/(Cn+Ccc) is approximately equal to FB.
Therefore, we can estimate the base charge partition factor FB as the dashed line shown in
Fig. 415.
QBBJT uses the collector charge partition factor FC to account for the NQS effects
in the extended quasineutral base, when base pushout occurs at very high current. Base
pushout rarely occur for contemporary highspeed BJTs, since their collectors are usually
doped heavily. Also, base pushout is usually not a desired BJT operation. However, if
base pushout does occur, FC can be extracted by optimizing the model prediction to the
extracted Ccc/(C7i:tCcc) at very high current.
95
Figure 415 Extract the charge partition factor by comparÂ¬
ing the Ccc and (Ccc+Cn) extracted at high current. The
dashed line approximates the charge partition factor FB
used in QBBJT to account for the NQS carrier transport in
quasineutral base.
Figure 416 shows simulated sparameters and sparameters measured from 100
MHz to 7.5 GHz. Fig. 417 shows the yparameter fitting of extracted equivalent circuit.
Fig. 418 shows simulated unitygain frequency fT and fT computed from measured s
parameter data. We can observe that simulations using the equivalent circuit extracted
with the extraction methodology compared very well with measurements.
4.5 Conclusion
The results of this chapter remove a key limitation of the MMSPICE QBBJT
model. QBBJT was developed to allow prediction of device/circuit performance from
96
Sil
901
S12
S21
9015
S22
Figure 416 The simulated sparameters compared with the
measured sparameters. Symbol lines represent measureÂ¬
ment. Solid lines represent simulation.
Mag(y22) Mag(y21) Mag(yl2) Mag(yll)
97
Figure 417 The simulated yparameters compared with
the measured yparameters. Symbol lines represent meaÂ¬
surement. Solid lines represent simulation.
98
Figure 418 The simulated unitygain frequency fT comÂ¬
pared with the fT computed from the measured sparameter
data. Symbol lines represent measurement. Solid lines repÂ¬
resent simulation.
99
knowledge of the device structure alone. QBBJT was designed for use in a predictive
mode for concurrent engineering applications. This dissertation shows how those physical
parameters can be calibrated with the electrical measurements for the representative
transistors of a given process. With these calibrated parameters, QBBJT can now be used
either in a predictive mode for preliminary circuit design or in a precision mode for final
circuit verification.
The methodology begins with the extraction of the GummelPoonlike parameters.
With these GummelPoonlike parameters extracted using the existing techniques, the
collector doping profile parameters are extracted from optimizing the output
characteristics and the base doping grading factor is extracted from optimizing fT vs. VBE
plots. By demonstrating that the extracted doping profile parameters are highly correlated
to the SIMS profiles, this dissertation suggests that the model can be used in a â€œparameter
extractionâ€ mode where the doping profile parameters are inferred by finding the best fit to
electrical measurements.
To further calibrate the extrinsic resistance parameters for accurate ac simulation, a
new methodology to extract QBBJTâ€™s smallsignal equivalent circuit is developed. The
method directly extracts BJTâ€™s hybrid7t equivalent circuit with using any optimization of
multiple parameters. Therefore, this method is more computationally efficient than any
global optimization techniques and ensures the uniqueness of the extracted smallsignal
equivalent circuit. With the extraction of QBBJTâ€™s smallsignal equivalent circuit, we
present a method to extract BJTâ€™s charge partition factors for the first time.
Up to now, we have developed a systematic methodology to extract all the QBBJT
parameters listed in Table 41 and Table 42, except AI, BI, CIF, CIR, JCOP, JEOP, NEBP,
100
and WEM. AI and BI are the preexponential and exponential coefficients of the BJT
impact ionization rate. AI and BI can be assigned with their default values or extracted
from the characteristics of a BJT operating at weak impact ionization region. CIF and CIR
are the factors that finetune the coupling of the highinjection and low injection
components in the forward and reverse collector current. CIF and CIR are usually not
sensitive parameters to QBBJT simulation. Therefore, CIF and CIR can also be assigned
with their default values. JCOP is the substrate saturation current density, which can be
extracted from measuring the forwardbiased substrate current. QBBJT uses JEOP, NEBP,
and WE to simulate the extra base current injected from the vertical sidewall. These three
parameters can be disabled in QBBJT simulation by setting JEOP to zero, since JEO
extracted from the total base current has included the sidewall injection term.
This extraction procedure is also adaptable for automated extraction. The
measurements required are similar to those needed for extracting GummelPoon model
parameters. The resulting fits to measurements are at least as close as one normally gets
using the GummelPoon model, and are typically much better in some regions, such as
quasisaturation. However, with the new extraction methods, the QBBJT model has
become much more flexible than GummelPoon. Whereas the GummelPoon parameters
can come only from electrical data, accurate QBBJT parameter sets can be found from
many combinations of electrical and physical data, exploiting all available data. Also, the
physical QBBJT parameters extracted from one transistor geometry apply naturally and
directly to other geometries. Furthermore, as we have seen, the extracted physical
parameters can be used to infer dopant profiles, which also suggests a variety of new
applications.
CHAPTER 5
SPICE MODELING OF BJT SUBSTRATE RESISTANCE
5.1 Introduction
The previous chapter described a direct extraction method to extract BJT small
signal equivalent circuit parameters including substrate resistance. The method generates
an excellent model fit to measured smallsignal parameters over a very wide frequency
range. However, the extraction must be repeated even for identical BJTs if they have
different substrate contact configurations. This chapter presents an efficient modeling
methodology for BJT substrate resistance, which is applicable to any BJT model,
including QBBJT as well as GummelPoon. The methodology eases the burden of
repetitive parameter extraction and eliminates the modeling uncertainties caused by
substrate resistance.
Substrate resistance is inherent to all semiconductor devices. Substrate resistance
causes debiasing effects in MOSFETs and influences the highfrequency ac characteristics
of bipolar transistors [48], As bipolar transistor speed performance continues to improve
due to scaling, more and more RF monolithic ICs use silicon bipolar transistors as low
cost substitutes for compound semiconductor devices. Substrate resistance can be
important in such highfrequency applications. Substrate resistance is, however, not
internally accounted for in QBBJT and most compact bipolar transistor models. A
conventional approach to compensate for this inadequacy is to optimize the collectorto
101
102
substrate capacitance Ccs to empirically account for substrate resistance. This approach,
however, is accurate only over a limited frequency range, because it is not physical to
represent an RC network using only a collectortosubstrate capacitor [48].
Ignoring substrate resistance limits the accuracy of the simulated smallsignal
parameter s22 or z22 of BJTs. In broadband RF applications, accurate prediction of s22 or
z22 with respect to frequency is essential for design matching networks to compensate the
overall gain over a wide range of frequencies. Inclusion of BJT substrate resistance is also
essential for predicting the undesired power loss through the substrate and the noise figure
(NF) of a low noise amplifier (LNA).
Ignoring substrate resistance also causes numerous problems in BJT parameter
extraction. Parameters extracted from two identical transistors with different substrate
resistances can vary significantly. This causes difficulties in statistical modeling of
parameters and uncertainties in BJT circuit simulations. In addition, parameters extracted
from one transistor in a test chip are not necessarily accurate enough for other transistors
in a circuit, since these transistors can have various substrate resistance values.
Modeling of substrate resistance is complicated by the layout, i.e., the transistor
size, the substrate contact size, and the distance between the transistor and its substrate
contact. A typical approach to eliminate the uncertainties caused by different substrate
resistances is to carefully control or equalize the values of substrate resistance for all the
transistors in the circuit and in the test chip. This can be done by minimizing the values of
substrate resistance for all the transistors. To minimize BJT substrate resistance, a guardÂ¬
ring substrate contact can be placed for each bipolar transistor with a minimum contactto
device spacing. This approach is reliable and eliminates most of the modeling problems
103
caused by BJT substrate resistance. However, this method is not desirable because it
requires extra silicon area. Minimizing BJT substrate resistance also degrades the small
signal voltage gain at high frequency.
In this chapter, we present an efficient approach to include substrate resistance in
BJT parameter extraction and circuit simulation. The method is not associated with any
layout rule. It is, therefore, consistent with the goal to minimize the final die area. By
including substrate resistance, we can also eliminate the uncertainty caused by substrate
resistance in BJT parameter extraction and circuit simulation and ease the burden of
repetitive parameter extraction.
The methodology requires an algorithm to accurately calculate substrate resistance
from layout and process information. Unlike the current flowing through poly or metal
interconnects, which is confined in a welldefined region and can be treated as an oneÂ¬
dimensional current, the substrate current is threedimensional. This 3D current flow
must be considered in the calculation of substrate resistance.
Our approach is to extend Gharpureyâ€™s work on substrate noise coupling to the
calculation of BJT substrate resistance [50][51]. The demand for integrated solutions
stimulates a growth in mixedsignal products. This growth, in turn, stimulates research on
substrate noise coupling [49][50] [51]. Substrate noise coupling is a severe problem when
analog and digital modules are placed on a same chip [49] [50], The primary goal of this
substrate noise coupling research is to quantify the digital noise injection onto sensitive
analog nodes through the common substrate. As a key to quantify substrate noise
coupling, devicetodevice substrate resistance can be computed with the finite difference
techniques or boundary element methods. However, these methods are usually
104
computationally inefficient. Recently, Gharpurey developed a quick algorithm to calculate
devicetodevice substrate resistances. The algorithm employs the discrete Green function
and Fast Fourier Transform (FFT) [51]. Simulations show that Gharpureyâ€™s method is as
accurate as the finite difference technique, but is 23 orders faster [52],
This chapter demonstrates how the recent achievement in the characterization of
substrate noise coupling can be applied to the calculation of BJT substrate resistance. Our
goal is to eliminate the uncertainties resulting from the layoutdependent substrate
resistances and enhance accuracy in SPICE BJT simulation.
To confirm the accuracy of Gharpureyâ€™s algorithm in the calculation of BJT
substrate resistance, the direct extraction method developed in the previous chapter was
used to extract BJT substrate resistance from measured sparameters. Comparison of
extracted and the calculated substrate resistances verifies that the calculation algorithm is
accurate.
This chapter is organized as following. Gharpureyâ€™s algorithm to calculate
substrate resistance is introduced in section 6.2. Limitations of using the algorithm to
calculate BJT substrate resistance is discussed in section 6.3. An ac verification for the
algorithm is discussed in section 6.4. A systematic methodology to include the layout
dependent substrate resistance in BJT circuit simulation is presented in section 6.5.
Finally, a brief conclusion is summarized in section 6.6.
5.2 Algorithm to Calculate BJT Substrate Resistance
To apply Gharpureyâ€™s algorithm to the calculation of BJT substrate resistance, we
consider the crosssection of a vertical doublepoly selfaligned npn bipolar transistor with
105
a single substrate contact as shown in Fig. 51. Since the current flowing along the
interface between ndoped region and psubstrate is negligible, the interface (i.e.,
boundary A) can be assumed to be equipotential. Also, since the region beneath the
substrate contact (i.e., the region enclosed by boundary B and substrate contact) is usually
highly doped, boundary B can also be assumed to be equipotential. Therefore, we can treat
the BJT nwell and the p+ region beneath substrate contact as two metal contacts in our
analysis and define the BJT substrate resistance as the resistance between boundary A and
boundary B as shown in Fig. 51. To calculate the substrate resistance, we supply a unity
current into one contact and out of the other, and then calculate the voltage difference
between the two contacts. If the substrate doping is uniform, we can assume that the
diffusion current in the substrate is negligible. Therefore, substrate current can be
expressed as
Jsubstrate = WpPE = Â°E > t5'1)
where (ip is the hole mobility of the substrate and a is the conductivity of the substrate.
Taking the divergence on both sides of (51) yields
y#y 1/a at one contact
V2cp(x, y, z) = ~~L'a = _i/(j at the other contact Â» (52)
0 elsewhere
where tp is the electrostatic potential.
To compute the electrostatic potential resulting from the supplied unity current, we
can solve (52) using the finite difference method, which discretizes a differential equation
into a finite number of difference equations. These difference equations are then linearized
with the Newton method, and the resulting coefficient matrix is solved with the sparse
matrix method. The finite difference method is commonly employed in device and process
N â—„
106
Boundary B f
ptype substrate
Figure 51 Crosssection of a multilayered substrate with
two surface contacts.
107
simulators, such as MEDICI and SUPREME. However, this method is usually very
computationally intensive and is thus not feasible when used in the calculation of substrate
resistances for a large number of devices.
The Green function technique can also be used to solve (52). The Green function
technique is often used to obtain the electrostatic potential for a dielectric medium
subjected to certain boundary conditions. Green function is defined as the potential at any
point (x,y,z) due to a unit point charge locating at (xâ€™,yâ€™,zâ€™); that is
x=x, y=y, z=z
elsewhere
(53)
0
where e is the permeativity of a dielectric medium. To obtain the electrostatic potential at
any point in the space, we can simply sum up the potential contributions from all the
charges distributed in the space. That is
(54)
The above analysis is developed to compute the electrostatic potential for a
dielectric medium. To comprehend how the Green function technique can be applied to
computing the electrostatic potential for a resistive medium, we consider two similar
situations shown in Fig. 52: 1) a purely resistive medium with a unity current flowing
into one contact and out of the other; 2) a purely dielectric medium with a positive unity
charge on one contact and a negative unity charge on the other. Notice that the boundary
conditions of these two cases are mathematically identical except with different
proportional constants. Therefore, the electric field or potential of the two cases must be
proportional to one another. Therefore, we can also solve the electric potential for a
resistive medium with the Green function technique, although it is developed to solve the
108
y# j 1/a at one contact
\ J substrate
Vâ€œ(p(x, y, z) = l/a at the other contact
0 elsewhere
x,y, z) 
1/e at one contact
1/e at the other contact
0 elsewhere
Figure 52 The electric field distributions of a purely
resistive and a purely dielectric media.
109
electrostatic potential for a dielectric medium. With the Green function technique, the
potential at any point (x,y,z) on the contact j due to a unity current uniformly distributed
on the contact i is obtained by integrating the Green function over the whole contact area
aÂ¡ as
(?Â¡(x,y,z) = Gc(x,y,z,x',y',z')daÂ¡. (55)
â€™ a,
where G0 is the Green function for a resistive medium.
Using series expansion and separation of variables, the Green function for a
multilayered resistive medium can be expressed as
oo Â©o
Gc(x,y,x',y')z=z, = 0 = Go+ ]T Â£/mncos
m = On = 0
(56)
where a and b are the length and width of a rectangular substrate; G0 and fmn are constant
coefficients defined by the physical structure of the substrate [51].
The average potential on the contact j due to a unity current uniformly distributed
on the contact i can be obtained by integrating (55) again over the whole contact area aÂ¡ as
cPy = JJg(.*, y, z, x\ y', z')daÂ¡daj. (57)
' JaÂ¡aÂ¡
Substituting (56) into (57) yields
frmxx\ f mnx
cos
\ a J \ a
(nny\ (nny'\
C0\b JC0Slb )
I S /,
m = On = 0
â– â– f a2
sin m 7iâ€”
. V a
sin mn
a j yi
. ( ^4^ . ( a3
sin mitâ€”  sin mix,â€”
a yjL V
(a2  iij)(a4  a3)
110
sin muâ€”
\ b
 sin mn
sin muâ€”
V b JjL \ b
. ( b3*
sin muâ€”
V b
(bi ~ b\)(b* â€” Â¿>3)
(58)
Because (58) converges slowly, using (58) to compute the electrostatic potential
is still timeconsuming. To make (58) converge faster, Gharpurey manipulated (58) using
the identity of trigonometric function
â€¢ ( ai} â€¢ ( ai
sin muâ€” sin wmâ€”
V a) \ a
[
( ai~aj
cosl muâ€”â€”
+ cos mu
a i + a m
a
n ,
(59)
and obtained
p Q
% ~ Z Z bmn
m = On = 0
COS
â– fPl,2Â±P3,4
mu\ â€”â€”â€”â€”
cos
r[â€”Qâ€”
(510)
where
2,2
k â€” a â€” f
inn 2 2 4^ nin '
m n K
(511)
P and Q are integers defining the upper limit of the series summations of (510). pk and qk
represent the integer indexes corresponding to the contact location axes ak and bk. pk and
qk are defined as
pk = integer[{ak/a)P], qk = integer[(bk/b)Q]
(512)
Notice that (510) can be decomposed into several cosine series summations.
Consider the twodimensional discrete cosine transform (DCT) series Kpq of any series
^mn as
KM =
p Q
Z Z knm
in = On = 0
COS
W7t ^
COS
nu\ 4
QJ.Ãœ
(513)
Ill
We can express cpÂ¡j as a sum of several specific terms in the Kpq series. With P and Q
chosen to be powers of 2 (i.e., 2n, n = 1,2, 3....), we can use the fast fourier transform
(FFT) algorithm to facilitate the computation of Kpq. Therefore, this method can be much
faster than the conventional Green function technique.
5.3 Limitations of the Algorithm
From above discussions, two assumptions are made: 1) the current flowing through
a contact is uniformly distributed; and 2) all contacts are planar and located on the top
surface of substrate.
The current flowing into a large contact is nonuniformly distributed. Simulations
show that the current density around the edge is relatively higher than that in the center. To
account for such nonuniformity, a large contact should be partitioned into several small
contacts. The substrate resistance of this large contact can then be estimated through
evaluating the total conductance of all the small contacts. This technique can also be used
to partition a nonrectangular contact. For example, a guardring contact can be partitioned
as the dashedlines shown in Fig. 53.
As substrate contacts or transistors can penetrate into the substrate with a finite
depth as shown in Fig. 51, several methods have been developed to account for these
vertical extensions [51]. However, all these methods work at the expense of higher
computation cost. In practical situations, when the lateral dimension of a contact is much
greater than the vertical extension, the current following into the vertical sidewall region
can be neglected. Therefore, the substrate resistance can be estimated by accounting for
only the current flowing into the bottom area.
112
Figure 53 Partition of a guardring substrate contact
Primary error sources of the Green function technique are the truncation error and
the roundoff error. The truncation error results when the infinite series summation of (5
10) is truncated with the upper bound P and Q. The roundoff error results when (512) is
used to convert the real numbers ak and bk to the integer indices pk and qk, which define
the contact location in the discrete (integer) domain. To minimize the truncation error, the
truncation bound P and Q should be chosen as large as possible. However, this approach
makes this algorithm numerically inefficient. In this way, this algorithm loses its
advantages over the finite difference method. Experience with this method indicated that
(510) converges sufficiently when P and Q were equal to or greater than 210=1024.
The roundoff error can be corrected by scaling the calculated resistance by the
ratio of the actual contact size to the integerindexed contact size. The roundoff error can
113
also be completely avoided, if all (ak, bk)â€™s are located at grid points. In this case, we can
select substrate size, P, and Q as
grid
(514)
where dgrid is the minimum grid spacing, a and b are the length and width that define a
rectangular area to cover two contacts adequately as shown in Fig. 51. In this way, the
roundoff error can be completely eliminated.
5.4 AC Verification of the Algorithm
In [51], Gharpurey had done some dc verifications for the algorithm and applied
this algorithm to calculate the equivalent welltowell (devicetodevice) resistances in the
characterization of substrate noise coupling. Gharpurey measured the contacttocontact
resistances for various metal contact configurations on a multilayered resistive substrate,
and then compared these measured resistances with model predictions. In this section, we
present an ac verification for the algorithm applied to the calculation of BJT substrate
resistance. Here, we used the direct extraction method developed in the previous chapter.
We extracted the substrate resistances for transistors with a wide variety of emitter feature
sizes, and then compared these extracted resistances with model prediction.
Test bipolar transistors were manufactured with TI (Texas Instruments, Inc.) 0.6
pm BiCMOS processes. All the test bipolar transistors have a 2.6 pm wide guardring
substrate contact placed around the BJT nwell with a spacing of 1.5 pm as shown in Fig.
53. The detailed layout information is listed in Table 51. With the processspecified
substrate resistivity (15 ohmcm), substrate resistances were calculated for transistors of
114
various emitter feature sizes. Sparameters were measured from 100 MHz to 7.5 GHz and
deembedded with open structure.
Fig. 54 shows the calculated substrate resistances and the substrate resistances
extracted using the method described in the previous chapter. The error bars in Fig. 54
represent the standard deviation of the extracted substrate resistances from measurements
at different bias conditions. Notice that the extracted substrate resistance values compare
well with the calculated substrate resistances for transistors of various emitter feature
sizes. This confirms the accuracy of the algorithm used in the calculation of BJT substrate
resistance.
Table 51 Layout information of the test devices
Test Device
WexLe
(am x (xm
# of
parallel
devices in
test array
structure
# of fingers
length of N
well (fim)
width of N
well (p.m)
0.6x1.2
54
1
17.8
20.8
0.6x1.5
45
1
19.8
21.7
0.6x3.0
24
1
19.8
23.2
0.6x6.0
12
1
19.8
26.2
0.6x12
6
2
22.8
26.2
0.6x23.9
3
3
25.6
27.5
0.6x36
2
4
28.6
28.8
0.6x45
2
5
31.4
29.2
0.6x72
1
5
31.4
34.0
0.6x96
1
5
31.4
38.8
0.6x192
1
10
46.6
38.8
Substrate resistance
115
Extracted Rsub vs. calculated Rsub
Figure 54 Comparison between extracted substrate
resistance and calculated resistances for BJTs with a wide
variety of emitter feature sizes.
116
5.5 Including BJT Substrate Resistance in Parameter Extraction
We implemented the methodology in the automated parameter extraction tool of
Texas Instruments, Inc., MACH. MACH extraction is based on the conventional Gummel
Poon model. First, MACH extracts dc BJT parameters from Gummel plots and output
characteristics. At this stage, MACH also extracts the capacitance parameters from small
signal CV measurements. Secondly, MACH extracts the forward base transit time by
fitting the unitygain frequency fT plot. Finally, MACH optimizes RB, RBM, RC, RE,
XCJC, and CJS to get the best fit to the measured sparameters.
Since substrate resistances are not explicitly modeled in MACH, the CJS values
obtained from MACHâ€™s optimizations can vary even for identical BJTs with different
substrate resistances. This causes numerous difficulties in the statistical modeling of CJS.
For instance, CJS values obtained from MACHâ€™s optimization are often unreasonably
small compared to their physical value, because the values are empirically lowered in
optimization to account for the substrate resistance.
To include substrate resistance in MACH parameter extraction, we added into
MACH a new variable RSUB. To test the methodology, we compared the extraction result
of MACH for two different cases: 1) old methodology: set the substrate resistance to zero,
and optimize CJS along with RB, RBM, RC, RE, and XCJC; 2) new methodology: use the
physical CJS obtained from lowfrequency CV measurements and the calculated substrate
resistance, and optimize only RB, RBM, RC, RE, and XCJC. The test was performed on
the bipolar transistors manufactured using a TI 0.6 pm RF BiCMOS process.
Figure 55 shows the differences between extractions using the two different
methodologies. Besides the optimal value of RB, RBM, RC, RE, and XCJC, we also
117
20.0
g, 10.0
D
o
C
SC
Q
0.0
10.0
Figure 55 Comparison between extractions using new
methodology and old methodology. The difference is
normalized by the extracted value of the old methodology.
compared the RMS errors of extractions using these two different methodologies. The
RMS error is a measure of how well the model prediction fits to the measured data. The
percentage differences shown in Fig. 55 are normalized by the extraction using the old
methodology. Notice that the new methodology fits model predictions to measured data as
well as the old methodology. Also, the extraction sensitivities of parameters are very
small, which means the extraction of BJT substrate parasitics can be isolated from the
extraction of other parasitics. However, the new methodology uses the physical CJS
obtained from lowfrequency CV measurement, thereby eliminating in the statistical
analysis of CJS the uncertainty caused by the layout dependence of BJT substrate
resistance. In addition, with the calculated substrate resistance, we can use the physical
RMS RB
1 ' 1
RC 1 ] III!
RBM
i i i
RE XCJC
118
CJS in the extraction, thereby avoiding the need to extract different CJS even for identical
BJTs with different substrate resistances.
Notice from that the extraction sensitivity of RBM is relatively large compared
with other parameters. This is because RBM is not a sensitive parameter. Very small
variation in RB causes relatively large variation in the extracted value of RBM.
5.6 Conclusions
This chapter presents a systematic methodology to include substrate resistance in
BJT parameter extraction and circuit simulation. An efficient algorithm to calculate
substrate resistance is reviewed. To confirm the accuracy of the algorithm used in the
calculation of BJT substrate resistance, substrate resistances were extracted from
measured sparameters and compared with model predictions. The results suggest that the
substrate resistance calculated with the algorithm is accurate. The methodology was tested
with the TI inhouse parameter tool, MACH. The testing results show that with the
calculated substrate resistance, we can use the physical CJS in parameter extraction and
circuit simulation, thereby avoiding the need to repeat the extraction process for varying
substrate contact positions.
The physical collectortosubstrate capacitance and calculated substrate resistance
can also be used in QBBJT simulation. Since both the physical collectortosubstrate
capacitance and resistance can be calculated from layout and process information, QBBJT
thus gains a predictive way to model substrate parasitics.
The algorithm to calculate substrate resistance is very fast. Therefore, it is practical
to include the calculated substrate resistance in routine parameter extraction and circuit
CHAPTER 6
SUMMARY AND SUGGESTED FUTURE WORK
6.1 Summary
This dissertation addresses several critical model enhancements and a complete dc
and ac parameter extraction methodology for the MMSPICE/QBBJT model. Significant
achievements are summarized as follows.
First, selfheating is included in the MMSPICE/QBBJT model. Selfheating
significantly influences bipolar transistors even at modest power dissipation. Physical
modeling of selfheating depends on accurate simulation of deviceâ€™s local temperature. By
adding a thermal node into QBBJTâ€™s admittance matrix, the device local temperature is
iteratively calculated in the SPICE modified nodal analysis. The resulting currentvoltage
characteristics are consistent with the device power dissipation and the thermal network
attached to the thermal node, thus including the selfheating effect. The method is also
adaptable to dc, ac, and transient thermal analysis.
Second, the dc current crowding model is modified to better model the currentÂ¬
crowding induced highinjection effects. The original QBBJT model accounts for dc
current crowding around the emitter periphery using the effective baseemitter voltage.
This method significantly underestimates the peak injection level around the emitter
periphery, thus overlooking the high injection effect caused by dc current crowding. To
avoid the underestimation of the highinjection effects caused by dc current crowding, this
119
120
dissertation presents an alternative way using the effective emitter area to model dc current
crowding. The method is consistent with the previous dc current crowding analysis, but
ensuring a correct high injection analysis.
Third, a compact, physical model is developed to account for NQS ac current
crowding. This is the first compact model which accounts for NQS majority carrier
transport in the quasineutral base and the nonuniformity of baseemitter admittance
caused by dc current crowding. AC current crowding is observed in BJTs as a nonlinear
phase shift in the base input impedance. Adapting the RC transmission line model, this
dissertation presents an equivalent circuit which simulates the nonlinear phase shift in the
base input impedance. This equivalent circuit is simple and is implemented in the
MMSPICE QBBJT model. This model also accounts for the dc base resistance reduction
caused by dc current crowding. With the effective base width calculated from the dc
current crowding analysis, the equivalent circuit is extended to empirically account for the
baseemitter admittance nonuniformity caused by dc current crowding. The extended
model is verified with the numerical solution of a nonlinear distributed RC transmission
line and the MIDIO simulation.
Fourth, a complete methodology to extract QBBJTâ€™s model parameters is
presented. With the physical constants (i.e. mobility and carrier saturation velocity)
assigned with default values, parameters are extracted in accordance with their significant
impacts to transistorâ€™s electrical characteristics. The extraction methodology is based on
several local optimizations for different operating regions. Because the methodology
optimizes no more than two parameters at a time, the methodology ensures the uniqueness
of extracted parameters. The physically extracted parameters can, thus, be employed to
121
infer BJT doping profiles. By demonstrating that the extracted doping profile is closely
correlated to the experimental SIMS profile, we suggest that the extraction methodology
can be used as an easy, inexpensive, and nondestructive tool for the investigation of BJT
doping profiles. In addition, a direct method to extract BJT hybridn equivalent circuit is
developed. With the direct methodology, the charge partition factor can be extracted and
the extrinsic resistance parameters can be further calibrated for accurate ac simulation.
The methodology is suitable for an automated extraction process. With the extraction
methodology, QBBJT can now be used in either a predictive mode for concurrent
engineering applications, or in a precision mode for accurate circuit simulations.
Fifth, this dissertation presents a systematic methodology to include substrate
resistance in BJT parameter extraction and circuit simulation. Substrate resistance is often
neglected in compact BJT models because it is layoutdependent. Even in most
sophisticated parasitics evaluation programs, substrate resistance is still ignored because
its calculation involves computationintensive threedimensional simulations. Ignoring
substrate resistance often causes many uncertainties in circuit simulation and parameter
extraction. This dissertation extends Gharpureyâ€™s work on substrate coupling to the
calculation of BJT substrate resistance. With the calculated substrate resistance, we can
use the physical collectortosubstrate capacitance in BJT parameter extraction, thereby
eliminating the uncertainties in the statistical analysis of collectortosubstrate capacitance
and avoiding the need to repeat the extraction for even identical BJTs but with different
substrate resistances.
122
6.2 Suggested Future Work
This dissertation demonstrates how the MMSPICE/QBBJT model is used in a
predictive mode for concurrent engineering as well as in a precision mode as a normal
empirical model for routine circuit design verifications. The resulting advantages make
this model uniquely flexible. However, until recently, use of MMSPICE is still rather
limited. Although lacking advertisements could be one of the major reasons, there is
another even more critical factor that limits popularity of QBBJT. It is that the MMSPICE/
QBBJT is not as numerically robust as other empirical models. QBBJT calculates terminal
currents with different equations. In this way, QBBJT better accounts for moving
boundary conditions and transport physics for different operating regions. However, this
may cause discontinuities in the derivatives of QBBJTâ€™s terminal currents. These
discontinuities are harmful to QBBJTâ€™s numerical stability and produce problems in some
analog simulations, such as harmonic balance analyses. Empirical models usually do not
have this problems. However, as most empirical BJT models are also criticized for their
improper formulation of collector current and static charge storage in quasisaturation
regions, QBBJT faces the biggest dilemma in its model formulation: physics? or
numerical robustness?
To better understand this dilemma, let us first generate a wish list for a desired
compact SPICE BJT model and see how the two currently available models (VB1C and
QBBJT) can satisfy the list. The results are shown in Table 61. Here, VB1C is the
representative for empirical models and QBBJT is the representative for physical models.
The formulation of VBIC and QBBJT are based on same equations, i.e. de Graaffâ€™s base
minority transport equation and Kullâ€™s collector major transport equation. However, with
123
different perspectives, the resulting models and their parameters can be very different, i.e.
empirical or predictive. Numerical stability and efficiency are guaranteed for empirical
models. Scalability and physical plausibility are inherent to physical models. Because of
using a singlepiece equation to model the majority carrier transport in the epicollector,
VBIC is numerically stable and efficient, but it fails to properly simulate the collector
current for all different operating regions. Because of using different equations for
different operating regions, QBBJT is physical, but it fails to be continuously
differentiable. The major tradeoff is to use a singlepiece smooth function for numerical
robustness or use different equations for physics.
