|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 DESIGN OPTIMIZATION OF THIN CRYSTALLINE SILICON SOLAR CELLS By DABRAJ SARKAR A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCT OR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012
2 2012 Dabraj Sarkar
3 To my parents
4 ACKNOWLEDGMENTS I thank my advisor, Professor Jerry G. Fossum for his guidance and my committee members, Professors Mark E. Law, Gijs Bosman, Timothy J. Ande rson, and Leo Mathew for helping me complete this work I thank Professor Sanjay K Banerjee at the Unviersity of Texas at Aust in for his advising and allowing me to use the device fabrication facility at the Microelectronics Research Center I also thank Dr. Rajesh Rao and Dr. Dharmesh Jawarani of AstroWatt Inc. for experimental setup and guidance. I thank AstroWatt Inc., and DOE for funding me and this work, and I thank my colleagues in lab for discussion and friendship: Emmanuel Onyegam, Nicole Rowsey, Dan Cummings, Sayan Saha, Dewei Xu, Scott Smith, Siddharth Chowksey, Shishir Agrawal, Zhenming Zhou, Zhichao Lu, Hema Chandra Prakash, Mike Ramon, Ricardo Garcia, Moses Ainom, Ariam Gurmu and Rachel Stout. I would like to thank Shannon Chillingworth Edwi na McKay, Stephenie Sparkman, March Lee, Cheryl Rhoden Marcus Moore and Teresa Stevens of Electrical and Computer Engineering department for their support during my graduate study at the University of Florida. Finally, I would like to thank my parents (J yotish Chandra Sarkar, Shova Sarkar, and Sheela Sarkar) and siblings ( Durga, Sujoy, and Chinmoy) for their continuous support, inspiration and sacrifice.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 2 PHYSICS BASED FLOODS SETUP FOR SOLAR CELL SIMULATION ................ 17 2.1 Solar Cell Devic e Physics and Model Development ................................ ...... 18 2.2 Model Application and Verification ................................ ................................ 23 2.3 Summary ................................ ................................ ................................ ....... 26 3 LOCAL BSF INSIGHTS FOR THIN SOM SOLAR CELLS ................................ ...... 31 3.1 Back Surface Field and Simulation Setup in FLOODS ................................ 32 3.2 Thin Cells with Uniform BSF Diffusions ................................ ......................... 33 3.3 Local BSF: Physical Insights and Design ................................ ...................... 36 3.4 Performance Dependence on Thickness Variation ................................ ....... 37 3.5 Summary ................................ ................................ ................................ ....... 38 4 BACK CONTACT SOLAR CELLS IN THIN CRYSTALLINE SILICON .................... 51 4.1 Novel Thin Silicon BC Solar Cell Process ................................ ..................... 51 4.2 2 Dimensional Numerical Simulations ................................ ........................... 53 4.3 BC Cell Design and Performance Projections ................................ ............... 54 4.4 Further Consideration of V OC ................................ ................................ ......... 61 4.5 Summary ................................ ................................ ................................ ....... 63 5 HETEROJUNCTION SOM SOLAR CELLS ................................ ............................ 73 5.1 Background ................................ ................................ ................................ ... 75 5.2 Exfoliation Process ................................ ................................ ........................ 79 5.3 Indium Tin Oxide Process Development ................................ ....................... 80 5.4 Overview of Remote Plasma Chemical Vapor Deposition System ............... 83 5.5 RPCVD Process Development and Cell Fabrication ................................ ..... 84 5.6 RPCVD Process Optimization ................................ ................................ ....... 86 5.7 Sing le Side Heterojunction Cell ................................ ................................ ..... 87 5.8 Double Side Heterojunction Cell ................................ ................................ ... 90
6 5.9 Summary ................................ ................................ ................................ ....... 92 6 SUMMARY AND FUTURE WORK ................................ ................................ ....... 113 6.1 Summary ................................ ................................ ................................ ..... 113 6.2 Future Work ................................ ................................ ................................ 114 6.2.1 Carrier Transport Model across HJ ................................ ................... 115 6.2.2 Surface Recombination Velocity Characterization ............................ 115 6.2 .3 Light Trapping by Surface Texturing ................................ ................. 116 LIST OF REFERENCES ................................ ................................ ............................. 117 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 122
7 LIS T OF TABLES Table page 2 1 FLOODS predicted performance of thick p + n cell versus base doping density. 27 2 2 FLOODS pre dicted performance of the p + nn + cell versus minority hole lifetime. ................................ ................................ ................................ ............... 27 3 1 FLOODS predicted performance versus back metal contact width for p + nn + uniform BSF cell (Structure A). ................................ ................................ ........... 40 3 2 FLOODS predicted performances of three thin BSF cell design architectures. .. 40 3 3 FLOODS predicted performance versus back contact pitc h for local BSF cell (Structure C) ................................ ................................ ................................ ....... 41 4 1 FLOODS predicted BC cell performance in AM1 (92.5mW/cm 2 ) sunlight versus front surface recombination velocity. ................................ ....................... 65 4 2 FLOODS predicted performance of the S f =10cm/s BC cell in Table 4 1 versus fractional metal contact coverage on the back n + /p + regions. ................. 65 4 3 FLOODS predicted p erformances of the near optimal 25 m BC cell design, with w (w n =w p =300 m ) widened to trade off BCSOM process simplicity versus lifetime ................................ ................................ ................................ ..... 66 4 4 FLOODS predicted performance of the w=150 m BCSOM cell in Table 4 3 with reduced N S =10 19 cm 3 versus the metal contact width ................................ 66 5 1 Overview of HJ device structures and corresponding performances .................. 94 5 2 Measured Suns Voc from cells fabricated on 500m c Si wafers for varying sample positions and temperatures ................................ ................................ .... 94
8 LIST OF FIGURES Figure page 2 1 Effective and actual doping density profiles implemented in FLOODS for minority carrier transport ................................ ................................ .................... 28 2 2 Schematic cross sections of conventional p + n and p + nn + solar cells .................. 29 2 3 FLOODS predicted illuminated JV and PV characteristics of thick p + n cell ........ 30 3 1 Cross section of pn junction solar cell w ith uniform BSF. ................................ ... 42 3 2 FLOODS predicted V OC and J SC performance as a function of back contact pitch width for uniform BSF cell (Structure A) ................................ ..................... 43 3 3 FLOODS predicted FF, and efficiency performance as a function of back contact pitch width for uniform BSF cell (Structure A) ................................ ........ 44 3 4 Cross sections of three thin BSF solar cell architectures. ................................ .. 45 3 5 FLOODS predicted V OC and J SC performance versus S b with back contact pitch as a parameter for local BSF cell (Structure C) ................................ ......... 46 3 6 FLOODS predicted V OC and J SC performance versus back contact pitch with S b as a parameter for local BSF cell (Structure C) ................................ ............. 47 3 7 FLOODS predicted FF and efficien cy versus back contact pitch with S b as a parameter for local BSF cell (Structure C) ................................ .......................... 48 3 8 FLOODS predicted comparative performance analysis versus hole lifetime with device thickness as a pa rameter for local BSF cell (Structure C) ............... 49 3 9 FLOODS predicted local BSF cell (Structure C) efficiency versus hole lifetime for different cell thicknesses. ................................ ................................ .. 50 4 1 A fabricated thin (t Si ~ 25 m) crystalline silicon BCSOM solar cell .................... 67 4 2 Demonstration of the flexibility of the SOM foil, which can broaden its possibl e applications. ................................ ................................ ......................... 68 4 3 A measured 1 sun current voltage characteristic of the BCSOM solar cell ........ 69 4 4 Basic cross section of the BC solar cell structure, or the FLOODS domain assumed for the 2 D numerical simulations. ................................ ....................... 70 4 5 FLOODS predicted open circuit voltage versus the back nonmetal surface recombination velocity f or varying n + /p + surface doping density. ........................ 71
9 4 6 FLOODS predicted loss of short circuit current density versus increasing p + region width, for different w n /w p ratios. ................................ ................................ 72 5 1 Cross section of a typical double side heterojunction cell on a n type crystalline silicon SOM wafer. ................................ ................................ ............. 95 5 2 Process flow for fabricating exfoliate d single heterojunction solar cells. ............. 96 5 3 SIMS profile showing Hydrogen incorporation into the Si substrate during the electroplating process. ................................ ................................ ........................ 97 5 4 Photographs of exfoliated SOM foils ................................ ................................ 98 5 5 Variation of sheet resistance of ITO films as function of substrate temperature ................................ ................................ ................................ ........ 99 5 6 Variation of o ptical transmittance characteristics of ITO films as function of substrate temperature ................................ ................................ ...................... 100 5 7 Variation of sheet resistance of ITO films as a functio n of deposition pressure 101 5 8 Optical transmittance spectra of ITO films as a function of Ar pressure. .......... 102 5 9 Schematic diag ram of RPCVD deposition chamber ................................ ......... 103 5 10 Photograph of SOM foil with fabricated HJ cells, and cross section of fabricated single side HJ cells ................................ ................................ ......... 104 5 11 Raman spectra of a Si:H(p + ) films deposited at different temperatures. ........... 105 5 12 High resolution TEM cross se ction of thin HJ solar cell on SOM foil. ............... 106 5 13 Illuminated JV and EQE characteristics of single side HJ solar cells ............... 107 5 14 Cross section of the proposed single side heterojunction (S HJ) cell on a n type SOM wafer ................................ ................................ ................................ 108 5 15 EQE data indicating carrier recombination at the back due to uniform metal contacts and no surface passivation. ................................ ................................ 109 5 16 FLOODS predicted dependence of intrinsic V OC J SC thickness for varying bulk lifetime. ................................ ................................ .... 110 5 17 Illuminated JV and EQE characteristics of double side HJ solar cells .............. 111 5 18 Comparison of dark saturation current densities vs. open circuit voltages for different cell structures. ................................ ................................ .................... 112
10 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 DESIGN OPTIMIZATION OF THIN CRYSTALLINE SILICON SOLAR CELLS By Dabraj Sarkar Aug ust 2012 Chair: Jerry G. Fossum Major: Ele ctrical and Computer Engineering Solar cells in thin crystalline silicon are of much interest because of their potential high efficie ncies and low cost. However, many previous attempts to attain thin silicon for photovoltaic applications have involved costly processes, such as high energy implants and epitaxial growths, and impractical ways of handling the thin films. Hence, development of cost effective crystalline silicon solar cells with optimal thickness of 10 100m has remained an unfulfilled goal for many years. In this work, we report a novel kerfless exfoliation techno logy capable of producing ultra thin 25m flexible mono crystalline silicon foils from thick Si wafers. We set up an object oriented 2 D devic e simulator (FLOODS), and augmented it for reliable physics based numerical simulation of thin Si solar cells. This tool provides flexibility for general simulations that commercial tools do not. The setup includes (i) characterization of the electron hole generation rate (ii) internal photon reflection, ( i ii) modeling of SRH and Auger carrier recombination rates (iv ) modeling of carrier mob ilities and ( v) physical accounting for energy bandgap narrowing and Fermi Dirac statistics in heavily doped region s.
