Chemical vapor deposition of thin films for diffusion barrier applications :

MISSING IMAGE

Material Information

Title:
Chemical vapor deposition of thin films for diffusion barrier applications :
Physical Description:
ix, 419 leaves : ill. ; 29 cm.
Language:
English
Creator:
Bchir, Omar James
Publication Date:

Subjects

Subjects / Keywords:
Chemical Engineering thesis, Ph. D   ( lcsh )
Dissertations, Academic -- Chemical Engineering -- UF   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 2004.
Bibliography:
Includes bibliographical references.
Statement of Responsibility:
by Omar James Bchir.
General Note:
Printout.
General Note:
Vita.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 003100349
sobekcm - AA00004688_00001
System ID:
AA00004688:00001

Full Text










CHEMICAL VAPOR DEPOSITION OF THIN FILMS FOR DIFFUSION BARRIER
APPLICATIONS

















By

OMAR JAMES BCHIR


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2004




























I would like to dedicate this dissertation to my nephew, Matthew Edward Brown, and to
the memory of my grandparents.















ACKNOWLEDGMENTS

On a professional note, I would like to thank my supervisory committee

(especially my advisor, Dr. Tim Anderson, and Dr. McElwee-White) for their constant

guidance and assistance throughout my Ph.D. work. I would also like to thank the past

and present members of Dr. Anderson's, Dr. McElwee-White's and Dr. Norton's

research groups, with whom I have collaborated during my time at UF, and Dr. Jianyun

Shen, with whom I collaborated on thermodynamic modeling. I would also like to thank

Dr. Holloway for use of his 4-point probe device, and the staff at the Major Analytical

Instrumentation Center (MAIC), including Eric Lambers, Wayne Acree and Brad

Willenburg, for valuable assistance and training with various characterization techniques.

On a personal note, I would like to thank my parents (Sara and Hachemi Bchir)

for their encouragement, love and support throughout my life, and for their strong

emphasis on the importance of education. I would also like to thank my sister, Annissa

Brown, my brother-in-law Eddie, and my nephew Matthew, for their encouragement and

support. Thanks also go to my girlfriend Luciana Manfrim for her love and support.

Lastly, thanks go to all of the friends that I have made through the years while growing

up in Florida, attending Georgia Tech, working at Fluor Daniel and attending UF.














TABLE OF CONTENTS

Page
A CKN OW LED GM ENT S......................................................................................... iii

ABSTRACT .............................................. .................... .................................... viii

1 INTRODUCTION
1.1 Background ..................................................... .......................... 1
1.2 Comparison of Diffusivities for Al and Cu in Si ............................ 4
1.3 Diffusion Barrier Requirements and Properties............................ 6
1.3.1 Film Structure.......................... .................................... 9
1.3.2 Electrical Properties ......................................... .......... ... 11
1.3.3 Conform ality ................................................. .............. 14
1.3.4 A dhesion............................................ ............................ 16
1.4 Dielectric Material Considerations............................. ............ ... 17
1.5 Diffusion Barrier Failure Mechanisms.......................... .......... 17
1.6 Diffusion Barrier Deposition Techniques ....................................... 21
1.6.1 Physical Vapor Deposition (PVD) ................................... 21
1.6.2 Chemical Vapor Deposition............................................. 24
1.6.3 Atomic Layer Deposition (ALD) ......................................... 37
1.7 Copper Deposition Methods...................................... ............ .... 44
1.8 Statement of Problem ................................................ ........ ..... 49
1.9 H ypothesis...................................................... ........................... 51

2 REVIEW OF THE LITERATURE
2.1 Cu-Si Interconnects without a Barrier............................................ 53
2.2 Refractory Metal-Based Barriers .............................................. 54
2.2.1 Unary Refractory Metal Barriers...................................... 54
2.2.2 Binary Refractory Metal-Based Barriers.......................... 56
2.2.3 Ternary Refractory Metal-Based Barriers............................ 63
2.3 Justification for WNx (and WNxCy) as the Barrier Material............. 71
2.4 WNx Film Properties ............................................................... 72
2.5 WNxCy Film Properties ............................................... ........... 75
2.6 Amorphous WNx Film Deposition .................................................. 77
2.7 Demonstrated Uses of WN ........................................... ........... .. 78
2.8 WNx Deposition Techniques....................................... ............ .. 85
2.8.1 Annealing W in NH3 .............................................. ........... .. 85
2.8.2 Plasma Nitridation / Ion Implantation of W...................... 86
2.8.3 Pulsed Laser Deposition................................... ........... 86
2.8.4 Reactively Sputtered WNx Deposition.............................. 87









2.8.5 LPCVD WNx Deposition ............................................... 87
2.8.6 PECVD WNx Deposition ............................................... 91
2.8.7 MOCVD WNx Deposition ............................................... 93
2.8.8 ALD WNx Deposition .................................... ............. 97
2.8.9 ALD WNxCy Deposition ................................... .......... 99
2.9 Conclusions ..................................................................................... 100

3 EQUILIBRIUM MODELS FOR WNx-BASED BARRIERS
3.1 Motivation and Method..................................................................... 101
3.2 Constraints..... ....................................... 111
3.3 Degrees of Freedom ........................................................................ 111
3.4 Computational Methods .................................................................. 112
3.5 Phase Equilibrium in the W-N System .......................................... 114
3.6 Previous Studies of the W-N Phase Diagram................................. 125
3.7 W-N Optimization Results and Discussion.................................... 126
3.8 Stable Solid Phases in the W-C System......................................... 139
3.9 Previous Study of the W-C Phase Model....................................... 142
3.10 Equilibrium in the W-C-C1-H-N System....................................... 145
3.11 Optimization of the FCC P-WNxCy Gibbs Energy ........................ 153
3.12 Degrees of Freedom (DOF) Analysis.............................................. 168
3.12.1 Homogeneous Gas Phase Speciation .................................. 171
3.12.2 Heterogeneous Gas Phase Speciation ................................. 177
3.12.3 Heterogeneous Solid Phase Equilibrium............................. 182
3.13 Predicted W-C-N Ternary Phase Diagram.................................... 186
3.14 Predicted P-WCo.5 P-WNo.5 Pseudobinary Equilibrium............. 190

4 EXPERIMENTAL APPROACH FOR CVD
4.1 Substrate Preparation......................................................................... 194
4.2 Solvent Tests ................................................................................... 195
4.3 Description of CVD System Components ...................................... 196
4.4 Start-Up........ ....................................... 201
4.5 Copper D eposition............................................................................. 201
4.6 Analysis Techniques ....................................................................... 202
4.7 Precursor Screening Procedure ....................................................... 208

5 EVALUATION OF C14(CH3CN)WN-i-Pr AS A SUITABLE
WNx PRECURSOR
5.1 Synthesis of Isopropyl [C14(CH3CN)WN-i-Pr] Precursor............... 210
5.2 Solvent Selection............................................................................... 211
5.3 Precursor Mass Spectral Pre-Screen .............................................. 212
5.4 Film Structure ................................................................................... 217
5.4.1 X RD Results........... .............................................................. 217
5.4.2 TEM .. ................................................................................... 223
5.5 Film Com position.............................................................................. 225
5.5.1 A ES ..................................................................................... 225
5.5.2 AES Depth Profiling ........................................................... 229









5.5.3 XPS...................................................................................... 233
5.5.4 SIM S Depth Profiling............................................................ 253
5.6 Film Growth Rate (XSEM )............................................................. 255
5.7 Film Incubation Time and M orphology............................................ 259
5.8 Film Electrical Properties.................................................................. 264
5.9 Effect of NH3 and N2 Addition to Film Growth ............................. 267
5.9.1 Film Structure........................................................................ 267
5.9.2 Film Composition...... ... ............................................................ 270
5.9.3 Growth Rate ........................................................................ 277
5.9.4 Film Resistivity ................................................................... 279
5.9.5 Sheet Resistance.................................................................. 280
5.10 Conclusions for Use of NH3 and N2 with i-Pr................................ 281
5.11 Effect of Solvent Change on Carbon Content................................... 282
5.11.1 Film Structure........................................................................ 284
5.11.2 Film Composition...................................................................... 287
5.11.3 Carbon Deposition Rate ...................................................... 291
5.11.4 Conclusions On Effect of Solvent Change.......................... 297
5.12 Conformality Tests............................................................................ 298
5.13 Adhesion Tests .................................................................................. 298
5.14 Conclusions on Use of C14(CH3CN)WN-i-Pr to Deposit WNx....... 299