Table 61 The wish list for a SPICE BJT model
WISH LIST
VBIC
QBBJT
The model is continuously differentiable.
Yes
No
The model can physically model majority carrier
transport in collector for various bias conditions.
No
Yes
The model can include as many secondary effects as
possible.
Yes
Yes
The model parameters are scalable.
No
Yes
The model parameters are evaluated from transistorâ€™s
layout and process information.
No
Yes
Actually, the reason why QBBJT uses different equations is to account for
different SCR formations in the epicollector. Therefore, we suggest to develop a unified
collector majority transport equation, which can account for various SCR formations in
the epicollector and ensure the numerical smoothness of resulting terminal currents. In
this way, we retain all the advantages of QBBJT, but eliminate all its numerical problems.
124
Another way to overcome this modeling problem is to investigate whether all SCR
formations need to be considered, because they may be insignificant under some
circumstances. For instance, base pushout may not happen for some very highspeed
BJTs.
Although QBBJT takes several geometric effects, such as current crowding, into
considerations, it is predominantly a ID model. Since 3D device simulations are still very
time consuming, especially in accounting for various layout and extrinsic parasitics, we
suggest to further enhance QBBJT to account for various quasi3D geometric effects. This
approach can be much more computation efficient than 3D device simulations, but still
useful in layout designs.
In Chapter 5, BJT substrate parasitics is modeled as a lumped resistance. The
model is experimentally justified up to 7.5 GHz. However, since the substrate parasitics is
actually distributed, we expect that the model should be insufficient at higher frequencies.
Therefore, we suggest to develop a more realistic model for BJTâ€™s substrate parasitics
used for applications at higher frequencies.
APPENDIX
THE SUBSTRATE RESISTANCE
CALCULATION PROGRAM SUBTEST
SUBTEST is a program written in C to numerically calculate the resistances
between two or many plannar contacts locating on the top surface of substrate. It accepts
either inputs written in a batch file or interactive inputs. From the specification of the
substrate physical structure, the program first calculates the series kmn of (611) and its 2
dimensional DCT series Kpq using FFT. The program then calculates the resistance
between contacts by summing the specific terms of the Kpq series in accordance with the
contact location indexes (refer to (612)). Since kmn and its DCT series Kpq depend only
on the substrate physical structure regardless of the contact configurations, the Kpq series
can be reused to calculated the substrate resistances for different contact configurations on
the same substrate. This attribute makes the technique especially efficient in such tasks as
optimizing contact configuration to minimize the noise injection to some sensitive node
through common substrate or calculating substrate resistances for many devices across a
very big region. For example, to calculate substrate resistance for all the devices on a big
chip as shown in Fig. Al, we can first divide this big chip into many small regions and
calculate the Kpq series for one of the subregions. Then, the DCT series is repeatedly used
to calculate the substrate resistances for all the active devices all over the chip area.
125
126
A sample input file to calculate the substrate resistance of a 0.6pmx96pm BJT
with a guardring substrate contact is provided in Fig. A2. The first three lines describe
the physical structure of the substrate. It is 102.4pm long, 102.4pm wide, 500pm thick.
Line 2 specifies the number of layers. In deriving (66), the bottom boundary of a substrate
is assumed to be grounded. Therefore, one more layer representing highresistive air must
be mounted to the bottom of a substrate, if the substrate is not grounded on bottom. Line 3
contains the information of the zaxis of the bottom of each layer. Since the zaxis of the
first layer bottom is always assumed to 0, this line contains only (# of layers minus 1)
numbers specifying the zaxis of the bottoms of the rest of the layers. Line 4 specifies the
resistivity of each layer. Line 5 specifies the output file name.
Line 6 and the following lines specify the contact configuration. Line 6 specifies
the numbers of contacts. Line 716 specify the x and y axes and the number of internal
partitions for each contact. The numbers are arranged as xl, x2, yl, y2, # of partitions in
xdirection, # of partitions in ydirection. The last line specifies whether to run another
contact configuration. If yes, another contact configuration specifications similar to those
contained in line 616 must be provided. Note that SUBTEST does not allow any text
comments in the input files.
, â€¢ lililÃ*
m
[fflO Active device area
â– Substrate contact
Figure Al Divide a big area into many small areas.
102.4 102
2
400
le8 le2
sub.out
9
4 500
7 :
39
6
62
8
35
9
66
5
1
1
8:
35
5
38
1
31
8
34
4
1
1
9 :
35
5
38
1
34
4
68
0
1
1
10
35
5
38
1
68
0
70
6
1
1
11
38
1
64
3
31
8
34
4
1
1
12
38
1
64
3
68
0
70
6
1
1
13
64
3
66
9
31
8
34
4
1
1
15
64
3
66
9
34
4
68
0
1
1
16
64
3
66
9
68
0
70
6
1
1
17
n
Figure A2 Sample input file of SUBTEST
REFERENCES
[1], J. Wamock, â€œSilicon bipolar device structures for digital applications: technology
trends and future directions,â€ IEEE Tran. Electron. Dev., Vol. ED42, pp377389,
1995.
[2]. T. Nakamura, and H. Nishizawa, â€œRecent progress in bipolar transistor technology,â€
IEEE Tran. Electron. Dev., Vol. ED42, pp390398, 1995.
[3], A.R. Alvarez (Ed.), â€œBiCMOS technology and applications,â€ Kluwer Academic
Publishers, Boston, 1989.
[4], L. Treitinger and M. MiuraMattausch (Ed.), â€œUltrafast silicon bipolar transistor,â€
SpringVerlag, New York, 1988.
[5], T. Yamagunchi, S. Uppili, and G. Kawamoto, â€œProcess and device optimization of a
30GHz fT submicrometer double polysi bipolar technology,â€ IEEE Bipolar
Circuits and Technology Meeting, pp. 136139, 1993.
[6]. J.J Ebers and J. L. Moll, â€œLargesignal behavior of junction transistors,â€ Proc. IRE,
Vol. 42, pp. 17611772, 1956.
[7]. I. E. Getreu, â€œModeling the bipolar transistor,â€ Elsevier, New York, 1978.
[8], H. K. Gummel and H. C. Poon, â€œAn integral charge control model of bipolar
transistors,â€ The Bell Sys. Tech. J., Vol. 49, pp. 827852, 1970.
[9], J. G. Fossum, â€œModeling issues for advanced bipolar device/circuit simulation,â€
IEEE Bipolar Circuits and Technology Meeting, pp. 234241, 1989.
[10]. G. Kull, L. Nagel, S. Lee, P. Lloyd, J. Prendergast, and H. Dirks, â€œA unified circuit
model for bipolar transistors including quasisaturation effects,â€ IEEE Tran.
Electron. Dev., Vol. ED32, ppl 1031113, 1985.
[11]. C. McAndrew, J. Seitchit, D. Bowers, M. Dunn, M. Foisy, I. Getreu, M. McSwain,
S. Moinian, J. Parker, P. van Wijnen, and L Wagner, â€œVBIC95: An improved
128
129
vertical, IC bipolar transistor modelâ€, IEEE Bipolar Circuits and Technology
Meeting, pp. 170177, 1995
[12]. H. C. de Graaff and W. J. Kloosterman el. al., â€œExperience with the new compact
MEXTRAM model for bipolar transistors,â€ IEEE Bipolar Circuits and Technology
Meeting, pp. 246249, 1989.
[13]. H. C. de Graaff and W. J. Kloosterman, â€œModeling of the collector epilayer of a
bipolar transistor in the MEXTRAM model,â€ IEEE Tran. Electron. Dev., Vol. ED
42, pp. 274282, 1995.
[14]. H. Jeong and J.G. Fossum, â€œA chargebased largesignal bipolar transistor model
for device and circuit simulation,â€ IEEE Tran. Electron. Dev., Vol. ED36, ppl24
131, 1989.
[15]. H. C. de Graaff and W. J. Kloosterman, â€œNew formulation of the current and charge
relations in bipolar transistors modeling for CAD purposes,â€ IEEE Tran. Electron.
Dev., Vol. ED32, pp. 24152419, 1985.
[16], J. Jin, â€œAssessment and modeling of nonquasistatic, nonlocal, and multiÂ¬
dimensional effects in advanced bipolar junction transistors,â€ Ph.D Dissertation,
University of Florida, Gainesville, 1992.
[17]. G. B. Hong, â€œModeling, simulation, and design of SiGebase heterojunction bipolar
transistors for highperformance integrated circuit,â€ Ph.D Dissertation, University
of Florida, Gainesville, 1994.
[18]. T. Y. Lee, â€œExtraction of thermal resistance for fullydepleted MOSFETs,â€ IEEE
Int. SOI Conf., pp. 7879, 1995.
[19]. R. M. Fox, S. G. Lee, and D. T. Zweidinger, â€œThe effects of bjt selfheating on
circuit behavior,â€ IEEE J. SolidState Circuits, Vol. 28, pp. 678685, 1993.
[20]. M. Reisch, â€œSelfheating in BJT circuit parameter extraction,â€ SolidState
Electron., Vol.35, pp. 677679, 1992.
[21], G. B. Gao, â€œThermal design studies of highpower heterojunction bipolar
transistors,â€ IEEE Tran. Electron. Dev., Vol. ED36, pp. 854863, 1989.
[22], J. S. Brodsky, D. T. Zweidinger and Rober M. Fox, â€œPhysicsbased multipole
models for BJT selfheating,â€ IEEE Bipolar Circuits and Technology Meeting, pp.
249252, 1993.
130
[23]. D. T. Zweidinger and R. M. Fox, â€œThermal impedance extraction for bipolar
transistors,â€ Accepted for publication by IEEE Tran. Electron. Dev.
[24]. R. L. Pritchard, â€œTwodimensional current flow in junction transistors at high
frequencies,â€ Proceedings of the IRE, pp. 11521160, 1958.
[25], H. N. Ghosh, â€œA distributed model of the junction transistor and its application in the
prediction of the emitterbase diode characteristics, base impedance, and pulse
response of the device,â€ IEEE Tran. Electron. Dev., Vol. ED12, pp. 513, 1965.
[26]. J. R. Hauser, â€œThe effects of distributed base potential on emittercurrent injection
density and effective base resistance for stripe transistor geometries,â€ IEEE Tran.
Electron. Dev., Vol. ED11, pp. 238242, 1964.
[27]. H. S. Rhee, S. Lee, and B. R. Kim, â€œDC and AC current crowding effects model
analysis in bipolar junction transistors using a news extraction method,â€ SolidState
Electron., Vol.38, pp. 3135, 1995.
[28]. M. P. J. G. Versleijen, â€œDistributed high frequency effects in bipolar transistors,â€
IEEE Bipolar Circuits and Technology Meeting, pp. 8588, 1991.
[29]. H. Cho and D. E. Burk, â€œA simple model for distributed base impedance with ac
verification using Sparameter measurements,â€ IEEE Bipolar Circuits and
Technology Meeting, pp. 106109, 1990.
[30], J. G. Fossum, and S. Veeraraghavan, â€œPartitionchargebase modeling of bipolar
transistors for nonquasistatic circuit simulation,â€ IEEE Electron. Dev. Lett., Vol.
EDL7, 1986.
[31]. D. J. Roulston, â€œBipolar semiconductor devices,â€ McGrawHill, pp. 251, New
York, 1990.
[32], W. C. Johnson and P. Panousis, â€œThe influence of debye length on the CV
measurement of doping profiles,â€ IEEE Tran. Electron. Dev., Vol. ED18, pp. 965,
1971.
[33]. D. Costa, W. U. Liu, and J. S. Harris, â€œDirect extraction of the AlGaAs/GaAs
heterojunction bipolar transistor smallsignal equivalent circuit,â€ IEEE Tran.
Electron. Dev., Vol. ED38, pp. 20182024, 1991.
131
[34], M. Kim, D. Yoon, S. Cha, and J. Jin, â€œA global optimization of bipolar model
parameters using stimulated diffusion,â€ Proc. IEEE 1994 Int. Conf. on
Microelectronic Test Structures, Vol.7, March 1994.
[35]. D. R. Pehlke and D. Pavlidis, â€œEvaluation of the factors determining HBT high
frequency performance by direct analysis of Sparameter data,â€ IEEE Tran.
Electron. Dev., Vol. ED39, pp. 23672373, 1992.
[36], S. J. Spiegel, D. Ritter, R. A. Hamm, A. Feygenson, and P. R. Smith, â€œExtraction of
the InP/GalnAs Heterojunction Bipolar Transistor SmallSignal Equivalent
Circuit,â€ IEEE Tran. Electron. Dev., Vol. ED42, pp. 10591064, 1995.
[37], S. Lee, B. R. Ryum, and S. W. Kang, â€œA new parameter extraction technique for
smallsignal equivalent circuit of polysilicon emitter bipolar transistors,â€ IEEE
Tran. Electron. Dev., Vol. ED41, pp. 233238, 1994.
[38]. T. Ning and D. Tang, â€œMethod of determining the emitter and base resistance of
bipolar transistors,â€ IEEE Tran. Electron. Dev., Vol. ED31, pp. 409412, 1984.
[39]. R. S. Muller and T. I. Kamins, â€œDevice electronics for integrated circuits,â€ 2nd Ed.,
Wiley, New York, 1986.
[40], J. W. Slotboom and H. C. de. Graaff, â€œMeasurements of bandgap narrowing in si
bipolar transistors,â€ SolidState Electron., Vol. 19, pp. 857862, 1976.
[41]. D. B. M. Klaassen, J. W. Slotboom and H. C. de Graaff, â€œUnified apparent bandgap
narrowing in n and ptype silicon,â€ SolidState Electron., Vol.35, pp. 125129,
1992.
[42]. D. B. M. Klaassen, â€œA unified mobility model for device simulationi. model
equations and concentration dependence,â€ SolidState Electron., Vol.35, pp. 953
959, 1992.
[43]. M. Ugagin and J. G. Fossum, â€œSignificant Time constants defined by high current
charge dynamics in advanced siliconbased bipolar transistors,â€ IEEE Tran.
Electron. Dev., Vol. ED41, pp. 17961799, 1994.
[44], K. Suzuki, â€œEmitter and base transit time of the polycrystalline silicon emitter
contact bipolar transistors,â€ IEEE Tran. Electron. Dev., Vol. ED38, pp. 25122518,
1991.
132
[45], G. Gonzalez, â€œMicrowave transistor amplifiers  analysis and design,â€ Prentice
Hall, Englewood Cliffs, 1984.
[46]. G. A. Armstrong and W. D. French, â€œA model for the dependence of maximum
oscillation frequency on collector to substrate capacitance in bipolar transistors,â€
SolidState Electron., Vol.38, pp. 15051510, 1995.
[47], A. P. Laser and D. L. Pulfrey, â€œReconciliation of methods for estimating fmax for
microwave heterojunction transistors,â€ IEEE Tran. Electron. Dev., Vol. ED38, pp.
16851692, 1991.
[48]. M. Pfost, H. Rein, and T. Holzwarth, â€œModeling of the substrate effect in high
speed sbipolar ICs,â€ IEEE Bipolar Circuits and Technology Meeting, pp. 182185,
1995.
[49]. N. Verghese, D. Allstot, M. Wolfe, â€œVerification techniques for substrate coupling
and their application to mixedsignal IC design,â€ IEEE Journal of SolidState
Circuits, Vol. 31, pp 354365, 1996.
[50]. R. Gharpurey, and R. Meyer, â€œModeling and analysis of substrate coupling in
integrated circuits,â€ IEEE Journal of SolidState Circuits, Vol. 31, pp 344353,
1996.
[51], R. Gharpurey, PhD. Dissertation, University of California, Berkeley, 1995.
[52], R. Gharpurey, and R. Meyer, â€œAnalysis and simulation of substrate coupling in
integrated circuit,â€ International Journal of Circuit theory and Applications, Vol.
23,381394, 1995.
BIOGRAPHICAL SKETCH
TzungYin Lee was bom in Kaohsiung, Taiwan in 1969. He received his B.S.
degree from the Department of Electronics Engineering at National Chiao Tung
University, Hsinchu, Taiwan, in 1991 and M.E. from the Department of Electrical
Engineering at the University of Florida, Gainesville, Florida, in 1992.
After obtaining his M.E., TzungYin continued his study in the same department to
pursue the degree of Ph.D. TzungYin is currently engaged in the physical modeling for
the advanced highspeed bipolar transistors.
133
I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philosophy.
Robert M. Fox, Chairman
Associate Professor of Electrical and Computer
Engineering
I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philosophy.
Jtjrry G. Fossum
Professor of Electrical and Computer Engineering
I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philosophy.
Mark E. Law,
Associate Professor of Electrical and Computer
Engineering
I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philosophy.,
Gijs Bosman
Professor of Electrical and Computer Engineering
I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philosophy.
Hr
U\Q
Mang Tia
Professor of Civil'Engineering
This dissertation was submitted to the Graduate Faulty of the College of
Engineering and to the Graduate School and was accepted as partial fulfillment of the
requirements for the degree of Doctor of Philosophy.
May 1997
Winfred M. Phillips
Dean, College of Engineering
Karen A. Holbrook
Dean, Graduate School
xml version 1.0 encoding UTF8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchemainstance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EOHE4XR18_EFKNHI INGEST_TIME 20140812T19:46:19Z PACKAGE AA00024910_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
PAGE 1
02'(/ (1+$1&(0(176 $1' 3$5$0(7(5 (;75$&7,21 )25 7+( 0063,&(4%%7 02'(/ %< 7=81*<,1 /(( $ ',66(57$7,21 35(6(17(' 72 7+( *5$'8$7( 6&+22/ 2) 7+( 81,9(56,7< 2) )/25,'$ ,1 3$57,$/ )8/),//0(17 2) 7+( 5(48,5(0(176 )25 7+( '(*5(( 2) '2&725 2) 3+,/2623+< 81,9(56,7< 2) )/25,'$
PAGE 2