11 Using FLOODS simulations, we present physical insights into the various recombination mechanisms and how they affect performances of thin BSF cells. We have explored novel design t echniques and engineering trade offs such as base doping density, local BS F, local back contacts, and contact width and pitch to reduce the recombination losses. As the crystalline silicon thickness is reduced to attain substantial cost reduction, excellent surface passiv ation on both sides is required to fabricate higher effic iency solar cells We fabricated back contact solar cells and a Si:H/c Si heterojunction solar cells to achieve this goal. Numerical simulations with FLOODS were used to identify losses in these devices and optimum device structures were designed and per formance predicted with numerical simulations. A novel remote plasma CVD (RPCVD) based process was developed for fabrication of a Si:H/c Si heterojunction (HJ) photovoltaic cells. In the RPCVD system, during the deposition process there is no direct expos ur e of the sample to the plasma This can reduce the plasma damage to the silicon surface and improve passivation quality Very high open circuit voltage measured from fabricated heterojunction cells suggests that RP CVD is a potential technology for achievin g improved passivation in HJ cells
12 CHAPTER 1 INTRODUCTION Crystalline silicon solar cells are of much interest because of their potential high efficiencies and low cost . In order to reduce cost/watt of crystalline silicon solar cells and to achieve grid electricity cost parity, it is necessary to achieve high conversion efficiency with thinner crystalline silicon wafers because the silicon accounts for a large part of the module cost . However, cell manufacturers are struggling to reduce the wafer manufacturing very thin Si wafers and such thin Si wafers impose stringent handling requirements such as wafer breakage and yield loss that impact final module cost. Hence, develo pment of cost effective crystalline silicon solar cells with optimal thickness of 10 100m has remained an unfulfilled goal for many years . The research described herein seeks to explore modeling, simulation, and design of thin (25m) crystalline Si so lar cells along with experimental corroboration based on a kerf f ree semiconductor on m etal (SOM) exfoliation process . Historically, manufacture r s have focused their research more on other cost elements than efficiency, and efficiency improvements in c ommercial cells were largely stagnated until 2002 . However, there is a recent resurgence in interest in improved efficiency particularly for thinner cells. The t heoretical maximum possible efficiency of a single junction silicon solar cell is about 29 % [4 ] [ 5]. Actual cell efficiencies fall quite short of this ideal limit due to various extrinsic lo sses, e.g., reflection losses at front grid and front surface, optical absorption losses in back contact, carrier recombination losses in the bulk Si at f ront/back surfaces and at front/back contacts, and resistive losses in diffusion regions, at contacts and metal grids . Identification and reduction of
13 these losses have become very significant factors for the design and manufacturing of thinner high ef laboratory solar cells have eliminated much of these extrinsic losses, and have yielded efficiencies of 23.0% and 24.2%, respectively. As the efficiency of solar cells has become premi um, numerical solutions of the carrier transport problem in the solar cells are desirable to find and analyze where and how improvements in the cell design can be achieved. Crystalline silicon solar cells can be simulated reliably because of the vast knowl edge of the silicon properties that has evolved for more than 50 years of integrated circuit developments. These properties must be physically accounted for. For thin cells, the features of heavily doped silicon (i.e. bandgap narrowing, carrier degeneracy, or Fermi Dirac (F D) statistics, and Auger recombination) are most important since much of the thin cell is heavily doped. In Chapter 2, we discuss how we augment and set up FLorida Object Oriented Device Simulator (FLOODS) , a TCAD tool under developm ent at the University of Florida, for reliable physics based numerical simulation of thin crystalline silicon solar cells in AM1 sunlight. The reliability of the simulation tool and analyses is corroborated by comparing results for conventional cells with published results in the literature. W e note in Chapter 2, that the high recombination velocities near surfaces can limit the cell efficiency. Moreover, the unavoidable Auger recombination in the back surface field (BSF) region limits the cell efficiency. Rear surface recombination can be reduced by reducing the back metal contact area and passivating most of the back surface with thermal oxides, and Auger recombination in the BSF region can be reduced by introducing local BSF doping and point contacts . Local back doping was
14 employed in 280m thick PERL cells , b ut there is currently little insight available in the literature on how local contact/doping works for thin cells with thickness of about 25um. In Chapter 3, we present new physical insights derived from FLOODS simulations, into the benefits of local doping/point contact and thus reduce the effect of back surface recombination on BSF cell performance. The back contact (BC) solar cell has the highest possibility of increased performance close to the intrinsic performance of 29% efficiency in the near future . Originally intended for concentrator systems , BC cells have attracted broad interests due to its several advantages over conventional cells. Front grid reflection and shadowing lo sses, and front grid resistance losses can be eliminated in BC solar cells . The back junction can be optimized for electrical performance (e.g., dar k saturation current density ) and the top surface can be optimized for optical performance . Since m etal contacts are on the back side, the trade off between grid shading and series resistance is no longer present in BC cells . Clearly, BC cells require starting materi al with a long minority carrier lifetime. However, since the exfoliated SOM foils a re of thickness 20 30um, the BCSOM cell is a potentially suitable technology for achieving high efficiency solar cells. In Chapter 4, fabrication of preliminary BCSOM solar cells is described along with experimental data, and ultimate efficiency of BCSOM cell is projected. We identify the important device parameters for efficient cell design: the front surface recombination velocity, the back nonmetal surface recombination velocity, back surface doping density, the widths of interdigitated n + and p + region s, and the widths of the respective metal contact lines. We gain physical insights into the front surface passivation
15 requirements, and tradeoff between the widths of interdigitated n + and p + regions and optimum back contact geometry for 25m thick BCSOM c ells. Key recombination mechanisms in back diffusion regions limiting the ultimate BCSOM cell performance are identified and insights on optimum doping profile are revealed. The projected performance of the ultimate BCSOM is compared to that of SunPower Co To reduce the noted surface recombination and to achieve better minority carrier suppression near the surface, an a Si :H /c Si heterojunction (HJ) with a bandgap larger than Si is put between the metal and sili con , using a transparent front electrode to keep the series resistance low. The a Si :H /c Si heterojunction with a thin intrinsic a Si :H buffer layer was first reported in 1977, and later developed and commercialized by Sanyo Electric Co. as HIT solar c ells . The effective passi vation and the reduced minority carrier concentra tion achieved by the a Si:H /c Si heterojunction enables high open circuit voltage V OC and high efficiency. In addition, t he a Si:H /c Si heterojunction is a low temperature proc ), and it reduces crystal damage and production costs . Thus, thin HJ is an obvious choice with SOM wafers to manufacture low cost solar cells. Fabrication of single side heterojunction (SHJ) and double side heterojunction (DHJ) cell s on SOM wafer s is described in Chapter 5 Insights about HJ cell losses and design for performance improvements are presented by analyzing current voltage (IV) quantum efficiency (QE), and Suns Voc characteristics. In HJ solar cells, a transparent conduc ting oxide (TCO) layer such as indium tin oxide (ITO) is used to reduce the lateral resistance associated with thin a Si:H emitters. An R.F. magnetron reactive
16 sputtering system was use d to develop an optimized process for the ITO films. A remote plasma chemical vapor deposition (RPCVD) based process was developed to fabricate a Si:H/c Si heterojunction cell. In the RPCVD system, the sample is held distant from the plasma source and the deposition chamber is in ultra high vacuum (UHV) both can lead to r educed surface damage to the c Si surface. Sample to plasma distance, deposition temperature, and thickness of a Si layers were characterized in order to achieve improved surface passivation. A 13.4% efficiency cell was fabricated and V OC as high as 662mV was obtained. Com prehensive study of the carrier transport mechanisms across the junction hetero interf ace is necessary for efficient heterojunction (HJ) cel l design. FLOODS upgraded by reliable engineeri ng based models for the carrier transport mechanis m and optical absorption across the heterojunction, can aid such a study The local doping and point contact BSF scheme in the single side HJ SOM cell, with optimal design derived from FLOODS simulations should be incorporated
17 CHAPTER 2 PHYSICS BASED F LOODS SETUP FOR SOLA R CELL SIMULATION Solar cells are bipolar devices with high doping levels in the emitter and BSF regions to achieve better device performance. Reliable modeling of heavy doping (HD) effects is important to predict solar cell performance In this chapter, FLOODS , an object oriented 2 D device simulator, is augmented and set up for reliable physics based numerical simulation of thin crystalline silicon solar cells in AM1 sunlight, with emphasis on the modeling of HD effects. Reliable m odeling and simulation tool s for solar cell design are warranted as the conversion efficiency of solar cell is at a premium today. This was neglected in the past largely due to shift in research trends to process and technology. FLOODS for solar cells prov ides flexibility for general simulations that commercial tools do not. Models we have implemented in FLOODS are: (i) characterization of the electron hole generation rate [g(x) in 1 D] defined by AM1 photon absorption, with or without internal photon refle ction, (ii) physical accounting for energy bandgap narrowing and Fermi Dirac (F D) statistics in heavily doped regions, (iii) modeling of SRH and band band Auger carrier recombination rates, and (iv) modeling of carrier mobilities dependent on doping densi ties. One and two dimensional simulations can be done in FLOODS. The chapter begins with modeling and implementation of the above mentioned device physics in FLOODS. Later the reliability of implemented models and FLOODS simulations were verified by deri ving physical insights into the back surface field (BSF). BSF consists of a high low ( HL) junction at the back of the device and it is an important design technique to reduce recombination at the back of the solar cell, and improve
18 device performance. BSF becomes very effective if the base of the solar cell is transparent to the minority carrier flow (i.e. minority carrier diffusion length longer than cell thickness). The built in electric field induced by the HL junction opposes the motion of minority carr iers to the back surface where the recombination velocity is very high. Moreover, the increase in integrated doping density in the base reduces minority carrier injection into the base. Thus BSF improves both the short circuit current density (J SC ) and ope n circuit voltage (V OC ). 2.1 Solar Cell Device Physics and Model Development For thinner cells, the features of heavi ly doped silicon [i.e., bandgap narrowing ( ), carrier degeneracy, or F D statistics, and Auger recombination] are m ost important since much of the silicon solar structure must be heavily doped. In FLOODS, we physically account for and F D statistics by using an effective doping density (e.g., for n + Si) to model the minor ity carrier (hole) transport : (2 1) where is the actual doping density and is an effective intrinsic carrier density that depends on and the F D integral of order 1/2: (2 2) with being the relative position of the Fermi level. Both the and the F 1/2 terms in Eq. 2 2 depend on the majority electron density, or We note that the former term increases with while the latter term decreases, implying that
19 in (2 1), which is generally less than does not vary with the doping density in a simple way [16 ]. Further, we note that the modeling of vs. (~linear for typical high ) differs from that of vs. for p + Si (~constant for typical hi gh ). Fig. 2 1 shows effective donor and acceptor doping densities for an analytical Gaussian doping profile ( ). We note that is itself a Gaussian profile with modified surface doping concentration ( ) and standard deviation (straggle ), i.e., To physically account for heavy doping effe cts and predict minority carrier transport, we use the effective net doping density ( ) in Poisson equation (2 3) which is solved in FLOODS with the carrier continuity equation. However, SRH and Auger band band carrier recombination rates are dependent on the actual doping densities (e.g., N D (x), N A (x)). SRH carrier recombination is given by  (2 4) where are doping dependent lifetime parameters : (2 5) (2 6)
20 wit h and and chosen based on the material quality. Auger band band carrier lifetime fundamentally varies with the square of the majority carrier density or doping density. Auger band band recombination is modeled in terms of actual carrier density as (2 7) (2 8) (2 9) (2 10) where and and Auger hole, electron lifetimes are respectively [18 ] [ 20]. Effective hole and electron lifetimes ( ) are given by (2 11) (2 12) Electron and hole mobilities are also based on actual dopi ng densities [21 ] [ 22] and modeled as (2 13)
21 where (min) =68.5cm 2 /V.s, (max)= 1414cm 2 /V.s for electron, and (min)= 44.9cm 2 /V.s, (max)= 470cm 2 /V.s for hole. For FLOODS simulations, physical characterizations of surface s are done via surface recombination velocities (SRV). For 1 D simulations, a q uasi ohmic boundary condition is defined to enable specification of recombination velocities at the surfaces through which current must flow by (2 14) where S is the surface recombination velocity (SRV). The charge neutrality at the surface (2 15) completes the set of surface boundary conditions. For 2 D simulations, a t surfaces through which n o current can flow (e.g., oxide or nitride passivated surfaces), the surface recombination velocity (SRV) is specified by (2 16) where S n and S p are the electron and hole SRVs respectively. For S n =S p =S (2 17) Electron hole generation rate g(x) induced by photon is defined in FLOODS based on an AM1 photon spectrum with an incident power level of 92.5mW/cm 2 . However, the current standard testing condition for terrestrial measuremen ts is AM1.5 (global) which was adopted in 2008 as IEC 60904 3 Ed. 2 (also equivalent to ASTM G173) with an input power level of 100mW/cm 2 . Thus, AM1 performance of solar cell predicted
22 by FLOODS can easily be normalized to global AM1.5 standard testin g conditions by multiplying J SC by 1.08, and increasing V OC by 2mV with FF unaltered. Unde r normal solar illumination, the photon flux falls off exponentially into the semiconductor, i.e. (2 18) defined by the photon absorption S o, the electron hole pairs created, per unit volume in the energy range E ph to (E ph + dE ph ) at x is (2 19) The total generation rate at x is just the integral of the above expression over the entire ener gy spectrum : (2 20) The generation rate distribution in silicon under one sum, air mass one (AM1) illumination at 300K, for zero reflection at the exposed surface, has been obtained by numerically evaluating the integral in ( 2 20) for values of x ranging from zero upward. For 1 pass internal photon reflection the total electron hole generation rate is assumed to be (2 21) For multiple internal reflections, the total generation rate is similarly given by (2 pass) (2 22)
23 (3 pass) (2 23) Note complete light trapping is modeled by for a large n such th at (2 24) 2.2 Model Application and Verification We used our FLOODS setup to simulate simple 1 D version of the p + n and p + nn + solar cells shown in Fig 2 2 to verify it, based on published results in the literature. The top p + n junction forms the emitter of the device and the bottom nn + high low (HL) diffused junction is used as a back surface field (BSF). The front and bottom interfaces are passivated by thermally grown oxide, nitride, or by PECVD /RPCVD intrinsic amorphous silic on (a Si :H ). The BSF can bring significant improvement in the cell performance, if the minority hole lifetime is long enough to ensure that the diffusion length (L p ) is much longer than the base thickness (W b ) . Typical thickness of commercially availa ble Si solar cells is about 200m 350m [2 ] [ 25 ] [ 29]. If the dark current can be reduced (and V OC improved) by the inc orporation of a back HL junction. Also the HL junction improves J SC as the built in electric field in the nn + d iffusion region forces the minority holes away from the high S b back contact. We initially assume a 250m thick ( ) p + n solar cell with no BSF and optimize the base donor doping density (N B ), with respect to minority hole lifetime ( ) according to Eq. (2 5), (2 10), and (2 11). The top surface acceptor doping density (N S )
24 is assumed to be 10 19 cm 3 with front surface recombination velocity S f =10 3 cm/s which is achievable with good thermal oxide pa ssivation . Th ough minority carrier lifetime as high as 1ms is attainable [7 ] [ 35], we assume in Eq. (2 5) and (2 6). Simulation is done under AM1 photon spectrum with input incident power of 92.5mW/cm 2 . FLOODS predicted current voltage (JV ) and power voltage cha racteristics are shown in Fig 2 3 The intrinsic efficiency ( ) of solar cell is calculated as (2 25) where V OC J SC and FF are intrinsic open circuit voltage, short circuit current den sity and fill factor respectively, and P int =92.5mW/cm 2 at AM1  incident optical power. There are extrinsic losses that can be minimized by process optimization. These losses are shadowing (and reflecting) losses (~2.2%) at the front metal contact grid and front surface, resistive losses (~0.7%) associated with the base, emitter, BSF regions, and contacts, metal grids, and photon absorption losses (~1.4%) at back metal contact . The near optimal intrinsic FF predicted by 1 D FLOODS simulations is also dependent on the base resistivity and on 2 D carrier flow through the device. FLOODS predicted intrinsic performance ( ) and the estimated extrinsic cell efficiency ( ) taking into account the extrin sic losses a re shown in Table 2 1. We note in Table 2 1 that for p + n cells (without BSF), the cell performance is limited by the degree of back surface carrier recombination which is dependent on the base doping density. With base doping density N B =10 15 cm 3 the hole diffusion length the base is transparent to the minority holes. Thus the dark
25 current is high due to higher recombination of holes at the high S b back surface. As the base doping density is increased, increased bulk recombination m akes the base less transparent to the minority carriers. That is the effect of back surface minority carrier recombination on V OC is reduced with shorter diffusion length ( ) as the base doping density is increased. The increase in V O C with increased N B is due to reduction in hole injection level ( ) in the base and reduction in back surface recombination due to non transparent base (where W b is base thickness and N D is base doping profile) However, photo generat ed carrier collection efficiency (i.e., J SC ) decreases with increased N B due to increased bulk recombination in addition to the back surface recombination. Decrease in FF at lower N B is due to increase in high injection component of dark current (larger id eality factor, n). There is a trade off between base transparency and back surface recombination and 10 16 cm 3 is found to be near optimum base doping density. The BSF is incorporated into the device by adding an nn + HL junction near the back with N S =3x10 20 cm 3 for the optimized N B =10 16 cm 3 cell. V OC is improved by about 20mV (dark current density J 0 halved) and the loss in J SC i s recovered as shown in Table 2 1. The improvement in V OC results from reduced carrier injection into the base by increased integra ted base doping density ( ) and less carrier recombination near back surface due to suppression of minority carriers. Minority carrier suppression near the back surface and built in electric field ( ) induced by BSF doping profile effectively reflects the carriers away from the back surface resulting in improved
26 J SC V OC in this p + nn + device is limited by the Auger recombination in the heavily doped BSF region which can be ameliorated by utilizing the trade off be tween Auger recombination and back surface recombination. We also note that if due to short the BSF is les s effective as shown in Table 2 2. The results predicted by FLOODS are consistent with those published in the literature [25 ] [ 27 ], [ 35]. 2.3 Summary We have overviewed the heavy doping effects and the device physics for thin silicon solar cells, and implemented it in FLOODS for 1 D and 2 D simulations. We have used our physics based 1 D FLOODS setup to ve rify physical operation of the conventional back surface field solar cells under AM1 sunlight (92.5mW/cm 2 input power). We believe our physical modeling for the heavily doped regions is quite reliable for bipolar devices like solar cells and is in agreemen t with those published in the literature.