6 EVALUATION OF C14(PhCN)W(NPh) AS A SUITABLE WNx PRECURSOR
6.1 Film Growth Studies ....................................................................... 303
6.2 Synthesis of Phenyl [C14(PhCN)W(NPh)] Precursor ........................ 303
6.3 Solvent Selection............................................................................... 303
6.4 Precursor M ass Spectral Pre-Screen .............................................. 304
6.5 Film Structure ................................................................................... 308
6.5.1 XRD .. ................................................................................... 308
6.5.2 Lattice Parameter................................................................... 313
6.5.3 Polycrystal Grain Size......................................................... 315
6.6 Film Composition.............................................................................. 316
6.6.1 AES ..................................................................................... 316
6.6.2 Film Growth Rate (XSEM )................................................. 320
6.7 Film Electrical Properties.................................................................. 322
6.7.1 Film Resistivity ................................................................... 322
6.7.2 Film Sheet Resistance ......................................................... 323
6.8 Conclusions on Use of C14(PhCN)W(NPh) to Deposit WNx............ 324

7 EVALUATION OF C14(CH3CN)WN-C3H5 AS A SUITABLE
WNx PRECURSOR
7.1 Film Growth Studies ....................................................................... 331
7.2 Synthesis of Allyl [C14(CH3CN)WN-C3Hs] Precursor..................... 332
7.3 Solvent Selection............................................................................... 333
7.4 Precursor M ass Spectral Pre-Screen .............................................. 334
7.5 Film Structure ................................................................................... 336
7.5.1 XRD .. ................................................................................... 336









7.5.2 Lattice Parameter................................................................... 339
7.5.3 Polycrystal Grain Size......................................................... 341
7.6 Film Com position.............................................................................. 342
7.7 Film Growth Rate (XSEM)............................................................. 344
7.8 Film Electrical Properties................................................................ 345
7.8.1 Film Resistivity ................................................................... 345
7.8.2 Film Sheet Resistance ......................................................... 346
7.9 Conclusions on Use of C14(CH3CN)WN-C3H5 to Deposit WNx...... 347

8 COPPER TESTS ON BARRIER FILMS FROM C14(CH3CN)WN-i-Pr
8.1 Copper Deposition Results............................................................... 357
8.2 Electrical Measurements ................................................................. 357
8.3 Scotch Tape Tests............... ................................................................ 358
8.4 Barrier Integrity Tests ..................................................................... 358
8.4.1 X RD .. ................................................................................... 359
8.4.2 Electrical Measurements ..................................................... 364
8.4.3 Depth Profiling Analysis..................................................... 365
8.5 Conclusions Regarding Cu Testing of Deposited Barrier Layers..... 366

9 RECOMMENDATIONS FOR FUTURE WORK
9.1 Control Film Composition .............................................................. 368
9.2 Decrease Deposition Temperature .................................................. 371
9.3 Decrease Barrier Thickness............................................................... 371
9.4 Further Conformality Testing............................................................ 372
9.5 Further Adhesion Tests ................................................................... 372
9.6 Further Barrier Integrity Analyses .................................................. 373
9.7 Deposition on Alternate Substrate Materials .................................. 375
9.8 Testing of Films for X-ray Absorption Mask Applications.......... 376
9.9 Deposition of Alternate Barrier Materials....................................... 376
9.10 System Modifications........................................................................ 376

REFERENCES................................................................................ 378

APPENDIX

A CVD SYSTEM PROCESS FLOW DIAGRAMS ........................................ 409

B OTHER PRECURSORS TESTED............................................................. 415


BIOGRAPHICAL SKETCH ...................................................................................419




F-


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

CHEMICAL VAPOR DEPOSITION OF THIN FILMS FOR DIFFUSION BARRIER
APPLICATIONS

By

Omar James Bchir

August 2004

Chairman: Timothy J. Anderson
Major Department: Chemical Engineering

The semiconductor industry is transitioning from aluminum to copper as the

interconnect material for integrated circuits. Diffusion barriers are essential for

preventing copper migration into silicon in the resulting integrated circuits. Metalorganic

chemical vapor deposition (MOCVD) is a useful technique for conformal deposition of

thin barrier films. By manipulating the molecular structure of the MOCVD precursor(s),

it is possible to control the structure and properties of the deposited film.

The focus of this work has been to examine novel precursors for use in MOCVD

of thin film diffusion barriers. To date, the suitability of a variety of novel precursors for

deposition of tungsten nitride (WNx) for diffusion barrier applications has been tested.

The WNx precursors were of the form C14(CH3CN)WNR, where R represents the

isopropyl, phenyl, and allyl group, among others. Mass spectrometry fragmentation

patterns for each precursor were studied to pre-screen candidate precursors. Film

properties were examined by several different characterization techniques, including








XRD, AES, XPS, SEM, and 4-Point Probe. Data from these techniques were then

correlated to pre-screen fragmentation patterns to determine the impact of the imido (R)

group on film deposition. Apparent activation energies for film growth from the allyl,

isopropyl and phenyl precursors, for example, were 0.15, 0.84 and 1.41 eV, respectively,

and were directly related to the strength of the N-R bond in the precursors.

In addition to experimental testing of new precursors for WNx deposition,

thermodynamic phase equilibria for the W-N-H-C-C1 deposition system was assessed.

Modifications were made to the previously reported W-N binary model to include new

experimental data from the literature on the desired FCC WNx phase as well as the SHP

WN phase. Once the binary W-N diagram was reassessed, this model was merged with

the existing W-H-C-Cl database to create an initial model for solid-gas equilibrium in

our system at our experimental conditions. Using a combination of XRD, AES, and XPS

data from our films, the W-N-H-C-C1 model was modified to include

carbon-nitrogen-vacancy interactions in the face-centered cubic (FCC) WNxCy solid

phase, so that the new model reflected our experimental results. Finally, the

low-temperature ternary W-C-N phase diagram was predicted.















CHAPTER 1
INTRODUCTION

1.1 Background

The semiconductor industry continues to shrink the size of transistors on

integrated circuits (ICs) to increase the number of transistors per chip, which translates to

increased IC device speed. To shrink these transistors, the sizes of the device features

(trenches, vias, etc.) and the metal interconnects ("wires" that connect the individual

transistors on the chip) must be decreased. Recent accomplishments have reduced

feature sizes to 100 nm, yielding more than 400 million transistors per chip. The

International Technology Roadmap for Semiconductors [ITR02] predicts IC feature sizes

to shrink continuously over the next several years, in the following steps: 100 nm (in

2003), 90 nm (2004), 80 nm (2005), 70 nm (2006), 65 nm (2007), 45 nm (2010), 32 nm

(2013), and 22 nm (2016).

Aluminum (Al) and Al-based alloys have historically been used as the metal

interconnect materials on integrated circuits. Al is reaching its usability limit as industry

continues to shrink IC features. For a given current flow, decreasing the interconnect

cross-section leads to an increase in current flow density. At high current flow density,

Al suffers from electromigration, where electrons flowing through the metal interconnect

impart enough momentum to carry the metal atoms with them. Once current flow density

rises above the critical value (106 A/cm2 for Al), electromigration leads to voids

(openings) and hillocks (pileups) in the interconnect wiring (Figure 1-1). Formation of









voids in the Al interconnect leads to an open circuit, which causes device failure. Al

doped with Cu (termed Al(Cu)) was used to increase the electromigration resistance, but

this scheme is also reaching its usability limit. A different interconnect metal is required

to meet present and future current flow density demands.


electrons- (a)

hill k
void


electrons (b) .