$&.12:/('*(0(176 ZRXOG OLNH WR H[SUHVV P\ VLQFHUH JUDWLWXGH DQG DSSUHFLDWLRQ WR P\ DGYLVRU 'U 5REHUW 0 )R[ IRU KLV GHYRWHG JXLGDQFH SDWLHQW HQFRXUDJHPHQW DQG VXSSRUW WKURXJKRXW P\ 3K' VWXG\ ,W LV D JUHDW SOHDVXUH WR ZRUN ZLWK KLP ZRXOG DOVR OLNH WR JLYH WKDQNV WR P\ VXSHUYLVRU\ FRPPLWWHH 'U *LMV %RVPDQ 'U HUU\ )RVVXP 'U 0DUN /DZ DQG 'U 0DQJ 7LD IRU WKHLU ZLOOLQJQHVV WR SURYLGH PH WKHLU WLPH DQG JXLGDQFH DP JUDWHIXO WR WKH 6HPLFRQGXFWRU 5HVRXUFH &RPSDQ\ 65&f 0RWRUROD DQG 7H[DV ,QVWUXPHQWV ,QF IRU WKHLU WHFKQLFDO DQG ILQDQFLDO VXSSRUW WKURXJKRXW P\ UHVHDUFK IHHO HVSHFLDOO\ REOLJDWHG WR H[SUHVV P\ GHHS DSSUHFLDWLRQ WR 'U 0DUN )RLV\ DW 0RWRUROD IRU KLV LQYDOXDEOH FRPPHQWV DQG GLVFXVVLRQV DP DOVR JUDWHIXO WR 0U 7RP 9URWVRV 'U .HLWK *UHHQ DQG 'U =KLOLDQJ &KHQ IRU WKHLU JXLGDQFH GXULQJ P\ LQWHUQVKLS LQ 7H[DV ,QVWUXPHQW ,QF :LWKRXW WKHLU VXSSRUW WKH ZRUN FRXOG QHYHU KDYH EHHQ ILQLVKHG ZRXOG DOVR OLNH WR WKDQN 'U 5DQMLW *KDUSXUH\ IRU WKH LQYDOXDEOH GLVFXVVLRQ RQ %7 VXEVWUDWH UHVLVWDQFH FDOFXODWLRQ ZRXOG OLNH WR UHFRJQL]H LQ WKLV DFKLHYHPHQW 'U 0LQJ&KDQJ /LDQJ DQG 0HVVUV 'DYLG =ZLGLQJHU RQDWKDQ %URGVN\ 0LQJ
PAGE 3
:DQJ 'U %UXFH /LX HQIHQJ
PAGE 4
7$%/( 2) &217(176 SDJH $&.12:/('*(0(176 LL $%675$&7 YL &+$37(56 ,1752'8&7,21 %$&.*5281' 3+<6,&6 $1' 02'(/ )2508/$7,21 2) 7+( 0063,&( &+$5*(%$6(' %,32/$5 75$16,6725 02'(/ 4%%7 %DFNJURXQG 5HYLHZ (EHUV0ROO 0RGHO *XPPHO3RRQ 0RGHO 1RYHO &RPSDFW 0RGHOV IRU %LSRODU 7UDQVLVWRUV 7KH ([WHQGHG *XPPHO3RRQ 0RGHO 0(;75$0 0063,&(4%%7 0RGHO 9HUWLFDO %LSRODU ,QWHU &RPSDQ\ 0RGHO 9%,&f 0RGHO )RUPXODWLRQ RI 4%%7 %7 2SHUDWLRQ 0RGHV )RUPXODWLRQ RI '& &XUUHQW9ROWDJH (TXDWLRQV 4%%7fV 6PDOO6LJQDO (TXLYDOHQW &LUFXLW 02'(/ (1+$1&(0(176 2) 7+( 4%%7 02'(/ 6HOIKHDWLQJ 0RGHOLQJ '& &XUUHQW &URZGLQJ 0RGHOLQJ $& &XUUHQW &URZGLQJ 0RGHOLQJ 6LPSOH 0RGHO )RUPXODWLRQ $FFRXQWLQJ IRU '& &XUUHQW &URZGLQJ LQ $& &XUUHQW &URZGLQJ $QDO\VLV 0RGHO ,PSOHPHQWDWLRQ DQG 9HULILFDWLRQ &RQFOXVLRQ DQG 'LVFXVVLRQ LY
PAGE 5
3$5$0(7(5 (;75$&7,21 )25 0063,&(4%%7 ,QWURGXFWLRQ ([WUDFWLRQ RI *XPPHO3RRQ/LNH 3DUDPHWHUV ([WUDFWLRQ RI 'RSLQJ 3URILOHV &ROOHFWRU 'RSLQJ 3URILOH %DVH 'RSLQJ 3URILOH ,QWHJUDWLRQ RI 'RSLQJ 3URILOH ([WUDFWLRQ 3URFHGXUHV ([WUDFWLRQ 5HVXOWV DQG 'LVFXVVLRQ ([WUDFWLRQ RI 4%%7fV 6PDOO6LJQDO (TXLYDOHQW &LUFXLW 0HWKRGRORJ\ &DOLEUDWLRQ RI 4%%7fV ([WULQVLF 5HVLVWDQFH 3DUDPHWHUV (YDOXDWLRQ RI 4%%7fV &KDUJH 3DUWLWLRQ )DFWRUV &RQFOXVLRQ 63,&( 02'(/,1* 2) %7 68%675$7( 5(6,67$1&( ,QWURGXFWLRQ $OJRULWKP WR &DOFXODWH %7 6XEVWUDWH 5HVLVWDQFH /LPLWDWLRQV RI WKH $OJRULWKP $& 9HULILFDWLRQ RI WKH $OJRULWKP ,QFOXGLQJ %7 6XEVWUDWH 5HVLVWDQFH LQ 3DUDPHWHU ([WUDFWLRQ &RQFOXVLRQV 6800$5< $1' 68**(67(' )8785( :25. 6XPPDU\ 6XJJHVWHG )XWXUH :RUN $33(1',; 7+( 68%675$7( 5(6,67$1&( &$/&8/$7,21 352*5$0 68%7(67 5()(5(1&(6 %,2*5$3+,&$/ 6.(7&+
PAGE 6
$EVWUDFW RI 'LVVHUWDWLRQ 3UHVHQWHG WR WKH *UDGXDWH 6FKRRO RI WKH 8QLYHUVLW\ RI )ORULGD LQ 3DUWLDO )XOILOOPHQW RI WKH 5HTXLUHPHQWV IRU WKH 'HJUHH RI 'RFWRU RI 3KLORVRSK\ 02'(/ (1+$1&(0(176 $1' 3$5$0(7(5 (;75$&7,21 )25 7+( 0063,&(4%%7 02'(/ %\ 7]XQJ
PAGE 7
FXUUHQW FURZGLQJ DQDO\VLV WKH PRGHO FDQ DOVR EH H[WHQGHG WR DFFRXQW IRU WKH QRQXQLIRUPLW\ LQ WKH EDVHHPLWWHU DGPLWWDQFH DORQJ WKH EDVH 7KLV GLVVHUWDWLRQ SUHVHQWV D FRPSOHWH SDUDPHWHU H[WUDFWLRQ PHWKRGRORJ\ IRU WKH 0063,&(4%%7 PRGHO 4%%7 LV GHVLJQHG WR WDNH WKH PRGHO SDUDPHWHUV GHULYHG IURP WUDQVLVWRUfV VWUXFWXUDO LQIRUPDWLRQ VXFK DV OD\RXW DQG GRSLQJ SURILOH ,Q VXFK D ZD\ 4%%7 HQDEOHV VLPXODWLRQ HYHQ ZKHQ WKH SURFHVV LV QRW \HW ILQDOL]HG 7KLV DWWULEXWH PDNHV 4%%7 HVSHFLDOO\ XVHIXO LQ FRQFXUUHQW HQJLQHHULQJ DSSOLFDWLRQV +RZHYHU 4%%7 VWLOO QHHGV D PHWKRGRORJ\ WR FDOLEUDWH RU WXQH LWV SDUDPHWHUV ZKHQ UHSUHVHQWDWLYH PHDVXUHPHQWV DUH DYDLODEOH 7KLV GLVVHUWDWLRQ GHPRQVWUDWHV D V\VWHPDWLF PHWKRGRORJ\ WR H[WUDFW 4%%7f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
PAGE 8
&+$37(5 ,1752'8&7,21 6LQFH WKH ELSRODU MXQFWLRQ WUDQVLVWRU %7f ZDV LQYHQWHG LQ LW KDV EHHQ HPSOR\HG LQ DSSOLFDWLRQV RI DQDORJ DPSOLILFDWLRQ DQG GLJLWDO VZLWFKLQJ &RPSOHPHQWDU\ PHWDOR[LGHVHPLFRQGXFWRU &026f WHFKQRORJ\ EHFDXVH RI LWV FKDUDFWHULVWLFV RI ORZ SRZHU FRQVXPSWLRQ DQG KLJKSDFNLQJ GHQVLW\ LV SUHIHUUHG LQ IDEULFDWLRQ RI XOWUD ODUJH VFDOH LQWHJUDWHG 8/6,f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fV VSHHG DGYDQWDJHV ZKLOH DFKLHYLQJ KLJKHU LQWHJUDWLRQ OHYHO O
PAGE 9
2QH DSSURDFK LV WR PHUJH KLJKSHUIRUPDQFH ELSRODU GHYLFHV LQWR &026 SURFHVVHV DQG IRUP %LSRODU&026 %L&026f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
PAGE 10
DQG DV D GLIIXVLRQ VRXUFH WR IRUP ORZUHVLVWLYH H[WULQVLF EDVH /LNH WKH EDVH FRQWDFW SRO\ WKH VHFRQGOD\HU SRO\ IRUPV WKH HPLWWHU FRQWDFW DQG VHUYHV DV D GLIIXVLRQ VRXUFH WR IRUP D VKDOORZ HPLWWHU MXQFWLRQ $ W\SLFDO VWUXFWXUH IRU D GRXEOHSRO\ VHOIDOLJQHG ELSRODU WUDQVLVWRUV LV VKRZQ LQ )LJ )LJXUH &URVVVHFWLRQ RI FRPPRQ DGYDQFHG GRXEOHn SRO\ VHOIDOLJQHG ELSRODU WUDQVLVWRUV 8VLQJ KLJKO\ GRSHG SRO\6L DV DQ HPLWWHU FRQWDFW UHGXFHV SURFHVV FRPSOH[LW\ DQG PDLQWDLQV WKH VHSDUDWLRQ EHWZHHQ WKH PHWDO FRQWDFW DQG WKH YHU\ VHQVLWLYH PRQRFU\VWDOOLQH VLOLFRQ UHJLRQ 7KH VHSDUDWLRQ SUHYHQWV VLQWHULQJ HIIHFWV DQG PHWDO VSLNHV >@ 7KH SRO\n HPLWWHUFRQWDFW %7 DOVR H[KLELWV FKDUDFWHULVWLFV VXSHULRU WR LWV PHWDOFRQWDFW FRXQWHUSDUW )LUVW D VKDOORZ EDVH MXQFWLRQ FDQ EH IRUPHG ZLWK VXIILFLHQW FRQWURO RQ VKDOORZ HPLWWHU MXQFWLRQ IRUPDWLRQ 7KH VKDOORZ EDVH UHGXFHV EDVH WUDQVLW WLPH DQG LPSURYHV VSHHG SHUIRUPDQFH RI WKH WUDQVLVWRU 6HFRQGO\ ELSRODU WUDQVLVWRUV ZLWK SRO\ HPLWWHU FRQWDFWV
PAGE 11
XVXDOO\ KDYH EHWWHU HPLWWHU HIILFLHQF\ WKDQ ELSRODU WUDQVLVWRUV ZLWK PHWDO FRQWDFWV EHFDXVH RI WKH ORZHU FDUULHU PRELOLW\ LQ SRO\ VLOLFRQ DQG WKH LQWHUIDFLDO R[LGH OD\HU LQWHUSRVHG EHWZHHQ DFWLYH HPLWWHU DQG SRO\ HPLWWHU $V WKH WKLFNQHVV RI HPLWWHU FRQWLQXHV WR GHFUHDVH IRU VKDOORZHU EDVH MXQFWLRQV FRQYHQWLRQDO PHWDOFRQWDFW WUDQVLVWRUV ZRXOG VXIIHU D VHULRXV FXUUHQWJDLQ UHGXFWLRQ 7KLV LV EHFDXVH WKH HPLWWHU EHFRPHV WUDQVSDUHQW WR WKH LQMHFWHG PLQRULW\ FDUULHUV IURP WKH EDVH ZKHQ LWV WKLFNQHVV LV OHVV WKDQ WKH FKDUDFWHULVWLF GLIIXVLRQ OHQJWK RI PLQRULW\ FDUULHUV )LQDOO\ WKH SRO\ HPLWWHU VHUYHV DV D JHWWHULQJ VRXUFH IRU PHWDO LPSXULWLHV >@ :LWK GRXEOHSRO\ VHOIDOLJQHG WHFKQRORJ\ %7fV ODWHUDO GLPHQVLRQV DUH DOVR VLJQLILFDQWO\ UHGXFHG 7KLV LPSURYHV %7fV VSHHG SHUIRUPDQFH LQ VHYHUDO DVSHFWV )LUVW HPLWWHU ZLGWK LV QDUURZHU WKDQ WKH PLQLPXP OLQHZLGWK JLYHQ E\ OLWKRJUDSK\ 7KH UHGXFWLRQ LQ HPLWWHU ZLGWK UHGXFHV WKH LQWULQVLF EDVH UHVLVWDQFH WKXV HDVLQJ WKH GF FXUUHQW FURZGLQJ HIIHFW IRU %7 RSHUDWLQJ DW KLJK FXUUHQW GHQVLWLHV 6HFRQGO\ WKH GLVWDQFH EHWZHHQ WKH HGJHV RI HPLWWHU DQG EDVH FRQWDFWV ZKLFK LV XVXDOO\ GHWHUPLQHG E\ WKH ZLGWK RI D VSDFHU LV UHGXFHG VR WKDW %7f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fV SRZHU JDLQ ,Q DGGLWLRQ UHGXFLQJ EDVH WKLFNQHVV DOVR UHGXFHV WKH YROWDJH DW ZKLFK WUDQVLVWRUV EUHDN GRZQ EHFDXVH RI EDVH SXQFKWKURXJK $ SRVVLEOH ZD\ WR FRPSURPLVH VSHHG SRZHU JDLQ
PAGE 12
DQG EUHDNGRZQ YROWDJH LQ GHVLJQLQJ D %7 LV WR LQFUHDVH WKH WUDQVLVWRUfV EDVH GRSLQJ +RZHYHU LQFUHDVLQJ EDVH GRSLQJ VDFULILFHV WUDQVLVWRUfV FXUUHQW JDLQ 8VLQJ 6L*H HSLWD[LDO ILOP IRU WKH WUDQVLVWRU EDVH LQ VLOLFRQEDVHG SURFHVV LV D FRVW HIIHFWLYH VROXWLRQ WR WKH DERYH GLOHPPD %HFDXVH WKH UDWLR EHWZHHQ WUDQVLVWRUfV EDVH DQG HPLWWHU FXUUHQW LQMHFWLRQV LV H[SRQHQWLDOO\ DPSOLILHG E\ WKH EDQGJDS GLIIHUHQFH DW WKH HPLWWHUEDVH MXQFWLRQ D 6L*H KHWHURMXQFWLRQ ELSRODU WUDQVLVWRU +%7f FDQ KDYH WR WLPHV WKH FXUUHQW JDLQ RI D KRPRMXQFWLRQ VLOLFRQ ELSRODU WUDQVLVWRU 7KLV H[WUD FXUUHQW JDLQ FDQ EH WUDGHG IRU WKH VSHHG SHUIRUPDQFH E\ RSWLPL]LQJ EDVH GRSLQJ DQG WKLFNQHVV ,Q DGGLWLRQ DV WKH EDQGJDS RI 6L*H FDQ EH PDQLSXODWHG E\ YDU\LQJ WKH *H FRQFHQWUDWLRQ LQ WKH 6L EDVH D EXLOWLQ HOHFWULF ILHOG FDQ EH FUHDWHG WR DVVLVW PLQRULW\ WUDQVSRUW LQ EDVH ZKLFK IXUWKHU HQKDQFH WUDQVLVWRUfV VSHHG SHUIRUPDQFH 5HFHQWO\ 6L*H +%7V ZLWK *+] XQLW\ JDLQ IUHTXHQF\ KDYH EHHQ UHSRUWHG (YHQ IRU SXUHO\ VLOLFRQ WHFKQRORJ\ DJJUHVVLYHO\ VFDOHG GRXEOHSRO\ VHOIDOLJQHG ELSRODU WUDQVLVWRUV QRZ FDQ UHDFK XQLW\JDLQ IUHTXHQFLHV DV KLJK DV *+] >@ WKXV EHLQJ VXLWDEOH IRU YHU\ KLJKVSHHG GLJLWDO DQG DQDORJ DSSOLFDWLRQV $ GHFDGH DJR WKHVH KLJKn VSHHG DSSOLFDWLRQV ZHUH UHVHUYHG IRU RQO\ FRPSRXQG VHPLFRQGXFWRU GHYLFHV +RZHYHU WKH GHYLFH PRGHOV ZKLFK DUH HVVHQWLDO WR VLPXODWLRQ GHVLJQ DQG RSWLPL]DWLRQ RI ELSRODU GHYLFHV DQG FLUFXLWV KDYH QRW NHSW SDFH ZLWK WKH WUHPHQGRXV SURJUHVV DFKLHYHG LQ SURFHVV WHFKQRORJ\ ,Q GHVLJQLQJ %LSRODU%L&026 GHYLFHV DQG FLUFXLWV WHFKQRORJ\ &$' 7&$'f ZKLFK H[SORUHV DQG HYDOXDWHV YDULRXV GHVLJQ WUDGHRII ZLWKRXW WLPHFRQVXPLQJ DQG FRVWO\ GHYLFH IDEULFDWLRQ LV ZLGHO\ XVHG LQ WRGD\fV VWDWHRIWKHDUW GHVLJQ SURFHVV ,Q WKH SDVW WHFKQRORJ\ GHVLJQ ZDV PRVWO\ EDVHG RQ H[SHULPHQWEDVHG LH WULDODQGHUURUf
PAGE 13
RSWLPL]DWLRQV +RZHYHU DV WKH FRVWV DQG GXUDWLRQ RI GHYHORSPHQW LQFUHDVH ZLWK WKH FRPSOH[LWLHV RI WRGD\f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fV PRGHO SDUDPHWHUV DUH PRVWO\ GHULYHG IURP GHYLFHfV OD\RXW DQG GRSLQJ SURILOH 7KHUHIRUH FRUUHODWLRQV EHWZHHQ PRGHO SDUDPHWHUV DQG SURFHVV LQIRUPDWLRQ DUH DXWRPDWLFDOO\ PDLQWDLQHG +RZHYHU WKH XVH RI 0063,&(4%%7 PRGHO KDV EHHQ UDWKHU OLPLWHG 7KLV LV EHFDXVH WKDW WKH 0063,&(4%%7 PRGHO ODFNV D ZHOOGHILQHG SDUDPHWHU H[WUDFWLRQ PHWKRGRORJ\ ,Q SUDFWLFH WKH 0063,&(4%%7 PRGHO LV XVHG LQ D YDULHW\ RI PRGHV ,Q FRQFXUUHQW HQJLQHHULQJ DSSOLFDWLRQV 4%%7 WDNHV SDUDPHWHU GHULYHG IURP
PAGE 14
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fV DGPLWWDQFH PDWUL[ ZKLFK LV XVHG LQ 63,&(fV PRGLILHG QRGDO DQDO\VLV 6HOIKHDWLQJ LV LQKHUHQW WR DOO VHPLFRQGXFWRU GHYLFHV 6HOIKHDWLQJ LV D VLJQLILFDQW HIIHFW HYHQ IRU EXONVLOLFRQ ELSRODU WUDQVLVWRUV RSHUDWLQJ DW PRGHVW SRZHU GLVVLSDWLRQ ,JQRULQJ VHOIKHDWLQJ LQ WKH %7 PRGHO FDQ FDXVH SUREOHPV LQ VLPXODWLQJ DQDORJ FLUFXLWV EHFDXVH %7fV RXWSXW FRQGXFWDQFH LV HLWKHU RYHUHVWLPDWHG IRU KRPRMXQFWLRQ VLOLFRQ ELSRODU WUDQVLVWRUV RU XQGHUHVWLPDWHG IRU VRPH FRPSRXQG KHWHURMXQFWLRQ ELSRODU WUDQVLVWRUV '& FXUUHQW FURZGLQJ LV DQ LPSRUWDQW WZRGLPHQVLRQDO HIIHFW 7KH RULJLQDO 4%%7 PRGHO DFFRXQW IRU FXUUHQW FURZGLQJ DURXQG WKH HPLWWHU SHULSKHU\ E\ XVLQJ WKH HIIHFWLYH
PAGE 15
EDVHHPLWWHU YROWDJH ZKLFK FDQ VLJQLILFDQWO\ XQGHUHVWLPDWH %7f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f HIIHFW DQG LV RIWHQ RYHUORRNHG LQ WKH %7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW 7KLV LV EHFDXVH %7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLWV DUH XVXDOO\ GHULYHG XVLQJ TXDVLVWDWLF 46f DSSUR[LPDWLRQV $V WKH WRWDO EDVH FKDUJH VWRUDJH LV PRGHOHG E\ D OXPSHG FDSDFLWRU WKH WLPH UHTXLUHG IRU WKH PDMRULW\ FKDUJHV LQ WKH EDVH WR UHGLVWULEXWH ODWHUDOO\ DORQJ WKH EDVH LV RYHUORRNHG 7KH 46 DSSUR[LPDWLRQV LV HTXLYDOHQW WR DVVXPLQJ WKDW WKH FDUULHUV FDQ WUDYHO ZLWK LQILQLWH YHORFLW\ LQ FHUWDLQ UHJLRQV 7KLV GLVVHUWDWLRQ SUHVHQWV D QHZ FRPSDFW PRGHO WR DFFRXQW WKH 146 DF FXUUHQW FURZGLQJ 7KLV PRGHO LV FRQVLVWHQW ZLWK 4%%7f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fV VXEVWUDWH UHVLVWDQFH LV OD\RXW GHSHQGHQW ,Q WKLV GLVVHUWDWLRQ D V\VWHPDWLF PHWKRGRORJ\ WR LQFOXGH VXEVWUDWH UHVLVWDQFH LQ 63,&( %7 VLPXODWLRQ LV
PAGE 16
SUHVHQWHG :LWK WKH PHWKRGRORJ\ %7fV VXEVWUDWH UHVLVWDQFH LV FRQVLVWHQWO\ LQFOXGHG LQ %7fV SDUDPHWHU H[WUDFWLRQ DQG FLUFXLW VLPXODWLRQV 7KH GLVVHUWDWLRQ LV RUJDQL]HG DV IROORZV &KDSWHU UHYLHZV VHYHUDO FODVVLF DQG QRYHO FRPSDFW PRGHOV $W WKH HQG RI &KDSWHU 4%%7fV PRGHO IRUPXODWLRQ LV EULHIO\ GLVFXVVHG WR LQWURGXFH VHYHUDO FRQFHSWV LQ WKH IROORZLQJ FKDSWHUV &KDSWHU DGGUHVVHV VHYHUDO DGGLWLRQV DQG UHILQHPHQWV RI 4%%7 &KDSWHU SUHVHQWV D FRPSOHWH SDUDPHWHU PHWKRGRORJ\ IRU 4%%7 &KDSWHU SUHVHQWV D V\VWHPDWLF PHWKRGRORJ\ WR LQFOXGH VXEVWUDWH UHVLVWDQFH LQ %7 SDUDPHWHU H[WUDFWLRQ DQG FLUFXLW VLPXODWLRQ &KDSWHU FRQFOXGHV WKH GLVVHUWDWLRQ ZLWK D VXPPDU\ DQG VXJJHVWHG IXWXUH ZRUN
PAGE 17
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
PAGE 18
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f >@ 0DQ\ RI WKH *XPPHO3RRQ PRGHOfV SDUDPHWHUV DUH ELDV GHSHQGHQW 7KH *XPPHO3RRQ PRGHO LQFOXGHV VHYHUDO HPSLULFDO SDUDPHWHUV VXFK DV WKH NQHH FXUUHQW ,. WR DFFRXQW IRU KLJK LQMHFWLRQ HIIHFWV (YHQ WKRXJK WKH *XPPHO3RRQ PRGHO LV WKH PRVW SRSXODU PRGHO HPSOR\HG LQ D ZLGH YDULHW\ RI FLUFXLW VLPXODWRUV LW KDV UHPDLQHG PRVWO\ XQFKDQJHG IRU \HDUV DQG LV LQVXIILFLHQW WR PRGHO WKH DGYDQFHG ELSRODU WUDQVLVWRUV 7KH FRQYHQWLRQDO *XPPHO3RRQ PRGHO KDV WKH IROORZLQJ PDMRU GHILFLHQFLHV f ,W PRGHOV EDVHZLGWK PRGXODWLRQ (DUO\f HIIHFWV XVLQJ FRQVWDQW (DUO\ YROWDJHV 9$) DQG 9$5f ZKLFK UHVXOWV LQ XQDFFHSWDEOH HUURUV LQ VLPXODWLQJ RXWSXW FRQGXFWDQFHV IRU %7V ZLWK WKLQ EDVHV f ,W XVHV FRQVWDQW FROOHFWRU UHVLVWDQFH WKHUHE\ RYHUORRNLQJ FROOHFWRU PRGXODWLRQ DQG TXDVLVDWXUDWLRQ HIIHFWV f ,WV VPDOOVLJQDO HTXLYDOHQW FLUFXLW LV GHULYHG EDVHG RQ TXDVLVWDWLF 46f SULQFLSOHV ,W WKXV IDLOV WR DFFRXQW IRU QRQTXDVLVWDWLF 146f HIIHFWV LQ WUDQVLVWRUV RSHUDWLQJ DW YHU\ KLJK IUHTXHQFLHV f ,W IDLOV WR DFFRXQW IRU VRPH
PAGE 19
H[WULQVLF SDUDVLWLFV ZKLFK DUH QHJOLJLEOH LQ ODUJH GHYLFHV EXW LPSRUWDQW IRU VFDOHG GHYLFHV HJ WKH EDVHHPLWWHU VLGHZDOO FDSDFLWDQFHV f ,W LV QRW VFDODEOH ZLWK JHRPHWU\ 1RYHO &RPSDFW 0RGHOV IRU %LSRODU 7UDQVLVWRUV $ SUHGLFWLYH PRGHO IRU VFDOHG ELSRODU WUDQVLVWRUV PXVW DFFRXQW IRU f PLQRULW\ FDUULHU WUDQVSRUW LQ TXDVLQHXWUDO EDVH UHJLRQ IRU W\SLFDO EDVH GRSLQJ SURILOH DQG DOO LQMHFWLRQ OHYHOV f PDMRULW\ FDUULHU WUDQVSRUW LQ TXDVLQHXWUDO FROOHFWRU DFFRXQWLQJ IRU QRQOLQHDU YROWDJH GURSV DFURVV WKH HSLFROOHFWRU DQG IRU TXDVLVDWXUDWLRQ f IRUZDUG DQG UHYHUVH EDVHZLGWK PRGXODWLRQ (DUO\f HIIHFWV f EDVH ZLGHQLQJ .LUNf HIIHFWV f LPSDFW LRQL]DWLRQ ZLWK FRQVLGHUDWLRQ RI QRQORFDO FDUULHU HQHUJ\ WUDQVSRUW UHODWLRQV f WZR GLPHQVLRQDO JHRPHWULF HIIHFWV VXFK DV VLGHZDOO FXUUHQW LQMHFWLRQ DQG FXUUHQWFURZGLQJ HIIHFWV f SK\VLFDO DQG FRQWLQXRXVO\ GLIIHUHQWLDEOH DF GLIIHUHQWLDO UHVLVWDQFHV DQG FDSDFLWDQFHV >@ DQG f JOREDO DQG ORFDO WKHUPDO HIIHFWV LH VHOIKHDWLQJ 0DQ\ FRPSDFW SK\VLFDO PRGHOV EDVHG RQ WKHVH UHTXLUHPHQWV KDYH EHHQ SURSRVHG LQ WKH SDVW WHQ \HDUV 7KH H[WHQGHG *XPPHO3RRQ PRGHO >@ 9%,& >@ 0(;75$0 >@ DQG WKH 0063&( FKDUJHEDVHG ELSRODU WUDQVLVWRU PRGHO 4%%7f >@ DUH UHYLHZHG LQ WKH IROORZLQJ VHFWLRQV 7KH ([WHQGHG *XPPHO3RRQ 0RGHO 8VLQJ .XOOfV PDMRULW\ FDUULHU WUDQVSRUW HTXDWLRQ LQ WKH TXDVLQHXWUDO FROOHFWRU >@ WKH H[WHQGHG *XPPHO3RRQ PRGHO DFFRXQWV IRU TXDVLVDWXUDWLRQ HIIHFW E\ FDOFXODWLQJ QRQOLQHDU YROWDJH GURS DFURVV WKH TXDVLQHXWUDO FROOHFWRU 7KH H[WHQGHG *XPPHO3RRQ PRGHO LV EDVHG RQ WKH IRUPXODWLRQ RI WKH RULJLQDO *XPPHO3RRQ PRGHO DQG LQKHULWV PRVW
PAGE 20
RI LWV SDUDPHWHUV 7KH H[WHQGHG *XPPHO3RRQ PRGHO LV HDVLO\ LPSOHPHQWHG DQG LV IDPLOLDU WR PRVW FLUFXLW GHVLJQHUV &RQVHTXHQWO\ WKH H[WHQGHG *XPPHO3RRQ PRGHO KDV EHHQ LQFOXGHG LQ D ZLGH YDULHW\ RI FLUFXLW VLPXODWRUV 7R LQFOXGH .XOOfV FROOHFWRU WUDQVSRUW HTXDWLRQ LQ WKH H[WHQGHG *XPPHO3RRQ PRGHO RQH DUWLILFLDO QRGH LV LQWURGXFHG DW WKH EDVHFROOHFWRU MXQFWLRQ :KLOH WKH RULJLQDO LQWHJUDO FKDUJHFRQWUROOHG UHODWLRQ ,&&5f LV HPSOR\HG WR VLPXODWH WKH PLQRULW\ FDUULHU WUDQVSRUW LQ WKH TXDVLQHXWUDO EDVH .XOOfV FROOHFWRU FXUUHQW HTXDWLRQ LV DSSOLHG LQ WKH TXDVLn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f 0(;75$0 GRHV QRW XVH WKH LQWHJUDO FKDUJH FRQWURO UHODWLRQ ,&&5f ,W PRGHOV PLQRULW\ FDUULHU WUDQVSRUW DQG FKDUJH VWRUDJH IRU DQ H[SRQHQWLDOO\ GRSHG EDVH ZKLFK LV FRPPRQ LQ DGYDQFHG ELSRODU WUDQVLVWRUV f 7KH WHUPLQDO FXUUHQWV DQG FKDUJH
PAGE 21
VWRUDJHV RI 0(;75$0 PRGHO DUH IRUPXODWHG DV IXQFWLRQV RI WKH GRSLQJ SURILOH DQG ELDV FRQGLWLRQ f %DVHZLGWK PRGXODWLRQ LH (DUO\ HIIHFWf LV QRW PRGHOHG ZLWK WKH FRQVWDQW (DUO\ YROWDJHV EXW ZLWK WKH EDVHHPLWWHU DQG EDVHFROOHFWRU MXQFWLRQ GHSOHWLRQ FKDUJHV 7KH ODWHVW YHUVLRQ RI WKH 0(;75$0 PRGHO DOVR DGGV D XQLILHG PDMRULW\ FDUULHU WUDQVSRUW HTXDWLRQ LQ WKH FROOHFWRU DFFRXQWLQJ IRU YDULRXV FRQGLWLRQV RI TXDVLVDWXUDWLRQ >@ 7KH 0(;75$0 PRGHO XVHG LQ WKH FLUFXLW VLPXODWRU 3VWDU 3KLOOLSVf DQG 0LFURZDYH 'HVLJQ 6\VWHP +HZOHWW3DFNDUGf KDV PDQ\ DGYDQWDJHV RYHU WKH *XPPHO3RRQ PRGHO IRU VLPXODWLQJ %7fV TXDVLVDWXUDWLRQ DQG KLJK FXUUHQW FKDUDFWHULVWLFV 0063,&(2%%7 0RGHO 7KH 0063&(4%%7 PRGHO ZDV GHYHORSHG E\ HRQJ DQG )RVVXP >@ DW 8QLYHUVLW\ RI )ORULGD :KLOH 4%%7 ZDV EHLQJ ILUVW LPSOHPHQWHG LQ 63,&(* 4%%7 LV QRZ DYDLODEOH LQ DW OHDVW RQH FRPPHUFLDO 63,&( VLPXODWRU 4%%7 LV VLPLODU WR 0(;75$0 LQ PDQ\ DVSHFWV %RWK 4%%7 DQG 0(;75$0 XVH GH *UDDIIfV HTXDWLRQ WR PRGHO PLQRULW\ FDUULHU WUDQVSRUW LQ WKH TXDVLQHXWUDO EDVH >@ DQG .XOOfV HTXDWLRQ WR PRGHO PDMRULW\ HDUQHU WUDQVSRUW LQ WKH TXDVLQHXWUDO FROOHFWRU >@ 4%%7 KRZHYHU KDV VHYHUDO XQLTXH IHDWXUHV 4%%7 XVHV PRYLQJ ERXQGDU\ FRQGLWLRQV IRU GLIIHUHQW ELDV FRQGLWLRQV WKXV LQFOXGLQJ PDQ\ SK\VLFDO HIIHFWV VXFK DV WKH EDVHZLGWK PRGXODWLRQ (DUO\f HIIHFWV DQG WKH EDVH ZLGHQLQJ (DUO\ .LUNf HIIHFW 4%%7 DOVR DOORZV IRU SRVVLEOH 6&5 IRUPDWLRQ LQ HSLFROOHFWRU WKXV HOLPLQDWLQJ WKH QRQSK\VLFDO DVVXPSWLRQ XVHG LQ WKH H[WHQGHG *XPPHO3RRQ PRGHO 4%%7 SK\VLFDOO\ VLPXODWHV %7fV FXUUHQWYROWDJH FKDUDFWHULVWLFV IRU DOO SRVVLEOH RSHUDWLQJ UHJLRQV