27 Table 2 1. FLOODS predicted performance of thick p + n cell versus base doping density Device thickness top surface doping density 10 19 cm 3 minority hole lifetime Performance enhancement via BSF (p + nn + ) for the base doping density, N B =10 16 cm 3 cell is shown as well. A back HL nn + junction with N S =3x10 20 cm 3 is used as BSF with back surface recombination velocity S b =3x10 4 cm/s. is intrinsic ef ficiency and is estimated extrinsic efficiency. Device Type N B (cm 3 ) L p ( m) V OC (mV) J SC (mA/cm 2 ) FF (%) (%) (%) NO BSF 10 15 635 561 35.2 80 9 17.3 15.7 NO BSF 10 16 423 617 34.9 82 9 19.3 17.1 NO BSF 10 17 143 664 32.0 83 6 19.2 16.8 BSF 10 16 423 637 36.2 83 1 20.7 18.2 Table 2 2 FLOODS predicted performance of the p + nn + cell versus minority hole lifetime Substrate thickness base doping density N B =10 1 7 cm 3 top surface doping density 10 19 cm 3 back N S =3x10 20 cm 3 ,and back surface recombination velocity S b =3x10 4 cm/s. ( s) V OC (mV) J SC (mA/cm 2 ) FF (%) (%) (%) 160 637 36.2 83 1 20.7 1 8.2 16 604 34.5 82 6 18.6 16.5
28 Figure 2 1. Effective and actual doping dens ity profiles implemented in FLOODS for minority carrier transport Actual doping is a Gaussian doping profile with N S = 3 x 10 20 cm 3
29 Figure 2 2. S chematic cross sections of conventional p + n and p + nn + solar cell s The high low (HL) nn + junction acts as a back surface field (BSF).
30 Figure 2 3 FLOO DS predicted illuminated JV and PV characteristics of thick p + n cell. Cell thickness with no BSF at AM1 (92.5mW/cm 2 ) sunlight. Base donor doping density N B =10 15 cm 3 emitter N S =10 19 cm 3 back surface is unpassivated (S b =10 6 cm/s) and front surface recombination velocity S f =1000cm/s
31 CHAPTER 3 LOCAL BSF INSIGHTS FOR THIN SOM SOLAR C ELLS As described in Chapter 2, the minority carrier recombination in the base of the device can be reduced by introducing a back surface field (BSF) if the minority carrier diffusion length is longer than the cell thickness. H owever, with thinner devices (t Si <100m), BSF ca n be effective even if the minority carrier lifetime in the base is moderately short i.e., even for cheaper silicon. But, the BSF must be optimized for high efficiency. There is an unavoidable Auger recombi nation in the BSF region that can limit the cell efficiency, and this must be traded off against the back surface recombination. Device engineering to effect this tradeoff is warranted . Back surface recombination can be reduced by reducing the back met al contact area and passivating most of the back surface with thermal oxide, and Auger recombination in the BSF region can be reduced by introducing local BSF In this chapter, we present a more comprehensive analysis of BSF engineering with FLOODS  T he chapter begins with a quick back ground of BSF Various design parameters to red uce the carrier recombination at the back side of the solar cell are: back contact width (w m ), contact pitch (w), surface recombination velocity of Si SiO 2 interface (S b ), wi dth of the BSF diffusion region (uniform or local), minority carrier lifetime ( ) in the base and device thickness (t Si ) These design parameters are varied systematically to gain physical insights on the recombination mechanism s in the solar cell and to design optimum device structures. For thinner cells, design optimization of the BSF region and back contacts is very important. In order to isolate the carrier recombination at the front surface front surface recombination velocity was assumed low (S f = 1cm/s).
32 3.1 Back Surface Field and Simulation Setup in FLOODS The back surface field (BSF) can bring significant improvement in the ce ll performance, if the minority carrier diffusion length ( ) is longer than t he base thickness W b . For a pn junction diode in Fig. 3 1 the minority hole current density (J p ) injected into the base from a p + emitter is can be modeled as (3 1) (3 2) where is the average hole diffusivi ty in the base and is the base Gummel number (N G ) V EB is emitter base forward bias, k B Boltzmann constant, and T is absolute temperature We see that dark current can be reduced ( and V OC improved) if a high low (HL) nn + diffusion layer is introduced in the base to increase N G. The BSF doping e ffectively reduces the minority carrier concentration near the back surfaces, thus reducing the recombination. Also the HL junction improves the short circuit current density J SC as the built in electric field in the nn + diffusion region forces the minority holes away from the high S p back contact. To get physical insights about local BSF and back contact design we set up FLOODS  for reliab le physics based numerical simulation in AM1 (92.5mW/cm 2 ) sunlight. Modeling of optical carrier generation rate g(x), recombination rates (SRH,
33 band to band Auger ), mobility, energy bandgap narrowing ( ) and F D statistics in heav ily doped regions wa s done as discussed in Chapter 2. An n type base ( ) wa s used in our simulation; similar results for p starting wafers can be projected from the simulations discussed here. A shallow p + n diffused junction on top wa s a ssumed, as we are interested in recombination at BSF diffusion region, back contact and back non metal surface (i.e. Si SiO 2 interface) Front surface acceptor doping density with a junction depth of wa s ass umed to minimize recombination losses in front p + region. To eliminate front surface carrier recombination and effect of front contacts, and thus to focus on the back losses, a uniform quasi ohmic top contact with low surf ace recombination velocity ( ) was assumed. 3.2 Thin Cells w ith Uniform BSF Diffusions The cross section of a pn junction solar cell with uniform BSF and the basic 2 D FLOODS domain are illustrated in Fig. 3 1 The surface recombination velocity (SRV) at the back me tal contact was assumed infinite (S=10 6 cm/s), and SRV at the Si SiO 2 interface (i.e. non metal back surface) S b which depend s on the surface doping density, was taken as a variable design parameter. The value of S b was determined from an empirical formula (S ~ N S x10 16 cm 3 ) . Initially, we assumed minority hole lifetime in the n type base, At this long lifetime, minority carrier (hole) diffusion length which is much longer than silicon thickness Thus carrier recombination in the BSF region, and at the non metal back surface and back contact s determines Later in the chapter, w e also present the physical insights on shorter
34 minority carrier lifetime on devic e performance. We initially assume d a uniform n + diffusion at the back of the cell (Gaussian profile, junction depth x j ~ 0.3 m) to understand depende nce of dark current density on Auger recombination in back n + region and on minor ity carrier recombination at back metal contacts to the n + region. We show in Table 3 1 the FLOODS predicted performance of the uniform BSF p + nn + cell as described in Fig. 3 1 under AM1 optical illumination. The back contact widths (w m ) were varied as a d esign parameter for fixed S b =10 3 cm/s, and pitch w=300 m. We see increases by 23mV (and dark current density is reduced by a factor of about 2.5) when back contact width (w m ) is reduced from 10 m to 2 m This i s largely due to the decrease in minority carrier recombination at the back metal contact. Note that the minimum contact width is limited by process technology which is mostly do ne with a shadow mask, and narrower contact increases series resistance that c an decrease FF. However, (as well as ) is relatively unaffected by back non metal surface recombination velocity as the uniform BSF suppresses minority c arrier density at back Long diffusion length L p makes insensitive to the variation of back contact width We call this uniform BSF cell Structure A. Next, we varied the back contact pitch width (w) as a design parameter for the uniform BSF cell (Structure A). The solar cell performance und er AM1 illumination is shown in Fig. 3 2 and Fig 3 3 As the pitch width (w) is increased, recombination at the back metal contact decreases as reflected by improved V OC with wider pitch es Improvement in V OC however, tends to saturate at wider pi tches (w >1mm) due to increased Auger recombination in the n + BSF and recombination at non metal back
35 surface which tend to offset reduction in the back contact recombination. J SC improves slightly with wider pitch (w), due to reduction ca rrier recombination at the back contact and J SC improves when w m /w ratio is minimized. Fill factor (FF) is limited by pitch; as wider pitch means higher lateral ohmic drops which d ecreases FF. So there is a design trade off between V OC and FF when designing pitch width. The optimum pitch width is for 2 m wide back contact opening and base minority carrier lifetime of 1ms A t this point of the uniform BSF cell (Structure A) is limited by the Auger recombination in the n + BSF region, and recombination at the back metal contact. Recombination at the back metal contact can be minimized further by incorporating a shallow heavy n ++ doping profile around the back contact as shown in Fig. 3 4 (Structure B) Minority hole concentration ( ) is inversely proportional to the doping density and there is a built in electric field induced by the gradient of doping profile ( ). Thus, introduction of a shallow heavy doping profile (N S =3x10 20 cm 3 x j ~0.16 m) aro und the back contact suppresses the minority carrier concentration and thereby it reduces ca rrier recombination at the back metal contact We call this design Structure B. As we see from FLOODS simulation in Table 3 2 V OC increases to 702mV if a shallow heavy doping profile is introduced around the back contact regions. J SC and FF remain unaffected with this optimization. However, a complex process is required to fabricate Structure B design. Open circuit voltage is still limited by the Auger recombinatio n in the n + diffusion region. Auger band band recombination can be eliminated by removing the uniform n + BSF region leaving only the shallow heavy
36 doping profile near the back contact as shown in Fig. 3 4. The shallow heavy doping profile around the back m etal contact acts as a local BSF and we call this design Structure C. 3.3 Local BSF: Physical Insights and Design We now extend our insights learned so far to design local BSF cells by completely removing the uniform BSF with only local doping around the back contact as illustrated in Fig. 3 4. This design (Structure C) eliminates Auger recombination near the back of cell and thus improves V OC and since minority carrier diffusion length carrier collection efficiency (J SC ) is also good as we will see later. However, the absence of a uniform BSF turns the non metal back surface (Si SiO 2 interface) very critical the interface could act as a recombination sink to the minority holes if the Si SiO 2 interface is not well passivated Sin ce local BSF width is small compared to the device width, a heavy doping profile is required to achieve sufficient built in electric field. We initially assume a local BSF with N S =3x10 20 cm 3 and junction depth x j =0.3 m. For lower surface doping density N S = 10 19 cm 3 deeper junction depth (x j 1 to 1.5 m) is required to achieve similar performance. For our FLOODS simulation, w e assumed infinite SRV at the back contact ( S=10 6 cm/s), base doping density N B =10 16 cm 3 minority hole lifetime of and contact width w m =2 m. The SRV (S b ) of the non metal back surface (Si SiO 2 interface), and pitch width (w) were varied as a design parameter for the local BSF cell (Structure C). We note that V OC and J SC are strongly dependent on S b as shown in Fig. 3 5 Voc decreases sharply for due to heavy recombination at the non metal back surface; similar decrease in J SC is also noticeable for For our subsequent
37 studies we assume d which is achievable with thermal oxide passivation. Fig 3 6 and Fig. 3 7 show the performance dependence of local BSF cells on pitch width with Si SiO 2 interface recombination velocity (S b ) as a parameter Table 3 3 shows the predicted perfo rmance of local BSF cells for different contact pitch width s with Si SiO 2 interface SRV S b =10cm/s. We see increased V OC due to elimination of Auger recombination from the back of the device (but poor non metal back surface passivations offsets this dark current reduction with pitch) and relatively smaller FF due to increased lateral ohmic drops as conductivity decreases when uniform BSF i s removed. As we have seen earlier for uniform BSF cases, V OC (and J SC slightly) increases and F F decreases with wider pitc h giving optimum efficiency at pitch width w=600 m in local BSF cells. Optimum pitch width decreases if non metal back surface is not passivated ( ). For thicker cells ( ) however, J SC remain s almost independent of pitch variations. In Table 3 2, we show a comparative performance analysis for the uniform BSF cell (Structure A), uniform BSF cell with shallow doping around metal (Structure B), and local BSF cells (Structure C) for contact width w=2 m, pitch width w m =300 m, and minority carrier lifetime of 1ms 3.4 Performance Dependence on Thickness Variation We now try to understand effects of varying silicon thickness (t Si ) on the local BSF device performance, which is influenced by the minor ity carrier lifetime, non metal back surface passivation, pitch (w) and contact width (w m ). For this analysis we assume, pitch w=600 m, contact width w m =2 m that give optimal performance for silicon thicknesses of t Si =80 m and 150 m at back surface r ecombination velocity, S b =10cm/s and base doping density, N B =10 16 cm 3 We show in Fig. 3 8 the device performance
38 parameters for 25 m, 80 m and 150 m thick cells with and without 1 pass internal photon re flection for different minority carrier lifetimes. V OC ( and dark current density) depends strongly on non metal back surface passivation and bulk SRH lifetime (and hence device thickness), on the other hand, J SC is weakly dependent on bu lk lifetime as long as minority carrier diffusion length is sufficien tly longer than the device thickness and non metal back surface is passivated (S b ~10cm/s) or opaque by uniform BSF. We note in Fig. 3 8 V OC decreases with increase in device thickness due to increase bulk SRH recombination, however, optically generated e lectron hole collection (or J SC ) is good as long as For long er hole lifetimes (~ 1ms) in the base the thicker (150 m) cells performs better than the thinner cells (25 m and 80 m), see Fig. 3 9 This is due to higher J SC in thicke r devices (150 m) where and due to comparatively less fractional decrease in V OC for thicker device arising from bulk SRH recombination. For short er lifetime s ( ~ 16 s ), the thinner cells perform better, probably because of higher V OC and comparable J SC with or without 1 pass internal reflection. Here we argue that for low cost silicon st arting materials (i.e. minority carrier lifetime short), V OC can be improved by increasing base doping density (N B >10 16 cm 3 ) which decreases dark c urre nt density by reducing minority carrier injection in the base without affecting J SC much as long as Thus we argue that 25 m is a near optimum thickness for low grade starting silicon materials with local BSF designs 3.5 Summary We have gained physical insights about recombination mechanism in a thin c Si solar cell and how a BSF doping profile can improve cell performance by reducing minority carrier concentration near back of the surface. We have systematically
39 analyzed differe nt BSF and back contact design structures. Effectiveness of the BSF can be significant if local doping and local contact designs are adopted. This is particularly important for thin (t Si =20 30 m) c Si solar cells. We have also shown dependence of device pe rformance on back local contact width and pitch, and on device thickness with or without internal photon reflection.
40 Table 3 1 FLOODS predicted performance versus back metal contact width for p + nn + uniform BSF cell (Structure A) Back contact pitch back base lifetime ,S f =1cm/s. Note, S f is assumed low to minimize front surface recombination. w m ( m) (mA/cm 2 ) (mV) FF (%) (%) 10 32.47 666 84 .0 19.54 2 32.51 689 83 0 20.17 Table 3 2 FLOODS predicted performances of three thin BSF cell design architectures Performances of u niform BSF cell (Structure A) uniform BSF with a shallow local doping (Structure B) and local BSF cell (Structure C) are shown Assumed parameters are uniform BSF shal low doping N S(local) =3x10 20 cm 3 base lifetime Structure Shallow d oping around back contact Uniform BSF w m ( m) (mA/cm 2 ) (mV) FF (%) (%) A NO YES 2 32.51 689 83 0 20.17 B YES YES 2 32.52 702 83.0 20.48 C YES NO 2 32.53 726 82.4 21.03
41 Table 3 3 FLOODS predicted performance versus back contact pitch for local BSF cell (Structure C) Assumed parameters are: lifetime Si SiO 2 interface SRV S b =10cm/s, SRV at back contact S=10 6 cm/s, contact width w m =2m, uniform BSF N S =1x10 19 cm 3 shallow doping N S(local) =3x10 20 cm 3 Device Structure Pitch w ( m) J SC (mA/cm 2) V OC (mV) FF (%) int (%) Loc al BSF 1000 32.53 736 81.0 20.97 Local BSF 600 32.53 733 81.7 21.06 Local BSF 300 32.53 726 82.4 21.03
42 Figure 3 1 Cross section of pn junction solar cell with uniform BSF. The basic FLOODS domain for 2 D numerical simulations is also shown The f ront emitter is an n + layer with x j =0.16 m. A low front surface recombination velocity (S f =1cm/s) is assumed to minimize front surface recombination, and to focus on carrier recombination at the back of the device.