Figure 1-1. Before (a) and after (b) electromigration induced hillock and void formation
in an Al interconnect.

Industry has already transitioned to copper as the interconnect material for the

intermediate and upper wiring levels on IC devices, due to copper's higher

electromigration resistance and 40% lower bulk resistivity (1.67 p&2-cm for Cu vs. 2.65

tQ-cm for Al) [Cha89, Kol91]. Copper has ten orders of magnitude higher

electromigration resistance than Al, which will increase device lifetime [Mur95], and its

lower resistivity will decrease the resistance-capacitance (RC) time constant, which is a

large factor in chip performance for ICs with feature sizes below 250 nm [Hu98]. The

following relation defines the resistance-capacitance time delay:


RC-= PI (1-1)
tM* t*

where p is resistivity (S2-cm), I is the length (cm) of the interconnect line, e is the electric

permittivity (C/V*cm) of the insulating film, tM is the thickness of the metal interconnect

(cm) and tiLD is the thickness of the neighboring insulator (cm). A decrease in the RC









time constant leads to higher device switching speed [Pau87], and the switch from Al to

Cu may lead to a 15% increase in overall microprocessor speed [IBM97]. While bulk Cu

has a resistivity of 1.67 i.Q-cm, the resistivity of IC Cu lines with thickness above 50 nm

ranges from 1.7 to 2.0 .Q-cm [Kal00]. In comparison, Cu-doped Al lines AI(Cu) have

resistivities ranging from 3.0 to 3.5 pQ-cm. In addition to resistivity and

electromigration improvements, Cu is said to provide higher IC production yield than

Al-based devices with similar design [Sin02a]. Other advantages of switching to Cu

interconnects include a decrease in the number of interconnect levels, a roughly 30%

decrease in power consumption for operation at a given frequency, and a cost savings of

roughly 30% per interconnect level due to integration of dual damascene processing

[Nov00]. Silver (Ag), which has an even lower bulk resistivity (1.59 uQ-cm) than Cu,

has been considered for interconnect metallization, but it suffers from poor

electromigration resistance [Mur95], making its use as the metallization material

unlikely.

The improved electrical properties of Cu are highly desirable, but with these

benefits come several drawbacks. These include copper's tendency to corrode under

standard fabrication conditions, copper's lack of a stable self-passivating oxide (like

A1203 on Al), the inability to chemically etch Cu, poor adhesion to low dielectric constant

(low-k) materials, and rapid diffusion/reaction with Si and Si02 [Hu98, MerOl]. Copper

diffusion into neighboring layers can cause an increase in contact resistance, a change in

the barrier height, leaky p-n junctions, embrittlement of the contact layer, deep level

traps and destruction of electrical connections to the chip [Kal00, Wit80]. Copper


M









contamination levels in Si at and above 1013 cm-3, for example, are believed to reduce

device yield [IstOO].

1.2 Comparison of Diffusivities for Al and Cu in Si

At high temperatures, both Al and Cu diffuse into Si, and when put into direct

contact with Si, they will intermix rapidly. While Si is non-reactive with Al [RamOO], it

has a small solubility range in Al (0.25 to 1.5% by weight) [Jae88]. At high temperatures

(>10000C), Al diffuses substitutionally into Si, where Al replaces a Si atom in the Si

lattice.

Copper, though, has much higher mobility in Si, and is very reactive with Si and

SiO2 substrates [Bau97], forming copper silicide compounds, such as Cu3Si and CusSi

[Bau97, Hym98, Kwo95, RamOO, Rei92], which cause strong deterioration of contact

systems [WanOlb]. Copper's high mobility in the Si lattice extends over a wide

temperature range. Cu moves into the Si lattice by substitutional diffusion at high

temperature (>8000C), but also by interstitial diffusion at low temperature (<7000C),

where Cu atoms diffuse between the Si atoms in the lattice (through the interstices). Both

interstitial and substitutional diffusion of Cu into the Si lattice occur simultaneously, with

interstitial diffusion dominating at low temperature and substitutional diffusion

dominating at high temperature.

While the diffusivity of Al in Si at 5000C is less than Ix10-20 cm2/s [IstOO], the

diffusivity for Cu in Si depends both on type and concentration of dopants in the Si

substrate. The nature of p-type silicon is such that acceptor atoms (such as boron, B)

substituted on Si lattice sites have a negative charge (B-). Since Cu diffuses through Si

as a positive ion (Cu') [Bai96, IstOO], the negatively charged B- ions tend to "trap" some









of the diffusing Cu ions, giving Cu an apparent diffusivity that is lower than that in

intrinsic (undoped) Si. In p-type Si with a boron concentration of 5.0 x 1020/cm3, for

example, the diffusivity of Cu is 2.0 x 106 cm2/sec at 5000C, while that in intrinsic Si is

2.0 x 10-5 cm2/sec [Ist00]. At room temperature, the diffusivity of Cu in p-type and

intrinsic Si, measured by the transient ion drift method [Hei97], ranged from 3.3 x 10-13

cm2/s in p-type Si up to 2.7 x 10-7 cm2/s in intrinsic Si [Ist98]. This indicates that after

30 minutes at room temperature, Cu can diffuse a distance of 0.5 pm in p-type Si and

400 glm in intrinsic Si.

Since the diffusivity of Al in Si is so small at the processing temperature, the main

problem with Al arises when it penetrates the barrier material and accumulates at the

barrier-Si interface. This accumulation can lead to formation of Al spikes at the

interface, which short out shallow junctions (Figure 1-2a). Copper's high diffusivity in

Si, however, means that Cu will not build up at the barrier-Si interface until the Si has

reached its saturation value of Cu. Hence, Cu can diffuse into the bulk of the Si wafer

and move laterally to contaminate a large area on the wafer [IstOO], as depicted in Figure

1-2b.

While Cu has high diffusivity in Si at room temperature, its solubility is very low

(less than 1 Cu atom per cm3) [IstOO], so formation of a Cu-Si solid solution is unlikely.

The diffused Cu will therefore form complexes, precipitates and agglomerates with Si.

Four typical reactions can occur for Cu in Si. These include formation of point

defects/complexes, copper silicide precipitates, addition to existing defects, and

outdiffusion to the Si surface. More information on Cu interdiffusion and reaction

mechanisms in Si is available elsewhere [IstOO]. In addition to Si, Cu also diffuses









rapidly through Si02 under thermal and electrical stress, leading to high leakage currents

and dielectric breakdown [Jai99]. Diffusion of Cu atoms into SiO2 typically occurs due

to application of an external electric field, which causes Cu to be ionized (to Cu') and

accelerated into the SiO2 [Kri01]. Copper's high diffusivity in Si and SiO2 makes the

need for a robust thin film diffusion barrier to separate Cu from neighboring layers on the

IC very critical. Figure 1-2c depicts a stable device structure resulting from use of a

diffusion barrier.


ILD Al ILD D LD LD Al or Cu ILD
Si a Si b Si c

Metal Diffusion Diffusion
Spikes Region Barrier


Figure 1-2. Metal to silicon contact layers: (a) Al on Si without a diffusion barrier
suffers from junction spiking. (b) Cu on Si without a barrier suffers from
massive diffusion (c) Metal to silicon contact with a diffusion barrier
remains intact.