PAGE 22
0RVW RI 4%%7fV SDUDPHWHUV DUH GHULYHG IURP GHYLFH VWUXFWXUH OD\RXWVf DQG SURFHVV LQIRUPDWLRQ GRSLQJ SURILOHVf 7KLV PDNHV 4%%7 VXLWDEOH IRU 7&$' DQG FRQFXUUHQW HQJLQHHULQJ DSSOLFDWLRQV ZKHUH D VLPSOH SUHSURFHVVRU 6800f LV XVHG WR JHQHUDWH 4%%7fV SDUDPHWHUV GLUHFWO\ IURP WKH SURFHVV VLPXODWRU RXWSXW RU IURP PHDVXUHG GRSLQJ SURILOHV 4%%7 SUHGLFWV FLUFXLW SHUIRUPDQFH HYHQ LQ WKH DEVHQFH RI IDEULFDWHG GHYLFHV 7KHUHIRUH 4%%7 FDQ EH XVHG LQ FRQFXUUHQW HQJLQHHULQJ DSSOLFDWLRQV ZKLFK HIILFLHQWO\ VKRUWHQV SURGXFWLRQ F\FOHV :LWK SDUDPHWHUV GHULYHG IURP WKH HVWLPDWLRQ RI SURFHVV LQIRUPDWLRQ FLUFXLW GHVLJQV FDQ EHJLQ EHIRUH UHSUHVHQWDWLYH WHVW FKLSV DUH DYDLODEOH 4%%7 LV DOVR VXLWHG IRU VWDWLVWLFDO FLUFXLW GHVLJQ DSSOLFDWLRQV 6LQFH PRVW RI 4%%7fV PRGHO SDUDPHWHUV FRPH GLUHFWO\ IURP WKH GHYLFH VWUXFWXUH DQG SURFHVV LQIRUPDWLRQ VWDWLVWLFDO FRUUHODWLRQV EHWZHHQ FLUFXLW SHUIRUPDQFH DQG SURFHVV YDULDWLRQV DUH DXWRPDWLFDOO\ PDLQWDLQHG 6XFK FRUUHODWLRQV DUH QRW XVXDOO\ PDLQWDLQHG LQ HPSLULFDO PRGHOV VXFK DV *XPPHO3RRQ 0063&( LV PXFK PRUH FRPSXWDWLRQDOO\ HIILFLHQW WKDQ GHYLFH VLPXODWRUV 7KLV PDNHV DFFXUDWH VWDWLVWLFDO DQDO\VHV WKDW UHTXLUH PDQ\ VLPXODWLRQV VXFK DV 0RQWH &DUORf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
PAGE 23
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Â9%,&f 9%,& ZDV SURSRVHG LQ E\ D JURXS RI ,& DQG &$' LQGXVWULDO UHSUHVHQWDWLYHV DV D UHSODFHPHQW IRU WKH FRQYHQWLRQDO *XPPHO3RRQ PRGHO 7KH JURXS HYDOXDWHG FRPSDFW %7 PRGHOV GHYHORSHG VLQFH *XPPHO3RRQ PRGHO 7KHLU REMHFWLYH ZDV WR SLQSRLQW WKH EHVW PRGHO DQG EULQJ LW WR WKH PDLQVWUHDP RI ELSRODU ,& GHVLJQV +RZHYHU QR PRGHOV WKH\ HYDOXDWHG VDWLVILHG DOO WKH UHTXLUHPHQWV IURP WKH FRPPLWWHH VR WKH\ GHFLGHG WR GHYHORS WKHLU RZQ 9%,&fV IRUPXODWLRQ LV VLPLODU WR WKH H[WHQGHG *XPPHO3RRQ PRGHO DOWKRXJK ZLWK VHYHUDO XQLTXH IHDWXUHV f 9%& XVHV EDVHHPLWWHU DQG EDVHFROOHFWRU MXQFWLRQ
PAGE 24
GHSOHWLRQ FKDUJHV WR PRGHO EDVHZLGWK PRGXODWLRQ HIIHFW WKHUHE\ HOLPLQDWLQJ WKH HPSLULFLVP RI XVLQJ FRQVWDQW (DUO\ YROWDJHV f 9%,& DGGV D SDUDVLWLF SQS PRGHOHG ZLWK D VLPSOLILHG *XPPHO3RRQ PRGHO f 7R DYRLG QXPHULFDO SUREOHPV 9%,& XVHV VLQJOHn SLHFH VPRRWK IXQFWLRQV ZKHQHYHU SRVVLEOH WR PRGHO %7fV FXUUHQWV DQG VPDOOVLJQDO FDSDFLWDQFHV f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fV PRGHO IRUPXODWLRQ LV UHYLHZHG DOO SRVVLEOH RSHUDWLRQ PRGHV RI D %7 PXVW EH GHILQHG 'HSHQGLQJ RQ ZKHWKHU D VSDFHFKDUJHUHJLRQ 6&5f LV IRUPHG DW EDVHFROOHFWRU PHWDOOXUJLFDO MXQFWLRQ %7 RSHUDWLRQ LV GLYLGHG LQWR WZR GLVWLQFW UHJLRQV VDWXUDWLRQ DQG IRUZDUG UHJLRQV 'LIIHUHQW VDWXUDWLRQ FRQGLWLRQV FDQ EH GHILQHG EDVHG RQ WKH YHORFLW\ RI PDMRULW\ FDUULHUV LQ HSLFROOHFWRU :KHQ WKH PDMRULW\ FDUULHUV LQ HSLFROOHFWRU PRYH ZLWK D YHORFLW\ SURSRUWLRQDO WR WKH ORFDO HOHFWULFDO ILHOG WKH %7 LV RSHUDWLQJ LQ RKPLF TXDVLVDWXUDWLRQ UHJLRQ DQG WKH HSLFROOHFWRU DFWV OLNH D UHVLVWRU :KHQ WKH PDMRULW\ HDUQHUV UHDFK WKHLU VDWXUDWLRQ YHORFLW\ H[FHVV PDMRULW\ HDUQHUV PXVW EH LQMHFWHG LQWR WKH
PAGE 25
7DEOH &RPSDULVRQ EHWZHHQ FRPSDFW %7 PRGHOV ([WHQGHG *3 0(; 75(0 4%%7 9%,& %DVH DQDO\VLV 8VH *XPPHO3RRQ ,&&5 ; 8VH GH *UDDIIV EDVH WUDQVSRUW HTXDWLRQ ; ; ; 8VH .QHH FXUUHQW WR PRGHO LQMHFWLRQ ; ; $FFRXQW IRU EDVH SXVKRXW ; ; $FFRXQW IRU EDQGJDS YDULDWLRQ LQ EDVH 6L*H +%7 VLPXODWLRQV ; (DUO\ HIIHFW 8VH (DUO\ YROWDJHV ; 8VH GHSOHWLRQ FKDUJHV ; ; 8VH PRYLQJ ERXQGDULHV ; &ROOHFWRU DQDO\VLV 8VH .XOOfV HTXDWLRQ WR PRGHO TXDVLVDWXUDWLRQ ; ; ; ; $FFRXQW IRU YDULRXV 6&5 IRUPDWLRQV LQ HSLFROOHFWRU ; ; $FFRXQW IRU ILHOG GHSHQGHQW PRELOLW\ ; ; $& DQDO\VLV 6PDOOVLJQDO HTXLYDOHQW FLUFXLW GHULYHG ZLWK TXDVLn VWDWLF 46f DSSUR[LPDWLRQ ; ; ; ; %LDV GHSHQGHQW VPDOOVLJQDO GLIIHUHQWLDO UHVLVWDQFHV DQG FDSDFLWDQFHV ; ; ; ; $FFRXQW IRU YHUWLFDO QRQTXDVLVWDWLF 146f HIIHFWV ; ; $FFRXQW IRU ODWHUDO QRQTXDVL VWDWLF 146f HIIHFWV ;
PAGE 26
7DEOH fÂ§ FRQWLQXHG 2WKHU HIIHFWV '& FXUUHQW FURZGLQJ ; 7UDQVLHQW FXUUHQW FURZGLQJ ; 6HOIKHDWLQJ ; ; &ROOHFWRU VSUHDGLQJ HIIHFW ; 7ZR GLPHQVLRQDO JHRPHWULF OD\RXWf HIIHFWV ; *HQHUDO DWWULEXWHV &RQWLQXRXV VPRRWK WHUPLQDO FXUUHQWV ; ; &RQWLQXRXV VPRRWK GLIIHUHQWLDO UHVLVWDQFHV DQG UHVLVWDQFHV ; ; 6FDODEOH SDUDPHWHUV ; 1XPHULFDO UREXVWQHVV ; ;
PAGE 27
TXDVLQHXWUDO FROOHFWRU ZKLFK LQGXFH D VSDFHFKDUJHG UHJLRQ 6&5f QHDU WKH EXULHG OD\HU ,Q WKLV VLWXDWLRQ %7 LV RSHUDWLQJ LQ QRQRKPLF TXDVLVDWXUDWLRQ UHJLRQ 'LIIHUHQW RSHUDWLQJ PRGHV FDQ DOVR EH GHILQHG IRU GLIIHUHQW ELDV FRQGLWLRQV RQ DQ ,Fn9%F SODQH DV VKRZQ LQ )LJ ZKHUH DOO WKH ERXQGDU\ OLQHV DUH GHILQHG DV >@ 'LIIHUHQW UHJLRQV LQ )LJ DUH GHILQHG DV IROORZLQJ ,f VDWXUDWLRQ ,,f RKPLF TXDVLn VDWXUDWLRQ ZKHUH HQWLUH EDVH DQG FROOHFWRU DUH TXDVLQHXWUDO ,,,f QRQRKPLF TXDVLn VDWXUDWLRQ ZKHUH H[FHVVLYH FKDUJH LV LQMHFWHG IURP EDVH UHJLRQ DQG DQ 6&5 IRUPV DW WKH HQG RI HSLFROOHFWRU ,9f IRUZDUGDFWLYH ZKHUH WKH EDVHFROOHFWRU MXQFWLRQ LV UHYHUVH ELDVHG DQG DQ 6&5 IRUPV DW EDVHFROOHFWRU PHWDOOXUJLFDO MXQFWLRQ 9f IRUZDUGDFWLYH ZLWK WKH HQWLUH HSLFROOHFWRU GHSOHWHG GXH WR DGHTXDWHO\ KLJK 9&% ,Q )LJ WKH WKLFN VROLG ERXQGDU\ OLQHV ODEHOHG ,6&& DQG ,6&&:(3, GHILQH WKH RQVHW RI 6&5 IRUPDWLRQ DW EDVHFROOHFWRU MXQFWLRQ $ERYH WKHVH WZR OLQHV QR 6&5 LV IRUPHG DW WKH EDVHFROOHFWRU MXQFWLRQ DQG %7 LV RSHUDWLQJ DW HLWKHU RKPLF RU QRQRKPLF TXDVLVDWXUDWLRQ %HORZ WKHVH WZR OLQHV DQ 6&5 IRUPHG DW EDVHFROOHFWRU MXQFWLRQ DQG WKH %7 HQWHUV IRUZDUGDFWLYH RSHUDWLRQ 7KH WKLFN GDVKHG ERXQGDU\ OLQH T Â‘ $F Â‘ 1HSL Â‘ YVDW LV XVXDOO\ GHILQHG DV WKH RQVHW RI EDVH SXVKRXW +RZHYHU IRU FROOHFWRU FXUUHQWV JUHDWHU WKDQ WKH TXDVLQHXWUDO EDVH GRHV QRW QHFHVVDULO\ H[WHQG LQWR WKH HSLFROOHFWRU $FWXDOO\ EDVH SXVKRXW GHSHQGV RQ WKH FROOHFWRU FXUUHQW OHYHO DV ZHOO DV WKH EDVHFROOHFWRU ELDV YROWDJH UHSUHVHQWV WKH PD[LPXP FXUUHQW WKDW WKH HSLFROOHFWRU FDQ FDUU\ EHIRUH LWV PDMRULW\ FDUULHUV UHDFK WKHLU VDWXUDWLRQ YHORFLW\ DQG DERYH ZKLFK H[WUD FDUULHUV PXVW EH LQMHFWHG LQWR TXDVLQHXWUDO FROOHFWRU WR VXVWDLQ H[FHVV FXUUHQW
PAGE 28
)LJXUH 2SHUDWLRQ PRGH ERXQGDULHV RQ ,&9%& SODQH
PAGE 29
)RUPXODWLRQ RI '& &XUUHQW9ROWDJH (TXDWLRQV 7KH IRUPXODWLRQ RI 4%%7fV FXUUHQWYROWDJH HTXDWLRQV DQG FKDUJH VWRUDJHV LV EDVHG RQ GH *UDDIIfV HTXDWLRQ WR PRGHO PLQRULW\ FDUULHU WUDQVSRUW LQ WKH TXDVLQHXWUDO EDVH >@ DQG .XO7V HTXDWLRQ WR PRGHO PDMRULW\ FDUULHU WUDQVSRUW LQ WKH TXDVLQHXWUDO FROOHFWRU >@ ,Q GHULYLQJ WKH PLQRULW\ FDUULHU WUDQVSRUW DQG FKDUJH VWRUDJH HTXDWLRQV GH *UDDII DVVXPHG TXDVLQHXWUDOLW\ IRU DQ H[SRQHQWLDO EDVH GRSLQJ SURILOH DV QD QDRn H[SQr:%$If f ZKHUH 1DT LV WKH H[WUDSRODWHG SHDN EDVH GRSLQJ U_ LV WKH EDVH GRSLQJ JUDGLQJ IDFWRU DQG :%0 LV WKH EDVH PHWDOOXUJLFDO ZLGWK 7R DFFRXQW IRU LQWHQGHG RU XQLQWHQGHG EDQGJDS YDULn DWLRQV ZKLFK UHVXOW LQ DQ DLGLQJ RU UHWDUGLQJ ILHOG LQ WKH TXDVLQHXWUDO EDVH GH *UDDIIfV FXUn UHQW HTXDWLRQV DUH UHIRUPXODWHG DV QL fOI a Q/5 IrU ARZ LQMHFWLRQ FRQGLWLRQV f Q+ Q+) Q+5 IRU KLJK LQMHFWLRQ FRQGLWLRQV f ZKHUH f H[3QnZEf 1 Q/)'QO7? H[SAn:AOA Q[Ff Q/5 H[SAn:Kf? f f H[SU?:f Q+5 V L'Q+AaSAZ f ;Q^[(f f .+5 a Q[Uf H[SU? :Kf f DQG WKH EDVH GRSLQJ JUDGLQJ IDFWRU U_f DQG Uff DUH GHILQHG DV
PAGE 30
W7 $ (N7f:%0 f LI $(N7:%0 f WR DFFRXQW IRU EDQGJDS YDULDWLRQV 7R HQVXUH D VPRRWK DQG FRQWLQXRXV WUDQVLWLRQ IURP ORZ LQMHFWLRQ WR KLJK LQMHFWLRQ OLQHDU H[WUDSRODWLRQ LV XVHG WR JHW D VLQJOH DQDO\WLF FXUUHQW HTXDWLRQ IRU DOO LQMHFWLRQ OHYHOV B Q/) n 1D[Hf Q+) n Q[(f Q/5 n 1$A;&! Q+5 n Q[&f A _If? 1$[(f Q^[(f 1$[Ff Q[Ff ,Q WKH TXDVLQHXWUDO FROOHFWRU 4%%7 XVHV DQ HTXDWLRQ VLPLODU WR .XOOfV IRUPXODWLRQ IRU WKH FROOHFWRU PDMRULW\ FDUULHU WUDQVSRUW HTXDWLRQ OF a T$?[Q9 71 HSL : 415 .:%&f."g%DfOQ L .9"%&2\ /O.2fVDf A%&2 a A%&Â 9 7 f ZKHUH Â‘ L! DQG :T15 LV WKH ZLGWK RI TXDVLQHXWUDO FROOHFWRU 4%%7 XVHV EDVHHGJH DQG FROOHFWRUHGJH TXDVL)HUPL OHYHO VHSDUDWLRQV DW WKH EDVHFROOHFWRU MXQFWLRQ DQG +nJ4 WR OLQN WKH EDVH DQG FROOHFWRU WUDQVSRUW DQDO\VHV 4%%7 WKHQ VROYHV WKH FROOHFWRU FXUUHQW E\ IRUFLQJ FXUUHQW FRQWLQXLW\ DW WKH EDVHFROOHFWRU MXQFWLRQ :KHQ 7nJ4 LV JUHDWHU WKDQ EDVHFROOHFWRU MXQFWLRQ EDQGHU 2F WKH %7 RSHUDWHV LQ VDWXUDWLRQ RU TXDVLVDWXUDWLRQ PRGHV DQG 9)%&, 7n%&1R 6&5 H[LVWV EHWZHHQ WKH EDVH DQG FROOHFWRU 2Q WKH RWKHU KDQG %7 RSHUDWHV LQ WKH IRUZDUGDFWLYH PRGH ZKHUH 1.J4 LV PRGHOHG WR EH HTXDO WR 2F DQG n)J4 LV PRGHOHG DV
PAGE 31
H[S AEF$ 49EF? N7 U H[SO\UUY H[S T! F N7 H[S IA%& ^ N7 >F f n&5,7 ZKHUH ,FULW GHQRWHV WKH FULWLFDO ERXQGDU\ FXUUHQWV ,VFF RU ,V&5:)3Of VHSDUDWLQJ WKH TXDVLVDWXUDWLRQ DQG IRUZDUGDFWLYH RSHUDWLRQV DV VKRZQ LQ )LJ 4%%7 VROYHV IRU WKH ZLGWKV RI EDVHHPLWWHU DQG EDVHFROOHFWRU 6&5V WR SURYLGH PRYLQJ ERXQGDU\ FRQGLWLRQV DQG WR DFFRXQW IRU EDVHZLGWK PRGXODWLRQ LH (DUO\ HIIHFWf LQ EDVH DQDO\VLV 7KH ZLGWK RI EDVHFROOHFWRU 6&5 LV FDOFXODWHG EDVHG RQ WKH SRWHQWLDO GURS DFURVV WKH EDVHFROOHFWRU 6&5 LH n5EFInA%&Mf :KLOH WDNLQJ LQWR FRQVLGHUDWLRQ WKH EDFNJURXQG FKDUJH PRGXODWLRQ LQGXFHG E\ FROOHFWRU FXUUHQW LQMHFWLRQ 4%%7 FDOFXODWHV WKH ZLGWK RI EDVHFROOHFWRU 6&5 XVLQJ WKH GHSOHWLRQ DSSUR[LPDWLRQ 7KH EDVHHPLWWHU 6&5 ZKLFK FRQWUROV WKH UHYHUVH (DUO\ HIIHFW LV PRGHOHG IRU ERWK IRUZDUG DQG UHYHUVH ELDV :LWK WKH ERXQGDU\ FRQGLWLRQV GHILQHG SK\VLFDOO\ WKH 4%%7 PRGHO SURSHUO\ OLQNV EDVH FROOHFWRU WUDQVSRUW HTXDWLRQV DQG HQVXUHV FXUUHQW FRQWLQXLW\ WKURXJKRXW DOO RSHUDWLQJ UHJLPHV 2%%7fV 6PDOO6LJQDO (TXLYDOHQW &LUFXLW %HIRUH FRQVWUXFWLQJ WKH VPDOOVLJQDO HTXLYDOHQW FLUFXLW FKDUJH VWRUDJHV PXVW EH IRUPXODWHG 7KH IRUPXODWLRQ RI H[FHVV FKDUJH VWRUHG LQ WKH TXDVLQHXWUDO EDVH LV EDVHG RQ GH *UDDIIfV ZRUN :LWK D OLQHDU LQWHUSRODWLRQ IRU LQWHUPHGLDWH LQMHFWLRQ H[FHVV FKDUJH VWRUDJH LQ WKH TXDVLQHXWUDO EDVH LV IRUPXODWHG DV 4QO 1D[Hf f/ 4Q% Q^[(f 1D[Hf Q[(f Â‘ 4 + Q% f ZKHUH / Q % DQG 4 + Q% DUH H[FHVV FKDUJH VWRUDJHV IRU ORZ DQG KLJK LQMHFWLRQ FDVHV
PAGE 32
([FHVV FKDUJH VWRUHG LQ WKH HPLWWHU LV IRUPXODWHG XVLQJ WKH HPLWWHU PLQRULW\ WUDQVLW WLPH 7( DQ HPSLULFDO SDUDPHWHU GHILQHG DV 4H 7($(HR>F[ S9%(97fO@ f ([FHVV FKDUJH VWRUHG LQ WKH EDVHHPLWWHU GHSOHWLRQ UHJLRQ LV PRGHOHG DV ;( 4MH Ln$n Mr1$[fLU f ZKHUH 1$[f LV WKH H[SRQHQWLDOO\ JUDGHG EDVH GRSLQJ SURILOH DQG [( LV WKH EDVHVLGH GHSOHn WLRQ ZLGWK RI EDVHHPLWWHU MXQFWLRQ PRGHOHG IRU ERWK IRUZDUG DQG UHYHUVH ELDV ([FHVV FKDUJH VWRUDJH LQ WKH TXDVLQHXWUDO FROOHFWRU LV PRGHOHG DFFRUGLQJ WR WKH HOHFWULF ILHOG GLVWULEXWLRQ VKRZQ DV )LJ ,Q VDWXUDWLRQ DQG TXDVLVDWXUDWLRQ FDVH $ RI )LJ f FKDUJH VWRUHG LQ WKH XQPRGXODWHG UHJLRQ LV ]HUR EHFDXVH WKH HOHFWULF ILHOG JUDGLHQW LV ]HUR 7KH H[FHVV FKDUJH VWRUHG LQ WKH FROOHFWRU LV RQO\ WKH FKDUJH LQ WKH PRGXODWHG UHJLRQ ZKLFK LV H[SUHVVHG DV 4 415 T $ S 97Qc T$Q a>H[S\VFRfH[S9VFf@ fÂ§: F 1 415 O\(3, f ,Q QRQRKPLF TXDVLVDWXUDWLRQ FDVH % RI )LJ f UHJLRQ WKH H[FHVV FKDUJH LQ WKH FROOHFWRU LV IRUPXODWHG DV WKH VXP RI FKDUJH VWRUHG LQ WKH PRGXODWHG UHJLRQ DV f DQG LQ 6&5 DW WKH HQG RI WKH HSLFROOHFWRU DV 4HSL6&5f a ZVFUnAQHSLaAF$[!VAn f ZKHUH :6&5 LV FDOFXODWHG ZLWK FRQVLGHUDWLRQ RI WKH EDFNJURXQG FKDUJH PRGXODWLRQ FDXVHG E\ FROOHFWRU FXUUHQW LQMHFWLRQ
PAGE 33
(OHFWULFDO ILHOG )LJXUH (OHFWULFDO ILHOG GLVWULEXWLRQ RI HDFK RSHUDWLRQ PRGH $ VDWXUDWLRQ t RKPLF TXDVLVDWXUDWLRQ % QRQRKPLF TXDVLVDWXUDWLRQ & IRUZDUGDFWLYH ZLWK HSLFROOHFWRU QRW IXOO\ GHSOHWHG IRUZDUGDFWLYH ZLWK HQWLUH HSLFROOHFWRU GHSOHWHG
PAGE 34
,Q IRUZDUGDFWLYH PRGH FDVH & RI )LJ f H[FHVV FKDUJH VWRUDJH LQ WKH FROOHFWRU LV IRUPXODWHG DV FKDUJH VWRUHG LQ EDVHFROOHFWRU GHSOHWLRQ UHJLRQ 4HSL6&&f AVFFn T1(SL ,FAVf f f ZKHUH :VFF LV DOVR FDOFXODWHG ZLWK FRQVLGHUDWLRQ RI EDFNJURXQG FKDUJH PRGXODWLRQ FDXVHG E\ FROOHFWRU FXUUHQW LQMHFWLRQ ,Q IRUZDUGDFWLYH PRGH ZLWK WKH HQWLUH HSLFROOHFWRU GHSOHWHG FDVH RI )LJ f LQ DGGLWLRQ WR FKDUJH VWRUHG LQ WKH HSLFROOHFWRU ZKLFK LV PRGHOHG DV f ZLWK :6&( UHSODFHG E\ :(3, H[FHVV FKDUJH LQ WKH FROOHFWRU PXVW LQFOXGH WKH FKDUJH VWRUHG LQ WKH EXULHG OD\HU DV F $ : ISU 4EO 9%&f A9R !Ff f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f f ,F DQG 4(3, :[!Vf Â‘ O4 ZKHUH : LV WKH 6&5 ZLGWK LQ WKH HSLFROOHFWRU
PAGE 35
7KH UHVXOWLQJ FKDUJH VWRUDJHV LQ EDVHHPLWWHU DQG EDVHFROOHFWRU WHUPLQDOV DUH IRUPXODWHG DV 4EH a e% n 4Q% &F Â‘ 4T15 4H 4H3, ( f 4EF a Af f 4fE B &Ff n 4TQU 4HSL 4EO f ZKHUH 4TQU LV VHW WR ]HUR IRU FDVH &f DQG 'f 4%/ LV VHW WR ]HUR IRU FDVH $f %f DQG &f :LWK DOO WHUPLQDO FKDUJHV GHILQHG IRU DOO ELDV FRQGLWLRQV WKH VPDOOVLJQDO HTXLYDOHQW FLUFXLW LV IRUPXODWHG DV VKRZQ LQ )LJ 7KLV VPDOOVLJQDO HTXLYDOHQW FLUFXLW KDV VHYHUDO XQLTXH IHDWXUHV f ,PSOHPHQWDWLRQ RI WKH VPDOOVLJQDO HTXLYDOHQW FLUFXLW LV HDV\ VLQFH DOO WKH UHJLRQDO FKDUJH VWRUDJHV DUH GHULYHG IURP WKH GF VROXWLRQ f 7KH GLIIHUHQWLDO FRQGXFWDQFH LH GOG9 WHUPVf DQG FDSDFLWDQFHV LH G4G9 WHUPVf DUH ELDV GHSHQGHQW f 7KH WUDQVFDSDFLWDQFHV DUH LQFOXGHG H[SOLFLWO\ f 7KH H[WUD GHOD\ SKDVH VKLIWf FDXVHG E\ EDVH ZLGHQLQJ LV LQFOXGHG LPSOLFLWO\ ZLWK 4TQU PRGHOHG LQ TXDVLQHXWUDO FROOHFWRU f 7KH 146 FDUULHU WUDQVSRUW LQVLGH WKH TXDVLQHXWUDO EDVH LV WDNHQ LQWR FRQVLGHUDWLRQ WKURXJK FKDUJH SDUWLWLRQLQJ
PAGE 36
)LJXUH 4%%7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW
PAGE 37
&+$37(5 02'(/ (1+$1&(0(176 2) 7+( 4%%7 02'(/ 6HOIKHDWLQJ 0RGHOLQJ 7KLV FKDSWHU GLVFXVVHV VHYHUDO DGGLWLRQV DQG UHILQHPHQWV WR WKH 0063,&(4%%7 PRGHO 6HOIKHDWLQJ RFFXUV LQ DOO VHPLFRQGXFWRU GHYLFHV DQG FLUFXLWV 6HOIKHDWLQJ LV DQ LQFUHDVH RI GHYLFH ORFDO WHPSHUDWXUH FDXVHG E\ GHYLFHf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
PAGE 38
H[SRQHQWLDOO\ ZLWK DEVROXWH WHPSHUDWXUH VLJQLILFDQW EDVH DQG FROOHFWRU FXUUHQW FKDQJHV LQ ELSRODU GHYLFHV FDQ EH REVHUYHG D VPDOO FKDQJH LQ WHPSHUDWXUH 6HOIKHDWLQJ DIIHFWV DQDORJ FLUFXLWV VLJQLILFDQWO\ E\ LQFUHDVLQJ %7fV RXWSXW FRQGXFWDQFH 7KLV FDQ OHDG WR D VLJQLILFDQW UHGXFWLRQ LQ WKH YROWDJH JDLQ LQ KLJKJDLQ DPSOLILHUV >@ &LUFXLWV WKDW GHSHQG RQ FORVH PDWFKLQJ RI EDVHHPLWWHU YROWDJHV FDQ EH YHU\ VHQVLWLYH WR VHOIKHDWLQJ EHFDXVH VPDOO GLIIHUHQFHV LQ MXQFWLRQ WHPSHUDWXUHV FDQ FDXVH VLJQLILFDQW PLVPDWFK LQ FXUUHQWV 5HFHQW UHVHDUFK GHPRQVWUDWHV WKDW QHJOHFWLQJ VHOIn KHDWLQJ FDQ FDXVH VLJQLILFDQW HUURUV LQ SDUDPHWHU H[WUDFWLRQ >@ 6HOIKHDWLQJ FDQ FDXVH PRUH WKDQ b GLIIHUHQFH LQ H[WUDFWHG YDOXHV RI %7 NQHH FXUUHQW 7KHUHIRUH SURSHU PRGHOV VKRXOG LQFOXGH VHOIKHDWLQJ :KLOH WKH FXUUHQW JDLQ RI KRPRMXQFWLRQ %7V KDV D SRVLWLYH WHPSHUDWXUH FRHIILFLHQW WKH FXUUHQW JDLQ RI KHWHURMXQFWLRQ %7V +%7f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
PAGE 39
)LJXUH 6FKHPDWLF RI WKHUPDO LPSOHPHQWDWLRQ
PAGE 40
,SZ5 a ,Fn9FH ,cn9EH 7KH ORFDO WHPSHUDWXUH LV WKHQ HYDOXDWHG WKURXJK PXOWLSO\LQJ WKH SRZHU GLVVLSDWLRQ E\ WKH WKHUPDO LPSHGDQFH ,Q WKLV ZD\ VHOIKHDWLQJ VLPXODWLRQ FDQ EH PHUJHG LQWR WKH HOHFWULFDO VLPXODWLRQ ,Q WKH LPSOHPHQWDWLRQ RI VHOIn KHDWLQJ DQ H[WUD QRGH 7 ZDV DGGHG LQWR 4%%7 E\ VLPSO\ PRGLI\LQJ 4%%7fV DGPLWWDQFH PDWUL[ LQ 63,&(fV PRGLILHG QRGDO DQDO\VLV 7KH LPSOHPHQWDWLRQ LQYROYHV PLQRU PRGLILFDWLRQV LQ VXEURXWLQHV DQG DGGLWLRQ RI D QHZ VXEURXWLQH WR XSGDWH WHPSHUDWXUHGHSHQGHQW SDUDPHWHUV LQ 4%%7 PRGHO 7HPSHUDWXUH GHSHQGHQW SDUDPHWHUV DUH XSGDWHG EDVHG RQ WKH WHPSHUDWXUH REWDLQHG IURP WKH WKHUPDO QRGH 7 LQ HYHU\ 1HZWRQ 5DSKVRQ LWHUDWLRQ 7KH UHVXOWLQJ RSHUDWLQJ SRLQW LV WKXV FRQVLVWHQW ZLWK GHYLFHfV ORFDO WHPSHUDWXUH ,Q WKLV LPSOHPHQWDWLRQ WKH DPELHQW WHPHUDWXUH LV VLPXODWHG DV D YROWDJH VRXUFH 9WDPE r ALV ZD\ ZH FDQ VLPXODWH D %7 WUDQVLVWRU DW GLIIHUHQW DPELHQW WHPSHUDWXUHV E\ VLPSO\ YDU\LQJ WKH YROWDJH VRXUFH 97DPE 7KLV LPSOHPHQWDWLRQ DOORZV HDVLHU H[WUDFWLRQ RI WKH WHPSHUDWXUH GHSHQGHQFH RI %7 HOHFWULFDO FKDUDFWHULVWLFV WKDQ XVLQJ WKH 7(03 FRQWURO FDUG )RU LQVWDQFH ZH FDQ YDU\ WKH DPELHQW WHPSHUDWXUH IURP r& WR r& ZLWK D VWHS RI r& XVLQJ WKH '& VZHHS FRQWURO DV '& 97$0% 35,17 '& ,9&f ,Q WKLV ZD\ ZH REWDLQ GLUHFWO\ WKH WHPSHUDWXUH GHSHQGHQFH RI WKH FROOHFWRU FXUUHQW 7KLV DOVR DGGV WKH IOH[LELOLW\ WR YDU\ DPELHQW WHPSHUDWXUH LQ IUHTXHQF\ GRPDLQ RU WLPH GRPDLQ DQDO\VHV +RZHYHU WKLV LPSOHPHQWDWLRQ GLVDEOHV WKH RULJLQDO 7(03 FRQWURO FDUG
PAGE 41
)LJXUH 2XWSXW FKDUDFWHULVWLF SORW 6\PERO OLQHV GHQRWH PHDVXUHPHQW GDWD IRU %7 RI SP[_LP HPLWWHU IURP 0RWRUROD PP %L&026 SURFHVV 6ROLG OLQHV DUH WKH 0063&( VLPXODWLRQ UHVXOWV ZLWK VHOIKHDWLQJ 'DVKHG OLQHV DUH VLPXODWLRQ ZLWKRXW VHOIKHDWLQJ
PAGE 42
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fV ZLWK PRGHUDWH HPLWWHU ZLGWKV &XUUHQW FURZGLQJ ZDV ILUVW LQYHVWLJDWHG E\ 3ULWFKDUG LQ ODWH fV >@ %\ FDVFDGLQJ DQ LQILQLWH QXPEHU RI ELSRODU WUDQVLVWRUV ZLWK VHULHV EDVH UHVLVWDQFHV 3ULWFKDUG GHULYHG WKH HTXLYDOHQW EDVH UHVLVWDQFH DV RQH WKLUG RI WRWDO EDVH UHVLVWDQFH +RZHYHU WKLV UHVXOW LV LQYDOLG IRU WKH KLJK EDVHFXUUHQW OHYHOV ZKLFK LV W\SLFDOO\ LQYROYHG LQ FXUUHQW FURZGLQJ $QRWKHU DQDO\VLV RI HPLWWHU FXUUHQW FURZGLQJ ZDV GRQH E\ +DXVHU >@ +DXVHU PDQLSXODWHG WKH GLIIHUHQWLDO FXUUHQWYROWDJH UHODWLRQV DFURVV WKH LQWULQVLF EDVH UHJLRQ :LWK