43 Figure 3 2 FLOODS predicted V OC and J SC performance as a function of back contact pitch width for uniform BSF cell (Structure A). Performances of both the 10 m and 2 m wide (pitch) back contact cells are shown ; with
44 Figure 3 3 FLOODS predicted FF, and efficiency performance as a function of back contact pitch width for uniform BSF cell (Structure A). Performances of both the 10 m an d 2 m wide (pitch) back contact cells are shown ; with
45 Figure 3 4 Cross section s of three thin BSF solar cell architectures. Structure A: uniform BSF cell, Structure B: uniform BSF with a shallow heavy d oping around back metal contacts, and Structure C: local BSF cell. Top p + n junction depth 0.16m with top surface recombination velocity S f =1cm/s
46 Figure 3 5 FLOODS predicted V OC and J SC performance versus S b with back contact pitch as a parameter for local BSF cell (Structure C) S b is SRV at Si SiO 2 interface, h ole lifetime w m =2 m, N S =3x10 20 cm 3 N B =10 16 cm 3 under AM1 ( 92.5mW/cm 2 ) sunlight.
47 Figure 3 6. FLO ODS predicted V OC and J SC performance versus back contact pitch with S b as a parameter for local BSF cell (Structure C) S b is SRV at the Si SiO 2 interface, h ole lifetime t Si =25 m back N S =3x1 0 20 cm 3 ,contact width w m =2 m, under AM1(92.5mW/cm 2 ) sunlight.
48 Figure 3 7 FLO ODS predicted FF and efficiency versus back contact pitch with S b as a parameter for local BSF cell (Structure C). S b is S RV at the Si SiO 2 interface, hole lifetime t Si =25 m back N S =3x10 20 cm 3 ,contact width w m =2 m, under AM1(92.5mW/cm 2 ) sunlight
49 Figure 3 8 FLOODS predicted comparative performance analysis versus hole lifetime with device thic kn ess as a parameter for local BSF cell (Structure C) Effect of internal photon reflection is also shown. Assumed parameters are: pitch w = 600 m contact width w m = 2 m S b = 10cm/s local BSF N S = 3x10 20 cm 3 N B = 1x10 16 cm 3
50 Figure 3 9 FLOODS predicted local BSF cell (Structure C) efficiency versus hole lifetime for different cell thicknesses Assumed parameters: w=600 m, w m =2 m, S b =10cm/s, N S =3x10 20 cm 3 N B =10 16 cm 3 under AM1(92.5mW/cm 2 ) sunli ght. (a) W/O internal photon reflection, (b) W/ 1 pass internal reflection.
51 C HAPTER 4 BACK CONTACT SOLAR CELLS IN THIN CRYSTALLINE SILICON Back contact (BC) s olar c ells are of much interest because of their potential high efficiencies. The main features of back contact solar cells are that the front grid reflection and shadowing losses and front grid resistance losses are eliminated. Hence, the optical performance optimization of top surface and the electrical performance optimization of back junctions and contacts are isolated and independent from each other. However, BC cells require starting material with longer minority carrier lifetimes and most of these cells are fabricated on thicker substrates (~150 m) with relatively costly process technologies  Henc e, development of BC solar cells on thinner substrates is necessary to offset the higher cost associated with device fabrication The work reported in this chapter is based on a novel kerf free process  for achieving crystalline silicon with on metal (SOM), which does not involve expensive steps like ion implantation and epitaxy, and on fabrication of back contact (BC) solar cells in the thin silicon wafers, which has never been done before. The bed, preliminary experimental results are reported, and 2 D numerical simulations are used to define the optimal BCSOM cell design and to project t he ultimate one sun performance 4.1 Novel Thin Silicon BC Solar Cell Process The first ever BCSOM cell, un refined but fully processed, is pictured in Fig. 4 1. Prior to the SOM exfoliation, interdigitated phosphorous doped n + and boron doped p + regions were diffused into a 3.5 inch p silicon wafer using POCl 3 and solid source boron, respectively. The surface was then passivated with a thin thermal SiO 2 layer, and LPCVD nitride was deposited over the oxide. The metal contact regions were defined
52 lithographically (although we plan to ultimately use a novel simplified back contact process). A thin ( ) silicon layer was then separated from the mother wafer using a novel exfoliation process ; the exposed surface shows some texturing. The SOM foil, flexible a s shown in Fig. 4 2, was further processed through a nitride layer deposition for the ARC and formation of the metal contacts to the diffused regions. The BCSOM cell shown in Figs. 4 1 and 4 2 does not include three planned major refinements: front surface passivation, via a diffused FSF  and/or amorphous Si PECVD , , optimally minimized back metal contact areas (as described below, and which will be done in the mentioned simplified BC process), and optimized front and back surface texturing for enhanced photon absorption and internal reflection. Measured current voltage charac teristics of the 25m BCSOM solar cell, albeit unrefined, imply the viability of our n ovel process. We show in Fig. 4 3 a characteristic obtained with a conventional solar simulator (which is not yet modified for BC cells), for back side one sun illuminati on ( 100mW/cm 2 ), with the current normalized to the unshaded area. We get J SC 33mA/cm 2 which implies good collection efficiency; based on this result and our simulations (see below), good collection efficiency can be inferred for front side illumination as well. The measured characteristic reveals significant shunt conductance in our unrefined cell, likely at the periphery of the wafer. When it is eliminated, we infer the actual V OC 620mV, which is less than what we anticipate (based on our simulations) mainly because the front surface passivation has not yet been integrated in the process. In fact, these data, when put in perspective based on our simulations (e.g., see Tables 4 1 and 4 2), imply a photovoltaic performance that is well in line with our projections.
53 wafer thicknesses greater than quality sili con, we believe, as argued herein, that our thin crystalline silicon can be used to better exploit the BC architecture even with shorter lifetimes. Our ultimate, simplified 25m BCSOM cell process will feature effective surface passivations, surface textur ing, optimal back side diffusions and metallization, internal photon reflection, and good yield. We project, based on numerical device simulations and optimal cell design criteria implied by them, and on the preliminary experimenta l results exemplified in Fig. 4 3, an ultimate efficiency, without internal photon reflection, near 20%. 4.2 2 D imensional Numerical Simulations To project thin BC cell efficiencies and to aid their o ptimal design, we have used 2 D object oriented numerica l device simulator FLOODS  As described in Chapter 2, FLOODS was first augmented for reliable physics based numerical simulation of thin Si solar cells in AM1 (92.5mW/cm 2 ) sunlight. This tool provides flexibility for general simulations that commercial tools do not. The setup includes (i) characterization of the electron ho le generation rate defined by the AM1 photon absorption, with or without internal photon reflection, (ii) modeling of SRH and (band band) Auger carrier recombination rates with lifeti mes dependent on doping densities, (iii) modeling of carrier mobilities (and diffusion lengths) dependent on doping densities and (iv) physical accounting for energy bandgap narrowing and Fermi Dirac statistics in heavily doped regions For thin cells, th e features of heavily doped silicon [i.e., Auger recombination, bandgap narrowing ( ), and carrier degeneracy, or Fermi Dirac (F D) statistics] are
54 most important since much of the Si cell structure must be heavily do ped. In FLOODS, w e physically account for and F D statistics by using an effective doping density to m odel the minority carrier tr ansport The Auger minority carrier lifetime fundamentally varies inversely with the square of the majority carrier, or doping density. We believe our physical m odeling for the heavily doped regions is quite reliable for bipolar devices like solar cells. It is in good agreement with the physically well interpreted experimental results repo rted in  for heavily doped reg ions. 4.3 BC Cell Design and Performance Projections The basic BC solar cell structure (i.e., the 2 D FLOODS domain) that we assume is illustrated in Fig. 4 4. The front p + (or n + ) region, or FSF, which is essential in thick BC cells , is a BCSOM opti on, albeit with some process complexity. However, it can likely be avoided in optimal design as we show. The SOM process  yields silicon thickness which we assume as nominal, with about +/ 5m variations. Whereas an n type base induced type base , our preliminary BCSOM fabrication has been done with unintentionally doped p starting wafers. We thus assume here that the cell base is p with which implies a best case (after process) minority electron lifetime ( ) possibly longer than 1ms , [30 ], . (One sun operation of the cell with this doping density produces low injection, which renders the relevant lifetime.) Lower doping a t ~ could yield longer lifetime directly , and indirectly because of high injection. However, our simulations show that the degradation of the fill factor (FF) due to the high injection is predominant, implying that
55 is optimal. For this doping density, we initially assume = 1ms, but later show good thin BC cell performance for shorter lifetimes. Other impor tant device parameters are the front surface recombination velocity ( ), the back, nonmetal surface recombination veloc ity ( ), which depends on the heavy surface doping density ( ) , th e widths of the back interdigitated n + and p + regions ( and ), and the widths of the respective metal contact lines ( and ). Later we discuss the possible benefit of spaced contact windows in the z direction as noted in Fig. 4 4. Analogous results for an n base can be inferred from the simulations discussed here. We initially assume moderate n + and p + diffusions at the back of the cell (i.e., gaussian profiles with and junction depth of about 0.3 m), which trade off the bulk Auger and surface recombinations for high open circuit voltage (V OC ) ,  , although not optimally. Later we derive optimal, lighter doping densities, and discuss the ultimate attainable. Such design optimization is possible with BC cells since the metal shadowing is not an issue, although and must be restricted to avoid excessive carrier recomb ination at the metal contacts as we discuss later. For thin the only requirement is that it be long enough to render the minority electron diffusion length ( as defined by the assumed and the electro n diffusivity) in the base much longer than Then, surface recombination losses must be suppressed for maximum efficiency. The front surface of the cell, which is the surface exposed after the th in silicon exfoliation , can be subsequently passivated at low temperature via an amorphous Si PECVD process , , prior to the ARC deposition. Without an FSF, effective surface passivation is crucial for high
56 efficiency (see below). The nonmetal back surface can be well passivated, before the thin silicon formation, via a thermal oxide , . We find that which is commensurate with the assumed , is adequate, but lower can yield some improvement to the BC cell performance; higher degrades the short circuit current density ( ) as well as significantly (see Fig.4 5). To stress the importance of we give in Table 4 1 thin BC cell AM1 performances predicted via FLOODS 2 D simulations for varying and no FSF, with and, initially, (about 13% metal contact coverage of the n + /p + regions). The assumed domain width of only 150m is seemingly quite narrow, given the noted long However, we make this assumption initially to isolate the effects of and later as a basis for ch ecking performances for wider and with possible asymmetry. In Table 4 1 we give the predicted intrinsic efficiency ( ) and the actual effici ency ( ), derived using estimated ARC associated losses of 5% and assuming negligible losses in the meta l (as implied by allowed liberal back metal coverage). We do not account for any internal photon reflectio n here, although preliminary data suggest that our BCSOM process yields some. We see that which is seemingly doable , , is needed for near maximum efficiency. The predicted performance is substantially improved by the low due to reduced recombination losses at the top surface that impact as well as as shown experimentally in . For S f = 10cm/s, t he predicted current in Table 4 1 reflects a 97% internal collection efficiency.