1.3 Diffusion Barrier Requirements and Properties

A diffusion barrier is a material used to separate two layers that, if put in direct

contact, would interdiffuse and/or react with one another. Diffusion barriers are typically

classified into one of three types: passive, sacrificial and stuffed. Passive (ideal) barriers

are inert with respect to the layers that they separate, and have low solubility for the

neighboring semiconductor and metal. Sacrificial barriers react with one or both of the

neighboring layers, and are eventually consumed. The rate of reaction between the

barrier and neighboring layers must be slow enough that the lifetime of the barrier is

longer than that of the device. If the barrier has a shorter lifetime than the device, the

sacrificial barrier will be totally consumed and the device will fail prematurely. Stuffed









barriers are polycrystalline films that have their grain boundaries stuffed with a material

that blocks diffusion [Wol86]. By stuffing these boundaries, the dominant

low-temperature route for diffusion through the barrier is minimized. Stuffing occurs

due to both physical and chemical effects. The stuffing "agent" should physically reduce

the amount of pore space in the films and should chemically repel the Cu from diffusing

through the grain boundary [Par95a]. As an alternative to grain boundary stuffing, high

temperature annealing has been used to "cure" deposited barrier films, so that

micro-defects such as grain boundaries are removed. Although this technique does

decrease the number of grain boundaries in the film, the high temperatures also increase

diffusivity of Cu in the barrier film, and can cause Cu penetration [Bai96]. The "curing"

properties of a high temperature thin-film anneal are therefore tempered by the increased

mobility of Cu in the barrier film at higher temperatures.

In addition to stuffing the grain boundaries, barrier effectiveness can be further

enhanced by minimizing the number of grain boundaries in the barrier film. Deposition

at low temperatures provides this benefit and several more. First, low temperature

deposition minimizes the number of grain boundaries by favoring amorphous film

growth. Second, in nitride films, for example, excess nitrogen is retained at lower

deposition temperature, and is therefore available to stuff any grain boundaries that may

form during thermal cycling. Third, the likelihood of damaging temperature sensitive

components (such as low-k dielectric polymers) on the device during barrier deposition

is minimized, provided that a temperature ceiling of 4000C is maintained [Eis00a, Ele02,

HauOO, Kim03b, Les02, SunOlb]. Deposition above 4500C can increase the compressive

stress in Cu films, which is relieved by formation of Cu hillocks [Jai99]. Fourth, low









deposition temperatures require a smaller thermal budget. The advantages provided by

low temperature deposition can greatly reduce the probability of Cu penetration through

the barrier and decrease power consumption. In practice, low deposition temperatures are

desirable, but depositing the barrier films at the lowest possible temperature may not be

the optimal approach. Depositing the barrier film at the highest allowable processing

temperature, assuming that the film structure remains amorphous at this temperature,

would prevent significant changes to barrier film structure during subsequent thermal

cycling.

During IC processing, the diffusion barrier material is typically deposited between

a metal layer (usually Al or Cu) and a semiconductor layer (usually Si or GaAs), a metal

and a dielectric (e.g., SiO2 or low-k), or two metal layers [Lin98a]. The requirements for

an ideal diffusion barrier are listed below. Film structure, electrical properties,

conformality and adhesion property requirements will be discussed further in the

following sections.

Barrier must prevent diffusion between and be non-reactive with
neighboring device layers.

Barrier should have amorphous film structure to eliminate grain
boundaries, which are facile paths for Cu diffusion through the barrier.

Barrier should be deposited at low temperature to prevent damage to
temperature sensitive components on the IC device and to enable
amorphous film growth.

Barrier should have good conformality and uniformity across the wafer, to
ensure good barrier coverage over small device features and uniform
barrier deposition on all devices across the wafer.

Barrier should have low bulk electrical resistivity and low contact
resistivity to Cu, to minimize resistance to current flow.

Barrier should have good thermal conductivity to minimize heating of the
barrier layer.









Barrier should have minimal contamination levels; halide impurities can
cause corrosion of the Cu layer, while oxygen and free carbon impurities
can increase film's electrical resistivity.

Barrier should have minimum thickness to maximize copper's
cross-section in the interconnect and to foster low contact resistance in the
metallization stack.

Barrier should promote adhesion between device layers, as poor adhesion
leads to electromigration of Cu at the de-adhered interface.

Barrier should have high thermal and structural stability, to prevent failure
of the barrier during exposure to thermal and mechanical stresses at
processing conditions.

Barrier should enable direct Cu electroplating to eliminate the need for a
Cu seed layer, and promote nucleation of the Cu (111) orientation, which
has the greatest resistance to electromigration.

Barrier should be compatible with chemical mechanical planarization
(CMP) processes, and act as a good CMP stop layer to eliminate need for
a separate stop layer deposition step.

1.3.1 Film Structure

The ideal diffusion barrier structure is a defect free single crystal film [Nic78].

This structure does not contain grain boundaries, which are interface regions

("micro-defects") between crystal grains in a polycrystalline material. Grain boundaries

are a "short circuit" path for rapid intermixing of the two neighboring layers, and are the

dominant cause for barrier failure at low temperature [Bai96]. In a single-crystal barrier

film, all diffusion occurs through bulk defects (including vacancies and dislocations),

which is inherently much slower than diffusion through grain boundaries. Deposition of

single crystal barrier films is impractical, however, due to low growth rates, lattice

mismatch with the underlying substrate, and deposition temperature limitations [Kal00].

The next best solution is to deposit an amorphous, dense, smooth, defect free film [Bai96,

Kwo95]. Amorphous films have short-range order (< 5 A), but no long-range order (>








20 A) [E1190]; therefore no polycrystalline grain boundaries exist to enable rapid

diffusion. In addition to having an amorphous microstructure, the density of the barrier

material should be as close to the bulk value as possible, to eliminate voids as possible

diffusion paths. Polycrystalline films, especially those with a columnar microstructure or

those with equiaxial grains of size similar to film thickness, are the poorest performers in

diffusion barrier applications. These films contain grain boundaries that can extend from

one side of the barrier to the other, meaning an easy diffusion path for Cu through the

barrier. Nano-crystalline films, which are polycrystalline films with grain size below

-50 A [Kal00], are more effective than polycrystalline films with larger grains, but are

not as effective as amorphous diffusion barriers. Examples of single-crystal,

polycrystalline and amorphous films are shown in Figure 1-3.


Metal t, ..... Metal M Meta Meta




Substrate a Substrate / I Substrate c Substrate d

Polycrystals Grain Polycrystals Grain
Boundary Boundary
Boundary

Figure 1-3. Diagram of (a) Single crystal barrier. (b) Polycrystalline barrier with
equiaxial grains. (c) Polycrystalline barrier with columnar grains. (d)
Amorphous barrier without grain boundaries.

Several different methods are used to deposit amorphous films [Siv95]. The first

is to deposit films at low temperature, which enhances sticking probability and

suppresses surface diffusion, forcing films toward a disordered state. The second is to

deposit films at high nucleation (arrival) rates, so that many small nuclei form on the

substrate surface, preventing coalescence of small nuclei into larger polycrystals. The









third is to add contaminants (such as N, C, etc.) to impede surface diffusion of the metal

species, imparting disorder and decreasing the ability of nuclei to coalesce and grow. In

addition to these deposition methods, careful selection of materials can assist formation

of an amorphous film. A material's composition may be chosen, for example, so that

several stable phases exist together at equilibrium. In a binary system, a film

composition should be chosen in a two-phase equilibrium region, while in a ternary

system, a three-phase equilibrium region should be selected [RamOO]. The competing

stable phases neighboring these regions should be line compounds (have minimal solid

solubility regions), have very different compositions, and have complex crystal structures

[RamOO]. Compared to a broad solid solution phase, line compounds are inflexible to

stoichiometry deviations, which can disrupt polycrystal formation and trigger amorphous

growth. In a binary system, then, choosing a film composition in a two-phase

equilibrium region bounded by two line compounds should increase the likelihood of

amorphous film deposition.

1.3.2 Electrical Properties

To obtain any benefit from the lower bulk resistivity of Cu, the barrier (liner)

thickness should be less than 10% of the overall linewidth [Hu98]. A thicker barrier

means decreased cross-sectional area of Cu available to conduct current, which causes an

increase in the effective resistivity of the Cu line. For a Ti/TiN/AI(Cu)/Ti/TiN sandwich

structure, the effective resistivity was estimated to be 4.1 pQ-cm, due to the high

resistivities of the Ti and TiN layers, as well as that for TiA13, which forms during

annealing at the Ti-AI(Cu) interface [Hu98]. For Cu lines with widths ranging from 0.1









to 1 p.m, keeping barrier thickness < 10 % of the overall linewidth results in a constant

effective resistivity < 2.3 pLQ-cm for the Cu line [Hu98].