PAGE 43
(PLWWHU )LJXUH +DOI FURVVVHFWLRQ RI D YHUWLFDO ELSRODU WUDQVLVWRUV
PAGE 44
DSSURSULDWH ERXQGDU\ FRQGLWLRQV IRU WKH SHULSKHU\ DQG DW WKH FHQWHU +DXVHU GHULYHG DQ DQDO\WLFDO VROXWLRQ IRU WRWDO EDVH FXUUHQW +DXVHUfV PHWKRG LV XVHIXO EXW LQDGHTXDWH IRU DGYDQFHG ELSRODU WUDQVLVWRUV EHFDXVH LW DVVXPHV WKDW WKH EDVH VKHHW UHVLVWDQFH LV FRQVWDQW WKXV RYHUORRNLQJ WKH EDVH PDMRULW\ FRQGXFWLYLW\ PRGXODWLRQ ZKLFK RFFXUV LQ KLJK LQMHFWLRQ RU EDVH ZLGHQLQJ 4%%7fV GF FXUUHQW FURZGLQJ PRGHO LV PRVWO\ EDVHG RQ +DXVHUfV DSSURDFK +RZHYHU 4%%7 LQFOXGHV FRQVLGHUDWLRQ RI EDVH ZLGHQLQJ DQG PDMRULW\ FDUULHU PRGXODWLRQ 4%%7 FDOFXODWHV TXDVLQHXWUDO EDVH ZLGWK DQG PDMRULW\ HDUQHU FRQFHQWUDWLRQ IRU HDFK GF ELDV SRLQW 7KH ELDVGHSHQGHQW EDVH VKHHW UHVLVWDQFH WKXV HOLPLQDWHV WKH LQDGHTXDF\ RI +DXVHUf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
PAGE 45
3RVLWLRQ \ cLPf )LJXUH 7ZR GLIIHUHQW PRGHOV WR DSSUR[LPDWH QRQXQLIRUP FXUUHQW GLVWULEXWLRQ DORQJ LQWULQVLF EDVH f HIIHFWLYH EDVHHPLWWHU YROWDJH 9%)HI)f PRGHO f HIIHFWLYH HPLWWHU DUHD $(HW)O PRGHO
PAGE 46
,Q RXU FXUUHQW FURZGLQJ DQDO\VLV WKH EDVHHPLWWHU MXQFWLRQ LV WUHDWHG DV DQ LQILQLWH QXPEHU RI GLRGHV DQG UHVLVWRUV FDVFDGHG LQ VHULHV DV VKRZQ LQ )LJ 7KHUHIRUH WKH EDVHHPLWWHU MXQFWLRQ YROWDJH DV D IXQFWLRQ RI SRVLWLRQ \ Y\f FDQ EH H[SUHVVHG DV Y\f 9%Ha Â rrefn 2 ZKHUH S LV WKH VSHFLILF EDVH UHVLVWLYLW\ DQG LV H[SUHVVHG DV 3 :e Â‘ 3S f 4%E 4TQUfff f f 4EE LV WKH WRWDO FKDUJH RI PDMRULW\ FDUULHU LQ TXDVLQHXWUDO EDVH DQG 4T15 LV WKH WRWDO FKDUJH RI PDMRULW\ FDUULHU LQ H[WHQGHG TXDVLQHXWUDO EDVH 4%% FDQ 4T15 DUH FDOFXODWHG IURP GF RSHUDWLQJ SRLQW WKHUHIRUH PDNLQJ EDVH UHVLVWLYLW\ ELDVGHSHQGHQW )URP )LJ WKH EDVH FXUUHQW DV D IXQFWLRQ RI SRVLWLRQ \ FDQ EH H[SUHVVHG DV ]VWf a ]ILffnf /H (2HIIfn H[S>fA@ f ZKHUH /( LV WKH OHQJWK RI WKH HPLWWHU DUHD DQG (RHIIf LV WKH HIIHFWLYH VDWXUDWLRQ FXUUHQW GHQVLW\ RI GLRGH ZKLFK LV WKH VWHDG\VWDWH GLRGH VDWXUDWLRQ FXUUHQW SOXV WKH WUDQVLHQW WHUP REWDLQHG IURP SUHYLRXV WLPHVWHS 6XEVWLWXWLQJ f LQWR f DQG GLIIHUHQWLDWLQJ WKH UHVXOWLQJ HTXDWLRQ WZLFH ZH REWDLQ GL B G O% S A YU%An % f ZKHUH ZH DVVXPH WKDW H[SY\f9f } 7KH JHQHUDO VROXWLRQ WR f FDQ EH H[SUHVVHG DV L%\f $ Â‘ WDQ f
PAGE 47
ZKHUH $ DQG % DUH WZR XQNQRZQV $W \ :( L%:(f VKRXOG GHFUHDVH WR ]HUR 7KHUHIRUH L%:(f f 6XEVWLWXWLQJ f LQWR f ZH FDQ REWDLQ WKH XQNQRZQ % DQG f EHFRPHV LnQ22 $ WDQ f ZKHUH ] $S:(f97 +HQFH WKH WRWDO EDVH FXUUHQW DW \ FDQ EH H[SUHVVHG DV L%f $ f WDQ]f f 7R VROYH WKH ODVW LQGHSHQGHQW XQNQRZQ $ ZH FDQ VXEVWLWXWH f LQWR WKH ILUVW WZR FRXSOHG LQWHJUDO HTXDWLRQV f DQG f DQG REWDLQ cf /H:((H[S f %\ HTXDWLQJ f DQG f ZH FDQ VROYH $ DQG WKH GHSHQGHQW YDULDEOH ] WKHUHE\ REWDLQLQJ WKH WRWDO EDVH FXUUHQW IRU D JLYHQ EDVHHPLWWHU YROWDJH 9%( 1RWLFH IURP f WKDW FXUUHQW FURZGLQJ PDNHV WKH WRWDO EDVH FXUUHQW VPDOOHU WKDQ WKDW H[SHFWHG IURP ELSRODU WUDQVLVWRUV ZLWK ]HUR EDVH VSUHDGLQJ UHVLVWDQFH ,Q WKH SUHYLRXV YHUVLRQ RI 4%%7 GF FXUUHQW FURZGLQJ LV PRGHOHG ZLWK DQ HIIHFWLYH EDVHHPLWWHU YROWDJH Y%(HIIff WKDW LV %f OHZHMHR][Â‘ f ZKHUH H[39IO(HIIfA H[3A%efn = f n f
PAGE 48
7KH UHVXOW IURP WKH 9%(HIIf DSSURDFK LV VLPLODU WR WKH UHVXOW XVLQJ EDVH UHVLVWDQFH LQ 63,&( %RWK XQGHUHVWLPDWH WKH SHDN FXUUHQW LQMHFWLRQ OHYHO $ PRUH SK\VLFDO PRGHO RI FXUUHQW FURZGLQJ VKRXOG HYDOXDWH HIIHFWLYH HPLWWHU DUHD WKHUHE\ PDLQWDLQLQJ FXUUHQW GHQVLW\ LQFUHDVH LQ WKH SHULSKHU\ GXH WR WKH UHGXFWLRQ LQ HIIHFWLYH HPLWWHU DUHD DV )LJ 7R DFFRXQW IRU FXUUHQW FURZGLQJ HIIHFW XVLQJ HIIHFWLYH HPLWWHU DUHD ZH LQKHULW PRVW RI WKH RULJLQDO PRGHO IRUPXODWLRQV +RZHYHU ZH XVH DQ HIIHFWLYH HPLWWHU DUHD $(HIIf LQVWHDG RI HIIHFWLYH EDVHHPLWWHU YROWDJH 9%(HIIf LQ f 7KH UHVXOWLQJ HLIHFWLYH HPLWWHU DUHD $(HIIf LV $(^HIIf D VLQ]f B e ] a ZHOH VLQ]f ] f :LWK $H UHSODFHG E\ $(HI(f LQ WKH VXEURXWLQH 4%&7 4%%7f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f WKH QHZ FXUUHQW FURZGLQJ PRGHO DFWXDOO\ LPSURYHV FRQYHUJHQFH VWDELOLW\
PAGE 49
9%( 9f )LJXUH &XUUHQW JDLQ YV EDVHHPLWWHU YROWDJH 9%( )URP WKLV ILJXUH ZH FDQ VHH FXUUHQWFURZGLQJ LQGXFHG KLJK LQMHFWLRQ HIIHFW LV EHWWHU PDQLIHVWHG XVLQJ WKH QHZ FXUUHQW FURZGLQJ PRGHO
PAGE 50
)LJXUH 8QLW\FXUUHQWJDLQ IUHTXHQF\ I7 YV EDVH HPLWWHU YROWDJH 9%( )URP WKLV ILJXUH ZH FDQ VHH FXUUHQWn FURZGLQJ LQGXFHG KLJKLQMHFWLRQ HIIHFW LV EHWWHU PDQLIHVWHG XVLQJ WKH QHZ FXUUHQW FURZGLQJ PRGHO
PAGE 51
$& &XUUHQW &URZGLQJ 0RGHOLQJ 6LPSOH 0RGHO )RUPXODWLRQ 4%%7fV DF DQDO\VLV GRHV QRW LQFOXGH VPDOOVLJQDO LQWULQVLF EDVH UHVLVWDQFH ZKHQ GF FXUUHQW FURZGLQJ DQDO\VLV LV DFWLYH ZKLFK XQGHUHVWLPDWHV WKH WRWDO VPDOOVLJQDO EDVH LPSHGDQFH 7R PRGHO VPDOOVLJQDO EDVH UHVLVWDQFH ZH PXVW DFFRXQW IRU GLVWULEXWHG EDVH UHVLVWDQFH DV ZHOO DV EDVHHPLWWHU FDSDFLWDQFH LQFOXGLQJ ERWK GHSOHWLRQ DQG GLIIXVLRQ FDSDFLWDQFHVf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f WKH GLIIXVLRQ FDSDFLWDQFH LQ WKH SHULSKHU\ LV SURSRUWLRQDOO\ KLJKHU WKDQ WKDW LQ WKH FHQWHU 7KHUHIRUH DF FXUUHQW FURZGLQJ LV PRUH VHYHUH ZLWK VLJQLILFDQW GF FXUUHQW FURZGLQJ >@ 7KH SKHQRPHQRQ RI GF FXUUHQW FURZGLQJ ZDV UHSRUWHG PRUH WKDQ \HDUV DJR >@ DQG KDV EHHQ ZLGHO\ VWXGLHG >@ $& FXUUHQW FURZGLQJ KRZHYHU KDV UHFHLYHG PXFK OHVV DWWHQWLRQ WKDQ LWV GF FRXQWHUSDUW $& FXUUHQW FURZGLQJ LV D QRQTXDVLVWDWLF 146f HIIHFW 7KH 146 HIIHFW LV WKH WLPHGHSHQGHQW EHKDYLRU FDQQRW EH REWDLQHG E\ H[WUDSRODWLQJ IURP VWHDG\VWDWH DQDO\VLV ,Q TXDVLVWDWLF 46f FKDUJHEDVHG PRGHOLQJ GLIIHUHQW FKDUJLQJ DQG GLVFKDUJLQJ G\QDPLFV DUH PRGHOHG ZLWK WUDQVLW WLPH FRQVWDQWV ZKLFK DUH FDOFXODWHG IURP D GF VROXWLRQ IRU H[FHVV FKDUJH VWRUDJH )RU LQVWDQFH WKH EDVH WUDQVLW WLPH FDQ EH FDOFXODWHG
PAGE 52
)LJXUH 6PDOOVLJQDO HTXLYDOHQW FLUFXLW RI WKH EDVH UHVLVWDQFH SOXV EDVHHPLWWHU FDSDFLWDQFH SRVLWLRQ \ SPf )LJXUH 1RUPDOL]HG VPDOOVLJQDO DF EDVH FXUUHQW GLVWULEXWLRQ DORQJ WKH LQWULQVLF EDVH &DOFXODWHG IRU W\SLFDO %7 ZLWKRXW VLJQLILFDQW GF FXUUHQW FURZGLQJ
PAGE 53
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
PAGE 54
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cf f ZKHUH @ 8VLQJ WKH VHULHV H[SDQVLRQ RI FRWK[f L [ [ f FDQ EH H[SUHVVHG DV [
PAGE 55
W5f 5\ 5aO Mp&.f f 7KURXJK VLPSOH DOJHEUDLF PDQLSXODWLRQ f FDQ EH IXUWKHU WUDQVIRUPHG DV f 7KHQ WKH VLPSOLILHG HTXLYDOHQW FLUFXLW RI WKH GLVWULEXWHG 5& EDVH LPSHGDQFH LV VKRZQ LQ )LJ 7KH HTXLYDOHQW FLUFXLW RI )LJ LV VLPLODU WR WKDW LQ >@ EXW FRQWDLQV DQ H[WUD FRPSRQHQW RI 65A $W ORZ FXUUHQW WKH EDVHHPLWWHU UHVLVWDQFH 5. LV XVXDOO\ PXFK ODUJHU WKDQ 5E VR WKDW WKLV 5UW WHUP FDQ EH QHJOHFWHG +RZHYHU VLQFH 5UW GHFUHDVHV ZLWK EDVH FXUUHQW WKH PDJQLWXGH RI 5Q FDQ EHFRPH PXFK VPDOOHU WKDQ 5E DW KLJK FXUUHQW 7KHUHIRUH 5VKRXOG EH NHSW LQ WKH HTXLYDOHQW FLUFXLW &r ,I )LJXUH (TXLYDOHQW FLUFXLW RI f
PAGE 56
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f LQ SDUDOOHO EHFDXVH &Q FDQ EH WKRXJKW RI DV DQ RSHQ HOHPHQW 7KLV LPSOLHV WKDW QR VLJQLILFDQW DF FXUUHQW FURZGLQJ RFFXUV DW ORZ IUHTXHQF\ +RZHYHU WKH HIIHFWLYH EDVH LPSHGDQFH EHFRPHV VPDOOHU DW KLJK IUHTXHQF\ EHFDXVH WKH DF FXUUHQW WHQGV WR FURZG DURXQG WKH SHULSKHU\ WKXV QDUURZLQJ WKH HIIHFWLYH EDVH :LWK &UZKLFK VKXQWV WKH HIIHFWLYH GF EDVH UHVLVWDQFH 65 DQG 5E LQ SDUDOOHOf WKH HTXLYDOHQW FLUFXLW RI )LJ SURSHUO\ VLPXODWHV DF FXUUHQW FURZGLQJ +RZHYHU WKH &Q WHUP VKRXOG QRW EH WKRXJKW RI DV D SK\VLFDO FRPSRQHQW EXW DV DQ HPSLULFDO FRPSRQHQW WR DSSUR[LPDWHO\ VLPXODWH WKH QRQTXDVLVWDWLF 146f GHOD\ RU SKDVH VKLIW $FFRXQWLQJ IRU '& &XUUHQW &URZGLQJ LQ $& &XUUHQW &URZGLQJ $QDO\VLV ,Q WKH DERYH GLVFXVVLRQ WKH LPSHGDQFH RI WKH EDVHHPLWWHU SQ MXQFWLRQ ZDV DVVXPHG WR EH XQLIRUPO\ GLVWULEXWHG WKDW LV WKH VPDOOVLJQDO EDVHHPLWWHU UHVLVWDQFH DQG FDSDFLWDQFH 5Q DQG &A DUH FRQVWDQW DORQJ WKH ODWHUDO EDVH GLPHQVLRQ 7KLV DVVXPSWLRQ LV YDOLG RQO\ IRU ORZ FXUUHQWV ZKHUH 9%( LV DOPRVW D FRQVWDQW DORQJ WKH ODWHUDO EDVH LH
PAGE 57
WKHUH LV QR VLJQLILFDQW GF FXUUHQW FURZGLQJ 6LQFH 5Q DQG &M DUH H[SRQHQWLDOO\ ELDV GHSHQGHQW WKH\ DUH QRQXQLIRUPO\ GLVWULEXWHG DORQJ WKH ODWHUDO EDVH GLPHQVLRQ HYHQ IRU YHU\ VPDOO YDULDWLRQV LQ 9%( GXH WR GF FXUUHQW FURZGLQJ 7KH QHW HIIHFW RI WKH QRQXQLIRUPO\ GLVWULEXWHG EDVHHPLWWHU LPSHGDQFH RQ DF FXUUHQW FURZGLQJ FDQ EH GHVFULEHG DV IROORZV 6LQFH WKH GF EDVHHPLWWHU YROWDJH 9%( DW WKH SHULSKHU\ LV KLJKHU WKDQ WKDW DW WKH FHQWHU WKH VPDOOVLJQDO EDVHHPLWWHU LPSHGDQFH LV VPDOOHU WKDQ WKDW LQ FHQWHU WKHUHE\ FDXVLQJ PRUH RI WKH DF EDVH FXUUHQW WHQGV WR IORZ WKURXJK WKH SHULSKHU\ LQVWHDG RI WKH FHQWHU +HQFH GF FXUUHQW FURZGLQJ HQKDQFHV DF FURZGLQJ ,W LV QRW HDV\ WR PRGHO WKH QRQXQLIRUPO\ GLVWULEXWHG EDVHHPLWWHU LPSHGDQFH VLQFH XQOLNH WKH XQLIRUP WUDQVPLVVLRQ QR DQDO\WLF VROXWLRQ H[LVWV 1HYHUWKHOHVV OLWWOH PRGLILFDWLRQ LQ WKH VROXWLRQ RI XQLIRUP WUDQVPLVVLRQ OLQH LV QHHGHG WR DSSUR[LPDWH WKH QRQXQLIRUP WUDQVPLVVLRQ OLQH IRU WKH SXUSRVH RI FRPSDFW PRGHOLQJ )RU D SK\VLFDO DSSUR[LPDWLRQ ZH ILUVW H[DPLQH WKH QHW HIIHFW RI GF FURZGLQJ RQ DF FXUUHQW FRQGXFWLRQ 7KH QHW HIIHFW RI GF FXUUHQW FURZGLQJ LV WR SXVK DF FXUUHQW WRZDUG WKH SHULSKHU\ 7KLV LV HTXLYDOHQW WR FRPSUHVV WKH DUHD ZKHUH PRVW RI WKH DF EDVH FXUUHQW IORZV +HQFH ZH HPSOR\ WKH FRQFHSW RI HIIHFWLYH DUHD IRUPXODWHG IRU GF FXUUHQW FURZGLQJ WR DSSUR[LPDWH WKH QRQXQLIRUP WUDQVPLVVLRQ OLQH 7R DFFRPPRGDWH WKH HIIHFWLYH DUHD IDFWRU LQ WKH SUHYLRXV DF FXUUHQW FURZGLQJ DQDO\VLV ZH DGMXVW WKH HIIHFWLYH OHQJWK RI WKH 5& WUDQVPLVVLRQ OLQH XVHG LQ RXU SUHYLRXV DF FXUUHQW FURZGLQJ DQDO\VLV 7R YHULI\ WKH DF FXUUHQW PRGHO ZH FRPSDUH WKH VROXWLRQV RI D QRQXQLIRUP WUDQVPLVVLRQ OLQH DQG D XQLIRUP WUDQVPLVVLRQ OLQH ZLWK DQ HIIHFWLYH OHQJWK 7KH UHDVRQ ZH XVH WKLV FRPSDULVRQ DV D SUHOLPLQDU\ YHULILFDWLRQ LV WKDW WKH VROXWLRQV RI WUDQVPLVVLRQ OLQHV FDQ EHWWHU LOOXVWUDWH WKH FRQFHSW RI HIIHFWLYH OHQJWK WKDQ QXPHULFDO GHYLFH VLPXODWLRQ WKH
PAGE 58
UHVXOWV RI ZKLFK DUH QRW DOZD\V HDVLO\ LQWHUSUHWHG ZLWK UHJDUG WR WKH LQWULQVLF SK\VLFV :H ZLOO XVH QXPHULFDO GHYLFH VLPXODWLRQ DV D ILQDO FKHFN WR WKH PRGHO LPSOHPHQWDWLRQ 7R GHULYH WKH GLIIHUHQWLDO HTXDWLRQ ZKLFK FRQWUROV FXUUHQWYROWDJH UHODWLRQ RI WKH QRQXQLIRUP WUDQVPLVVLRQ OLQH WKH VPDOOVLJQDO EDVHHPLWWHU DGPLWWDQFH LV PRGHOHG DV \A\f "r1H[S>fÂ§A f f ZKHUH \. 5Q ML!&. \QR LV D FRQVWDQW DQG Q LV WKH LQMHFWLRQ LGHDOLW\ IDFWRU Q HTXDOV WR RQH DW PHGLXP LQMHFWLRQ OHYHO $W KLJK LQMHFWLRQ Q LV EHWZHHQ RQH DQG WZR +RZHYHU IRU VLPSOLFLW\ ZH XVH Q HTXDO WR RQH IRU DOO WKH LQMHFWLRQ OHYHOV 7R REWDLQ WKH EDVHHPLWWHU YROWDJH DV D IXQFWLRQ RI SRVLWLRQ ZH FDQ VXEVWLWXWH f LQWR f DQG REWDLQ \%(\f 9%(f97?Q FRV >] \ 9Uef@ FRV]f f ZKHUH ] LV GHILQHG LQ f 7KHQ XVLQJ f LQ f WKH VPDOOVLJQDO EDVHHPLWWHU LPSHGDQFH DV D IXQFWLRQ RI SRVLWLRQ LV H[SUHVVHG DV \.\f \M n VHF>]O\:(f@< f ZKHUH \ L]R \rRH[S>9ILef9U@>FRV]f@ )URP )LJ ZH FDQ GHULYH D SDLU RI ILUVWRUGHU GLIIHUHQWLDO HTXDWLRQV IRU WKH VPDOOVLJQDO FXUUHQWYROWDJH UHODWLRQ DORQJ WKH ODWHUDO EDVH DV GLG\f fÂ§ \Z\fY\f \QRn Â‘ VHF>]O \:(f?`a Â‘ YK\f f GYG\f aUK Â‘ ccÂ\f G\ f
PAGE 59
6XEVWLWXWLQJ f LQWR f ZH FDQ REWDLQ fÂ§M9IF22 n VHF>]O \:(f@` Â‘ YE\f f G\ 7KH DERYH VHFRQG RUGHU GLIIHUHQWLDO HTXDWLRQ LV VLPLODU WR WKDW RI D XQLIRUP WUDQVPLVVLRQ OLQH EXW GRHV QRW KDYH DQ DQDO\WLFDO VROXWLRQ EHFDXVH RI WKH VHFDQW IXQFWLRQ f FDQ EH VROYHG QXPHULFDOO\ XVLQJ WKH ERXQGDU\ YDOXH PHWKRG 7KH ERXQGDU\ FRQGLWLRQV RI f DUH GHVFULEHG DV IROORZLQJ $W \ YÂ\f O LI D XQLW\ DF YROWDJH VRXUFH LV DSSOLHG $W \ :( LK\f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fV DF DGPLWWDQFH PDWUL[ LQ 63,&(fV PRGLILHG QRGDO DQDO\VLV 0(',&, ZDV XVHG WR YHULI\ WKH DF FXUUHQW FURZGLQJ PRGHO DQG LWV LPSOHPHQWDWLRQ LQ 0063,&(4%%7 PRGHO ,Q VSLWH RI EHLQJ VWULFWO\ VSHDNLQJ D TXDVLVWDWLF 46f VLPXODWRU 0(',&, FDQ EH XVHG WR VLPXODWH WKH QRQTXDVLVWDWLF 146f H[WUD SKDVHVKLIW GXH WR WKH FKDUJH UHGLVWULEXWLRQ LQ WKH GHYLFH 7KH UHDVRQV DUH GHVFULEHG DV IROORZLQJ f 0(',&)V PHVK LV XVXDOO\ VPDOO HQRXJK WKDW WKH WLPH UHTXLUHG IRU WKH FKDUJH WR EH
PAGE 60
SKDVHYE\ff )LJXUH &RPSDULVRQ EHWZHHQ WKH QXPHULFDO VROXWLRQ RI f DQG WKH DQDO\WLFDO VROXWLRQ RI XQLIRUP WUDQVPLVVLRQ OLQH 6\PERO OLQHV UHSUHVHQW WKH H[DFW VROXWLRQ 7KLFN OLQHV UHSUHVHQW WKH DQDO\WLF VROXWLRQ RI WKH WUDQVPLVVLRQ OLQH ZLWK HIIHFWLYH OHQJWK 7KLQ OLQHV UHSUHVHQW WKH DQDO\WLF VROXWLRQ RI WKH WUDQVPLVVLRQ OLQH ZLWK QRUPDO OHQJWK
PAGE 61
IUHTXHQF\ +]f )LJXUH 7KH LQSXW DGPLWWDQFH RI WKH QRQXQLIRUP WUDQVPLVVLRQ OLQH V\PERO OLQHVf WKH XQLIRUP WUDQVPLVVLRQ OLQH ZLWK HIIHFWLYH OHQJWK WKLFN OLQHVf DQG WKH XQLIRUP WUDQVPLVVLRQ OLQH ZLWK QRUPDO OHQJWK WKLQ OLQHVf
PAGE 62
UHGLVWULEXWHG LQ LW FDQ EH DVVXPHG PXFK VPDOOHU WKDQ DQ\ RI GHYLFHfV FKDUDFWHULVWLF WUDQVLW WLPH FRQVWDQWV WKHUHIRUH ZH FDQ DVVXPH WKDW FKDUJHV DUH LQVWDQWDQHRXVO\ UHGLVWULEXWHG LQVLGH D PHVK f 0(',&,f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fV VLPXODWLRQV 1RWLFH WKDW WKH 146 DF FXUUHQW FURZGLQJ DQDO\VLV PD\ QRW EH FRQVLVWHQW ZLWK WKH H[LVWLQJ WUDQVLHQW FURZGLQJ DQDO\VLV 7KH DF DQG WUDQVLHQW FXUUHQW FURZGLQJ PRGHOV HPSKDVL]H GLIIHUHQW GHYLFH SK\VLFV IRU GLIIHUHQW DSSOLFDWLRQV ,Q IDVW VZLWFKLQJ GLJLWDOf DSSOLFDWLRQV HYHQ ZKHQ GF FXUUHQW FURZGLQJ LV QRW VLJQLILFDQW WKH WUDQVLHQW EDVH FXUUHQW
PAGE 63
)UHTXHQF\ *+]f )LJXUH &RPSDULVRQ EHWZHHQ WKH 0(',&, VLPXODWHG EDVH LQSXW DGPLWWDQFH DQG WKH 0063,&( VLPXODWHG EDVH LQSXW DGPLWWDQFH ZLWK DQG ZLWKRXW DF FXUUHQW FURZGLQJ 7KH V\PERO OLQH UHSUHVHQWV WKH 0(',&, VLPXODWLRQ 7KH WKLFN VROLG OLQH UHSUHVHQWV WKH 0063,&( VLPXODWLRQ ZLWK DF FXUUHQW FURZGLQJ 7KH WKLQ OLQH UHSUHVHQWV WKH 0063,&( VLPXODWLRQ ZLWKRXW DF FXUUHQW FURZGLQJ
PAGE 64
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
PAGE 65
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fV LQSXW SDUDPHWHU DUH GHULYHG IURP WKH FROOHFWRU DQG EDVHGRSDQW SURILOHV 7KH PRGHO SDUDPHWHU H[WUDFWLRQ FDQ EH XVHG WR LQYHVWLJDWH WKH GHYLFH VWUXFWXUH 'HYLFHfV
PAGE 66
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f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f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fV IDEULFDWLRQ SURFHVV
PAGE 67
&XUUHQW JDLQ )LJXUH &XUUHQW JDLQ YV FROOHFWRU FXUUHQW GHQVLW\ F IRU GLIIHUHQW HPLWWHU VL]HV
PAGE 68
I7 *+]f )LJXUH 8QLW\FXUUHQWJDLQ IUHTXHQF\ I7 YV FROOHFWRU FXUUHQW GHQVLW\ F IRU GLIIHUHQW HPLWWHU VL]HV
PAGE 69
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fÂ§
PAGE 70
7DEOH 0RGHO FDUG SDUDPHWHUV 1DPH 'HVFULSWLRQ 8QLWV 'HIDXOW 7\SLFDO 81(3, (OHFWURQ PRELOLW\ LQ HSL FROOHFWRU FPm?7mVn H H 1(3, (SL GRSLQJ GHQVLW\ FPn H H :(3, (SL FROOHFWRU ZLGWK P H H :%0 0HWDOOXJLFDO EDVH ZLGWK P H H 1$2 ([WUDSRODWHG SHDN EDVH GRSLQJ FPn H H (7$ %DVH GRSLQJ JUDGLHQW $O 3UHH[SRQHQWLDO FRHIILFLHQW IRU LPSDFW LRQL]DWLRQ UDWH FPn H %O ([SRQHQWLDO FRHIILFLHQW IRU LPSDFW LRQL]DWLRQ UDWH 9}FP H 96 (OHFWURQ VDWXUDWHG GULIW YHORFLW\ FP}Vn H H (2 (PLWWHU VDWXUDWLRQ FXUUHQW GHQVLW\ $rPn 2H 2H 6(2 %( 6&5 VDWXUDWLRQ FXUUHQW GHQVLW\ $}Pn H H 1(% %( 6&5 HPLVVLRQ FRHIILFLHQW :6(2 =HURELDV %( 6&5 ZLGWK P 2H 2H 3( %( MXQFWLRQ SRWHQWLDO EDUULHU 9 0( %( MXQFWLRQ H[SRQHQWLDO IDFWRU f 3& %& MXQFWLRQ SRWHQWLDO EDUULHU 9 '1% (OHFWURQ GLIIXVLYLW\ LQ EDVH FP 36 &6 MXQFWLRQ EDUULHU 9 06 &6 MXQFWLRQ H[SRQHQWLDO IDFWRU &6 =HURELDV &6 MXQFWLRQ FDSDFLWDQFH )}FPn 2H &5%, ,QWULQVLF EDVH UHVLVWDQFH FRHIILFLHQW 9rVrFPn H H 7% &DUULHU OLIHWLPH LQ EDVH V 2H H
PAGE 71
7DEOH 0RGHO FDUG SDUDPHWHUV &RQWLQXHGf 7& &DUULHU OLIHWLPH LQ FROOHFWRU V 2H H &,) )RUZDUG FXUUHQW FRXSOLQJ FRHIILFLHQW &,5 5HYHUVH FXUUHQW FRXSOLQJ FRHIILFLHQW f 5& ([WULQVLF FROOHFWRU UHVLVWDQFH : 5% ([WULQVLF EDVH UHVLVWDQFH : 5( ([WULQVLF HPLWWHU UHVLVWDQFH Z )% %DVHFKDUJH SDUWLWLRQ IDFWRU )& &ROOHFWRUFKDUJH SDUWLWLRQ IDFWRU f 7( (PLWWHU KROH WUDQVLW WLPH V 2H H (23 3HULSKHUDO %( 6&5 VDWXUDWLRQ FXUUHQW GHQVLW\ $}P H 1(%3 3HULSKHUDO %( 6&5 HPLVVLRQ FRHIILFLHQW 83%$6( +ROH PRELOLW\ LQ LQWULQVLF EDVH FP}9f}Vn /%( :LGWK RI LQWULQVLF EDVH UHJLRQ P H &23 6XEVWUDWH VDWXUDWLRQ FXUUHQW GHQVLW\ $rPn H '1%+ (OHFWURQ GLIIXVLYLW\ LQ EDVH IRU KLJK LQMHFWLRQ FPrV '(*( %DQGJDS UHGXFWLRQ LQ EDVH HPLWWHU MXQFWLRQ H9 '(*& %DQGJDS UHGXFWLRQ LQ EDVH FROOHFWRU MXQFWLRQ H9 :(0 (PLWWHU MXQFWLRQ GHSWK P '(*$ $YHUDJH EDQGJDS UHGXFWLRQ LQ HPLWWHU H9
PAGE 72