57 We checked the benefit of the p + (and n + ) FSF as well. For which is commensurate with the FSF surface doping density , the predicted performance is a bit inferior to that for in T able 4 1, due mainly to reduced and some lowering. These losses can be mitigated some by using very high surface doping density and shallow junction depth for the FSF, but the optimal passivation appears to b e the noted (intrinsic) a Si process ,  without the FSF, if, indeed, can be achieved. For the cell in Table 4 1, the predicted performance ( and ) is limi ted by the noted collection efficiency, and by or the dark current, which comprises largely minority carrier recombination at the back metal contacts to the n + /p + regions. In fact, we find that as well as is limited by recombination sensitive to the metal contact width. The predict ed cell performances in Table 4 2 for decreasing fractional metal contact width ( where ) reflect this insight, and suggest that should be only ~1% to ensure near optimal perfor mance. For in Table 4 2, is near maximum (99% collection efficiency) for the 25m cell with no internal photon reflection and is now limited by dark current defined by nonmetal surface recombination and Auger recombination in the n + /p + regions. To gain insight on this limitation, we plot in Fig. 4 5 predicted versus for varying We note that for the assumed values of perhaps with x1/2 and x2 variations corresponding to the same variations
58 in . So, the impli cation of Fig. 4 5 is that cannot be reduced enough, for a specific to yield a significant increase in and that varying to reduce the Auger recombination, via a tradeoff between the Auger lifetime and the carrier injection level , cannot yield much benefit either. Thus, the (which is tant amount to 1 2%) cell in Table 4 2 seems to be near optimal, at least for We note also that predicted sensitivity to the expected variations around 25m, due mainly to integrated g(x) variations, is not strong; varies between 31.4 and 32.6mA/cm 2 We now check effects of wider and which may be needed in our simplified BC process. Actually is the more critical parameter here, but we initially still assume as a basis for later checking n + /p + r egion width asymmetry in the ultimate BCSOM cell design. The planned back diffusion/metallization processing of our thin Si BC cell is novel, and is much simpler than that used for thick BC cells in , provided and are sufficiently wide. The optimal BCSOM cell design, regarding processing as well as performance, thus will maximize and The former is subject to the achievable and, as we hav e found, parameters associated with the p + region; the latter is subject to the lateral majority hole transport and ohmic losses in the p base. FLOODS simulations of the cell in Table 4 2 for increasing show that continua lly decreases. We plot in Fig.4 6 the decrease, normalized to the maximum possible AM1 for a 25 m cell without internal photon reflect ion, versus w p (= w n =
59 w). The dependence is nearly linear, clearly revealing the predominant loss, in the p + region, of minority electrons generated above the p + region. The loss for increasing in Fig. 4 6 can be mitigated by using an asymmetric BC design with since virtually 100% of the electrons generated above the n + region are collected. Such mitigat ion is also indicated in Fig. 4 6 where we show, versus how the loss is reduced as is increased relative to These simulation results depend, of course, on our FLOODS modeling of the heavy doping effects in the p + region (with ), which we believe is representative. So, the optimal BCSOM cell design could b e based on a tradeoff in Fig. 4 6 between a wider needed for the BC process and the associated loss, and a wider that must be limited because of fill factor (F F) reduction due to lateral series resistance mainly in the p base, and possibly reduction due to added recombination in the wider n + region. We believe that our simple BC process can be done with If so, the n (with = 2 3m) is viable, without significant loss of as indicated by the cell in Table 4 2. However, a wider may be needed to ensure t he integrity of the process, possibly calling for For example, if is needed to be used (with = 3 6m, which could yield less contact resistance), Fig. 4 6 suggests that using = 2 3 would ameliorate the loss significantly. However, our simulations for such wider also show FF reductions that offset the higher (The corresponding reductions are not significant.) Thus, the asymmetric design is
60 not beneficial, and would give the best BC process performance tradeoff in this ca se. The FLOODS prediction of the overall performance of this wider cell is given in Table 4 3. Its efficiency ( ) is a bit less than that of the near optimal cell in Table 4 2 (and also given in Table 4 3 for comparison) due to the reduced (31 .2mA/ cm 2 ) and some FF (0.821) reduction as well. Further, we stress that the optimal BCSOM cell design yields good performance even for electron lifetimes much shorter than 1ms. We include in Table 4 3 the FLOODS predicted performance of the cell with ( ), which could be a more pragmatic lifetime dependent on . The efficiency ( ) is comparable to that of the cell w ith long We also include in Table 4 3 the predicted dependent performances for the near minimum allowed by the BCSOM process. For the narrower cells, the low ef ficiency ( ) is close to that for long ( ) as well. Our design optimization has led to narrow (2 3 m) metal contact line widths, which are d oable in our simple BC process. We nonetheless note that wider could be allowed (for example, to minimize contact resistance which we have neglected) and nearly the same cell performance could be achi eved, provided the contact lines are replaced by strings of contact windows, optimally spaced in the z direction in Fig. 4 4 to keep the total contact area fixed (i.e., 1 2% of the n + /p + region area). The perf ormance losses noted in Table 4 2 for wider would be ef fectively offset by use of the contact
61 windows, but there could be a slight reduction of FF due to added ohmic drops in the n + / p + regions along the z direction. 4.4 Further Consideration of V OC The projected BCS OM cell performances in Table 4 3 are near optimal, being limited significantly only by or the pre exponential current density [ ] of the dark current voltage characteristic; (without internal photon reflection) and FF are virtu ally maximum. Indeed, substantively higher was thick A measured at AM1.5 (100mW/cm 2 ), corresp onding to was achieved. Normalizing this result back to AM1 (92 .5mW/cm 2 ), which we have assumed, and factoring in the reduction of for thinner as implied by g(x), we get for comparison with our predictions in Table 4 3. Comparing with the for the cell with and (w hich we is almost 8x higher than that of the SunPower cell. In this section we do more FLOODS simulations to identify the source(s) of the excess recombination in our optimal BCSOM cell, and thus, perhaps, to further optimize the design. We first note, bas ed on our previous simulations, that (or ) in our DUT is not significantly affected by (see Table 4 3), (see Fig. 4 5), (see Table 4 2), nor (see Table 4 1). We infer then that the Auger recombination in the n + /p + regions is the predominant component in Reducing and and introducing a gap betwee n the n + and p + regions, albeit via more complex BC processing, yields some
62 incremental increase due to less Auger recombination. However, then becomes quite sensitive to the back surface recombination veloci ty in the gap and FF degrades due to lateral ohmic drops in the p base. Such a BC design is not a good tradeoff. The key design parameter has to be the n + /p + surface doping density Reducing will increase the Auger carrier lifetime, tending to decrease but will also increase the carrier injection level and render more critical, tending to offset the reduction. The design tradeoff is best effected, as noted in , by using low (<10 18 cm 3 ) in the nonmetal, surface passivated areas (for minimal Auger recombination) and moderate (>10 19 cm 3 ) in the metal areas (for minimal recombination at the ve ry high S ohmic contacts). Of course, this renders the processing more complex. We check a simpler process with a uniform, but lower, more optimal than we previously assumed. With reduced accordingly , we now predict for the DUT, still with as given in Table 4 4 ( ). We have reduced by only a factor of by decreasing by an order of magnitude, which suggests, in accord with , that recombination at the metal contacts is now important due to the transparency of the n + /p + regions. We therefore thin and predict the increas e given in Table 4 4, along with the modified DUT performance. For (only 0.07% metal contact coverage), is increased to 694mV, reflecting another 2x reduction in Further, is increased a bit due to reduced loss in the p + region (related to Fig.
63 4 6). We note that the thin in the modified DUT, with predicted (versus 19.1% in Table 4 3) could be effected by point cont acts in the z direction, as discussed previo usly. At this point, in the DUT is x higher than that in the SunPower cell. We surmise, based on the insight we have given, that the main component now is recombinat ion at the oxidized surfaces the n + /p + regions, defined by Indeed, simulations confirm this; with lowered to 10cm/s, we predict implying that is comparable to that of the SunPower cell. However, such a low is seemingly not possible with , and so the ultimate and cell efficiency, achieved by SunPower can be attributed to the noted optimi zation of the n + /p + surface doping densities . With simpler processing, our projected thin Si BCSOM cell performance (in Table 4 cell due to lower (without internal photon reflect ion) as well as lower Further, we stress that the optim al cell in Table 4 4 is not strongly dependent on as we discussed with reference to Tab le 4 3. 4.5 Summary We have overviewed a novel kerf free SOM pr ocess  for making thin crystalline silicon wafers with thickness and have described a process for fabricating BC solar cells in the thin Si wafers (which we plan to simplify via a novel BC technique). We have set up FLOODS  f or reliable physics based 2 D numerical simu lation of the BCSOM cells, and have used it to define the optimal design of the cells and to project their performance in AM1 sunlight. We predicted, without internal photon reflection, an
64 efficiency near 20%. Wi th significant light trapping, which we intend to incorporate, we thus can project an ultimate efficiency near 24% as has been achieved with thick Si BC cells . These projections, augmented by preliminary experimental data that we presented, lead us to believe that our thin Si BCSOM process can potentially yield a truly low cost/watt, reliable silicon solar cell.
65 Table 4 1. FLOODS predicted BC cell performance in AM1 (92.5mW/cm 2 ) sunlight versus front surface recombination velocity Device thickness no FSF, no internal photon reflection. was estimated from the predicted by assuming a 5% ARC associated loss. (cm/s) (mA/cm 2 ) (mV) (%) (%) 1 31.7 651 18.6 17.7 10 31.6 649 18.5 17.6 100 31.0 637 17.7 16.8 1000 26.3 595 13.9 13.2 Table 4 2 F LOODS predicted performance of the S f =10cm/s BC cell in Table 4 1 versus fractional metal contact coverage on the back n + /p + regions. The results for 1 2%, are near optimal. w m /w (% ) (mA/cm 2 ) (mV) (%) (%) 13.1 31.6 649 18.5 17.6 5.1 31.9 657 18.9 18.0 1.1 32.1 662 19.1 18.1 0.2 32.1 663 19.2 18.2
66 Table 4 3 FLOODS predicted performances of the near optimal 25 m BC cell design, with w (w n =w p =300 m ) widened to trade off BCSOM proce ss simplicity versus lifetime. With The predicted performances for the near minimum ( ), with allowed by the BCSOM process are given as well. ( m) ( s) (mA/cm 2 ) (mV) (%) (%) 300 1000 0.821 31.2 661 18.3 17.4 300 160 0.817 31.0 658 18.0 17.1 150 1000 0.832 32.1 662 19.1 18.1 150 160 0.830 31.9 659 18.9 18.0 Table 4 4 FLOODS predicted performance of the w=150 m BCSOM cell in Tabl e 4 3 with reduced N S =10 19 cm 3 versus the metal contact width ( m) (mA/cm 2 ) (mV) (%) 1.6 0 32.1 677 0.832 19.6 0.38 32.2 686 0.831 19.9 0.18 32.3 691 0.831 20.0 0.10 32.3 694 0.830 20.1
67 Figure 4 1. A fabricated thin ( t Si ~ 25 m ) cryst alline silicon BCSOM solar cell. The cell was processed with backside n + /p + d iffusions, surface passivation, and metal, and front side SOM defined texturing and ARC but without passivation (and no FSF). The back and front sides of the 3.5 inch flexible wafer are pictured, along with a schematic profile of the cell Photo courtesy o f Leo Mathew, AstroWatt Inc.
68 Figure 4 2 Demonstrat ion of the flexibility of the SOM foil which can broaden its possible applications. Photo courtesy of Leo Mathew, AstroWatt Inc.
69 Figure 4 3 A measured 1 sun current voltage characteristic of th e BCSOM so lar cell Illumination level AM1.5G (100mW/cm 2 ), measured with back (metal) side illumination and the current normalized to the unshaded area. The characteristic shows significant shunt conductance that yields an apparent V OC less than the actual value of 620mV.
70 Figure 4 4 Basic cross section of the BC solar cell structure, or the FLOODS domain assumed for the 2 D numerical simulations. The front p + (or n + ) region (FSF) is likely not needed in our thin Si BC cells. The actual cell, with assume d interdigitated back n + /p + regions and metal contact lines (in z, possibly over spaced contact windows), would comprise repetitions (in y) of this basic structure, as indicated. Note that the metal lines, which can be much wider than the contact lines, ar e not part of the domain.
71 Figure 4 5 FLOODS predicted open circuit voltage versus the back nonmetal surface recombination velocity for varying n + /p + surface doping density BC solar cell thickness is 25m with near optimal back metal contact coverage ( w m /w = 1.1% with w = w n = w p = 150 m with).
72 Figure 4 6 FLOODS predicted loss of short circuit current density versus increasing p + region width, for different w n /w p ratios Short circuit current loss is normalized to the 25 m AM1 maximum ( ).
73 CHAPTER 5 HETEROJUNCTION SOM S OLAR CELLS As we have noted in previou s chapters, increasing minority c arrier lifetime, minimizing the surface recombination velocity, and improving minority carrier suppression at the front an d back surfaces are essential in achieving better performance of solar cells The latter two criteria are most important for thinner cells. The s e can be achieved by using doped hydrogenated amorphous silicon (a Si:H) to form both the emitter and the back surface field (BS F) of crystalline Si solar cell. The resulting device structure is called s ilicon heterojunction (HJ) solar cell. Hydrogenated a Si can achieve substantial surface passivation (SRV ~ 10cm/s) of c Si interface and improve the minority carrier sup pression in the emitter and BSF regions significantly Insertion of an intrinsic a Si layer between the doped a Si:H an d c Si interface further enhance s the surface passivation quality of the heterojunction solar cell (Heterojunction with Intrinsic Thin layer or HIT , ,  ). Moreover, a Si:H deposition is a simple, low temperature, and inexpensive process technology Hence, the minority carrier lifetime degradation during fabrication steps is minimized. Sanyo has reported an efficiency of 23.0% on their cham pion HIT cells . Thus, a Si:H/c Si h e terojunction solar cells fabricated on thin semiconductor on metal ( SOM ) foils have the potential of achieving very high efficiency solar cells and reducing cost [ 39 ]  The main challenges in achieving such e fficient HJ solar cells on thin SOM substrates are to improve the passivation capabi lity of a Si :H /c Si interface to clean SOM substrates prior to a Si:H deposition to remove contaminants from the surface, to reduce the resistive and optical absorption lo sses in a Si :H and transparent conducting oxide (TCO), to control thicknesses of intrinsic and doped a Si:H layers, to suppress of
74 epitaxial growth at the hetero interface, to incorporate surface texture, an d to design optimum front grids For the first t ime, a remote plasma chemical vapor depos ition (RPCVD) based crystalline silicon heterojunction solar cell process was developed o n a novel thin semiconductor on metal (SOM) substrate. Thin crystalline silicon (c Si) is required to attain substantial cost reduction and fabricate higher efficiency cells. Hydrogenated amorphous silicon (a Si:H) fi lms are best suited to form low temperature hete rojunctions on thin crystalline silicon materials. As the silicon thickness decreases, the surface passivation qualit y dictates the performance of these cells. In RPCVD systems, deposition temperature, deposition rate, and the distance of the sample from the plasma source can be varied to minimize the surface damage from ion bombardment and enhance passiv ation quality. Heterojunction cells without intrinsic a Si:H layers were fabricated to stress the potential of achieving reduced plasma damage to the c Si surface and improved passivation with RPCVD. Two device architectures were fabricated on 25 m c Si SOM foils: 1) a single side heterojunction (SHJ) cell, and 2) a double side heterojunction (DHJ) cell. An efficiency of 13.4% with highest open circuit voltage of 662mV was measured on a device. Losses in these devices were identified and the optimum SHJ cell device stru cture was designed and performance predicted with numerical simulations. These results suggest that RPCVD is a potential alternative technology to achieve high effici ency heterojunction solar cells The chapter begins with a brief background and physical i nsights on minority carrier suppression in doped a Si:H emitter. The SOM exfoliation process flow to