The barrier material will typically be used in two types of scenarios on the

integrated circuit, which are depicted in Figure 1-4. In the first case, the barrier runs in

parallel with the Cu line, as with a trench structure, where the barrier separates the Cu

trench from neighboring dielectric layers. In this case, the thickness of the barrier must

be minimized, in order to have maximum cross-sectional area for the Cu conductor line.

The resistivity of the barrier in this scenario is not critical, however, as electrical current

will be flowing through the Cu and not directly through the barrier [Tra03]. In the

second case, the barrier will be separating one metal layer from another, as in a via

structure, hence the current will pass directly through this intermetal barrier layer. The

resistance associated with current flowing from one Cu line through a via (and barrier) to

another Cu line is called the via resistance. The barrier film's thickness and bulk

resistivity must be minimized to limit the barrier's impact on via resistance. The

suggested ideal film resistivity is < 500 p.-cm [Ele03], and barrier thickness should be

<30 nm [Bai96]. Specific via resistance (Q-pLm2 or 2-cm2), defined as the electrical

resistance through the via multiplied by the via's cross-sectional area, is a common

measure of barrier impact on current flow through a via. The current, acceptable specific

via resistance value for the 100 nm device node is 0.1 9-2-m2 (1 x 10~9 Q-cm2), and this

is projected to decrease to <0.01 Q-pLm2 (1 x 10-10 o2-cm2) for the 22 nm device node in

2016 [ITRO2, Nic95, Tra03].

As an example, the change in specific via resistance with ITRS feature diameter

for a fixed 15:1 via aspect ratio was calculated. The via was treated as a cylinder, with









Cu in the center and the barrier layer lining the inside edges of the cylinder. Current is

assumed to travel in the Cu cross section of the via, which is the path of least resistance,

and then passes through the barrier layer at the via's bottom, as depicted in Figure 1-5.


Trench
Barrier

- Via Barrier


Intermetal Barrier


Figure 1-4. Schematic of trench and via applications for barrier materials.


-- Cu


- Barrier


Figure 1-5. a) Cut away view of Cu trenches with via contact, b) Expanded, 3D view of
Cu via with barrier layer on sides and bottom.

The thickness of the barrier layer corresponded to ITRS projections for each device node,

as follows: 12 nm (in 2003), 10 nm (2004), 9 nm (2005), 8 nm (2006), 7 nm (2007), 5

nm (2010), 3.5 nm (2013), and 2.5 nm (2016). The results for varying barrier

resistivities, along with projected requirements from the ITRS Roadmap, are shown in

Figure 1-6. The diagram indicates that by the 65 nm node (2007), the barrier layer's









resistivity must be < 200 p.2-cm to meet ITRS projections. Likewise, this value must be

S25 pQ-cm by the 32 nm node (2013).

1.2e-9
^-*- 10 pL-cm
1.0e-9- -0- 25 I-cm
"-0- 200 Q-cm
a 8.0e-10 -- 200 .pQ-cm
-0- 300 .Q-cm iS^ /
,- ITRS Projections
6.0e-1 -

> 4.0e-10

i 2.0e-I&

0.0
22 32 45 6570 80 90 100
Feature Diameter (nm)

Figure 1-6. Specific via resistance as a function of feature diameter, shown for various
barrier layer resistivities. Calculation is for a via with a 15:1 aspect ratio.

1.3.3 Conformality

Conformality (or step coverage) in device microstructures, especially in trenches

and vias, is of utmost importance in barrier design. Highly conformal films (approaching

100%) have nearly uniform thickness at all points on a substrate surface (on both the

sidewalls and bottom). Film conformality is determined by measuring the film's

thickness on the wafer surface (ts), on feature sidewalls (tw) and at feature bottom (tb)

(Figure 1-7). Conformality on the sidewall and bottom can be calculated by Equations

1-2 and 1-3, respectively. The feature's aspect ratio, which is the ratio of the feature's

height to its width, can be calculated by Equation 1-4.

Sidewall conformality (%)= 100(1-2)
Sidewall conformality (%) = ('- 100
t s











Bottom conformality (%) = tb 1 100 (13)
t{ )


Aspect Ratio = (1-4)



te -- Barrier
film


i .b


Substrate


Figure 1-7. Sketch of film dimensions for calculation of conformality.

Barrier Film







Substrate Substrate Substrate


Figure 1-8. Diagram of (a) Ideal barrier conformality. (b) Good barrier conformality. (c)
Poor barrier conformality.

Barriers with poor conformality have uneven thickness, with the film being

thinner in certain spots than others. These thin spots are "weak links" in the barrier,

which are more susceptible to diffusion than the thicker parts of the layer. Figure 1-8a

shows an ideal barrier with 100% conformality. Figure 1-8b shows a practical barrier

with rounded edges, while Figure 1-8c shows a poor barrier, with a large overhang near








the trench opening and sparse coverage at the trench bottom. Cu can easily penetrate to

the underlying Si substrate through the thin barrier at the bottom of the trench.

While device dimensions continue to shrink, the aspect ratio continues to

increase, and the ability to conformally cover the bottom and sidewalls of these features

becomes even more challenging. Deposition methods and chemistries must be developed

to accomplish this task.

1.3.4 Adhesion

The adhesion of the barrier to both the Cu and the neighboring dielectric or

semiconductor layers, along with the crystallographic orientation of the deposited Cu

(which is affected by adhesion issues), all impact copper's electromigration behavior

[Jai99]. Strong Cu adhesion to the barrier layer contributes to good electromigration

resistance [Pet03], whereas poor adhesion can cause self-diffusion of Cu along the

barrier interface. Self-diffusion of Cu along the interface is even more likely than

diffusion along grain boundaries, because grain boundary self-diffusion has a higher

activation energy than interface self-diffusion [Llo95]. Copper has an activation energy

for bulk self-diffusion of 2.19 eV [Mur95], while the values for grain boundary and

interface self-diffusion are 1.2 and 0.8 eV, respectively [Llo95]. Cu (111) is the

preferred orientation for the metal, as this plane has minimum surface energy, which

results in formation of low-angle grain boundaries [Llo95]. These low-angle grain

boundaries minimize flux divergence of electrons flowing through the metal, and

therefore minimize electromigration [Jai99]. Aside from the barrier-Cu interface,

adhesion at the barrier-dielectric interface can be adversely affected by out-gassing of









moisture or other organic materials from the dielectric surface, which can lead to

anomalous Cu removal behavior.

1.4 Dielectric Material Considerations

In addition to the changeover to Cu, parallel research is pursuing new, low-k

materials, which will also decrease the RC time constant. Historically, SiO2, with a k

value ranging from 3.9 to 4.1, was the dielectric material of choice [Hu98]. Transitioning

from Al/SiO2 devices to Cu/low-k based devices (with k-2) may lead to more than a 400

% reduction in RC delay [Kal00]. New films (such as polyimide, SiLKT, SiOC, etc.),

with dielectric constants below 3.9, are being investigated for use as low-k materials, but

they cannot tolerate high temperatures during processing [Hu98, Pet03]. The processing

temperature limit for these materials, along with the need for good adhesion to Cu, are

factors that must be considered when selecting a barrier material to separate them.

Another factor influencing barrier material selection is its compatability with

high-k materials, such as those in MOSFET gates or in DRAM capacitors. Barrier

materials have been used as top electrodes on high-k gates and DRAM capacitors, and

their favorable performance may dictate further use in the future.