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fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW *HQHUDOO\ WKHUH DUH WZR GLIIHUHQW DSSURDFKHV WR H[WUDFW %7fV VPDOOVLJQDO FLUFXLW f 1XPHULFDO RSWLPL]DWLRQ ZKLFK REWDLQV SDUDPHWHUV WKURXJK ILWWLQJ WKH HTXLYDOHQW FLUFXLW WR PHDVXUHG VSDUDPHWHUV f 'LUHFW H[WUDFWLRQ ZKLFK REWDLQV SDUDPHWHUV VWHS E\ VWHS WKURXJK PDQLSXODWLQJ WKH VPDOOVLJQDO SDUDPHWHUV ] \ RU K SDUDPHWHUVf GHULYHG IURP WKH HTXLYDOHQW FLUFXLW 2SWLPL]DWLRQ LV D QXPHULFDO WHFKQLTXH XVHG WR GHWHUPLQH RQH RU PRUH PRGHO SDUDPHWHUV IURP PHDVXUHPHQW 2SWLPL]DWLRQ LV HVSHFLDOO\ XVHIXO LQ WKH ILQH WXQLQJ RI PXOWLSOH SDUDPHWHUV WR REWDLQ WKH EHVW PRGHO ILW WR PHDVXUHPHQWV +RZHYHU WKH RSWLPL]DWLRQ RI PXOWLSOH SDUDPHWHUV LV XVXDOO\ QRW GHVLUDEOH IRU H[WUDFWLQJ SDUDPHWHUV IRU D
PAGE 73
SK\VLFDO PRGHO EHFDXVH LW RIWHQ UHVXOWV LQ QRQXQLTXH RU QRQSK\VLFDO SDUDPHWHU VHWV ZKHQ JLYHQ GLIIHUHQW LQLWLDO JXHVVHV IRU SDUDPHWHUV 7KHUH DUH VHYHUDO PHWKRGV WKDW SUHYHQW RSWLPL]DWLRQ IURP JHQHUDWLQJ QRQSK\VLFDO SDUDPHWHU VHWV f SURYLGLQJ FORVHU LQLWLDO JXHVVHV EDVHG RQ GF SDUDPHWHU H[WUDFWLRQ f LPSRVLQJ SK\VLFDO FRQVWUDLQWV f XVLQJ VWHS E\VWHS GHHPEHGGLQJ WHFKQLTXHV ZLWK VSHFLDO WHVWLQJ VWUXFWXUHV >@ DQG f XVLQJ DGYDQFHG RSWLPL]DWLRQ DOJRULWKPV VXFK DV WKH VWLPXODWHG GLIIXVLRQ WHFKQLTXH >@ ,Q FRQWUDVW GLUHFW H[WUDFWLRQV DYRLG JHQHUDWLQJ QRQXQLTXH SDUDPHWHU VHWV 7KHUHIRUH D GLUHFW H[WUDFW PHWKRGRORJ\ LV SUHIHUUHG LQ REWDLQLQJ D XQLTXH VHW RI SDUDPHWHUV IRU D SK\VLFDO PRGHO $ GLUHFW H[WUDFWLRQ PHWKRG LV DOVR PRUH FRPSXWDWLRQDOO\ HIILFLHQW WKDQ JOREDO RSWLPL]DWLRQ %DVHG RQ WKH KSDUDPHWHUV DQG ]SDUDPHWHUV GHULYHG IURP %7fV K\EULG7 HTXLYDOHQW FLUFXLW VRPH GLUHFW H[WUDFWLRQ PHWKRGV KDYH EHHQ UHSRUWHG >@ >@ +RZHYHU WKHVH PHWKRGV DUH QRW XVHIXO WR GHWHUPLQH 4%%7fV VPDOO VLJQDO SDUDPHWHUV EHFDXVH WKH\ XVH WKH RYHUVLPSOLILHG K\EULG7 PRGHO 7KH H[WUDFWLRQ RI VPDOOVLJQDO SDUDPHWHUV LV YHU\ VHQVLWLYH WR WKH WRSRORJ\ RI WKH HTXLYDOHQW FLUFXLW 4%%7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW LV EDVHG RQ WKH K\EULG7 HTXLYDOHQW FLUFXLW VKRZQ LQ )LJ 5HFHQWO\ D PHWKRGRORJ\ WR H[WUDFW K\EULGL HTXLYDOHQW FLUFXLW ZDV UHSRUWHG >@ 7KH PHWKRG ZDV DOVR HPSOR\HG WR LQYHVWLJDWH VPDOOVLJQDO EDVH LPSHGDQFH >@ 7KH PHWKRG LV EDVHG RQ WKH IDFW WKDW %7fV VXEVWUDWH SDUDVLWLFV DUH QRW VLJQLILFDQW DW ORZ IUHTXHQFLHV 8VLQJ WKLV IDFW D %7fV ORZIUHTXHQF\ ]SDUDPHWHUV FDQ EH DQDO\WLFDOO\ GHULYHG IURP WKH VLPSOLILHG ORZIUHTXHQF\ HTXLYDOHQW FLUFXLW VKRZQ LQ WKH GDVKHG ER[ LQ )LJ 7KURXJK PDQLSXODWLRQ RI WKH DQDO\WLF IRUP RI %7fV ORZIUHTXHQF\ ]SDUDPHWHUV VPDOOVLJQDO UHVLVWDQFHV DQG FDSDFLWDQFHV VXFK DV UEE UHH UFF DQG &FF FDQ EH H[SUHVVHG H[SOLFLWO\ +RZHYHU >@ LV QRW FRPSOHWO\ GLUHFW H[WUDFWLRQ PHWKRG EHFDXVH LW GRHV QRW
PAGE 74
)LJXUH 7KH 4%%7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW
PAGE 75
SURYLGH D GLUHFW ZD\ WR H[WUDFW RU FDOFXODWH %7fV EDVHHPLWWHU FDSDFLWDQFH &Q ,Q DGGLWLRQ WKLV PHWKRG UHTXLUHV D ILQDO RSWLPL]DWLRQ WR REWDLQ WKH RSWLPDO YDOXH IRU WKH EDVHFROOHFWRU FDSDFLWDQFH SDUWLWLRQ IDFWRU ZKLFK DFFRXQWV IRU WKH 146 HIIHFWV FDXVHG E\ WKH GLVWULEXWHG EDVH LPSHGDQFH 7KLV FKDSWHU SUHVHQWV D FRPSOHWH SDUDPHWHU H[WUDFWLRQ PHWKRGRORJ\ IRU WKH 0063,&(4%%7 PRGHO :LWK WKH H[WUDFWLRQ PHWKRGRORJ\ 4%%7fV SDUDPHWHUV FDQ QRZ EH HYDOXDWHG HLWKHU IURP SURFHVV LQIRUPDWLRQ RU IURP PHDVXUHG GDWD 7KH H[WUDFWLRQ PHWKRGRORJ\ DOVR LQFOXGHV WKH H[WUDFWLRQ RI %7f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fV SK\VLFDO SDUDPHWHUV VXFK DV WKH FROOHFWRU DQG EDVH GRSLQJ SURILOHV 6HFWLRQ GLVFXVVHV WKH H[WUDFWLRQ RI 4%%7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW 6HFWLRQ GLVFXVVHV UHPDLQLQJ SDUDPHWHUV DQG FRQFOXGHV WKLV FKDSWHU ZLWK D VXPPDU\ ([WUDFWLRQ RI *XPPHO3RRQ/LNH 3DUDPHWHUV 6RPH RI WKH 4%%7 SDUDPHWHUV DUH DQDORJRXV WR VLPLODU SDUDPHWHUV LQ WKH *XPPHO3RRQ PRGHO DQG FDQ EH H[WUDFWHG IURP D *XPPHO SORW >@ 6DWXUDWLRQ FROOHFWRU
PAGE 76
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f DUH UHTXLUHG LQ VXEVHTXHQW H[WUDFWLRQ DQG RSWLPL]DWLRQV RI FROOHFWRU DQG EDVH GRSLQJ SURILOHV %HFDXVH 4%%7 FDOFXODWHV WKH YROWDJH GURSV DFURVV WKH HSLFROOHFWRU DQG LQWULQVLF EDVH EDVH UHJLRQ XQGHUQHDWK WKH HPLWWHUf WKHUHE\ LQFOXGLQJ WKH ELDVGHSHQGHQW LQWULQVLF FROOHFWRU UHVLVWDQFH DQG EDVH UHVLVWDQFH 5& DQG 5% UHSUHVHQW WKH H[WULQVLF SDUW RI WKH WRWDO FROOHFWRU DQG EDVH UHVLVWDQFHV 7KH LQWULQVLF FROOHFWRU UHVLVWDQFH XVXDOO\ GRPLQDWHV WKH WRWDO FROOHFWRU UHVLVWDQFH LQ WKH IRUZDUGDFWLYH UHJLRQ EXW LW EHFRPHV QHJOLJLEOH LQ KDUG VDWXUDWLRQ 7KH H[WULQVLF FROOHFWRU UHVLVWDQFH FDQ WKHUHIRUH EH H[WUDFWHG IURP WKH VORSHV RI ,T YV 9FH FXUYHV LQ KDUG VDWXUDWLRQ L H QHDU ,F f >@ (PLWWHU DQG EDVH UHVLVWDQFHV FDQ EH H[WUDFWHG VLPXOWDQHRXVO\ XVLQJ 1LQJ DQG 7DQJfV PHWKRG >@ $VVXPLQJ ,% GHJUDGHV RQO\ GXH WR WKH 9%( GURS DFURVV WKH EDVH DQG HPLWWHU UHVLVWDQFHV WKH SRWHQWLDO GURS FDQ EH H[SUHVVHG DV 97 Â‘ ,Q^,%,%f,& 5( 5( 5%f f 7KHUHIRUH WKH LQWHUFHSW RI WKH 97 Â‘ ?Q,%,%ff,F YV FXUYH JLYHV 5( DQG WKH VORSH JLYHV 5(5%f $ W\SLFDO EDVH UHVLVWDQFH YV ELDV SORW H[WUDFWHG XVLQJ WKLV PHWKRG LV VKRZQ LQ )LJ 7KH LQWULQVLF EDVH UHVLVWDQFH SUHGRPLQDWHV DW ORZ LQMHFWLRQ EXW LW EHFRPHV QHJOLJLEOH DW KLJK FXUUHQWV EHFDXVH RI EDVH PDMRULW\ FKDUJH PRGXODWLRQ DQG
PAGE 77
)LJXUH ([WUDFWHG EDVH UHVLVWDQFH YV 9%( SORW VKULQNLQJ RI WKH HIIHFWLYH HPLWWHU ZLGWK GXH WR FXUUHQW FURZGLQJ 7KHUHIRUH WKH H[WULQVLF EDVH UHVLVWDQFH FDQ EH HVWLPDWHG IURP WKH PLQLPXP KLJKLQMHFWLRQf YDOXH RI WKH WRWDO EDVH UHVLVWDQFH ([WUDFWLRQ RI 'RSLQJ 3URILOHV &ROOHFWRU 'RSLQJ 3URILOH &ROOHFWRU SDUDPHWHUV :(3, HSLOD\HU WKLFNQHVVf DQG 1(3, HSLOD\HU GRSLQJf GHWHUPLQH WKH HSLOD\HU UHVLVWDQFH ZKLFK LQ WXUQ FRQWUROV WKH TXDVLVDWXUDWLRQ ,9 FKDUDFWHULVWLFV 1(3, LV GLUHFWO\ UHODWHG WR EDVHFROOHFWRU EUHDNGRZQ YROWDJH DV ZHOO DV WKH WKUHVKROG YROWDJH RI 3026 GHYLFHV LQ Q ZHOO %L&026 WHFKQRORJLHV >@ 1(3, DOVR FRQWUROV WKH RQVHW RI EDVH SXVKRXW DQG WKH ZLGWK RI WKH EDVHFROOHFWRU 6&5 DQG WKXV
PAGE 78
VWURQJO\ DIIHFWV FROOHFWRU WUDQVLW WLPH $V EDVH ZLGWKV DUH SURJUHVVLYHO\ VFDOHG GRZQ FROOHFWRU WUDQVLW WLPH EHFRPHV DQ LQFUHDVLQJO\ VLJQLILFDQW SDUW RI WKH WRWDO WUDQVLW WLPH DQG KHQFH VWURQJO\ DIIHFWV KLJKIUHTXHQF\ DF FKDUDFWHULVWLFV 6LQFH WKH HOHFWULFDO SHUIRUPDQFH RI D ELSRODU WUDQVLVWRU RSHUDWLQJ LQ TXDVLn VDWXUDWLRQ LV SULPDULO\ FRQWUROOHG E\ LWV FROOHFWRU VWUXFWXUH ZH FDQ H[WUDFW FROOHFWRU SDUDPHWHUV E\ RSWLPL]LQJ WKHP WR ILW WKH 4%%7 PRGHO SUHGLFWLRQ WR WKH TXDVLVDWXUDWLRQ ,9 FKDUDFWHULVWLFV 8QOLNH PRVW FRPSDFW %7 PRGHOV IRU H[DPSOH WKH H[WHQGHG *XPPHO3RRQ PRGHOf PRGHO WKH TXDVLVDWXUDWLRQ RSHUDWLRQ E\ DVVXPLQJ WKDW WKH HQWLUH HSLFROOHFWRU LV TXDVLQHXWUDO 4%%7 DFFRXQWV IRU DOO SRVVLEOH 6&5 IRUPDWLRQ LQ WKH HSL FROOHFWRU 4%%7 VROYHV GH *UDDIIfV EDVH WUDQVSRUW HTXDWLRQ >@ DQG .XOOfV FROOHFWRU WUDQVSRUW HTXDWLRQ >@ VXEMHFW WR PRYLQJ ERXQGDU\ FRQGLWLRQV >@ 7KHUHIRUH 4%%7 SURSHUO\ VLPXODWHV FXUUHQWYROWDJH FKDUDFWHULVWLFV DFFXUDWHO\ LQ TXDVLVDWXUDWLRQ 7R LPSURYH WKH VWDELOLW\ DQG FRQYHUJHQFH UDWH RI WKH RSWLPL]DWLRQ 3& %& MXQFWLRQ SRWHQWLDO EDUULHUf DQG 5(3c HSLOD\HU UHVLVWDQFHf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f FDQ EH UHODWHG WR 3& DQG 5(3, DV f 3& f ,Q1(3,Qcf 5HSc :(3,T Â‘ 81(3 Â‘ $F Â‘ 1(3,f f
PAGE 79
1$2 1(3, Â‘ H[S(7$f f &DOFXODWLQJ :%0 PHWDOOXUJLFDO EDVH ZLGWKf DQG '1% DYHUDJH PLQRULW\ GLIIXVLYLW\ LQ EDVHf UHTXLUHV D YDOXH IRU WKH WRWDO EDVH GRSLQJ 4% 7KLV SDUDPHWHU FDQ EH H[WUDFWHG IURP WKH IRUZDUG (DUO\ YROWDJH LI VHOIKHDWLQJ LV HOLPLQDWHG >@ 7KH (DUO\ YROWDJH 9IR GHILQHG DV 9$ ,FfG,FG9&% FDQ EH H[SUHVVHG DV >@ 4ER 4MH a 4MF f $ HVnAVFE AVFFf ZKHUH HV LV WKH VLOLFRQ SHUPLWWLYLW\ :6&% DQG :VFF DUH WKH EDVHVLGH DQG FROOHFWRUVLGH GHSOHWLRQ ZLGWKV RI WKH EDVHFROOHFWRU MXQFWLRQ DQG 4( DQG 4& DUH WKH GHSOHWLRQ FKDUJHV RI WKH HPLWWHU DQG FROOHFWRU MXQFWLRQV 8QGHU ORZLQMHFWLRQ FRQGLWLRQV L H QHJOHFWLQJ EDFNJURXQG FKDUJH PRGXODWLRQ LQGXFHG E\ FROOHFWRU FXUUHQWf :VFF DQG :6&% FDQ EH FDOFXODWHG XVLQJ WKH GHSOHWLRQ DSSUR[LPDWLRQ 4& FDQ EH FDOFXODWHG DV T Â‘ 1(3, Â‘ : VFF Â‘ :LWK WKH EDVHHPLWWHU GHSOHWLRQ ZLGWK HVWLPDWHG DVVXPLQJ D RQHVLGH DEUXSW MXQFWLRQ 4( FDQ EH REWDLQHG E\ LQWHJUDWLQJ WKH EDVH GRSDQW DFURVV WKH EDVHHPLWWHU GHSOHWLRQ UHJLRQ 6LQFH 9$ LV ZHOOGHILQHG IRU HYHU\ LWHUDWLRQ YDOXH RI 4% 4%R FDQ EH REWDLQHG E\ RSWLPL]LQJ LW WR ILW WKH 9$9&% FXUYH :LWK 4E REWDLQHG IURP WKH SUHYLRXV VWHS :%0 DQG '1% FDQ EH FDOFXODWHG DV :%0 4ER (7$ 1$2 H[S ^(7$f F6 n 1J T f QLH f '1% f
PAGE 80
ZKHUH 1* LV WKH *XPPHO QXPEHU GHILQHG DV L9* 4ERa4MHa 4MFA4 f DQA A&6 LV H[WUDFWHG IURP WKH *XPPHO SORW DV GLVFXVVHG LQ 6HFWLRQ :LWK DOO WKH FRUUHODWHG SDUDPHWHUV UHFDOFXODWHG IRU HYHU\ LWHUDWLRQ WKH FROOHFWRU SDUDPHWHUV FDQ EH REWDLQHG IURP RSWLPL]LQJ WKHP WR ILW WKH PRGHO SUHGLFWLRQ WR WKH PHDVXUHPHQW GDWD +RZHYHU )URP f LW LV FOHDU WKDW YDULDWLRQ RI WKH EDQGJDS FDQ FULWLFDOO\ DIIHFW WKH FDOFXODWHG YDOXH RI '1% )RU D FRQWHPSRUDU\ ELSRODU WUDQVLVWRU ZLWK D YHU\ WKLQ DQG H[SRQHQWLDOO\ GRSHG EDVH WKH EDVH GRSLQJ DW WKH HPLWWHU VLGH FDQ EH DV KLJK DV FPn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f FDQ EH REWDLQHG XVLQJ (LQVWHLQfV UHODWLRQ ,I WKH LQWHUPHGLDWH YDOXHV RI '1% JR RXW RI WKH YDOLG UDQJH WKH EDQGJDS YDULDWLRQ SDUDPHWHUV FDQ EH ILQHWXQHG LQ DFFRUGDQFH ZLWK WKH FDOFXODWLRQ RI >@ 7KH IORZFKDUW RI WKH RSWLPL]DWLRQ SURFHGXUHV RI FROOHFWRU SDUDPHWHUV LV VKRZQ LQ )LJ
PAGE 81
7XQH EDQGJDS YDULDWLRQ SDUDPHWHUV $VVLJQ (7$ ,QLWLDO JXHVV RI 3& DQG 5(3c &DOFXODWH 1(3, :(3, 1$2 ([WUDFW 4%2 E\ ILWWLQJ 9$9&% FXUYH :LWK *XPPHO QXPEHU NQRZQ &DOFXODWH :%0 DQG '1% &DOFXODWH QH[W 3& DQG 5(3L XVLQJ /HYHQEXUJ 0DUTXDUW DOJRULWKP ,QYRNH 0063,&(4%%7 VLPXODWLRQ &DOFXODWH WKH VXP RI VTXDUH HUURUV H )LJXUH )ORZFKDUW RI FROOHFWRU GRSLQJ H[WUDFWLRQ
PAGE 82
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f FDQ EH HVWLPDWHG IURP WKH XQLW\ FXUUHQWJDLQ IUHTXHQF\ XVLQJ [7 N Â‘ I7f )RU QDUURZ EDVH ZLGWKV [F DQG [( FDQ EHFRPH FRPSDUDEOH WR [% DQG PXVW EH FDUHIXOO\ DFFRXQWHG IRU LQ RUGHU WR FRUUHFWO\ SUHGLFWM ,Q LWV UHJLRQDO FKDUJH DQDO\VLV 4%%7 PRGHOV [F DV 4F,F ZKHUH 4F f J9eSX4 8VLQJ 1(3, REWDLQHG IURP WKH FROOHFWRUSURILOH H[WUDFWLRQ :VFF FDQ EH FDOFXODWHG EDVHG RQ WKH GHSOHWLRQ DSSUR[LPDWLRQ PRGLILHG WR DFFRXQW IRU WKH EDFNJURXQG FKDUJH PRGXODWLRQ LQGXFHG E\ FROOHFWRU FXUUHQW ZKLFK DOORZV DQ DFFXUDWH HVWLPDWH RI [F 4%%7 XVHV WKH HPSLULFDO SDUDPHWHU 7( HPLWWHU PLQRULW\FDUULHU WUDQVLW WLPHf WR DFFRXQW IRU [( [( FDQ EH UHODWHG WR 7( DV [e 4((,& 7( Â‘ ,%f,Fa7( ZKHUH 4HH LV WKH H[WUD FKDUJH VWRUDJH LQ WKH HPLWWHU ,I WKH HPLWWHU MXQFWLRQ LV YHU\ VKDOORZ SPf DQG KDV D PHWDO FRQWDFW WKH H[WUD FKDUJH VWRUDJH LQ HPLWWHU LV JHQHUDOO\ YHU\ VPDOO DQG
PAGE 83
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
PAGE 84
FXUYHV 1H[W WKH HPLWWHU WUDQVLW WLPH b( LV HVWLPDWHG XVLQJ WKH IRUPXOD LQ >@ DQG WKH EDVHGRSDQW JUDGLQJ IDFWRU (7$ LV H[WUDFWHG IURP RSWLPL]DWLRQ RII7 YV 9%( FXUYHV DV LQ 6HFWLRQ ,,, $V D ILQDO VWHS WKH ]HURELDV EDVHHPLWWHU 6&5 ZLGWK :6(2 LV WXQHG WR ILW > YV 9%( FXUYHV ([WUDFWLRQ 5HVXOWV DQG 'LVFXVVLRQ 7KH SDUDPHWHU H[WUDFWLRQ SURFHGXUHV DUH WHVWHG RQ GRXEOHSRO\ VHOIDOLJQHG %7 GHYLFHV IURP D 0RWRUROD MDP %L&026 SURFHVV 7KUHH VHWV RI GDWD DUH WDNHQ 7KH\ DUH f *XPPHO SORWV HJ ORJ,Ff DQG ORJ,%f YV 9%( ZLWK 9&( DV SDUDPHWHU f ,F YV 9(( SORWV ZLWK 9%( DV SDUDPHWHU f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aVW>O VQf f Vf Ab@ >@ IM YV 9%( FXUYH FDQ EH REWDLQHG IURP H[WUDSRODWLQJ YV IUHTXHQF\ SORW WR XQLW\ JDLQ DW HDFK 9%(
PAGE 85
,Q VSLWH RI WDNLQJ HIIHFWV LQWR FRQVLGHUDWLRQ>@ HJ VLGHZDOO FXUUHQWV DQG H[WULQVLF SDUDVLWLF FDSDFLWDQFHV WKH 4%%7 PRGHO LV YLUWXDOO\ D PRGHO 7R PLQLPL]H WKH LQIOXHQFH RI HIIHFWV RQ SDUDPHWHUV H[WUDFWLRQ PHDVXUHPHQWV DUH GRQH RQ D FRQVLGHUDEO\ ODUJH SP [ SPf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b ODUJHU WKDQ WKDW LQGLFDWHG E\ 6,06 ZKLFK OHG WR DERXW D HUURU LQ WKH SHDN GRSLQJ 1$2 7KLV GLVFUHSDQF\ FDQ EH H[SODLQHG E\ QRWLQJ WKDW WKH 6,06 PHDVXUHPHQWV ZHUH SHUIRUPHG RQ YHU\ ODUJH DUHD GHYLFHV %HFDXVH RI WZRGLPHQVLRQDO GLIIXVLRQ HIIHFWV WKH HPLWWHU MXQFWLRQ RI VPDOO GHYLFHV LV H[SHFWHG WR EH VKDOORZHU WKDQ WKDW RI ODUJH GHYLFHV 7KHUHIRUH WKH EDVH RI WKH VPDOO GHYLFH VKRXOG EH ZLGHU DQG FRQWDLQ PRUH GRSDQWV WKDQ SUHGLFWHG E\ 6,06
PAGE 86
7DEOH &RPSDULVRQ EHWZHHQ 6,06 DQG H[WUDFWLRQ UHVXOWV 1(3, :(3, 1$2 (7$:%0 :%0 6,06 H H SP H SPn SP ([WUDFWLRQ H SP H SPn SP 'LIIHUHQFHb fÂ§ b b b b 7KH VLPXODWLRQV ZLWK H[WUDFWHG SDUDPHWHUV DUH VKRZQ LQ )LJ WKURXJK )LJ $ QHZ YHUVLRQ RI WKH 4%%7 PRGHO ZKLFK LQFOXGHV VHOIKHDWLQJ DQG PRGLILHG FXUUHQW FURZGLQJ PRGHO LV XVHG WR VLPXODWH GHYLFH FKDUDFWHULVWLFV ZLWK WKH H[WUDFWHG SDUDPHWHUV 7KH WKHUPDO LPSHGDQFH LV FDOFXODWHG WKHRUHWLFDOO\ DV LQ >@ RU HYDOXDWHG DV GLVFXVVHG LQ >@ 1RWLFH WKDW WKH H[WUDFWHG SK\VLFDO SDUDPHWHUV VXFK DV EDVH DQG FROOHFWRU GRSLQJ SURILOHV DUH DOVR DSSOLFDEOH WR WKH %7fV RI WKH VDPH VWUXFWXUH EXW ZLWK GLIIHUHQW HPLWWHU IHDWXUH VL]HV )RU LQVWDQFH 1$2 :%0 (7$ 1(3, DQG :(3, H[WUDFWHG IURP WKH SP[SP GHYLFH FDQ EH XVHG IRU WKH SP[SP GHYLFH ZLWKRXW UHSHDWLQJ WKH FRPSOHWH H[WUDFWLRQ SURFHGXUHV 7KLV VXJJHVWV WKDW WKH PRGHO LV WUXO\ VFDODEOH DQG WKXV FDQ UHGXFH WKH SDUDPHWHU H[WUDFWLRQ HIIRUWV VLJQLILFDQWO\ 2QH ERQXV IHDWXUH RI WKH SDUDPHWHU H[WUDFWLRQ PHWKRGRORJ\ LV WKDW ZLWK WKH SDUDPHWHUV FDOLEUDWHG ZLWK PHDVXUHPHQWV WKH ILUVWRGHU GHULYDWLYH GLVFRQWLQXLWLHV LQ 4%%7fV FROOHFWRU FXUUHQW FDQ EH UHPRYHG 'HSHQGLQJ RQ WKH H[LVWHQFH RI EDVHFROOHFWRU 6&5 4%%7 XVHV WZR GLIIHUHQW VHWV RI HTXDWLRQV WR FDOFXODWH WKH FROOHFWRU FXUUHQW 7KLV FDQ FDXVHV WKH ILUVWRUGHU GHULYDWLYH GLVFRQWLQXLW\ LQ 4%%7fV FROOHFWRU FXUUHQW ZKHQ RSHUDWLRQ WUDQVLWV EHWZHHQ TXDVLVDWXUDWLRQ DQG IRUZDUGDFWLYH PRGHV LI JLYHQ
PAGE 87
Df SP[_LP Ef SP[JP )LJXUH *XPPHO SORW ORJOFf t ORJO%f YV 9%( 6\Pn ERO OLQHV DUH WKH PHDVXUHPHQW GDWD 6ROLG OLQHV DUH VLPXODn WLRQ UHVXOWV
PAGE 88
Df _LP[_LP Ef JP[_LP ,& $f )LJXUH &XUUHQW JDLQ 3 YV 9%( 6\PERO OLQH LV WKH PHDn VXUHPHQW GDWD 6ROLG OLQH LV VLPXODWLRQ ZLWK FXUUHQW FURZGn LQJ HIIHFW 'DVKHGOLQH LV VLPXODWLRQ ZLWKRXW FXUUHQW FURZGLQJ
PAGE 89
,F P$f F P$! Df _LP[MLP Ef _DP[SP 9&( 9f )LJXUH ,F YV 9&( SORW 6\PERO OLQHV DUH WKH PHDVXUHn PHQW GDWD 6ROLG OLQHV DUH VLPXODWLRQ UHVXOWV ZLWK VHOIKHDWn LQJ 'DVKHGOLQH DUH VLPXODWLRQ UHVXOWV ZLWKRXW VHOIKHDWLQJ
PAGE 90
)LJXUH I7 YV 9%( SORW 6\PERO OLQH LV WKH PHDVXUHPHQW GDWD 6ROLG OLQH LV VLPXODWLRQ ZLWK FXUUHQW FURZGLQJ HIIHFW 'DVKHGOLQH LV VLPXODWLRQ ZLWKRXW FXUUHQW FURZGLQJ
PAGE 91
LQDSSURSULDWH SDUDPHWHUV %HFDXVH WKH FROOHFWRU SDUDPHWHUV DUH REWDLQHG IURP RSWLPL]DWLRQ ZKLFK LQ WXUQ FRQIRUP WKH PRGHO SUHGLFWLRQ WR ILW WKH PHDVXUHPHQW GDWD WKH GLVFRQWLQXLW\ SUREOHP FDQ EH DYRLGHG DV VKRZQ LQ )LJ ([WUDFWLRQ RI 2%%7fV 6PDOO6LJQDO (TXLYDOHQW &LUFXLW 0HWKRGRORJ\ ,Q SUHYLRXV VHFWLRQV PRVW RI 4%%7fV SDUDPHWHUV ZHUH H[WUDFWHG IURP GF FXUUHQW YROWDJH FKDUDFWHULVWLFV ORZIUHTXHQF\ &9 PHDVXUHPHQWV DQG I7 SORWV 7KH UHVXOWLQJ PRGHO SDUDPHWHUV JLYH YHU\ DFFXUDWH SUHGLFWLRQV RI %7fV GF FKDUDFWHULVWLFV +RZHYHU VRPH RI WKHVH SDUDPHWHUV PLJKW QRW EH RSWLPDO IRU DF VLPXODWLRQV EHFDXVH LVRODWLRQ RI WKHLU LQGLYLGXDO HIIHFWV IURP GF PHDVXUHPHQW LV GLIILFXOW )RU LQVWDQFH WKH EDVH DQG HPLWWHU UHVLVWDQFHV FDQ QRW EH DFFXUDWHO\ GHWHUPLQHG WKURXJK GF PHDVXUHPHQWV EHFDXVH WKHLU HIIHFWV DUH FRXSOHG WRJHWKHU DW GF $OVR IRU DFFXUDWH DF VLPXODWLRQ ZH VWLOO QHHG WR NQRZ WKH FKDUJH SDUWLWLRQ IDFWRUV ZKLFK DUH XVHG LQ 4%%7 WR HPSLULFDOO\ DFFRXQW IRU WKH QRQTXDVLVWDWLF 146f FDUULHU WUDQVSRUW LQ WKH TXDVLQHXWUDO EDVH 7KLV VHFWLRQ SUHVHQWV D QHZ PHWKRGRORJ\ WR GLUHFWO\ H[WUDFW 4%%7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW IURP ZKLFK ZH FDQ HYDOXDWH WKH RSWLPDO YDOXH RI 5( 5% DQG FKDUJH SDUWLWLRQ IDFWRUV IRU DFFXUDWH DF VLPXODWLRQ 7KH H[WUDFWLRQ PHWKRGRORJ\ EHJLQV ZLWK WKH FDOFXODWLRQ RI WKH GLIIHUHQWLDO FXUUHQW JDLQ DQG EDVHHPLWWHU UHVLVWDQFH IURP WKH GF FXUUHQW YROWDJH FKDUDFWHULVWLFV $W PHGLXP WR KLJKLQMHFWLRQ OHYHOV WKH EDVH FXUUHQW LV GRPLQDWHG E\ WKH PLQRULW\ FDUULHU GLIIXVLRQ FXUUHQW LQ WKH HPLWWHU JLYHQ DV AE a AVRH[3 9EH AUff f ZKHUH 9%(f LV LQWHUQDO ELDV YROWDJH 7KHUHIRUH U. FDQ H[SUHVVHG DV
PAGE 92
,F P$f G,FG9 P$9f )LJXUH 6LPXODWHG ,F YV 9&( DQG LWV ILUVW GHULYDWLYH
PAGE 93
U. ^G,%G9%(nf 97,% f %HFDXVH WKH LGHDOLW\ H[SRQHQWLDOf IDFWRU RI FROOHFWRU FXUUHQW LV QRW QHFHVVDULO\ HTXDO WR XQLW\ DW KLJK LQMHFWLRQ ZH FDQQRW GLUHFWO\ REWDLQ JP E\ GLUHFW GLIIHUHQWLDWLRQ RI WKH FROOHFWRU FXUUHQW ,F ZLWK UHVSHFW WR WKH EDVHHPLWWHU YROWDJH 9%(f ,QVWHDG ZH VKRXOG XVH f ZKHUH DF LV ORZIUHTXHQF\ FRPPRQHPLWWHU FXUUHQW JDLQ ZKLFK FDQ EH REWDLQHG IURP %7fV GF FROOHFWRU DQG EDVH FXUUHQWV ,Q >@ JP LV H[SUHVVHG DV P PR H[3 f f ZKHUH WKH H[SRQHQWLDO WHUP LV XVHG WR DFFRXQW IRU WKH SKDVHVKLIW FDXVHG E\ WKH QRQTXDVLn VWDWLF 146f PLQRULW\ WUDQVSRUW LQ WKH TXDVLQHXWUDO EDVH 6LQFH VXFK D 146 HIIHFW KDV EHHQ DFFRXQWHG IRU LQ 4%%7 ZLWK WKH FKDUJH SDUWLWLRQ WHFKQLTXH ZH XVH f LQVWHDG RI f 6LQFH FRQWHPSRUDU\ VLOLFRQ ELSRODU WUDQVLVWRUV DUH XVXDOO\ EXLOW RQ D YHU\ OLJKWO\ GRSHG VXEVWUDWH RU HSLOD\HU WKH UHVXOWLQJ VXEVWUDWH SDUDVLWLFV DUH LQVLJQLILFDQW DW ORZ IUHTXHQFLHV :LWK WKHVH VXEVWUDWH SDUDVLWLFV QHJOHFWHG DW ORZ IUHTXHQF\ %7 ]SDUDPHWHUV ZLWKRXW FRQVLGHUDWLRQ RI SUREH SDG SDUDVLWLFV DUH H[SUHVVHG DV A EE UHH A 6PAL]A = a UHH JP=Qf ]? UHH L=.OP=.f@APD&FFf ] UFF UHH =W p&FFf JP=Qf f f f f ZKHUH =b U. MD!UQ&.f 6XEWUDFWLQJ ] IURP ]c \LHOGV WKH EDVH UHVLVWDQFH UEE DV