75 produce thin (~25m) crystalline Si is described. Transparent conducting oxide (TCO) is a critical layer deposited on top of doped a Si:H film to enhance t he lateral conduction of minority carriers. Indium tin oxide (ITO) is an excellent TCO for silicon HJ solar cells and t he ITO process development is described. Subsequently, a Si:H process development using RPCVD system is described and the systematic opti mization of the process parameters e.g. sample to plasma distance, deposition temperature, and deposition time (i.e. a Si:H film thickness) is described Raman spectroscopy, TEM cross sections, and Suns Voc measurements were performed to characterize the deposited a Si:H films. Fabrication of single and double side HJ cell is described. Dark and light J V and external quantum efficiency ( EQE ) characteristics of the fabricated HJ solar cells were measured and key losses in the devices were understood. Final l y, optimum efficiency of HJ cells fabricated on thin SOM substrates, based on FLOODS simulations is projected. 5.1 Background Recombination of minority carriers at the cell surfaces and metal contacts, and in the emitter region can be min imized by reduc ing the minority carrier concentration. This can be achieved by replacing the diffused emitter with a thin doped wide bandgap material such as hydrogenated a Si:H (E g ~ 1.67eV). The a Si:H/c Si heterojunction solar cell was first demonstrated in 1977 and later developed and commercialized by Sanyo Electric Co. as HIT (Heterojunction with Intri nsic Thin layer) solar cells [37 ]. In addition, t he thin intrinsic amorphous silicon layer (i layer) deposited at low temperature enhances the surface passivation by terminating the dangling bonds at the crystalline Si interface The conversion efficiency of the HIT solar cell has been increased to 23.0% at a research level [37 ]. Fig. 5 1 shows t he schematic cross section of a typical double
76 side HJ (DHJ) solar cell wi th both the emitter and BSF regions formed by doped a Si:H layers A cross section of HIT structure is also shown in the same figure. For a pn homojunction diode, the s uppression of minority carriers (electrons) in the p + emitter depends on the doping dens ities of the emitter and base In dark, the electron current injected into the p + c Si emitter from t he n type base at maximum power point forward bias (V MPP ) can be modeled as mentioned in Chapter 3: (5 1) where is an effective intrinsic carrier density defined in Chapter 2, is the average elec tron diffusivity in the emitter and is the Gummel number of front emitter and V MPP is the maximum power point forward voltage From Chapter 2, is defined as: (5 2) where n i is the intrinsic carrier concentration in c Si, E g is the bandgap narrowing in c Si, and is the F D integral of order 1/2 with The intrinsic carrier concentration, n i is given by : (5 3) where N c and N v are the effective densities of states in the conduction and valence bands, respectively, and E g is the c Si bandgap. From Eq. (5 2) and (5 3), (5 4) The electron current injected into the p + c Si emitter can hence be written as
77 (5 5) For simplicity, we assume neglig ible bandgap narrowing and the p + do ping level is such that the hole Fermi level is few k B T/q away from the valence band edge so that we can ignore majority carrier degeneracy. Thus dark electron current density in p + emitter can be written as (5 6 ) Similarly for an a Si:H(p + )/c Si(n) heter ojunction diode, the dark electron current density d ue to the injection of minority electrons into the p + a Si :H emitter from n type crystalline Si base (assuming proper tunne ling of electrons through the hetero interface and neglecting heavy doping effec ts of p + a Si:H layer ) at maximum power point forward bias can be assumed as (5 7 ) where E g(a Si) is the bandgap of the a Si (p + ) emitter At 300K the bandgap of a Si :H (doped) is about 1.67 eV  and the bandgap of crystalline Si is about 1.12eV. Similarly, minority hole current density into the c Si(n) base due to the hole injection from a Si:H(p + ) emitter can be written as (5 8 ) Hence the ratio of emitter current to the base current can be written as
78 (5 9 ) So, in an a Si:H(p + )/c Si(n) heterojunction diode suppression of minority carrier concentration in the a Si:H emitter depends not only on the emitter doping density (N G(p + ) ) but also on the bandgap difference of amorphous Si and crystalline Si. Both of these two parameters (heavier emitter doping density and wider bandgap of a Si:H) enhances passivation quality of emitter by suppressing minority carrier concentration. In case of a homojunction pn dio de, the suppression of min ority carrier concentration depend s only on the ratio of the emitter to base doping densities. However, extremely low lifetime of minority carrier in a Si emitter, absorption of photons in a Si emitter and minority carrier transport (tunneling) across the hetero interface dictates the a Si:H thickness and hence, the HJ solar cell performance. Introduction of an intrinsic a Si:H layer further improves surface passivation of c Si interface and improves the tunneling of holes across the HJ. For simplicity and first order analysis carrier mobility differences in the emitter and base can be neglected and total integrated emitter and base doping densities can be assumed equal. For this simple case, the ratio of emitter current to the base current can be written as (5 10) We see the electron injection level into the a Si (p + ) emitter is almost 10 orders of magnitude lower than that into the c Si(n) base for same doping densities. However, the
79 hole current density (J p (c Si) ) injected in t o the base is high as defined by the recombination profile in the base, BSF region, and at the back contact and surface. Thus understanding the recombination mechanism in the base, BSF and back surface is very critical and should be reduced by clever desi gns such as local BSF/local contacts that we have discussed in detail in Chapter 2 and Chapter 3. T he a Si:H/c Si HJ solar cells on thin SOM substrates have potential of higher V OC due to minority carrier s uppression in the a Si:H emitter compared to a con ventional c Si based solar cell with emitter fabricated by thermal diffusion. 5.2 Exfoliation Process A kerf free Semiconductor on Metal (SOM) exf oliati on process , [15 ] wa s developed to produce ~ rystalline s i licon foil s from a parent wafer The process flow for exfoliation and fabrication of heterojunction cells is shown in Fig. 5 2. As shown in the figure, an exfoliated Si cell fabrication process begins by pre forming n + /n high low junction by POCl 3 or spin on diffusion on the starting waf er followed by silicon nitride (Si 3 N 4 ) passivation, and seed metal film stack deposition. A metal film is then deposited over the seed layer u sing an electrochemical deposition process. During this process hydrogen is incorporated from the electroplating b ath into the underlying substrate as shown in the SIMS data in Fig. 5 3 which is then diffused into the substrate using an optimized heat t reatment. Coefficients of thermal expansion (CTE) are different between Si and metal film. During the heat treatment or annealing process, internal stresses are created due to the thermal expansion mismatch stresses between the deposited metal film and the underlying Si substrate. The thermal expansion mismatch stresses results in exfoliation of a thin crystalline Si la yer from the parent wafer The exfoliation is aided and controlled to the required depth using a proprietary
80 Si exfoliation tool. A mechanical wedge is used which creates a fracture along a sub surface plane of the substrate. The thickness of the exfoliate d layer is controlled by varying the electroplated metal thickness, the annealing, and the mechanical wedge parameters. With this technology, w e have recently engineered the deposited metal and the thermal treatment to obtain 25 30m thin exfoliated foils from an 8 inch diameter 775m thick monocrystal line wafer, as shown in Fig. 5 4. S caling the exfoliation process to 8 inch diameter wafers enable s industry standard 125mm x 125mm pseudo square cells to be fabricated from the exfoliated foils. As seen in th e figure, the exfoliated foil has a curl associated with it due to the thermal expansion mismatch between the metal and the Si. However, the exfoliated foil is flexible due to the low thickness of Si used and is rugged and easy to handle due to the metal b acking. Therefore, the foil can be mounted flat by mechanical clamps or vacuum during processing. 5.3 Indium Tin Oxide Process Development In silicon heterojunction solar cells, low resistance lateral conduction through the a Si:H emitter is not possible. Hence, a transparent conducting oxide (TCO) layer is required to transport carriers laterally to the metallic front contact grid. Indium tin oxide (ITO) film is a degenerately doped n type semiconductor and has a high optical transmittance and a low elect rical resistivity Therefore, ITO is used as a transparent conducting oxide in a Si:H/c Si heterojunction solar cells In this study ITO films were deposited by a n R.F. magnetron sputtering process. We investigated t he effects of substrate temperature and Ar pressure on electrical and optical properties of ITO films At first, the ITO process was developed on glass substrates and then the optimized process was used for HJ device fabrication.
81 The ITO films were deposited on Corning glass slides by Kurt J. L esker R.F. magnetron sputtering tool using a 3 inch diameter ITO target with a purity of 99.99% and with weight percentage of 90 to 10 for In 2 O 3 and SnO 2 respectively. The target was bonded to a copper backing plate for cooling and protecting from fractur es. The target to substrate distance was kept at the same optimum value for each deposition. High purity Ar gas was introduced as sputtering gas and deposition pressure was controlled as a process parameter. The films were deposited without any oxygen gas into the deposition chamber due to the limitation of the sputtering system The substrate holder was heated by halogen lamps and the temperature was m onitored as a process control parameter. 13.56MHz R.F. power (100W) was introduced through a n automatic ma tching network to minimize the reflected power. The base pressure of the sputtering chamber was in the low 10 7 Torr. Pr ior to transferring into load lock chamber the glass substrates were degreased in acetone/IPA solution in an ultrasonic bath, 5 cycle r insed in DI water and then blow dried in N 2 gas. Prior to each deposition, a pre sputter etching of the target was carried out with shutter closed for 30 minutes. Deposited ITO film thicknesses were ~ 70nm and ~ 80nm. The thicknesses of the films were mea sured by surface profilometry. The sheet resistance and optical transmittance of the films were measured by four point probe and Thermo Scientific UV Visible spectrophotometer respectively. At first, we optimized our process for low substrate temperature (~200 C) and then we carried out depo sition pressure optimization to obtain low sheet resistance and high optical transmittance for our ITO films
82 The measured sheet resistances of the deposited ITO films (~70nm and ~80nm thick) for different substrate t emperatures (20 C 230 C) are shown in Fig. 5 5 The films were deposited at an Ar pressure of 10mTorr and R.F. power of 100W. It is observed that as the substrate temperature is in creased sheet resistance decreases. The decrease in sheet resistance is p robably due to the increase in oxygen vacancies with increase in substrate temperature [49 ]. For ITO film deposited at 20 C the optical transmittance is low in short and long wavelength range s as seen from Fig. 5 6 For purposes of comparison, the trans mittance of the glass substrate without an ITO coating is also shown in Fig. 5 6 The optical transmittances are almost identical for films deposited at 200 and 230 C for longer wavelength with a slight increase in transmittance in shorter wavelength s (U V Visual) for 230 C deposited film. It is important to note, that as the thickness of the film is increase d the sheet resistance decreases but on the other hand, optical transmittance decreases for a particular deposition temperature and Ar pressure. In this study, we optimized our ITO process for a film thickness of 70nm and substrate temperature of 200 C A low temperature ( 200 C) ITO process development was essential for our HJ solar cell fabrication process, as we discuss it later in this chapter. We then studied the effect of deposition pressure (i.e. Ar pressure) on sheet resistance and optical transmittance of the ITO films. T he deposition pressure was varied from 10 mTorr to 5mTorr and Fig. 5 7 shows the dependence of sheet resistance on deposit ion pressure. The substrate temperature was kept at constant 200 thickness was ~70nm for this experiment. It is observed that sheet resistance of the films decreases with decreased deposition pressure and at 5mTorr Ar pressure, the
83 sheet resistance of the film is ~ Lower sheet resistance of the film d eposited at lower pr essure indicates increased car r ier mobility in films deposited at lower Ar pressure [51 ] since the electron concentration s remain s almost the same at different deposition pressures The optical transmittance cha racteristics are shown i n Fig. 5 8 For purposes of comparison, t he transmittance of the glass substrate without an ITO coating is also plotted in the same figure The variation of the optical transmittance in the longer wavelength regions (near IR region) indicates the variation of the electron concentration in the ITO films [49 ]  The optical transmittance s in the longer wavelength region for ITO films d eposited at different Ar pressures do not show much difference This reflects that the electron concentrations in the fil ms deposited at different Ar pressures are almost the same. The variation in optical transmittances in shorter wavelengths (UV Visual) indicates variations in optical bandgap of ITO film. The increase in short wavelength optical transmittance of ITO films deposited at lower Ar pressure suggests widening of optical bandgap of ITO [ 49 ]  The ITO film deposited at 200 C and 5mTorr Ar pressure has an average optical transmittance of ~ 84% and sheet resistance of This proc ess is used for subsequent TCO layer deposition during a Si:H/c Si HJ solar cell fabrication. 5.4 Overview of Remote Plasma Chemical Vapor Deposition System RPCVD is a promising low Si film deposition with lower plasm a damage than conventional Plasma Enhanced Chemical Vapor Deposition (PECVD). In RPCVD system, during the deposition process there is no direct exposure of the sample to the plasma as the sample is not immersed in the plasma. Instead, the sample is held do wnstream from the glow discharge. This can
84 reduce the damage to the silicon surface from energetic particles in the plasma and improve passivation quality. The RPCVD system uses ultra high vacuum (UHV) and the plasm a is remote from the sample Both of thes e features can contribute to improved passivation; hence, in this study a majority of the cells have been fabricated without the intrinsic amorphous Si:H (i layer) passivation as used in typical heterojunction cells. Single side and double side silicon het erojunction cells were fabricated. Potential of improving surface passivation using intrinsic a Si:H is shown. The optimal distance of the wafers from the plasma source an d the temperature can be varied to obtain best passivation of c Si surface This proc ess with optimum sample to plasma distance and temperature can be transferred to SOM foils to form heterojunction for solar cells. A schematic of the deposition chamber for the RPCV D system [38 ] is shown in Fig. 5 9 5.5 RPC VD Process Development and Cell Fabrication After the exfoliation process was carried out, RPCVD technology was used to fabricate the cells on these SOM foils. The RPCVD tool was first used to develop the highest possible Suns Voc on silicon substrates that were 500m thick. The optimal distance of the wafers from the plasma source and the temperature were evaluated. These conditions were then used to deposit a stack of p + a Si:H films on the SOM foils. We fabricated solar cells with front heterojunction and back surface field (BSF) usin g the 25m c Si foils that were exfoliated from (100) monocrystalline CZ 10 20 ohm cm n ty pe Si wafers. As shown in Fig. 5 2, the single side heterojunction (SHJ) cell fabrication process begins with forming a uniform diffused n + c Si junction as BSF on th e starting wafer using POCl 3 diffusion. Subsequently, a nitride passivation layer was formed over the wafer surface, and it was patterned non lithographically [1 5 ] to open
85 contact holes to the n + diffusion. A metal film was then deposited electrochemically over the wafer and the exfoliation process was carried out. This metal film serves some key roles in our process: (a) it provides the mismatch stresses needed for exfoliation, (b) it serves as a backside contact to the cell and (c) it provides mechanical support to enable handling of the exfoliated Si during subsequent processing. After exfoliation cell fabrication is completed by deposition o f a heterojunction stack and Transparent Conducting Oxide (TCO) film followed by silver screen printing For the st udy of optimal distance of the samples from the plasma source and the temperature, the cells were completed without nitride passivation on the back surface and thermal evaporation of aluminum was used to form back metal electrodes. Prior to RPCVD a Si depo sition, the exfoliated Si SOM foils were degreased in an ultrasonic bath of acetone and isopropyl alcohol, followed by 5 cyle DI water rinse, piranha, and 5 cycle DI water rinse to remove the residual organic contaminants. Samples were then immersed in a d ilute HF solution to remove the native oxide and to leave the surface hydrogen terminated. Finally, samples were blow dried with nitrogen and immediately transferred into the RPCVD system. Argon plasma was inductively excited with ~7W of RF power (13.56 MH z). For a typical p + a Si:H layer deposition process, the base pressure was in the low 10 9 30mTorr deposition pressure, and pure SiH 4 and 100ppm B 2 H 6 /H 2 mixture were introduced as source gases. Intrinsic a Si:H layer was Si films were deposited on 136nm thick thermally grown oxide on Si control wafers to measure a Si:H film thickness by spectroscopic ellipsometry and to carry out Raman spectroscopy. Solar cells were completed by deposition of ~ 80nm thick indium tin oxide