1.5 Diffusion Barrier Failure Mechanisms

Thin films typically have higher diffusivity than thicker bulk films, due to a large

number of short circuit paths (grain boundaries, dislocations, etc.) distributed over a very

small volume [Bal75]. Thinning the barrier film increases the likelihood that these short

circuit paths will extend from one side of the film to the other, making it more susceptible

to diffusion. Thin film diffusion barriers typically fail in one of two ways. The first is

metallurgical failure, where the Cu content of the barrier increases to several atomic








percent, changing the barrier layer's composition. The second is electrical failure, where

little intermixing with the barrier occurs, but the Cu penetrates the barrier and is present

in the semiconductor/dielectric layer in sufficient quantity to alter the characteristics of

the device [IstOO]. Most barrier testing techniques, such as depth profiling or SECCO

etch tests [Iva99], are suited for detection of metallurgical failure. Use of XRD to detect

CuSix compounds, for example, is a fairly insensitive detection method, because the bulk

concentration of Cu in Si must already be above saturation before the silicide compounds

will form.

Measurement of electrical failure is a more sensitive way to determine if Cu has

penetrated the barrier and intermixed with the underlying substrate. To detect electrical

failure, the electrical properties of devices such as p-n junctions, Schottky diodes or

metal-oxide-semiconductor (MOS) capacitors, built under the Cu-exposed barrier or on

the wafer surface after the barrier has been removed, are measured. Leakage currents

are measured on p-n junctions and MOS devices and compared to Cu-free ones to

determine if Cu has penetrated the barrier. Changes in the capacitance and current-

voltage (I-V) characteristics of Schottky diodes are measured to determine if Cu has

penetrated into the active areas on devices.

Ideally, the performance of different diffusion barrier materials would be tested

by comparing the diffusivity of Cu in the different barriers. To determine diffusivity, the

migration of atoms, or diffusive flux (J, atoms/cm2-sec), through the barrier material

should be estimated. The flux can be described by Fick's law (Equation 1-5):

J=- C (1-5)
J = -D (1-5)
(dx








where D is the diffusivity (cm2/sec), C is the atomic concentration (atoms/cm3) and x is

the diffusion distance (cm). By measuring flux (J) through the barrier for a given

concentration gradient (dC/dx) across the barrier, the diffusivity can be determined.

Once this is known, the next level of analysis is to determine whether diffusion through

the bulk lattice or the grain boundary dominates. To distinguish between diffusion

through the bulk lattice and the grain boundary, two different expressions for diffusivity

have been given [Ohr92]. Equation 1-6 gives the expression for lattice diffusivity (DL):


DL = DoL exp(ERT (1-6)

where EL is energy per mole for atomic diffusion through the lattice, and DoL is the value

for lattice diffusivity at standard conditions. Equation 1-7 gives the expression for grain

boundary diffusivity (DB):


D, = DoB exp T) (1-7)

where EB is the energy per mole for atomic diffusion through the grain boundary, and DoB

is the value for grain boundary diffusivity at standard conditions. At low temperature, the

diffusion mechanism through a polycrystalline solid is typically controlled by grain

boundaries and other defects in the film [Nic78]. Since vacancy-assisted bulk diffusion

is negligible below about three-tenths of the solid's melting temperature (0.3Tm),

diffusion through the grain boundary is the path of least resistance at low temperature.

This is reflected by a lower activation energy for grain boundary diffusion relative to bulk

diffusion [Cha82, Nic78], with EB = 0.5EL for FCC metal thin films [Bal75]. As

annealing temperature is increased, a changeover from grain boundary controlled

diffusion to bulk diffusion occurs. The temperature at which this occurs, called the








Tammann temperature, is typically 0.5-0.66Tm [Nic78]. At higher temperature, the

values for DL and DB begin to approach each other, and flux through the bulk lattice

becomes the dominant route for diffusion, due to the much larger cross sectional area of

the bulk lattice relative to that of the grain boundaries.

Experimentally, determination of diffusivity from Cu flux measurements is

impractical. Exact measurement of the Cu flux through the barrier is difficult to

determine, as failure determination techniques usually focus on qualitative barrier failure.

Moreover, issues such as defect density can have a large impact on thin film diffusivity,

and can cloud efforts to distinguish between bulk lattice and grain boundary diffusion.

Diffusivity in thin barrier films may be estimated with Equation 1-8, which describes

diffusion from a limited source at a crystal surface [Wol86]:


Ck = exp(- d Dt (1-8)


where Ck is Cu concentration (cm-3) at distance d (in cm) from the surface, D is the

diffusivity (cm2/s), t is diffusion time (sec), and Qo (cm-3) is the concentration of Cu at

the surface. Since exact Cu concentrations at the two interfaces can be difficult to

measure, an apparent diffusivity (Dapp) is defined by Equation 1-9 [IstOO]:

d2
Dapp (1-9)
4t

where d is the film thickness (in cm) and t is the time (s) for the Cu to appear at the

barrier/substrate interface (i.e., the time it takes the barrier to fail). Comparison of

apparent diffusivities for different barrier films may be done to determine which barrier is

best.









1.6 Diffusion Barrier Deposition Techniques

Diffusion barrier films have been deposited by a variety of methods, including

physical vapor deposition (PVD) and chemical vapor deposition (CVD) techniques. The

required deposition temperature, substrate characteristics and feature size generally

dictate which deposition method is appropriate for a given application. A discussion of

some common PVD and CVD techniques to deposit metal nitrides and their inherent

advantages and/or disadvantages follows.

1.6.1 Physical Vapor Deposition (PVD)

Typical physical vapor deposition (PVD) methods to deposit barrier films include

electron beam evaporation and sputtering. In electron beam evaporation, an electron

beam is used to heat a small portion of the solid metal above its melting point. While at

high vacuum (10-3 to 10-4 Torr), the molten metal then evaporates atoms into the gas

phase. These atoms diffuse through the chamber's atmosphere and deposit on the

susbtrate. One or more reactive gases, such as NH3 or CH4, for example, may be

introduced into the chamber during evaporation to deposit multi-component films.

The sputtering process involves the use of a plasma gas (typically argon) to

physically "knock" atoms off of a metal target and onto a substrate. The sputter chamber

operates at low vacuum (-100-120 mTorr), and cations from the plasma impact the metal

target cathode to liberate metal atoms, which then deposit on the substrate. In the

reactive sputtering process, a reactive gas such as N2, for example, is introduced into the

sputter chamber along with the plasma gas. Once the atoms liberated from the metal

target cathode reach the substrate, they can interact with the nitrogen radicals formed by









the plasma and deposit either a metal nitride compound or a metal-nitrogen solid

solution.

Films produced by PVD processes suffer from high stress, a large number of

vacancy defects, excess free volume and poor conformality, leaving spots in the barrier

open to diffusion by neighboring species at via and trench sides/bottoms [Aff85, Bau98,

Bos91, Kel99, Lee93, Lu98b]. The poor conformality inherent to all PVD processes is

due to the directionality imparted to the atoms/clusters traveling toward the substrate.

This directionality causes step shadowing, where parts of the substrate surface are not

"seen" by the incoming sputter atoms. Step shadowing results in little or no film

coverage on certain sections of the substrate. These exposed substrate areas are then

vulnerable to reaction with subsequently deposited Cu atoms. Figure 1-9 depicts step

shadowing.

Directional Particles Directional Particles Directional Particles
0 4 Q 00 00 0 0 1
o .\ ** e




Substrate Substrate SuSubstrate Substrate Substrate
/Barrier
Film Shadowed Shadowed
/ wall t wall
Shadowed Barrier Barrier
walls Film Film
a b c


Figure 1-9. Step shadowing effects in small device features due to directional particles
arriving at the substrate. a) At an angle to the surface, b) Perpendicular to
surface, without resputter. c) Perpendicular, with resputter.

Three variations on sputtering have been developed to extend its applicability to

smaller device feature nodes. The first, collimated sputtering, involves the placement of









a plate (which contains small holes) between the sputter target and the substrate [Sin02b].