PAGE 94
6XEWUDFWLQJ ] IURP ] \LHOGV WKH EDVHFROOHFWRU FDSDFLWDQFH &FF DV &FF O>FR f ,P] =?f@ Â‘ f 7R H[WUDFW HPLWWHU UHVLVWDQFH UHH ZH PXOWLSO\ WKH QXPHUDWRU DQG GHQRPLQDWRU RI WKH VHFRQG WHUP RI f E\ JP DQG REWDLQ UHH 5H]DDFJPf f ZKHUH DDF DFDF f LV WKH ORZIUHTXHQF\ FRPPRQEDVH FXUUHQW JDLQ 7R FRPSXWH WKH FROOHFWRU UHVLVWDQFH UFF ZH FDQQRW GLUHFWO\ VXEWUDFW ] IURP ] EHFDXVH ] FRQYHUWHG IURP VSDUDPHWHUV LV XVXDOO\ XQUHDVRQDEO\ ODUJH DW ORZ IUHTXHQFLHV >@ 7R FRPSXWH FROOHFWRU UHVLVWDQFH UFF ZH VXEWUDFW ] IURP ] DQG REWDLQ UFF 5H =O =^ MQFFF9LJP]. 7KH DERYH HTXDWLRQ FRQWDLQV DQRWKHU XQNQRZQ &Q 7R FDOFXODWH &Q ZH XVH %7fV ILUVW GRPLQDQW SROH H[WUDFWHG IURP WKH XQLW\JDLQ IUHTXHQF\ I7 DV 3GRPLQDQWAD$A.IUf f )URP WKH K\EULGW HTXLYDOHQW FLUFXLW RI )LJ %7fV ILUVW GRPLQDQW SROH LV H[SUHVVHG DV 3GRPLQDQW &QUQ &F$UW] PU.AUHH UFFA f ff 1RWLFH WKDW f DQG f DUH LQGHSHQGHQW DQG FRQWDLQ RQO\ WZR XQNQRZQV UFF DQG &MM 7KHUHIRUH ZH FDQ VROYH UFF DQG &Q IURP f DQG f :LWK UQ JP UEE UHH UFF &FF DQG &Q FDOFXODWHG ZLWK f f DQG f f %7fV ORZIUHTXHQF\ HTXLYDOHQW FLUFXLW LH WKH FLUFXLW LQ WKH GDVKHGER[ LQ )LJ f LV GLUHFWO\ H[WUDFWHG IURP ORZIUHTXHQF\ ]SDUDPHWHUV 7KLV HTXLYDOHQW FLUFXLW ILWV FORVHO\ WR PHDVXUHG VPDOOVLJQDO SDUDPHWHUV DW ORZ IUHTXHQF\ +RZHYHU LW GRHV QRW DFFXUDWHO\ VLPXODWH %7fV VPDOOVLJQDO SDUDPHWHUV DW KLJK IUHTXHQFLHV EHFDXVH WKLV HTXLYDOHQW FLUFXLW GRHV QRW WDNH VXEVWUDWH SDUDVLWLFV LQWR FRQVLGHUDWLRQV 7KHUHIRUH ZH H[WUDFW VXEVWUDWH
PAGE 95
SDUDVLWLFV IURP WKH GLVFUHSDQF\ EHWZHHQ WKH VLPXODWHG DQG PHDVXUHG GDWD DW KLJK IUHTXHQFLHV 7KH H[WUDFWLRQ RI VXEVWUDWH SDUDVLWLFV EHJLQV ZLWK WKH FDOFXODWLRQ RI ] RI WKH ORZ IUHTXHQF\ HTXLYDOHQW FLUFXLW 7KHQ WKH FDOFXODWHG ] LV FRPSDUHG ZLWK WKH PHDVXUHG ] )URP WKH GLIIHUHQFH WKH VXEVWUDWH SDUDVLWLFV DUH H[WUDFWHG DV f ZKHUH ]f LV WKH VLPXODWHG ] RI WKH ORZIUHTXHQF\ HTXLYDOHQW FLUFXLW DQG ]f LV WKH PHDVXUHG ] ,Q SUDFWLFDO VLWXDWLRQV RQO\ 5VXE LV H[WUDFWHG IURP WKLV VWHS :H XVH WKH &FV PHDVXUHG ZLWK ORZIUHTXHQF\ &9 PHDVXUHPHQWV DQG H[WUDFW 5VXE IURP ILWWLQJ f WR WKH PHDVXUHG ] IURP 0+] WR *+] 6LQFH WKH RSWLPL]DWLRQ FRQWDLQV RQO\ RQH YDULDEOH ZH HQVXUH WKH XQLTXHQHVV RI 5VXE DQG PDLQWDLQ LWV SK\VLFDO PHDQLQJ 7KH H[WUDFWLRQ PHWKRGRORJ\ RI %7 VXEVWUDWH UHVLVWDQFH ZLOO DOVR EH XVHG LQ WKH QH[W FKDSWHU WR YHULI\ WKH FDOFXODWLRQ RI %7 VXEVWUDWH UHVLVWDQFH &DOLEUDWLRQ RI 2%%7f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
PAGE 96
[$ WR P$ DQG 9&( IURP O 9 WR 9 7KH UHVXOWLQJ VSDUDPHWHUV ZHUH GHHPEHGGHG ZLWK PHDVXUHPHQWV RI RSHQ VWUXFWXUHV 7KHQ %7fV ]SDUDPHWHUV ZHUH FRQYHUWHG IURP WKH GHHPEHGHG VSDUDPHWHUV 7KH EDVHHPLWWHU UHVLVWDQFH U. DQG WUDQVFRQGXFWDQFH JP DUH FDOFXODWHG GLUHFWO\ IURP %7fV GF ELDV FXUUHQWV XVLQJ f DQG f 7KHUHIRUH WKH UA DQG JP DUH FRQVLVWHQW ZLWK WKH SUHGLFWLRQ RI 4%%7 XVLQJ WKH SDUDPHWHUV H[WUDFWHG IURP &KDSWHU 2WKHU FRPSRQHQWV DUH H[WUDFWHG ZLWK WKH SURFHGXUHV GHVFULEHG SUHYLRXVO\ )LJXUH VKRZV WKH H[WUDFWLRQ RI UHH 4%%7 PRGHOV WKH HPLWWHU UHVLVWDQFH ZLWK D FRQVWDQW SDUDPHWHU 5( 5( FDQ EH H[WUDFWHG IURP 1LQJ7DQJf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fV KLJKIUHTXHQF\ VSDUDPHWHUV LV DOPRVW D
PAGE 97
)LJXUH 7KH H[WUDFWHG VPDOOVLJQDO HPLWWHU UHVLVWDQFH UHH IRU GLIIHUHQW ELDV FRQGLWLRQV )LJXUH 7KH H[WUDFWHG VPDOOVLJQDO FROOHFWRU UHVLVWDQFH UFF IRU GLIIHUHQW ELDV FRQGLWLRQV
PAGE 98
FRQVWDQW ZKHQ ,% LV JUHDWHU WKDQ D WKUHVKROG YDOXH :KHQ D %7 LV RSHUDWLQJ DW IRUZDUG DFWLYH UHJLRQ LWV LQWULQVLF FROOHFWRU UHVLVWDQFH FDQ EH DSSUR[LPDWHG DV WKH HSLFROOHFWRU UHVLVWDQFH 7KHUHIRUH 5& FDQ EH H[WUDFWHG DV 5& 5&WRWD>M 5HSL f ZKHUH 5(3, :(3,T Â‘ 81(3 Â‘ $F Â‘ 1(3,f )LJXUH LOOXVWUDWHV WKH H[WUDFWLRQ RI WRWDO EDVH UHVLVWDQFH UEE 1RWLFH WKDW UEE GHFUHDVHV ZLWK %7f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fÂ§ O[X LQWULQVLFf &5%, :( :( 4 f %% ZKHUH :( LV WKH HPLWWHU ZLGWK 4%% LV WKH WRWDO EDVH PDMRULW\ FKDUJH DQG &5%, LV DQ LQSXW SDUDPHWHU ,Q WKLV FDVH WKH LQWULQVLF EDVH UHVLVWDQFH LV D FRQVWDQW IRU DOO ELDV FRQGLWLRQV &5%, FDQ EH WKHRUHWLFDOO\ FDOFXODWHG DV &5%, Q Â‘ 83%$6(f f
PAGE 99
)LJXUH 7KH H[WUDFWHG VPDOOVLJQDO EDVH UHVLVWDQFH UEE IRU GLIIHUHQW ELDV FRQGLWLRQV ZKHUH 83%$6( LV WKH PDMRULW\ PRELOLW\ LQ WKH EDVH DQG Q LV HTXDO WR IRU VLQJOH EDVH FRQWDFW FRQILJXUDWLRQV DQG IRU GRXEOH EDVH FRQWDFW FRQILJXUDWLRQV :LWK 4%% FDOFXODWHG IURP WKH WRWDO EDVH GRSLQJ IRU ORZ DQG PHGLXP LQMHFWLRQ FRQGLWLRQV &5%, FDQ DOVR EH WXQHG WR ILW WKH DF H[WUDFWHG WRWDO EDVH UHVLVWDQFH (YDOXDWLRQ RI 2%%7fV &KDUJH 3DUWLWLRQ )DFWRUV )LJXUH VKRZV WKH H[WUDFWLRQ RI WKH EDVHHPLWWHU DQG EDVHFROOHFWRU FDSDFLWDQFHV &Q DQG &FF &FF LV URXJKO\ D FRQVWDQW DW ORZ FXUUHQWV EXW EHFRPHV SURSRUWLRQDO WR &Q IRU EDVH FXUUHQWV JUHDWHU WKDQ O22ML$ 7KLV SKHQRPHQRQ LV FDXVHG E\ QRQTXDVLVWDWLF 146f PLQRULW\ WUDQVSRUW LQ WKH TXDVLQHXWUDO EDVH 7KH GHULYDWLRQ RI WKH HTXLYDOHQW FLUFXLW RI )LJ LV PDLQO\ EDVHG RQ WKH TXDVLVWDWLF 46f DSSUR[LPDWLRQ
PAGE 100
,E $f )LJXUH 7KH H[WUDFWHG VPDOOVLJQDO EDVHHPLWWHU DQG EDVHFROOHFWRU FDSDFLWDQFHV &Q DQG &FF IRU GLIIHUHQW ELDV FRQGLWLRQV
PAGE 101
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f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f DW YHU\ KLJK FXUUHQW
PAGE 102
)LJXUH ([WUDFW WKH FKDUJH SDUWLWLRQ IDFWRU E\ FRPSDUn LQJ WKH &FF DQG &FF&Qf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
PAGE 103
6LO 6 6 6 )LJXUH 7KH VLPXODWHG VSDUDPHWHUV FRPSDUHG ZLWK WKH PHDVXUHG VSDUDPHWHUV 6\PERO OLQHV UHSUHVHQW PHDVXUHn PHQW 6ROLG OLQHV UHSUHVHQW VLPXODWLRQ
PAGE 104
0DJ\f 0DJ\f 0DJ\Of 0DJ\OOf )LJXUH 7KH VLPXODWHG \SDUDPHWHUV FRPSDUHG ZLWK WKH PHDVXUHG \SDUDPHWHUV 6\PERO OLQHV UHSUHVHQW PHDn VXUHPHQW 6ROLG OLQHV UHSUHVHQW VLPXODWLRQ
PAGE 105
)LJXUH 7KH VLPXODWHG XQLW\JDLQ IUHTXHQF\ I7 FRPn SDUHG ZLWK WKH I7 FRPSXWHG IURP WKH PHDVXUHG VSDUDPHWHU GDWD 6\PERO OLQHV UHSUHVHQW PHDVXUHPHQW 6ROLG OLQHV UHSn UHVHQW VLPXODWLRQ
PAGE 106
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fSDUDPHWHU H[WUDFWLRQf PRGH ZKHUH WKH GRSLQJ SURILOH SDUDPHWHUV DUH LQIHUUHG E\ ILQGLQJ WKH EHVW ILW WR HOHFWULFDO PHDVXUHPHQWV 7R IXUWKHU FDOLEUDWH WKH H[WULQVLF UHVLVWDQFH SDUDPHWHUV IRU DFFXUDWH DF VLPXODWLRQ D QHZ PHWKRGRORJ\ WR H[WUDFW 4%%7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW LV GHYHORSHG 7KH PHWKRG GLUHFWO\ H[WUDFWV %7fV K\EULGW HTXLYDOHQW FLUFXLW ZLWK XVLQJ DQ\ RSWLPL]DWLRQ RI PXOWLSOH SDUDPHWHUV 7KHUHIRUH WKLV PHWKRG LV PRUH FRPSXWDWLRQDOO\ HIILFLHQW WKDQ DQ\ JOREDO RSWLPL]DWLRQ WHFKQLTXHV DQG HQVXUHV WKH XQLTXHQHVV RI WKH H[WUDFWHG VPDOOVLJQDO HTXLYDOHQW FLUFXLW :LWK WKH H[WUDFWLRQ RI 4%%7fV VPDOOVLJQDO HTXLYDOHQW FLUFXLW ZH SUHVHQW D PHWKRG WR H[WUDFW %7fV FKDUJH SDUWLWLRQ IDFWRUV IRU WKH ILUVW WLPH 8S WR QRZ ZH KDYH GHYHORSHG D V\VWHPDWLF PHWKRGRORJ\ WR H[WUDFW DOO WKH 4%%7 SDUDPHWHUV OLVWHG LQ 7DEOH DQG 7DEOH H[FHSW $, %, &,) &,5 &23 (23 1(%3
PAGE 107
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
PAGE 108
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
PAGE 109
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f RI D ORZ QRLVH DPSOLILHU /1$f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n ULQJ VXEVWUDWH FRQWDFW FDQ EH SODFHG IRU HDFK ELSRODU WUDQVLVWRU ZLWK D PLQLPXP FRQWDFWWR GHYLFH VSDFLQJ 7KLV DSSURDFK LV UHOLDEOH DQG HOLPLQDWHV PRVW RI WKH PRGHOLQJ SUREOHPV
PAGE 110
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n GLPHQVLRQDO FXUUHQW WKH VXEVWUDWH FXUUHQW LV WKUHHGLPHQVLRQDO 7KLV FXUUHQW IORZ PXVW EH FRQVLGHUHG LQ WKH FDOFXODWLRQ RI VXEVWUDWH UHVLVWDQFH 2XU DSSURDFK LV WR H[WHQG *KDUSXUH\f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
PAGE 111
FRPSXWDWLRQDOO\ LQHIILFLHQW 5HFHQWO\ *KDUSXUH\ GHYHORSHG D TXLFN DOJRULWKP WR FDOFXODWH GHYLFHWRGHYLFH VXEVWUDWH UHVLVWDQFHV 7KH DOJRULWKP HPSOR\V WKH GLVFUHWH *UHHQ IXQFWLRQ DQG )DVW )RXULHU 7UDQVIRUP ))7f >@ 6LPXODWLRQV VKRZ WKDW *KDUSXUH\fV PHWKRG LV DV DFFXUDWH DV WKH ILQLWH GLIIHUHQFH WHFKQLTXH EXW LV RUGHUV IDVWHU >@ 7KLV FKDSWHU GHPRQVWUDWHV KRZ WKH UHFHQW DFKLHYHPHQW LQ WKH FKDUDFWHUL]DWLRQ RI VXEVWUDWH QRLVH FRXSOLQJ FDQ EH DSSOLHG WR WKH FDOFXODWLRQ RI %7 VXEVWUDWH UHVLVWDQFH 2XU JRDO LV WR HOLPLQDWH WKH XQFHUWDLQWLHV UHVXOWLQJ IURP WKH OD\RXWGHSHQGHQW VXEVWUDWH UHVLVWDQFHV DQG HQKDQFH DFFXUDF\ LQ 63,&( %7 VLPXODWLRQ 7R FRQILUP WKH DFFXUDF\ RI *KDUSXUH\fV DOJRULWKP LQ WKH FDOFXODWLRQ RI %7 VXEVWUDWH UHVLVWDQFH WKH GLUHFW H[WUDFWLRQ PHWKRG GHYHORSHG LQ WKH SUHYLRXV FKDSWHU ZDV XVHG WR H[WUDFW %7 VXEVWUDWH UHVLVWDQFH IURP PHDVXUHG VSDUDPHWHUV &RPSDULVRQ RI H[WUDFWHG DQG WKH FDOFXODWHG VXEVWUDWH UHVLVWDQFHV YHULILHV WKDW WKH FDOFXODWLRQ DOJRULWKP LV DFFXUDWH 7KLV FKDSWHU LV RUJDQL]HG DV IROORZLQJ *KDUSXUH\fV DOJRULWKP WR FDOFXODWH VXEVWUDWH UHVLVWDQFH LV LQWURGXFHG LQ VHFWLRQ /LPLWDWLRQV RI XVLQJ WKH DOJRULWKP WR FDOFXODWH %7 VXEVWUDWH UHVLVWDQFH LV GLVFXVVHG LQ VHFWLRQ $Q DF YHULILFDWLRQ IRU WKH DOJRULWKP LV GLVFXVVHG LQ VHFWLRQ $ V\VWHPDWLF PHWKRGRORJ\ WR LQFOXGH WKH OD\RXW GHSHQGHQW VXEVWUDWH UHVLVWDQFH LQ %7 FLUFXLW VLPXODWLRQ LV SUHVHQWHG LQ VHFWLRQ )LQDOO\ D EULHI FRQFOXVLRQ LV VXPPDUL]HG LQ VHFWLRQ $OJRULWKP WR &DOFXODWH %7 6XEVWUDWH 5HVLVWDQFH 7R DSSO\ *KDUSXUH\fV DOJRULWKP WR WKH FDOFXODWLRQ RI %7 VXEVWUDWH UHVLVWDQFH ZH FRQVLGHU WKH FURVVVHFWLRQ RI D YHUWLFDO GRXEOHSRO\ VHOIDOLJQHG QSQ ELSRODU WUDQVLVWRU ZLWK
PAGE 112
D VLQJOH VXEVWUDWH FRQWDFW DV VKRZQ LQ )LJ 6LQFH WKH FXUUHQW IORZLQJ DORQJ WKH LQWHUIDFH EHWZHHQ QGRSHG UHJLRQ DQG SVXEVWUDWH LV QHJOLJLEOH WKH LQWHUIDFH LH ERXQGDU\ $f FDQ EH DVVXPHG WR EH HTXLSRWHQWLDO $OVR VLQFH WKH UHJLRQ EHQHDWK WKH VXEVWUDWH FRQWDFW LH WKH UHJLRQ HQFORVHG E\ ERXQGDU\ % DQG VXEVWUDWH FRQWDFWf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re f ZKHUH MLS LV WKH KROH PRELOLW\ RI WKH VXEVWUDWH DQG D LV WKH FRQGXFWLYLW\ RI WKH VXEVWUDWH 7DNLQJ WKH GLYHUJHQFH RQ ERWK VLGHV RI f \LHOGV \\ D DW RQH FRQWDFW 9FS[ \ ]f B@D DW WKH RWKHU FRQWDFW } f HOVHZKHUH ZKHUH WS LV WKH HOHFWURVWDWLF SRWHQWLDO 7R FRPSXWH WKH HOHFWURVWDWLF SRWHQWLDO UHVXOWLQJ IURP WKH VXSSOLHG XQLW\ FXUUHQW ZH FDQ VROYH f XVLQJ WKH ILQLWH GLIIHUHQFH PHWKRG ZKLFK GLVFUHWL]HV D GLIIHUHQWLDO HTXDWLRQ LQWR D ILQLWH QXPEHU RI GLIIHUHQFH HTXDWLRQV 7KHVH GLIIHUHQFH HTXDWLRQV DUH WKHQ OLQHDUL]HG ZLWK WKH 1HZWRQ PHWKRG DQG WKH UHVXOWLQJ FRHIILFLHQW PDWUL[ LV VROYHG ZLWK WKH VSDUVH PDWUL[ PHWKRG 7KH ILQLWH GLIIHUHQFH PHWKRG LV FRPPRQO\ HPSOR\HG LQ GHYLFH DQG SURFHVV
PAGE 113
1 %RXQGDU\ % I SW\SH VXEVWUDWH )LJXUH &URVVVHFWLRQ RI D PXOWLOD\HUHG VXEVWUDWH ZLWK WZR VXUIDFH FRQWDFWV
PAGE 114
VLPXODWRUV VXFK DV 0(',&, DQG 6835(0( +RZHYHU WKLV PHWKRG LV XVXDOO\ YHU\ FRPSXWDWLRQDOO\ LQWHQVLYH DQG LV WKXV QRW IHDVLEOH ZKHQ XVHG LQ WKH FDOFXODWLRQ RI VXEVWUDWH UHVLVWDQFHV IRU D ODUJH QXPEHU RI GHYLFHV 7KH *UHHQ IXQFWLRQ WHFKQLTXH FDQ DOVR EH XVHG WR VROYH f 7KH *UHHQ IXQFWLRQ WHFKQLTXH LV RIWHQ XVHG WR REWDLQ WKH HOHFWURVWDWLF SRWHQWLDO IRU D GLHOHFWULF PHGLXP VXEMHFWHG WR FHUWDLQ ERXQGDU\ FRQGLWLRQV *UHHQ IXQFWLRQ LV GHILQHG DV WKH SRWHQWLDO DW DQ\ SRLQW [\]f GXH WR D XQLW SRLQW FKDUJH ORFDWLQJ DW [f\f]ff WKDW LV [ [b \ \ ] ] HOVHZKHUH f ZKHUH H LV WKH SHUPHDWLYLW\ RI D GLHOHFWULF PHGLXP 7R REWDLQ WKH HOHFWURVWDWLF SRWHQWLDO DW DQ\ SRLQW LQ WKH VSDFH ZH FDQ VLPSO\ VXP XS WKH SRWHQWLDO FRQWULEXWLRQV IURP DOO WKH FKDUJHV GLVWULEXWHG LQ WKH VSDFH 7KDW LV f 7KH DERYH DQDO\VLV LV GHYHORSHG WR FRPSXWH WKH HOHFWURVWDWLF SRWHQWLDO IRU D GLHOHFWULF PHGLXP 7R FRPSUHKHQG KRZ WKH *UHHQ IXQFWLRQ WHFKQLTXH FDQ EH DSSOLHG WR FRPSXWLQJ WKH HOHFWURVWDWLF SRWHQWLDO IRU D UHVLVWLYH PHGLXP ZH FRQVLGHU WZR VLPLODU VLWXDWLRQV VKRZQ LQ )LJ f D SXUHO\ UHVLVWLYH PHGLXP ZLWK D XQLW\ FXUUHQW IORZLQJ LQWR RQH FRQWDFW DQG RXW RI WKH RWKHU f D SXUHO\ GLHOHFWULF PHGLXP ZLWK D SRVLWLYH XQLW\ FKDUJH RQ RQH FRQWDFW DQG D QHJDWLYH XQLW\ FKDUJH RQ WKH RWKHU 1RWLFH WKDW WKH ERXQGDU\ FRQGLWLRQV RI WKHVH WZR FDVHV DUH PDWKHPDWLFDOO\ LGHQWLFDO H[FHSW ZLWK GLIIHUHQW SURSRUWLRQDO FRQVWDQWV 7KHUHIRUH WKH HOHFWULF ILHOG RU SRWHQWLDO RI WKH WZR FDVHV PXVW EH SURSRUWLRQDO WR RQH DQRWKHU 7KHUHIRUH ZH FDQ DOVR VROYH WKH HOHFWULF SRWHQWLDO IRU D UHVLVWLYH PHGLXP ZLWK WKH *UHHQ IXQFWLRQ WHFKQLTXH DOWKRXJK LW LV GHYHORSHG WR VROYH WKH
PAGE 115
\ M D DW RQH FRQWDFW 9fS[ M ]f D DW WKH RWKHU FRQWDFW HOVHZKHUH 8QLW\ SRVLWLYH FKDUJH 8QLW\ QHJDWLYH FKDUJH VXEVWUDWH ; \ ]f H DW RQH FRQWDFW H DW WKH RWKHU FRQWDFW HOVHZKHUH )LJXUH 7KH HOHFWULF ILHOG GLVWULEXWLRQV RI D SXUHO\ UHVLVWLYH DQG D SXUHO\ GLHOHFWULF PHGLD
PAGE 116
HOHFWURVWDWLF SRWHQWLDO IRU D GLHOHFWULF PHGLXP :LWK WKH *UHHQ IXQFWLRQ WHFKQLTXH WKH SRWHQWLDO DW DQ\ SRLQW [\]f RQ WKH FRQWDFW M GXH WR D XQLW\ FXUUHQW XQLIRUPO\ GLVWULEXWHG RQ WKH FRQWDFW L LV REWDLQHG E\ LQWHJUDWLQJ WKH *UHHQ IXQFWLRQ RYHU WKH ZKROH FRQWDFW DUHD Dc DV "c[\]f AM*F[\][n\n]nfGDL? f f D ZKHUH LV WKH *UHHQ IXQFWLRQ IRU D UHVLVWLYH PHGLXP 8VLQJ VHULHV H[SDQVLRQ DQG VHSDUDWLRQ RI YDULDEOHV WKH *UHHQ IXQFWLRQ IRU D PXOWLOD\HUHG UHVLVWLYH PHGLXP FDQ EH H[SUHVVHG DV RR RR *[ \[ff e e PQFRV P 2Q f ZKHUH D DQG E DUH WKH OHQJWK DQG ZLGWK RI D UHFWDQJXODU VXEVWUDWH DQG IPQ DUH FRQVWDQW FRHIILFLHQWV GHILQHG E\ WKH SK\VLFDO VWUXFWXUH RI WKH VXEVWUDWH >@ 7KH DYHUDJH SRWHQWLDO RQ WKH FRQWDFW M GXH WR D XQLW\ FXUUHQW XQLIRUPO\ GLVWULEXWHG RQ WKH FRQWDFW L FDQ EH REWDLQHG E\ LQWHJUDWLQJ f DJDLQ RYHU WKH ZKROH FRQWDFW DUHD Dc DV F3\ J[ \ ] [n \n ]nfGDcGDM f D cD L 6XEVWLWXWLQJ f LQWR f \LHOGV I P.[? PL[n FRV 9 D ? D QQ\A IQQ\n? &?E &6WE f 3\ 6 P n Â‘ I D VLQ P LfÂ§ 9 D VLQ PQ D M \L AA D VLQ PLWfÂ§ VLQ UUQWfÂ§ D \M/ 9 D LLMfD Df
PAGE 117
VLQ PXfÂ§ ? E VLQ PQ VLQ PXfÂ§ 9 E MO ? E EA VLQ PXfÂ§ 9 E EL a E?fEr fÂ§ ERf f %HFDXVH f FRQYHUJHV VORZO\ XVLQJ f WR FRPSXWH WKH HOHFWURVWDWLF SRWHQWLDO LV VWLOO WLPHFRQVXPLQJ 7R PDNH f FRQYHUJH IDVWHU *KDUSXUH\ PDQLSXODWHG f XVLQJ WKH LGHQWLW\ RI WULJRQRPHWULF IXQFWLRQ f DL` f DL VLQ PXfÂ§ VLQ UDQfÂ§ 9 Df ? D I FLM fÂ§ D $ M 4 Â‘ DNDf3@ TN LQWHJHU>ENEf4? f 1RWLFH WKDW f FDQ EH GHFRPSRVHG LQWR VHYHUDO FRVLQH VHULHV VXPPDWLRQV &RQVLGHU WKH WZRGLPHQVLRQDO GLVFUHWH FRVLQH WUDQVIRUP '&7f VHULHV .ST RI DQ\ VHULHV APQ DV .ST 3 4 = = NPQ P 2Q &26 PX? A FRV QX? 4ff
PAGE 118
,OO :H FDQ H[SUHVV FScM DV D VXP RI VHYHUDO VSHFLILF WHUPV LQ WKH .ST VHULHV :LWK 3 DQG 4 FKRVHQ WR EH SRZHUV RI LH Q Q f ZH FDQ XVH WKH IDVW IRXULHU WUDQVIRUP ))7f DOJRULWKP WR IDFLOLWDWH WKH FRPSXWDWLRQ RI .ST 7KHUHIRUH WKLV PHWKRG FDQ EH PXFK IDVWHU WKDQ WKH FRQYHQWLRQDO *UHHQ IXQFWLRQ WHFKQLTXH /LPLWDWLRQV RI WKH $OJRULWKP )URP DERYH GLVFXVVLRQV WZR DVVXPSWLRQV DUH PDGH f WKH FXUUHQW IORZLQJ WKURXJK D FRQWDFW LV XQLIRUPO\ GLVWULEXWHG DQG f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
PAGE 119
)LJXUH 3DUWLWLRQ RI D JXDUGULQJ VXEVWUDWH FRQWDFW 3ULPDU\ HUURU VRXUFHV RI WKH *UHHQ IXQFWLRQ WHFKQLTXH DUH WKH WUXQFDWLRQ HUURU DQG WKH URXQGRII HUURU 7KH WUXQFDWLRQ HUURU UHVXOWV ZKHQ WKH LQILQLWH VHULHV VXPPDWLRQ RI f LV WUXQFDWHG ZLWK WKH XSSHU ERXQG 3 DQG 4 7KH URXQGRII HUURU UHVXOWV ZKHQ f LV XVHG WR FRQYHUW WKH UHDO QXPEHUV DN DQG EN WR WKH LQWHJHU LQGLFHV SN DQG TN ZKLFK GHILQH WKH FRQWDFW ORFDWLRQ LQ WKH GLVFUHWH LQWHJHUf GRPDLQ 7R PLQLPL]H WKH WUXQFDWLRQ HUURU WKH WUXQFDWLRQ ERXQG 3 DQG 4 VKRXOG EH FKRVHQ DV ODUJH DV SRVVLEOH +RZHYHU WKLV DSSURDFK PDNHV WKLV DOJRULWKP QXPHULFDOO\ LQHIILFLHQW ,Q WKLV ZD\ WKLV DOJRULWKP ORVHV LWV DGYDQWDJHV RYHU WKH ILQLWH GLIIHUHQFH PHWKRG ([SHULHQFH ZLWK WKLV PHWKRG LQGLFDWHG WKDW f FRQYHUJHV VXIILFLHQWO\ ZKHQ 3 DQG 4 ZHUH HTXDO WR RU JUHDWHU WKDQ 7KH URXQGRII HUURU FDQ EH FRUUHFWHG E\ VFDOLQJ WKH FDOFXODWHG UHVLVWDQFH E\ WKH UDWLR RI WKH DFWXDO FRQWDFW VL]H WR WKH LQWHJHULQGH[HG FRQWDFW VL]H 7KH URXQGRII HUURU FDQ
PAGE 120
DOVR EH FRPSOHWHO\ DYRLGHG LI DOO DN ENffV DUH ORFDWHG DW JULG SRLQWV ,Q WKLV FDVH ZH FDQ VHOHFW VXEVWUDWH VL]H 3 DQG 4 DV JULG f ZKHUH GJULG LV WKH PLQLPXP JULG VSDFLQJ D DQG E DUH WKH OHQJWK DQG ZLGWK WKDW GHILQH D UHFWDQJXODU DUHD WR FRYHU WZR FRQWDFWV DGHTXDWHO\ DV VKRZQ LQ )LJ ,Q WKLV ZD\ WKH URXQGRII HUURU FDQ EH FRPSOHWHO\ HOLPLQDWHG $& 9HULILFDWLRQ RI WKH $OJRULWKP ,Q >@ *KDUSXUH\ KDG GRQH VRPH GF YHULILFDWLRQV IRU WKH DOJRULWKP DQG DSSOLHG WKLV DOJRULWKP WR FDOFXODWH WKH HTXLYDOHQW ZHOOWRZHOO GHYLFHWRGHYLFHf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f SP %L&026 SURFHVVHV $OO WKH WHVW ELSRODU WUDQVLVWRUV KDYH D SP ZLGH JXDUGULQJ VXEVWUDWH FRQWDFW SODFHG DURXQG WKH %7 QZHOO ZLWK D VSDFLQJ RI SP DV VKRZQ LQ )LJ 7KH GHWDLOHG OD\RXW LQIRUPDWLRQ LV OLVWHG LQ 7DEOH :LWK WKH SURFHVVVSHFLILHG VXEVWUDWH UHVLVWLYLW\ RKPFPf VXEVWUDWH UHVLVWDQFHV ZHUH FDOFXODWHG IRU WUDQVLVWRUV RI