86 (ITO) as an anti reflection coating (ARC) and transparent front conductive layer in an printed silver grid lines to complete the front electrodes. Fig. 5 10 (a) shows a photograph of completed 1.1cm 2 area heterojunction cells fabricated on a 25m thick flexible SOM foil. We also fabricated dual side heterojunction cells us ing the thin exfoliated foils. For a dual side heterojunction cell proc e ss, steps (b) and (c) in F ig. 5 2 were replaced with deposition of a heterojunction a Si film stack (intrinsic and doped layers) on a starting wafer followed by deposition of an ITO film. Formation of back intrinsic/n + a Si:H stack in double side heterojunction cells was carried ou t in an industrial PECVD system, while formation of front p + a Si:H stack for front emitter was carried out in RPCVD system. The rest of the processes including exfoliation and front si de heterojunction deposition were id entical in both process flows. Thus all processing after exfoliation was carried out at temperatures below 200 C. Cells were fabricated both with and without the intrinsic amorphous Si layer in the heterojunction stack. Table 5 1 shows the overview of device structures studied in this work with corresponding device performances. As seen from the table that the per formance of a HJ cell depends on thicknes s of doped a Si:H layer type of BSF (diffused for single side HJ cell or doped a Si :H for double side HJ cell ), and incorporation of intrinsic a Si:H layer. 5. 6 RPC VD Process Optimization First, a Si:H(p + )/c Si(n) heterojunction cells without intrinsic a Si:H layer were formed on 500m thick wafers to characterize the effective passivation quality of a Si:H film and to optimize the emitter for the front of the sin gle heterojunction cell (Fig. 5 10 (b)) in terms of S uns Voc. Table 5 2 shows the optimal conditions depend on sample temperature and distance of samples from plasma source. In order to understand the
87 variation of the RPCVD layer across the wafer, four different cells were formed on 2 square inch samples. Th ese 2 square inch samples were diced from larger wafers and from larger SOM foils due to limits of the RPCVD chamber size (Fig. 5 10 (a)). The data across these four cells suggest that an optimal distance from the plasma source exists. Wh en the wafers were fa rther away or closer than this optimal distance, the V OC of the cells were degraded or varied substantially for an optimal temperature. In addition, Raman spectroscopy of the p doped a Si:H layers at different deposition temperatures was used to characte rize the cry stallinity of the films (Fig. 5 11 ). The characteristic amorphous Si Gaussian peak (~480cm 1 ) shifts to the right with increasing temperature, making the film more microcrystalline. Therefore, p doped a preliminary studies showed that with the introduction of hydrogen, t he films become more microcrystalline at a particular temperature. Based on this analysis and the Su ns Voc results shown in Table 5 2, sample N2 4 showed the least variation, best surface passivation as evident from the highest Suns Voc numbers [ 3 9] [41 ], and thus its process was used to fabricate cells on the thin crystalline SOM foils. Fig. 5 12 shows a high resolution TEM cro ss section of a completed cell. As can be seen, the exfoliated Si is completely monocrystalline. Further, the interface between the deposited amorphous Si layers and the exfoliated Si is clean and sharp, showing no epitaxial growth of the deposited layers 5. 7 Single Side Heterojunction Cell Fig. 5 13 shows the JV characteristics and external quantum efficiencies (EQE) of a Si:H(p + )/ c Si(n) single side heterojunction cells (SHJ) fabricated on thin exfoliated SOM foil as measured under AM1.5G illumination. These cells have no intrinsic a Si:H
88 layer passivation with a uniform n + diffusion (surface doping concentration, N S =1x10 20 cm 3 an d junction depth x j =0.5m) as back surface field (BSF) and non optim ized local back contact (Fig. 5 10 (b)). The cell 41 04 1 had V OC =605mV, J SC =29.6mA/cm 2 2 and shadowing from front metal grid was measured ~11.5%. Since mid and long wavelength EQE of both devices are almost identical, the improvement in J SC by 0.8mA/cm 2 in sample 41 05 1 from sample 41 04 1 is due to less optical absorption in thinner a Si:H(p + ) layer [ 4 0], [41 ]. Thicker p + layer seems to improve V O C by improving surface passivation, but the cell performance is offset by decrease in J SC with increasing a Si:H(p + ) layer thickness. FLOODS , [42 ] simulation was used to determine the performance limiting factors in these single side heterojunction cel ls (SHJ). V OC in sample 41 04 1 is limited by excessive band to band Auger carrier recombination in the heavily doped back n + diffusion region. The overlap of EQE and relatively lower quantum efficiency in the long wavelength region for both samples sugges t that excessive recombination of photo generated carriers is taking place in the heavily doped n + diffusion region. Thus both V OC and J SC can be improved by incorporating an optimum doping density for the n + BSF regi on. For bulk Si resistivity of 1 cm, the optimal n + diffusion is found to be a profile with N S =10 19 cm 3 and x j = 0.3 0.5m. For thinner cells, the minority carrier diffusion length is usually longer than cell thickness (~25m) and hence, design optimiza tion of the BSF and back metal contact is important. At this optimum doping density for n + diffusion, recombination at the back metal contact and back surface is now important due to the transparency of the n + region  , , [43 ]. Hence, we propo se a device architecture based on local back doping and local back contact with
89 better surface passivation for thinner (sub 30m) SOM based single side heterojunction cell similar t o the PERL cell architecture , [44 ] as shown in Fi g. 5 1 4 Moreover, V O C and J SC can be improved further by incorporat ing an intrinsic a Si:H layer [39 ]. Two thin single heterojunction SOM solar cells with different back side contact schemes were studied. Sample 07 B had uniform n + diffusion as BSF, uniform back metal coverag e and no nitride passivation at back and Sample 36 10 had uniform n + diffusion as BSF, local back contact (non optimized) and nitride passivation at back. The a Si:H(p + )/c Si stack was identical for both cells as evident from the overlap of EQE at shorte r wavelengths as seen in Fig. 5 1 5 Relatively lower quantum efficiency at mid to longer wavelength range in sample 07 B is due to additional recombination in at the back metal contact. Local back contact and back passivation significantly improve device pe rformance at longer wavelengths (as seen from higher J SC in 36 10) and reduce recombination at back contact (as seen from higher V OC in 36 10). We performed numerical simulation in FLOODS to project performance of single side heterojunction cell with local back doping and local back contact scheme. We assumed ohmic back contact (with recombination velocity S=10 6 cm/s at the back contacts), base donor doping density N B =10 16 cm 3 and local n + diffusion in contact areas was a Gaussian doping pro file with N S =31 0 20 cm 3 x j coverage of 1% with 600m contact spacing (pitch). Back contact areas were assumed to be heavily doped to reduce recombination at these contac t areas by suppressing minority carrier co ncentration in these areas , [24 ]  Surf ace recombination velocities at the non metal front and back surfaces are assumed to be 10cm/s which we think is achievable [1 7 ] by thermal oxide passivation at the back surface and
90 improved a Si:H passivation by RPCVD at the front surface , [ 4 0 ]. Dev ice thi ckness was varied with minority carrier lifetime in the bulk as a parameter to proje ct the cell performance. Fig. 5 16 shows the simulated intrinsic device performance of the suggested device under AM1.5G solar spectrum. For thin cells (t Si SC has negligible sensitivity to the minority carrier lifetime which relaxes the requirement for the more expensive high lifetime substrates. V OC increases with decreasing substrate thickness due to a decrease in the bulk recombination For realis tic 7% loss due to front electrode shading a nd front surface reflectance , , [45 ] simulation predicts 18.6% and 17.4% cell efficiencies for 160s and 16s respectively without any internal light trapping. With 1 pass internal photon reflection, 2 0% efficiency cell is projected for a bulk lifetime of 160s. Enhanced light trapping can be achieved by front surface texturing and incorporating back surface reflectors [4 6 ]. 5. 8 Double Side Heterojunction Cell Double side heterojunction (DHJ) cells on SOM were also fabricated via RPCVD. JV characteristics under AM1.5G illumination fro m DHJ cells are shown in Fig. 5 1 7 We obtained a relatively high conversion efficiency of 13.4% (V OC =645mV, FF=66.2%, and J SC =31.35 mA/cm 2 ) on sample M 1 1 without incorpo rating surface texture and front i layer. More importantly, we found good relationships between the solar cell characteristics and thickness of a Si:H(p + ). It is seen from the light JV data that sample M 1 2 with 6nm thick a Si(p + ) has higher J SC and lower V OC while sample M 1 1 with 12nm thick a Si(p + ) has lower J SC and higher V OC The V OC data indicates the effectiveness of thickness of a Si:H (p + ) in passivating the Si surface. J SC decreases with increasing p layer thickness as the optical transmission through the p layer decreases expo nentially with its thickness [41 ]. It is evident that there is an optimum a
91 Si:H(p + ) thickness to achieve highest device performance. The FF of these devices is limited by the series resistance associated with screen print ed front contact. Heterojunction with intrinsic a Si:H layer (i layer) can improve the passivation of c Si interface and improve open circuit voltage [ 13], [ 4 0 ]. Hence we fabricated sample M 4 2 with an intrinsic 6.8nm thick a Si:H layer followed by 12nm t hick a Si:H(p + ) layer in front. The light JV data for samples M 1 1 and M 4 2 indicates that the i layer improves the surface passivation as evident from the increase in V OC of about 17mV and nearly unchanged J SC Slightly lower short wavelength external quantum efficiency in J SC is due to increased optical losses in the non optimized intrinsic a Si:H and a Si:H(p + ) layers. Relatively lower FF is due to non optimized a Si:H(p + ) and i layer thicknesses , [47 ] in addition to series resistance associated with screen printed front contacts. Fig. 5 18 shows dark saturation current density, J 0 vs. V OC obtained from the samples (07 B, 41 04 1, M 1 1, M 4 2) with same a Si:H(p + ) l ayer thicknesses The dark saturation current density was estimated from the foll owing equation: (5 11 ). As expected, the J 0 values are in agreement with that measured from Suns Voc, and there is a strong correlation between J 0 and device structure. In case of n + BSF, by reducing the back contact percentage and incor porating nitride passivation, J 0 reduces by a factor of ~3.5 (samples 07 B and 41 04 1). J 0 in sample M 1 1 decreases by a factor of 4.5 from that of sample 41 04 1 by incorporation of a Si:H(i)/a Si:H(n + ) on the back of the c Si SOM substrate instead of a heavily doped n + BSF. Incorporation of a Si:H(i)/a Si:H(n + ) layer is quite bene ficial in reducing the minority carrier recombination at the back of c Si SOM substrate. In sample M 4 2, J 0 decreases by a factor of ~2 from
92 that of sample M 1 1 by incorporat ing intrinsic a Si:H layer at the front, w hich suggests, in accord with , [ 4 0 ] that intrinsic a Si:H layer improves surface passivation. Efficiency of the cells fabricated with RPCVD can be improved further by using optimum thicknesses of a Si:H(p + ) a nd i layer which will lead to increased open circuit voltage and FF. The distance of the sample from plasma during i layer deposition can be studied further to reduce plasma damage for achieving better c Si surface passivation. Moreover, the degradation of FF and J SC in our fabricated cells can be minimized by using optimum front contact and transparent conductive oxide (TCO) to reduce external series resistance and optical losses. The Suns Voc measurements highlight the projected efficiency of our cells if external series resistance is minimized. Pseudo FFs of samples 41 04 1, M 1 1 and M 4 2 were measured to be 82.4%, 82.0%, and 76.1%, r espectively as shown in Table 5 1. This Suns Voc data indicate effectiveness of RPCVD in achieving high efficiency solar cells. Moreover, internal light trapping by front and back surface texturing and improved back reflection will lead to efficiency comparable to conventional thick Si high efficiency heterojunction solar cells. 5. 9 Summary We obtained physical insights in to the minority carrier injection into the c Si base from the a Si:H emitter and minority carrier suppression inside a Si:H emitter in a silicon heterojunction cell. A n R.F. magnetron sputtering system based process was developed to deposit ITO as a transp arent conducting layer. We dev eloped for the first time a low temperature RPCVD process to form a Si:H(p + ) /c Si heterojunction on thin crystal line silicon ( SOM ) foils that are only ~25 microns thick. We studied the dependence of sample to plasma distance a nd deposition temperature on plasma damage and passivation quality of thin crystalline cells. Two device architectures were fabricated with
93 the RPCVD process. A 13.4% efficiency heterojunction solar cell with the highest V OC of 662mV on these SOM foils wit hout any high efficiency features such as light trapping was demonstrated. Effect of p doped a Si:H thickness and incorporation of intrinsic a Si:H were studied. This work suggests that RPCVD is a potential technology to deliver lower plasma damage and imp roved pa ssivation for thin crystalline Si heterojunction solar cells
9 4 Table 5 1. Overview of HJ device structures and corresponding performances Sample i layer thickness (nm) p layer thickness (nm) Structure V OC (mV) J SC (mA/cm 2 ) FF (%) (%) pFF (%) 41 04 1 12nm SHJ 605 29.6 62.8 11.2 82.4 41 05 1 6nm SHJ 524 30.4 63.5 10.1 78.6 M 1 1 12nm DHJ 645 31.4 66.2 13.4 82.0 M 1 2 6nm DHJ 535 32.8 73.9 12.9 80.2 M 4 2 6.8nm 12nm DHJ 662 31.6 52.6 11.0 76.1 Table 5 2 Measured Suns Voc from cell s fabricated on 500m c Si wafers for varying sample positions and temperatures. Smaller position number indicates sample distance closer to the plasma source Sample Position Temp ( C) Suns Voc across dies (mV) Range (mV) Gas flow rate SiH 4 (%) B 2 H 6 (%) Ar (%) N1 2 3"48 250 571 535 531 508 63 5 50 80 N1 1 3"32 250 580 558 576 545 35 5 50 80 N1 3 3"23 250 577 536 614 592 78 5 50 80 N1 4 3"32 200 585 588 587 579 9 5 5 0 80 N2 1 3"23 200 557 572 557 565 15 5 50 80 N2 2 3"23 225 610 609 611 611 2 5 50 80 N2 4 200 629 634 632 631 5 5 100 80 N2 3 225 603 615 613 614 12 5 80 80
95 Figure 5 1. C ross section of a typical double side heterojunction cell on a n type crystalline silicon SOM wafer The doped a Si:H layers forms both the emitter a nd the BSF regions, and improve the surface passivation The indium tin oxide (ITO) film acts as a transparent conductive layer for the charge carriers, and helps reduc ing the later al resistance. The cross section of the HIT device structure is also shown.
96 Figure 5 2 Process flow for fabricating exfoliated single heterojunction solar cells
97 Figure 5 3 SIMS profile showing H ydrogen i ncorporation into the Si substrate during the electroplating process
98 Figure 5 4 Photograph s of exfoliated SOM foils (a) Picture of exfoliation from an 8 inch diameter wafer showing 25m thin Si foil along with the residual wafer after exfoliation and (b) 125mm x 125mm pseudosquare cut out from the foil. Photo courtesy of Leo Mathew, AstroWatt Inc and Dabraj Sarkar.
99 Fi gure 5 5 Variation of sheet resistance of ITO films as function of substrate temperature. Ar pressur e was 10mTorr and two different film thicknesses were analyzed for this experiment.
100 Figure 5 6 Variation of o ptical transmittance characteristics of ITO films as function of substrate temperature. Film thicknesses were ~ 70nm and the Ar pressure was 10mTorr.
101 Figure 5 7 Variation of sheet resistance of ITO films as a function of deposition pressure. Substrate temperature was 200 C and film thicknesses were ~70nm.
102 Figur e 5 8 Optical transmittance spectra of ITO films as a function of Ar pressure. Substrate temperature was 200 C and film thicknesses were ~70nm.
103 Figure 5 9 Schematic diagram of RPCVD deposition chamber The key features are that the plasma is remot e from the sample, and sample can be placed at different distances downstream from plasma source.
104 Figure 5 10 Photograph of SOM foil with fabricated HJ cells, and cross section of fabricated single side HJ cells. (a) Photograph of the completed 25m t hick SOM foil with 1.1cm2 heterojunction cells, (b) schematic device structure of single side HJ cell fabricated on 25m SOM foil without any a Si(i) layer, and (c) schematic device structure fabricated on 500m wafer Photo courtesy of Leo Mathew, AstroWa tt Inc. and Dabraj Sarkar.
105 Figure 5 11 Raman spectra of a Si:H(p + ) films deposited at different temperatures. Deposition conditions wer e same as that of sample N2 4 as in Tab le 5 2
106 z Figure 5 1 2 High resolution TEM cross section of thin HJ solar cell on SOM foil. The cross section clearly shows monocrystalli ne exfoliated Si with amorphous Si and ITO layers on top Photo courtesy of Leo Mathew, AstroWatt Inc. and Dabraj Sarkar.
107 Figure 5 13 Illuminated JV and EQE characteristics of single side HJ solar cells. (a) AM1.5G illuminat ed JV characteristics, and (b) EQE data of single side heterojunction solar cells.