The plate collimates the sputtered atoms/clusters, so that only those traveling parallel to

each other in a direction perpendicular to the plate reach the substrate. Once at the

substrate, the atoms can deposit or can resputter the metal film at the bottom of a feature

(Figure 1-9c), moving it up onto the feature walls to ensure sidewall conformality. The

second variation, long-throw sputtering, improves the performance of the sputter

deposition system by increasing the distance between the target and substrate, so that

only sputtered atoms/clusters travelling in parallel (and perpendicular to the substrate)

reach the substrate [Sin02b]. Ionized-PVD (I-PVD) is the third variation used to extend

the applicability of sputtering processes beyond the 100 nm device node. I-PVD

involves use of a second rf coil to ionize neutral metal atoms that have been liberated

from the target by sputtering. These ionized atoms are then accelerated toward the

substrate, which is biased to attract the newly ionized metal atoms [RosOl]. Impinging

metal atoms resputter the metal that has already deposited on the substrate, and these

resputtered atoms can move from the bottom of a feature to the sidewalls and coat them.

This leads to improved step coverage in small features compared to other PVD methods.

Conformality improvements within the feature from the I-PVD resputter effect are

tempered by resputter thinning bevelingg) of the feature's top edges (Figure 1-9c). This

beveled edge is a weak point in the barrier, and is susceptible to attack by Cu. In

addition, although resputter associated with I-PVD improves sidewall coverage, this

coverage is still tenuous for films in high aspect ratio features [Vij99]. These extended

sputtering techniques, being directional in nature, also have a limited lifespan [Sin02b],

and may reach their usability limit at the 45 nm device node in 2010. Among the barrier









deposition techniques, only a chemical vapor deposition (CVD) based process offers the

conformality required for future device generations, with 100% conformality being the

goal for all barriers [Fal98, Fle98, Lev98, LiuOO].

1.6.2 Chemical Vapor Deposition

Chemical vapor deposition (CVD) is a widely used film growth technique. It

involves the use of one or more gas phase reactants to deposit a solid film on a substrate.

The reactant flux at the substrate surface is non-directional in nature, eliminating the

possibility of step shadowing in small device features. The substrate promotes reaction

between components from the gas phase, and successive reactions lead to film growth.

CVD processes typically grow highly conformal films at high growth rates, making them

superior to PVD techniques for deposition on aggressive device features. Unlike PVD

methods, CVD is capable of area selective growth, which can eliminate a patterning step

during device fabrication [Gat96]. CVD techniques are also characterized by high

throughput, minimal downtime and easy source changeout, which makes them very

attractive for industrial applications. The steps of the CVD process are as follows [Siv95,

Vos78]:

Reactants in the gas phase diffuse through the boundary layer (defined
below) to the substrate surface

Reactants adsorb onto the substrate surface

Adsorbed species move around surface and settle into available surface
sites

Reactants adsorbed to available surface sites undergo final reactions,
which are often catalyzed by the substrate surface, and reaction products
are incorporated into the growing film

By-products from the reaction desorb from the substrate surface









Desorbed by-products diffuse through the boundary layer and away from
the substrate surface

Conversion of gas phase precursors into deposited film can involve both gas

phase and surface reactions. Gas phase reactions can involve formation of intermediates

and by-products as well as parasitic reactions, such as gas phase nucleation. Surface

reactions can involve adsorption/desorption, film formation, and side reactions, which

can cause contaminant incorporation into the film. The total pressure in the reactor

determines the degree of coupling between gas and surface chemistry [Gat96]. At a total

reactor pressure of 1 Torr or greater, homogenous and heterogeneous processes are

tightly coupled, while at pressures < 10-4 Torr (where Xmfp = 50 cm), the probability of

gas phase collisions becomes negligible and the process becomes strictly surface

controlled [Gat96]. This surface control at low pressure enables finer control of film

properties, such as smoothness and thickness. At higher reactor pressures (>1 Torr), one

or more of the gas phase precursors may begin to decompose and/or react to form

intermediate gas phase compounds [Gat96]. These precursors (and/or intermediates) can

then travel to the substrate surface.

When the precursor (or intermediate) molecules first reach the substrate surface,

they can weakly bind (physisorb) to the surface by van der Waals forces or they can

immediately react and deposit chemisorbb) [Mas96]. Physisorbed molecules have some

degree of mobility, and can diffuse along the substrate surface until they either chemisorb

or desorb back into the gas phase. High surface mobility, coupled with the

non-directional nature of the reactant flux, are the key CVD features that enable highly

conformal film deposition on substrates with varying topography. The probability that an

impinging precursor chemisorbs at the initial point of impact on the substrate surface is








known as the reactive sticking coefficient. A precursor with high reactive sticking

coefficient (approaching unity) and low surface mobility yields a rough film texture,

while a precursor with a low reactive sticking coefficient (10-3 or less) and high mobility

yields a very smooth film [Gat96]. Hence, to deposit smooth films by CVD, a low

reactive sticking coefficient coupled with high surface mobility is desirable.

Surface mobility and reactive sticking coefficient are highly dependent on

substrate temperature, which must be controlled closely to ensure deposition of films

with appropriate structure and properties. Temperature must be high enough to activate

the reactions leading to CVD, but cannot be increased indiscriminately. At lower

temperature, the sticking coefficient approaches unity and surface diffusivity is minimal

[Gat96]. As temperatures increase, the sticking coefficient decreases, while the surface

diffusivity and deposition rate typically increase. Higher deposition rates are likely on

the flat surface of the substrate, but aspect ratio dependent effects (ARDE) can inhibit

deposition in small features. ARDE occur when by-products emanating from the

sidewalls and bottom of a small feature inhibit in-diffusion of fresh precursor to the

feature. While increasing temperature increases reaction rate on the flat surface, ARDE

can lead to a starved condition on the interior walls of the feature. Hence, although

increased deposition temperature improves surface diffusivity and decreases sticking

coefficient, it can negatively impact film conformality in small diameter, high aspect ratio

features (Figure 1-8c).

In addition to optimizing surface mobility and reactive sticking coefficient, the

operating regime must be considered when selecting deposition temperature. At lower

temperature, the deposition rate is exponentially dependent on substrate temperature,








which is indicative of the kinetically controlled regime. Increasing temperature in this

regime causes a large increase in film deposition rate, and the slow step in the deposition

process is the reaction on the surface. At higher temperature, the deposition rate is

relatively insensitive to deposition temperature, which is indicative of the gas-phase

mass transfer controlled regime. In this region, the temperature is high enough that

essentially all reactants reaching the substrates surface immediately react, and the

rate-determining step in the deposition process is transport of reactants to the surface.

While the temperature dependence of the deposition rate in this regime is mild (varying

from T1 5-2.), the lack of kinetic (surface) control in this regime exacerbates conformality

problems. Moreover, higher deposition temperature can alter film structure, causing a

shift from amorphous to polycrystalline films, which introduces grain boundaries that can

kill barrier performance.

In small features that are to be conformally coated, the surface diffusion length of

the physisorbed precursor should be the same order of magnitude as the feature diameter

[Coo89]. In larger features, surface diffusion alone is inadequate for conformal coverage,

so it must work in tandem with reflection, whereby molecules are reflected back and

forth from the sides of the feature until they find a place to chemisorb. Decreasing the

precursor's sticking coefficient increases the chance of reflection in a feature. For good

coverage, the dimension of the feature should be smaller than the mean free path in the

gas phase, so higher pressure (which has a shorter mfp) can be used with smaller device

features [Tsa86]. Pressures at and below 500 Torr would be adequate for deposition in a

100 nm feature.








Molecules designed for use as CVD precursors should have high volatility,

thermal liability, and have easily (and cleanly) removable supporting ligands [Kod94].

The CVD chemistry should be selected to avoid formation of non-volatile reaction

by-products, which can cause particle deposition on the substrate surface and high point

defect density [Jai99]. In addition, selection of precursors with low reactive sticking

coefficient and high mobility surface species will enhance conformality of the deposited

film.