PAGE 121
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f ZLGWK RI 1 ZHOO SPf [ [ [ [ [ [ [ [ [ [ [
PAGE 122
6XEVWUDWH UHVLVWDQFH ([WUDFWHG 5VXE YV FDOFXODWHG 5VXE )LJXUH &RPSDULVRQ EHWZHHQ H[WUDFWHG VXEVWUDWH UHVLVWDQFH DQG FDOFXODWHG UHVLVWDQFHV IRU %7V ZLWK D ZLGH YDULHW\ RI HPLWWHU IHDWXUH VL]HV
PAGE 123
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fV RSWLPL]DWLRQV FDQ YDU\ HYHQ IRU LGHQWLFDO %7V ZLWK GLIIHUHQW VXEVWUDWH UHVLVWDQFHV 7KLV FDXVHV QXPHURXV GLIILFXOWLHV LQ WKH VWDWLVWLFDO PRGHOLQJ RI &6 )RU LQVWDQFH &6 YDOXHV REWDLQHG IURP 0$&+fV RSWLPL]DWLRQ DUH RIWHQ XQUHDVRQDEO\ VPDOO FRPSDUHG WR WKHLU SK\VLFDO YDOXH EHFDXVH WKH YDOXHV DUH HPSLULFDOO\ ORZHUHG LQ RSWLPL]DWLRQ WR DFFRXQW IRU WKH VXEVWUDWH UHVLVWDQFH 7R LQFOXGH VXEVWUDWH UHVLVWDQFH LQ 0$&+ SDUDPHWHU H[WUDFWLRQ ZH DGGHG LQWR 0$&+ D QHZ YDULDEOH 568% 7R WHVW WKH PHWKRGRORJ\ ZH FRPSDUHG WKH H[WUDFWLRQ UHVXOW RI 0$&+ IRU WZR GLIIHUHQW FDVHV f ROG PHWKRGRORJ\ VHW WKH VXEVWUDWH UHVLVWDQFH WR ]HUR DQG RSWLPL]H &6 DORQJ ZLWK 5% 5%0 5& 5( DQG ;&& f QHZ PHWKRGRORJ\ XVH WKH SK\VLFDO &6 REWDLQHG IURP ORZIUHTXHQF\ &9 PHDVXUHPHQWV DQG WKH FDOFXODWHG VXEVWUDWH UHVLVWDQFH DQG RSWLPL]H RQO\ 5% 5%0 5& 5( DQG ;&& 7KH WHVW ZDV SHUIRUPHG RQ WKH ELSRODU WUDQVLVWRUV PDQXIDFWXUHG XVLQJ D 7, SP 5) %L&026 SURFHVV )LJXUH VKRZV WKH GLIIHUHQFHV EHWZHHQ H[WUDFWLRQV XVLQJ WKH WZR GLIIHUHQW PHWKRGRORJLHV %HVLGHV WKH RSWLPDO YDOXH RI 5% 5%0 5& 5( DQG ;&& ZH DOVR
PAGE 124
J W! R F e 4 )LJXUH &RPSDULVRQ EHWZHHQ H[WUDFWLRQV XVLQJ QHZ PHWKRGRORJ\ DQG ROG PHWKRGRORJ\ 7KH GLIIHUHQFH LV QRUPDOL]HG E\ WKH H[WUDFWHG YDOXH RI WKH ROG PHWKRGRORJ\ FRPSDUHG WKH 506 HUURUV RI H[WUDFWLRQV XVLQJ WKHVH WZR GLIIHUHQW PHWKRGRORJLHV 7KH 506 HUURU LV D PHDVXUH RI KRZ ZHOO WKH PRGHO SUHGLFWLRQ ILWV WR WKH PHDVXUHG GDWD 7KH SHUFHQWDJH GLIIHUHQFHV VKRZQ LQ )LJ DUH QRUPDOL]HG E\ WKH H[WUDFWLRQ XVLQJ WKH ROG PHWKRGRORJ\ 1RWLFH WKDW WKH QHZ PHWKRGRORJ\ ILWV PRGHO SUHGLFWLRQV WR PHDVXUHG GDWD DV ZHOO DV WKH ROG PHWKRGRORJ\ $OVR WKH H[WUDFWLRQ VHQVLWLYLWLHV RI SDUDPHWHUV DUH YHU\ VPDOO ZKLFK PHDQV WKH H[WUDFWLRQ RI %7 VXEVWUDWH SDUDVLWLFV FDQ EH LVRODWHG IURP WKH H[WUDFWLRQ RI RWKHU SDUDVLWLFV +RZHYHU WKH QHZ PHWKRGRORJ\ XVHV WKH SK\VLFDO &6 REWDLQHG IURP ORZIUHTXHQF\ &9 PHDVXUHPHQW WKHUHE\ HOLPLQDWLQJ LQ WKH VWDWLVWLFDO DQDO\VLV RI &6 WKH XQFHUWDLQW\ FDXVHG E\ WKH OD\RXW GHSHQGHQFH RI %7 VXEVWUDWH UHVLVWDQFH ,Q DGGLWLRQ ZLWK WKH FDOFXODWHG VXEVWUDWH UHVLVWDQFH ZH FDQ XVH WKH SK\VLFDO 506 5% 5& _fÂ§L LOO 5%0 L / L 5( ;&&
PAGE 125
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
PAGE 126
&+$37(5 6800$5< $1' 68**(67(' )8785( :25. 6XPPDU\ 7KLV GLVVHUWDWLRQ DGGUHVVHV VHYHUDO FULWLFDO PRGHO HQKDQFHPHQWV DQG D FRPSOHWH GF DQG DF SDUDPHWHU H[WUDFWLRQ PHWKRGRORJ\ IRU WKH 0063,&(4%%7 PRGHO 6LJQLILFDQW DFKLHYHPHQWV DUH VXPPDUL]HG DV IROORZV )LUVW VHOIKHDWLQJ LV LQFOXGHG LQ WKH 0063,&(4%%7 PRGHO 6HOIKHDWLQJ VLJQLILFDQWO\ LQIOXHQFHV ELSRODU WUDQVLVWRUV HYHQ DW PRGHVW SRZHU GLVVLSDWLRQ 3K\VLFDO PRGHOLQJ RI VHOIKHDWLQJ GHSHQGV RQ DFFXUDWH VLPXODWLRQ RI GHYLFHfV ORFDO WHPSHUDWXUH %\ DGGLQJ D WKHUPDO QRGH LQWR 4%%7fV DGPLWWDQFH PDWUL[ WKH GHYLFH ORFDO WHPSHUDWXUH LV LWHUDWLYHO\ FDOFXODWHG LQ WKH 63,&( PRGLILHG QRGDO DQDO\VLV 7KH UHVXOWLQJ FXUUHQWYROWDJH FKDUDFWHULVWLFV DUH FRQVLVWHQW ZLWK WKH GHYLFH SRZHU GLVVLSDWLRQ DQG WKH WKHUPDO QHWZRUN DWWDFKHG WR WKH WKHUPDO QRGH WKXV LQFOXGLQJ WKH VHOIKHDWLQJ HIIHFW 7KH PHWKRG LV DOVR DGDSWDEOH WR GF DF DQG WUDQVLHQW WKHUPDO DQDO\VLV 6HFRQG WKH GF FXUUHQW FURZGLQJ PRGHO LV PRGLILHG WR EHWWHU PRGHO WKH FXUUHQWn FURZGLQJ LQGXFHG KLJKLQMHFWLRQ HIIHFWV 7KH RULJLQDO 4%%7 PRGHO DFFRXQWV IRU GF FXUUHQW FURZGLQJ DURXQG WKH HPLWWHU SHULSKHU\ XVLQJ WKH HIIHFWLYH EDVHHPLWWHU YROWDJH 7KLV PHWKRG VLJQLILFDQWO\ XQGHUHVWLPDWHV WKH SHDN LQMHFWLRQ OHYHO DURXQG WKH HPLWWHU SHULSKHU\ WKXV RYHUORRNLQJ WKH KLJK LQMHFWLRQ HIIHFW FDXVHG E\ GF FXUUHQW FURZGLQJ 7R DYRLG WKH XQGHUHVWLPDWLRQ RI WKH KLJKLQMHFWLRQ HIIHFWV FDXVHG E\ GF FXUUHQW FURZGLQJ WKLV
PAGE 127
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fV PRGHO SDUDPHWHUV LV SUHVHQWHG :LWK WKH SK\VLFDO FRQVWDQWV LH PRELOLW\ DQG FDUULHU VDWXUDWLRQ YHORFLW\f DVVLJQHG ZLWK GHIDXOW YDOXHV SDUDPHWHUV DUH H[WUDFWHG LQ DFFRUGDQFH ZLWK WKHLU VLJQLILFDQW LPSDFWV WR WUDQVLVWRUfV HOHFWULFDO FKDUDFWHULVWLFV 7KH H[WUDFWLRQ PHWKRGRORJ\ LV EDVHG RQ VHYHUDO ORFDO RSWLPL]DWLRQV IRU GLIIHUHQW RSHUDWLQJ UHJLRQV %HFDXVH WKH PHWKRGRORJ\ RSWLPL]HV QR PRUH WKDQ WZR SDUDPHWHUV DW D WLPH WKH PHWKRGRORJ\ HQVXUHV WKH XQLTXHQHVV RI H[WUDFWHG SDUDPHWHUV 7KH SK\VLFDOO\ H[WUDFWHG SDUDPHWHUV FDQ WKXV EH HPSOR\HG WR
PAGE 128
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fV ZRUN RQ VXEVWUDWH FRXSOLQJ WR WKH FDOFXODWLRQ RI %7 VXEVWUDWH UHVLVWDQFH :LWK WKH FDOFXODWHG VXEVWUDWH UHVLVWDQFH ZH FDQ XVH WKH SK\VLFDO FROOHFWRUWRVXEVWUDWH FDSDFLWDQFH LQ %7 SDUDPHWHU H[WUDFWLRQ WKHUHE\ HOLPLQDWLQJ WKH XQFHUWDLQWLHV LQ WKH VWDWLVWLFDO DQDO\VLV RI FROOHFWRUWRVXEVWUDWH FDSDFLWDQFH DQG DYRLGLQJ WKH QHHG WR UHSHDW WKH H[WUDFWLRQ IRU HYHQ LGHQWLFDO %7V EXW ZLWK GLIIHUHQW VXEVWUDWH UHVLVWDQFHV
PAGE 129
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fV WHUPLQDO FXUUHQWV 7KHVH GLVFRQWLQXLWLHV DUH KDUPIXO WR 4%%7f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f FDQ VDWLVI\ WKH OLVW 7KH UHVXOWV DUH VKRZQ LQ 7DEOH +HUH 9%& LV WKH UHSUHVHQWDWLYH IRU HPSLULFDO PRGHOV DQG 4%%7 LV WKH UHSUHVHQWDWLYH IRU SK\VLFDO PRGHOV 7KH IRUPXODWLRQ RI 9%,& DQG 4%%7 DUH EDVHG RQ VDPH HTXDWLRQV LH GH *UDDIIfV EDVH PLQRULW\ WUDQVSRUW HTXDWLRQ DQG .XOOfV FROOHFWRU PDMRU WUDQVSRUW HTXDWLRQ +RZHYHU ZLWK
PAGE 130
GLIIHUHQW SHUVSHFWLYHV WKH UHVXOWLQJ PRGHOV DQG WKHLU SDUDPHWHUV FDQ EH YHU\ GLIIHUHQW LH HPSLULFDO RU SUHGLFWLYH 1XPHULFDO VWDELOLW\ DQG HIILFLHQF\ DUH JXDUDQWHHG IRU HPSLULFDO PRGHOV 6FDODELOLW\ DQG SK\VLFDO SODXVLELOLW\ DUH LQKHUHQW WR SK\VLFDO PRGHOV %HFDXVH RI XVLQJ D VLQJOHSLHFH HTXDWLRQ WR PRGHO WKH PDMRULW\ FDUULHU WUDQVSRUW LQ WKH HSLFROOHFWRU 9%,& LV QXPHULFDOO\ VWDEOH DQG HIILFLHQW EXW LW IDLOV WR SURSHUO\ VLPXODWH WKH FROOHFWRU FXUUHQW IRU DOO GLIIHUHQW RSHUDWLQJ UHJLRQV %HFDXVH RI XVLQJ GLIIHUHQW HTXDWLRQV IRU GLIIHUHQW RSHUDWLQJ UHJLRQV 4%%7 LV SK\VLFDO EXW LW IDLOV WR EH FRQWLQXRXVO\ GLIIHUHQWLDEOH 7KH PDMRU WUDGHRII LV WR XVH D VLQJOHSLHFH VPRRWK IXQFWLRQ IRU QXPHULFDO UREXVWQHVV RU XVH GLIIHUHQW HTXDWLRQV IRU SK\VLFV 7DEOH 7KH ZLVK OLVW IRU D 63,&( %7 PRGHO :,6+ /,67 9%,& 4%%7 7KH PRGHO LV FRQWLQXRXVO\ GLIIHUHQWLDEOH
PAGE 131
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fV VXEVWUDWH SDUDVLWLFV XVHG IRU DSSOLFDWLRQV DW KLJKHU IUHTXHQFLHV
PAGE 132
$33(1',; 7+( 68%675$7( 5(6,67$1&( &$/&8/$7,21 352*5$0 68%7(67 68%7(67 LV D SURJUDP ZULWWHQ LQ & WR QXPHULFDOO\ FDOFXODWH WKH UHVLVWDQFHV EHWZHHQ WZR RU PDQ\ SODQQDU FRQWDFWV ORFDWLQJ RQ WKH WRS VXUIDFH RI VXEVWUDWH ,W DFFHSWV HLWKHU LQSXWV ZULWWHQ LQ D EDWFK ILOH RU LQWHUDFWLYH LQSXWV )URP WKH VSHFLILFDWLRQ RI WKH VXEVWUDWH SK\VLFDO VWUXFWXUH WKH SURJUDP ILUVW FDOFXODWHV WKH VHULHV NPQ RI f DQG LWV GLPHQVLRQDO '&7 VHULHV .ST XVLQJ ))7 7KH SURJUDP WKHQ FDOFXODWHV WKH UHVLVWDQFH EHWZHHQ FRQWDFWV E\ VXPPLQJ WKH VSHFLILF WHUPV RI WKH .ST VHULHV LQ DFFRUGDQFH ZLWK WKH FRQWDFW ORFDWLRQ LQGH[HV UHIHU WR ff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
PAGE 133
$ VDPSOH LQSXW ILOH WR FDOFXODWH WKH VXEVWUDWH UHVLVWDQFH RI D SP[SP %7 ZLWK D JXDUGULQJ VXEVWUDWH FRQWDFW LV SURYLGHG LQ )LJ $ 7KH ILUVW WKUHH OLQHV GHVFULEH WKH SK\VLFDO VWUXFWXUH RI WKH VXEVWUDWH ,W LV SP ORQJ SP ZLGH SP WKLFN /LQH VSHFLILHV WKH QXPEHU RI OD\HUV ,Q GHULYLQJ f WKH ERWWRP ERXQGDU\ RI D VXEVWUDWH LV DVVXPHG WR EH JURXQGHG 7KHUHIRUH RQH PRUH OD\HU UHSUHVHQWLQJ KLJKUHVLVWLYH DLU PXVW EH PRXQWHG WR WKH ERWWRP RI D VXEVWUDWH LI WKH VXEVWUDWH LV QRW JURXQGHG RQ ERWWRP /LQH FRQWDLQV WKH LQIRUPDWLRQ RI WKH ]D[LV RI WKH ERWWRP RI HDFK OD\HU 6LQFH WKH ]D[LV RI WKH ILUVW OD\HU ERWWRP LV DOZD\V DVVXPHG WR WKLV OLQH FRQWDLQV RQO\ RI OD\HUV PLQXV f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
PAGE 134
cPL P OLOL SL_S0 LOLOOOOV.SLtL IO +8 P PU VDDUW Â‘ P Lr P cDOLL PP rLQ >IIOL _Â33__Â JJJJ +%r W : A P ,OO VPPPPPPPP n V QU LL.6LL66nL,Jc ÂrpÂÂ UDR $FWLYH GHYLFH DUHD Â‘ 6XEVWUDWH FRQWDFW )LJXUH $O 'LYLGH D ELJ DUHD LQWR PDQ\ VPDOO DUHDV OH OH VXERXW Q )LJXUH $ 6DPSOH LQSXW ILOH RI 68%7(67
PAGE 135
5()(5(1&(6 >@ :DPRFN f6LOLFRQ ELSRODU GHYLFH VWUXFWXUHV IRU GLJLWDO DSSOLFDWLRQV WHFKQRORJ\ WUHQGV DQG IXWXUH GLUHFWLRQVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 7 1DNDPXUD DQG + 1LVKL]DZD f5HFHQW SURJUHVV LQ ELSRODU WUDQVLVWRU WHFKQRORJ\f ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ $5 $OYDUH] (Gf f%L&026 WHFKQRORJ\ DQG DSSOLFDWLRQVf .OXZHU $FDGHPLF 3XEOLVKHUV %RVWRQ >@ / 7UHLWLQJHU DQG 0 0LXUD0DWWDXVFK (Gf f8OWUDIDVW VLOLFRQ ELSRODU WUDQVLVWRUf 6SULQJ9HUODJ 1HZ @ 7 @ (EHUV DQG / 0ROO f/DUJHVLJQDO EHKDYLRU RI MXQFWLRQ WUDQVLVWRUVf 3URF ,5( 9RO SS >@ ( *HWUHX f0RGHOLQJ WKH ELSRODU WUDQVLVWRUf (OVHYLHU 1HZ @ + *XPPHO DQG + & 3RRQ f$Q LQWHJUDO FKDUJH FRQWURO PRGHO RI ELSRODU WUDQVLVWRUVf 7KH %HOO 6\V 7HFK 9RO SS >@ )RVVXP f0RGHOLQJ LVVXHV IRU DGYDQFHG ELSRODU GHYLFHFLUFXLW VLPXODWLRQf ,((( %LSRODU &LUFXLWV DQG 7HFKQRORJ\ 0HHWLQJ SS >@ .XOO / 1DJHO 6 /HH 3 /OR\G 3UHQGHUJDVW DQG + 'LUNV f$ XQLILHG FLUFXLW PRGHO IRU ELSRODU WUDQVLVWRUV LQFOXGLQJ TXDVLVDWXUDWLRQ HIIHFWVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SSO >@ & 0F$QGUHZ 6HLWFKLW %RZHUV 0 'XQQ 0 )RLV\ *HWUHX 0 0F6ZDLQ 6 0RLQLDQ 3DUNHU 3 YDQ :LMQHQ DQG / :DJQHU f9%,& $Q LPSURYHG
PAGE 136
YHUWLFDO ,& ELSRODU WUDQVLVWRU PRGHOf ,((( %LSRODU &LUFXLWV DQG 7HFKQRORJ\ 0HHWLQJ SS >@ + & GH *UDDII DQG : .ORRVWHUPDQ HO DO f([SHULHQFH ZLWK WKH QHZ FRPSDFW 0(;75$0 PRGHO IRU ELSRODU WUDQVLVWRUVf ,((( %LSRODU &LUFXLWV DQG 7HFKQRORJ\ 0HHWLQJ SS >@ + & GH *UDDII DQG : .ORRVWHUPDQ f0RGHOLQJ RI WKH FROOHFWRU HSLOD\HU RI D ELSRODU WUDQVLVWRU LQ WKH 0(;75$0 PRGHOf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ + HRQJ DQG * )RVVXP f$ FKDUJHEDVHG ODUJHVLJQDO ELSRODU WUDQVLVWRU PRGHO IRU GHYLFH DQG FLUFXLW VLPXODWLRQf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SSO >@ + & GH *UDDII DQG : .ORRVWHUPDQ f1HZ IRUPXODWLRQ RI WKH FXUUHQW DQG FKDUJH UHODWLRQV LQ ELSRODU WUDQVLVWRUV PRGHOLQJ IRU &$' SXUSRVHVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ LQ f$VVHVVPHQW DQG PRGHOLQJ RI QRQTXDVLVWDWLF QRQORFDO DQG PXOWLn GLPHQVLRQDO HIIHFWV LQ DGYDQFHG ELSRODU MXQFWLRQ WUDQVLVWRUVf 3K' 'LVVHUWDWLRQ 8QLYHUVLW\ RI )ORULGD *DLQHVYLOOH >@ % +RQJ f0RGHOLQJ VLPXODWLRQ DQG GHVLJQ RI 6L*HEDVH KHWHURMXQFWLRQ ELSRODU WUDQVLVWRUV IRU KLJKSHUIRUPDQFH LQWHJUDWHG FLUFXLWf 3K' 'LVVHUWDWLRQ 8QLYHUVLW\ RI )ORULGD *DLQHVYLOOH >@ 7 < /HH f([WUDFWLRQ RI WKHUPDO UHVLVWDQFH IRU IXOO\GHSOHWHG 026)(7Vf ,((( ,QW 62, &RQI SS >@ 5 0 )R[ 6 /HH DQG 7 =ZHLGLQJHU f7KH HIIHFWV RI EMW VHOIKHDWLQJ RQ FLUFXLW EHKDYLRUf ,((( 6ROLG6WDWH &LUFXLWV 9RO SS >@ 0 5HLVFK f6HOIKHDWLQJ LQ %7 FLUFXLW SDUDPHWHU H[WUDFWLRQf 6ROLG6WDWH (OHFWURQ 9RO SS >@ % *DR f7KHUPDO GHVLJQ VWXGLHV RI KLJKSRZHU KHWHURMXQFWLRQ ELSRODU WUDQVLVWRUVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 6 %URGVN\ 7 =ZHLGLQJHU DQG 5REHU 0 )R[ f3K\VLFVEDVHG PXOWLSROH PRGHOV IRU %7 VHOIKHDWLQJf ,((( %LSRODU &LUFXLWV DQG 7HFKQRORJ\ 0HHWLQJ SS
PAGE 137
>@ 7 =ZHLGLQJHU DQG 5 0 )R[ f7KHUPDO LPSHGDQFH H[WUDFWLRQ IRU ELSRODU WUDQVLVWRUVf $FFHSWHG IRU SXEOLFDWLRQ E\ ,((( 7UDQ (OHFWURQ 'HY >@ 5 / 3ULWFKDUG f7ZRGLPHQVLRQDO FXUUHQW IORZ LQ MXQFWLRQ WUDQVLVWRUV DW KLJK IUHTXHQFLHVf 3URFHHGLQJV RI WKH ,5( SS >@ + 1 *KRVK f$ GLVWULEXWHG PRGHO RI WKH MXQFWLRQ WUDQVLVWRU DQG LWV DSSOLFDWLRQ LQ WKH SUHGLFWLRQ RI WKH HPLWWHUEDVH GLRGH FKDUDFWHULVWLFV EDVH LPSHGDQFH DQG SXOVH UHVSRQVH RI WKH GHYLFHf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 5 +DXVHU f7KH HIIHFWV RI GLVWULEXWHG EDVH SRWHQWLDO RQ HPLWWHUFXUUHQW LQMHFWLRQ GHQVLW\ DQG HIIHFWLYH EDVH UHVLVWDQFH IRU VWULSH WUDQVLVWRU JHRPHWULHVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ + 6 5KHH 6 /HH DQG % 5 .LP f'& DQG $& FXUUHQW FURZGLQJ HIIHFWV PRGHO DQDO\VLV LQ ELSRODU MXQFWLRQ WUDQVLVWRUV XVLQJ D QHZV H[WUDFWLRQ PHWKRGf 6ROLG6WDWH (OHFWURQ 9RO SS >@ 0 3 9HUVOHLMHQ f'LVWULEXWHG KLJK IUHTXHQF\ HIIHFWV LQ ELSRODU WUDQVLVWRUVf ,((( %LSRODU &LUFXLWV DQG 7HFKQRORJ\ 0HHWLQJ SS >@ + &KR DQG ( %XUN f$ VLPSOH PRGHO IRU GLVWULEXWHG EDVH LPSHGDQFH ZLWK DF YHULILFDWLRQ XVLQJ 6SDUDPHWHU PHDVXUHPHQWVf ,((( %LSRODU &LUFXLWV DQG 7HFKQRORJ\ 0HHWLQJ SS >@ )RVVXP DQG 6 9HHUDUDJKDYDQ f3DUWLWLRQFKDUJHEDVH PRGHOLQJ RI ELSRODU WUDQVLVWRUV IRU QRQTXDVLVWDWLF FLUFXLW VLPXODWLRQf ,((( (OHFWURQ 'HY /HWW 9RO ('/ >@ 5RXOVWRQ f%LSRODU VHPLFRQGXFWRU GHYLFHVf 0F*UDZ+LOO SS 1HZ @ : & RKQVRQ DQG 3 3DQRXVLV f7KH LQIOXHQFH RI GHE\H OHQJWK RQ WKH &9 PHDVXUHPHQW RI GRSLQJ SURILOHVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ &RVWD : 8 /LX DQG 6 +DUULV f'LUHFW H[WUDFWLRQ RI WKH $O*D$V*D$V KHWHURMXQFWLRQ ELSRODU WUDQVLVWRU VPDOOVLJQDO HTXLYDOHQW FLUFXLWf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS
PAGE 138
>@ 0 .LP @ 5 3HKONH DQG 3DYOLGLV f(YDOXDWLRQ RI WKH IDFWRUV GHWHUPLQLQJ +%7 KLJK IUHTXHQF\ SHUIRUPDQFH E\ GLUHFW DQDO\VLV RI 6SDUDPHWHU GDWDf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 6 6SLHJHO 5LWWHU 5 $ +DPP $ )H\JHQVRQ DQG 3 5 6PLWK f([WUDFWLRQ RI WKH ,Q3*DOQ$V +HWHURMXQFWLRQ %LSRODU 7UDQVLVWRU 6PDOO6LJQDO (TXLYDOHQW &LUFXLWf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 6 /HH % 5 5\XP DQG 6 : .DQJ f$ QHZ SDUDPHWHU H[WUDFWLRQ WHFKQLTXH IRU VPDOOVLJQDO HTXLYDOHQW FLUFXLW RI SRO\VLOLFRQ HPLWWHU ELSRODU WUDQVLVWRUVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 7 1LQJ DQG 7DQJ f0HWKRG RI GHWHUPLQLQJ WKH HPLWWHU DQG EDVH UHVLVWDQFH RI ELSRODU WUDQVLVWRUVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 5 6 0XOOHU DQG 7 .DPLQV f'HYLFH HOHFWURQLFV IRU LQWHJUDWHG FLUFXLWVf QG (G :LOH\ 1HZ @ : 6ORWERRP DQG + & GH *UDDII f0HDVXUHPHQWV RI EDQGJDS QDUURZLQJ LQ VL ELSRODU WUDQVLVWRUVf 6ROLG6WDWH (OHFWURQ 9RO SS >@ % 0 .ODDVVHQ : 6ORWERRP DQG + & GH *UDDII f8QLILHG DSSDUHQW EDQGJDS QDUURZLQJ LQ Q DQG SW\SH VLOLFRQf 6ROLG6WDWH (OHFWURQ 9RO SS >@ % 0 .ODDVVHQ f$ XQLILHG PRELOLW\ PRGHO IRU GHYLFH VLPXODWLRQL PRGHO HTXDWLRQV DQG FRQFHQWUDWLRQ GHSHQGHQFHf 6ROLG6WDWH (OHFWURQ 9RO SS >@ 0 8JDJLQ DQG )RVVXP f6LJQLILFDQW 7LPH FRQVWDQWV GHILQHG E\ KLJK FXUUHQW FKDUJH G\QDPLFV LQ DGYDQFHG VLOLFRQEDVHG ELSRODU WUDQVLVWRUVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 6X]XNL f(PLWWHU DQG EDVH WUDQVLW WLPH RI WKH SRO\FU\VWDOOLQH VLOLFRQ HPLWWHU FRQWDFW ELSRODU WUDQVLVWRUVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS
PAGE 139
>@ *RQ]DOH] f0LFURZDYH WUDQVLVWRU DPSOLILHUV DQDO\VLV DQG GHVLJQf 3UHQWLFH +DOO (QJOHZRRG &OLIIV >@ $ $UPVWURQJ DQG : )UHQFK f$ PRGHO IRU WKH GHSHQGHQFH RI PD[LPXP RVFLOODWLRQ IUHTXHQF\ RQ FROOHFWRU WR VXEVWUDWH FDSDFLWDQFH LQ ELSRODU WUDQVLVWRUVf 6ROLG6WDWH (OHFWURQ 9RO SS >@ $ 3 /DVHU DQG / 3XOIUH\ f5HFRQFLOLDWLRQ RI PHWKRGV IRU HVWLPDWLQJ IPD[ IRU PLFURZDYH KHWHURMXQFWLRQ WUDQVLVWRUVf ,((( 7UDQ (OHFWURQ 'HY 9RO (' SS >@ 0 3IRVW + 5HLQ DQG 7 +RO]ZDUWK f0RGHOLQJ RI WKH VXEVWUDWH HIIHFW LQ KLJK VSHHG VELSRODU ,&Vf ,((( %LSRODU &LUFXLWV DQG 7HFKQRORJ\ 0HHWLQJ SS >@ 1 9HUJKHVH $OOVWRW 0 :ROIH f9HULILFDWLRQ WHFKQLTXHV IRU VXEVWUDWH FRXSOLQJ DQG WKHLU DSSOLFDWLRQ WR PL[HGVLJQDO ,& GHVLJQf ,((( RXUQDO RI 6ROLG6WDWH &LUFXLWV 9RO SS >@ 5 *KDUSXUH\ DQG 5 0H\HU f0RGHOLQJ DQG DQDO\VLV RI VXEVWUDWH FRXSOLQJ LQ LQWHJUDWHG FLUFXLWVf ,((( RXUQDO RI 6ROLG6WDWH &LUFXLWV 9RO SS >@ 5 *KDUSXUH\ 3K' 'LVVHUWDWLRQ 8QLYHUVLW\ RI &DOLIRUQLD %HUNHOH\ >@ 5 *KDUSXUH\ DQG 5 0H\HU f$QDO\VLV DQG VLPXODWLRQ RI VXEVWUDWH FRXSOLQJ LQ LQWHJUDWHG FLUFXLWf ,QWHUQDWLRQDO RXUQDO RI &LUFXLW WKHRU\ DQG $SSOLFDWLRQV 9RO
PAGE 140
%,2*5$3+,&$/ 6.(7&+ 7]XQJ
PAGE 141
, FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 5REHUW 0 )R[ &KDLUPDQ $VVRFLDWH 3URIHVVRU RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ BJU\ )RVVXP 3URIHVVRU RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORARQK\ 0DUN ( /DZ $VVRFLDWH 3URIHVVRU RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ *LMV %RVPDQ 3URIHVVRU RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ $ n 9DY?D O /7FY 0DQJ 7LD 3URIHVVRU RI &LYLOHA JLQHHUPJ
PAGE 142
7KLV GLVVHUWDWLRQ ZDV VXEPLWWHG WR WKH *UDGXDWH )DXOW\ RI WKH &ROOHJH RI (QJLQHHULQJ DQG WR WKH *UDGXDWH 6FKRRO DQG ZDV DFFHSWHG DV SDUWLDO IXOILOOPHQW RI WKH UHTXLUHPHQWV IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 0D\ :LQIUHG 0 3KLOOLSV 'HDQ &ROOHJH RI (QJLQHHULQJ .DUHQ $ +ROEURRN 'HDQ *UDGXDWH 6FKRRO