108 Figure 5 1 4 Cross section of the proposed single side heterojunction (SHJ) cell on a n type SOM wafer The back side has local BSF and local contact to reduce recombination losses (similar to PERL cell). Local BSF and local contact designs are described in Chapter 3.
109 Figure 5 1 5 EQE data indicating carrier recombina tion at the back due to uniform metal contacts and no surface passivation.
110 Figure 5 1 6 FLOODS predicted dependence of intrinsic V OC J SC i con thickness for varying bulk lifetime S f =S b =10cm/s and zero internal reflection
111 Figure 5 1 7 Illuminated JV and EQE characteristics of double side HJ solar cells. (a) The AM1.5G illuminat ed JV characteristics, and (b) EQE d ata of double side heterojunction solar cells
112 Figure 5 1 8 Comparison of dark saturation current densities vs. open circuit voltage s for different cell structures
113 CHAPTER 6 SUMMARY AND FUTURE WORK 6.1 Summary The main goal of this work was to design and fabricate high efficiency solar cells on exfoliated thin crystalline Si substrates that are about ~25 m thick In order to do this, we need ed to gain physical insights and indentify key loss mechanisms in a thin crystalline Si solar cell by device sim ulation. We upgraded FLOODS for photovoltaic device simulation with necessary physical models. Exfoliated semiconductor on metal (SOM) substrates were produced at AstroWatt Inc. and solar cells were fabricated on these substrates at the Microelectronic Re search Center at the University of Texas at Austin. The crystalline Si photovoltaic industry has been scaling down the Si wafer thickness in order to reduce costs and potentially attain higher efficiencies by minimizing bulk recombination As the silicon t hickness decreases, the surface passivation quality and recombination in the back surface field region and at contacts dictate the performance of these cells. With FLOODS, we were able to comprehensively study the effect of surface r ecombination velocity, minority carrier lifetime in the bulk, doping profile of back surface field region, and contact coverage on the performance of the solar cell in the sub 40 m thickness. We used physics based FLOODS simulations to understand and propose efficient device str ucture s for back contact and heterojunction solar cells. Thin crystalline silicon based heterojunction solar cells require excellent passivation on the front and back side of the cell. Plasma enhanced chemical vapor deposition (PECVD) and hot wire chemical vapor deposition (HWCVD) are typically
114 used to fabricate a Si:H/c Si based heterojunction so lar cells. In this work, we developed remote plasma chemical vapor deposition (RPCVD) as an alternative promising low Si:H fi lm deposition with lower plasma damage t han conventional PECVD. RPCVD was shown to be a potential technology to achieve low effective surface recombination velocity (SRV) Two device architec tures were fabricated on 25 m c Si SOM foils: 1) a single side h eterojunction (SHJ) cell, and 2) a double side heterojunction (DHJ) cell. An efficiency of 13.4% with highest open circuit voltage of 662mV was measured on a device. Based on our experimental work, it is evident that the re are two ways to achieve high effi ciency thin Si solar cells: 1) single side HJ cells with local back junction and local back contacts, and 2) double side HJ cells The proposed devic e architectures will have front and back texture for enhanced light trapping. Low SRV is possible with our RPCVD process as evident from very high open circuit voltage obtained from the fabricated cells High efficiency HJ solar cells are a chievable with thin crystalline Si substrates that are comparable to conventional thicker HJ solar cells. 6.2 Future Work Future work should include expansi on of FLOODS to include carrier transport models across the a Si:H/c Si heterojunction. On the experime ntal side, characterization of the passivation quality of RPCVD deposited a Si:H layers should be done by measuring eff ective surface recombination velocity. Q uantum efficiency and hence short circuit current density should be improved by incorporating light trapping schemes suc h as surface texturing and back surface reflectors.
115 6.2.1 Carrier Transport Model a cross HJ The critical transport mechanism is believed to be tunneling acro ss the heterojunction interface which depends on the band alignment and band offset . Minority carrier lifetime is extremely short in the a Si/c Si heterojunction, but the injected minority carrier concentration is suppressed significantly inside the heterojunc tion as shown in Sec. 5.2 The photon absorption profile and hence the optical carrier generation rate g(x) are modified in the heterojunction due to wider bandgap and different absorp tion coefficients of a Si. Moreover, the transparent conductive oxide (typically Indium Tin Oxide or ITO) is a degenerately doped n type semiconductor with bandgap of 3.7eV and can be treated as a metallic contact or as a part of the hetero interface   We propose to develop reliable engineering based models for the carrier transport mechanism and optical absorption across the heterojunction and implement the models in FLOODS 6.2.2 Surface Recombination Velocity Characterization Very high open c ircuit voltage (>662mV) measured from fabricated HJ cells suggests RPCVD is a potential technology to achieve improved surface passivation by reducing the plasma re lated damage to the crystalline silicon surface. We intend to characterize surface recombina tion velocity at the a Si:H/c Si interface by the quasi steady state photoconductivity (QSSPC) method [ 48 ]: (6 1 ) eff bulk is bulk lifetime and S eff is measured effective surface recombination velocit y.
116 6.2.3 Light Trapping by Surface Texturing We intend to improve short circuit current density by introducing surface texturing. We have optimized our RPCVD process for polished SOM foils. Our future work will include optimization of RPCVD process for te xtured surfaces. We have already developed a KOH based texturing process for our SOM foils.
117 LIST OF REFERENCES  valley of death to the golden decade: crystalli ne silicon solar cells from 10 m to 100 m Proc. 19th Workshop Cryst Si. Solar Cells and Modules Vail, CO, pp. 192 195, Aug. 2009.  Proc. IEEE Electron Devices Meeting pp. 359 362, Dec. 2007  using a separation U.S. Patent 7749884 Jul. 6 2010.  Proc. 31st IEEE Photovoltaic Spec. Conf pp. 889 894, Jan. 2005.  W. Shockley and H. J. Queisse balance limit of efficiency of p n junction solar cells J. Appl. Phys ., vol. 32 pp. 510 519, Mar. 1961.  Sanyo Electric Co., Ltd., News Release May 22, 2009.  P. J. Cousins, D. D. Smith, L. Hsin Chiao J. Manning, T. D. Dennis, A. Waldhauer K. E. Wilson, G. Harley, and W. P. Mulligan, : improved performance at lower cost Proc. 35th IEEE Photovoltaic Spec. Conf ., Honolulu, HI, pp. 275 278, Jun. 2010.  M. E. Law based modeling of silicon inte grated circuit processing:an object oriented approach Computational Materials Science vol. 12 p p 289 308, Nov 1998.  M. Stocks, A. Blakers, and A. Cuevas Proc. 26th IEEE Photovoltaic Spec. Conf ., Anaheim, CA pp. 67 70, Sep. 1997.  A. Wang, J. Zhao, and M. A. Green Appl. Phys. Lett ., vol. 57 pp. 602 604, Aug. 1990.  M. D. Lammert igitated back contact solar cell: A IEEE Trans. Electron Devices vol. 24, pp. 337 342, Apr. 1977.  R. A. Sinton backside contact solar cells IEEE Trans. Electron D evices vol. 37 pp. 348 352, Feb. 1990.  H. Sakata, Y. Tsunomura, H. Inoue, S. Taira, T. Baba, H. Kanno, T. Kinoshita, M. Taguchi and E. Maruyama progress of next generation very thin HIT solar cells Proc. 25th EUPVSEC Valencia, Spain, pp. 1102 1105, Sep. 2010
118  J. Appl. Phys ., vol. 105 094507 1 094507 7, May 2009  R. A. Rao, L. Mathew, S. Saha S. Smith, D. Sarkar, R. Garcia, R. Stout, A. Gurmu, D. Ahn, D. Xu, D. Jawarani, E. Onyegam, M. Hil ali, S. Banerjee, and J. Fossum olar cell technology Proc. 26th EUPVSEC Hamburg, Germany, pp. 2439 2442, Sep. 2011  J. G. Fossum analytic model for minority carrier transport in heavily doped regions of silicon devices IEEE Trans. Electron Devices vol. 28, pp. 1018 1025, Sep 1981  A. Cuevas, P. A. Basore, G. Gi roult Maltakowski, and C. Dubois recombination velocity of highly doped n type silicon J. Appl. Phys ., vol. 80 pp. 3370 3375, May 1996  S. M. Sze, and K. K. Ng semiconductor devices 3rd Edition, John Wiley & Sons Hoboken, New Jersey, 2007.  aided numerical analysis of silicon solar cells Solid State Electronics vol. 19, pp. 269 277, Apr. 1976.  J. G. Fossum, R.P. Mertens D.S. Lee, and J.F. Nijs recombination and lifetime in high ly doped silicon Solid State Electronics vol. 26 pp. 569 576, Jun.1983  unified mobility model for device simulation IEDM Tech. Dig. pp. 357 360, Dec. 1990.  G. Masetti, M. Severi and S. Solmi, M odeling of carrier mob ility against carrier concentration in arsenic phosphorus and boron doped silicon IEEE Trans. Electron Devices vol. 30 pp. 764 769, Jul. 1983.  on incident solar energy Proc. 20th Annu. Meet. Inst. Environ. Sci p. 21 Jan. 1974  path to 25% silicon solar cell efficiency: history of silicon cell evolution Prog. Photovoltaics: Res. Appl ., vol. 17 pp. 183 189, May 2009.  operation of back surface field silicon solar cells IEEE Trans. Electron Devices vol.24 Apr. 1977.  M. Sheoran, A. Upadhyayaa, and A. Rohatgi lifetime and efficiency enhancement due to gettering and hydrogenation of defects during cast multicrystalline silicon solar cell fabricati on Solid State Electronics vol. 52 pp. 612 617, May 2008.
119  N. Stem, C.A.S. Ramos and M. Cid circuit voltages: theoretical and experimental optimizations of rear passiva ted silicon solar cells using FZ and CZ wafers Solid State Electronics vol 54 pp. 221 225, Mar. 2010.  D. Kray, and of selective phosphorous laser doping in high efficiency solar cells IEEE Trans. Electron Devices vol. 56 pp. 1645 1650, Aug. 2009  A. Das, V. Meemongkolkiat D. Kim, S. Ramanathan, and A. Rohatgi % efficient screen printed cells with spin on dielectric passivated boron back surface field IEEE Trans. Electron Devices vol. 57, pp. 2462 2469 Jul. 2010.  W. P. Mulligan, D. H. Rose, M. J. Cudzinovic, D. M. De Ceuster, K. R. McIntosh, D. D. Smith, and R. M. Swanson of solar cells with 21 % efficiency Proc. 19th EUPVSEC Paris, France, p p 387 390 2004.  M. Schaper, J. Schmidt, H. Plagwitz, and R. Brende l efficient crystalline silicon solar cell with amorphous silicon rear surface passivation Photovoltaics: Res. Appl. vol. 13, p p 381 386, Aug 2005. [3 2] M. Lu, S. Bowden, U. Das and R. Birkmire, Si/c Si heterojunction for interdigitated back contact solar cell Proc. 22nd EUPVSEC Milan, Italy, p p 924 928 2007.  V. G. Weizer, H. W. Brandhorst, J.D. Broder, R. E. Hart, and J. H. Lamneck, degradation effects in terrestrial silicon solar cells J. Appl. Phys ., vol. 50 pp. 4443 4449, Jun. 1979.  J. G. Fossu m hysical model for the dependence of carrier lifetime on doping density in nondegenerate silicon Solid State Electron ., vol. 25, pp. 741 747, Aug. 1982.  J. G. Fossum e fficiency p + n n + back surface field silicon solar c Appl. Phys. Lett ., vol. 33, pp. 238 240, May 1978.  Y. Taur and T. H. Ning, Fundamentals of m odern VLSI d evices U.K.: Cambridge Univ. Press 1998. [37 ] t status of high Solar Energy Mater. & Solar Cells vol. 95, pp. 18 21, Jan. 2011 [38 ] plasma enhanced CVD of silicon: reaction kinetics as a function of growth J. Electronic Materials vol. 19, pp. 1089 1094, May 1990.
120  T. Swada, N. Terada, S. Tsuge, T. Baba, T. Takahama, K. Wakisaka, S. Tsuda, efficiency a Si/c Proc. 24th IEEE Photov oltaic Spec. Conf. Waikoloa, HI, pp. 1219 1226, Dec. 1994.  Prog. Photovoltaics: Res. Appl ., vol. 13, pp. 481 488, Sep. 2005.  Z. C. Holman, A. Descoeudres L. Barraud, F. Z. Fernandez, J. P. Seif, D. Wolf, IEEE J. Photovoltaics vol. 2, pp. 7 15, Jan. 2012. [42 ] J. G. Fossum, D. Sarkar, L. Mathew, R. Rao, D. Jawarani, and M. Proc. 35th IEEE Photovoltaic Spec. Conf ., Honolulu, HI, pp. 3131 3136, Jun. 2010. [43 ] Textured Multicrystalline a Appl. Phys. Lett. vol. 73, pp. 1991 1993, Jul. 1998. [44 ] PERL silicon solar cell: recent improvements in high efficiency silic on cell Solar Energy Mater. Solar Cells vol. 41 42, pp. 87 99, Jun. 1996. [45 ] optimisation of the emitter region and the metal Prog. Photovoltaics: Res. Appl ., vol. 8, pp. 603 616, Nov. 2000. [4 6 ] Appl. Phys. Lett ., vol. 70, pp. 390 392, Jan. 1997. [47 ] U. Das, S. Hegedus, L. Zhang, J. Appel, J. Rand, an hetero Proc. 35th IEEE Photovoltaic Spec. Conf ., Honolulu, HI, pp. 1358 1362, Jun. 2010.  F. M. Schuurmans, A. Schonecker, A. R. Burgers, and W. C. Sinke, Simplified evaluation method for light biased effective lifetime measurements Appl. Phys. Lett ., vol. 71 pp. 1795 1797 Sep 1997.  S. Ray, R. Banerjee, N. Basu, A. K. Batabyal, and A. K. Barua doped indium oxide thin J. Appl. Phys ., vol. 54, pp. 3497 3501, Jun. 1983  O. Tuna, Y. Selamet, G. Aygun, and L. Ozy 55402 55409, Jan. 2010.
121 [51 ] Films, vol. 303, pp. 151 155, Jan. 1997.
122 BIOGRAPHICAL SKETCH D abraj Sarkar received his B.Sc. in electrical and electronic engineering from Bangladesh University Engineering and Technology in 2003. After his baccalaureate, h e worked as an engineer trainee at GE Healthcare and then as an engineer at GrameenPhone Ltd. He is currently doing his Ph.D under Prof. Jerry G. Fossum in electrical engineering at t he University of Florida He collaborated with AstroWatt Inc. and Prof. Sanjay Banerjee in Austin, TX from June 2010 June 2012 where he developed a Si:H/c Si heteroju nction process and indium tin oxide(ITO) based transparent conducting oxide(TCO) process for its silicon heterojunction solar cell process line. His research interests include device modeling, simulation, remote plasma CVD, fabrication and characterization of thin a Si:H/c Si heterojunction and back contact solar cells.