1.6.2.1 Precursor Delivery

During CVD, one or more gas-phase chemical precursors must be delivered to

the reactor. The precursors themselves may exist in the gas phase, or they made be put

into the gas phase by a number of techniques. Liquid source precursors are typically put

into the gas phase by a liquid bubbler system, where a canister containing the liquid

precursor sits in a temperature controlled bath (which controls vapor pressure). Carrier

gas flows into the bottom of the canister, and bubbles up through the liquid. As the gas

bubbles up through the liquid precursor, it becomes saturated with precursor molecules.

This carrier gas, saturated with precursor, then flows into the reactor.

There are two typical methods to convey low volatility solid precursors to the

reactor: solid source and nebulizer delivery. In a solid source delivery system, the

precursor is held in a metal or glass tube. The tube and precursor are heated by thermal or

photon energy, and solid precursor near the top of the tube sublimes and is conveyed to

the reactor by a carrier gas. As a general rule, a solid precursor must have a vapor

pressure higher than 10-3 Torr at its melting point to be effectively sublimed and









conveyed by a solid source delivery system [Ohr92]. If the vapor pressure is below this

value at the melting point, minimal precursor delivery and film deposition will occur.

The second technique, nebulizer delivery, can be used to overcome this vapor

pressure limitation. This technique requires dissolution of the solid precursor in a liquid

solvent. The solvated precursor is pumped into the nebulizer, which contains a

piezoelectric quartz plate. This plate vibrates due to application of a high frequency

electrical current, and vaporizes the liquid droplets that contact it. Once the droplets are

vaporized into a mist, carrier gas flows through the nebulizer and conveys the mist into

the reactor. Figure 1-10 shows a schematic of a nebulizer.

Solvated Precursor /
Carrier Gas to CVD
Reactor


00
0 O
O0 Precursor
"Mist"


Vibrating
Quartz Plate Dissolved
S "' 4 i \ Precursor from
Plastic Syringe Pump
Tubing
Carrier Gas to
Nebulizer




Cable to Power Supply

Figure 1-10. Schematic of a nebulizer delivery system.

1.6.2.2 Variants of CVD

Several variants of CVD are typically used to deposit thin films. The first is

low-pressure chemical vapor deposition (LPCVD), which typically relies on metal halide









chemistry (e.g., TiCl4, WF6) to deposit films. This technique usually requires high

deposition temperature (>4500C) and suffers from halide incorporation into the barrier

films. The presence of halides in the barrier can lead to Cu corrosion, which decreases

Cu-barrier adhesion and reduces electromigration resistance [Huo02].

The same halide chemistries used for LPCVD are typically used for

plasma-enhanced chemical vapor deposition (PECVD), in which a plasma assists

fragmentation of the reactants, thereby lowering the deposition temperature. After

fragmentation by the plasma, the precursor fragments travel to the heated substrate, react

on its surface and deposit a film. Reduced conformality for PECVD films in high aspect

ratio trench features is due to the directional nature of the plasma [Tsa96], as shown in

Figure 1-9. Despite the low resistivity and deposition temperatures that PECVD offers,

its inability to deposit highly conformal films in high aspect ratio features makes its use

in future barrier deposition processes unlikely.

In metalorganic chemical vapor deposition (MOCVD), films are deposited by

reaction of one or more carbon-containing vapor phase precursor compounds. MOCVD

precursors typically have some or all of the halide ligands common to LPCVD and

PECVD replaced with carbon-containing ligands. These precursors afford little or no

halide incorporation into the deposited films, and MOCVD has been demonstrated to

deposit a variety of films at low temperature. By varying the structure of the precursor

molecule, the precursor can be optimized to dissociate in a "clean" fashion (i.e., with

minimal oxygen and carbon contamination) and at relatively low deposition temperature.

The presence of impurities (O, N, C) is reported to improve stability in contact structures,

because the impurities tend to stuff grain boundaries and inhibit diffusion [So88].








Oxygen's "stuffing" effectiveness, however, is lower for Cu than Al, because the

reactivity of Cu with O is less than that for Al with O [Kim99c]. Carbon bound to the

metal has also been reported to improve thermal stability of the barrier and to foster

growth of smaller grains [Wan01b], but free carbon in the films can scatter electrons,

increasing film resistivity. Among the challenges for MOCVD of refractory nitride

materials are controlling carbon and oxygen contamination, lowering the deposition

temperature, and minimizing film resistivity.

1.6.2.3 CVD Reactors

CVD reactors fall into two broad categories: hot wall and cold wall reactors. Hot

wall reactors have heated walls, with typically laminar flow profiles. These are preferred

for exothermic reactions, since the hot wall temperatures discourage or prevent unwanted

deposition on the reactor walls. In contrast, cold wall reactors have steep temperature

gradients surrounding the susceptor, which often leads to convection pattern formation in

the reactor. These are preferred for endothermic reactions, since the reaction will occur

most readily on the hottest surfaces in the system [Vos78].

CVD reactors are typically run in two modes: differential or starved. In a

differential reactor (similar to a CSTR), the ratio of reactant out to reactant in is

approximately one, so that the composition in the reactor is essentially constant. In a

starved (feed rate limited) reactor, the ratio of reactant out to reactant is much less than

one, meaning that there are large concentration gradients in the reactor caused by fast

reactions [Kod94]. ARDE can cause both differential and starved conditions to exist

simultaneously on a substrate with aggressive feature topography. Flat surfaces on the

substrate receive adequate precursor flux, and hence operate in a differential condition,








while ARDE in small features prevent good precursor flux, starving the sidewalls and

bottom. Increasing the deposition temperature increases the concentration disparity

between the inside of the feature and the flat surface of the substrate, further degrading

the conformality over these features.

1.6.2.4 CVD Transport Issues

Heat and mass transport in a cold-wall CVD reactor can be complex, due to large

temperature gradients. Depending on pressure in the reactor, mass transport may occur

by fluid flow and/or diffusion [Tim01]. At higher reactor pressures (near atmospheric),

both fluid flow and diffusion occur simultaneously. In fluid flow transport, gas

molecules follow streamlines through the reactor, while for diffusion transport,

concentration gradients drive transport across streamlines. Heat transport occurs by a

combination of convection, conduction and radiation in the reactor. Near the deposition

surface, the temperature, velocity and concentration vary significantly [Kod94].

Use of an impinging jet to feed reactants onto a small susceptor results in

stagnation point flow. The stagnation point is at the center (origin) of the surface, where

flow velocities are zero (where y=0 in Figure 1-11). With adequate pressure, a shear

layer of uniform thickness develops near the surface of the susceptor [Whi91]. The shear

layer, also called the momentum boundary layer, is the relatively stagnant region between

the surface of the substrate and the region where the fluid velocity (u) reaches 99 % of

the free stream velocity (U., cm/sec). Reactants and by-products travel to and from the

substrate surface, respectively, by diffusing through this layer. This boundary layer (SM)

displaces the outer inviscid flow away from the substrate surface, as depicted in Figure 1-

11. Thickening of the low-velocity shear (boundary) layer due to viscous diffusion is








balanced by thinning of this layer due to acceleration of the outer inviscid stream, which

leads to uniform boundary layer thickness.

The momentum boundary layer thickness (8M, cm) in stagnation point flow is

approximated by Equation 1-10:


6M = 2.4 VD5 (1-10)


where v is the kinematic viscosity (cm2/sec) and Ds is susceptor diameter (cm) [Whi91].

More information on the dynamics of stagnation point flow may be found elsewhere

[Whi91].












--- -- -- -- -- -- ---- -------- -- -





Figure 1-11. Stagnation-point flow [Whi91].

In addition to the momentum boundary layer, concentration and thermal boundary

layers, with thickness denoted 8c and &r, respectively, are also defined. The

concentration boundary layer extends from the substrate surface to a point where

precursor concentration is 99% of the bulk concentration, and likewise, the thermal

boundary layer extends to a point where the temperature is 99% of the bulk temperature.









The thickness of the boundary layers depends on several process variables,

including gas velocity, temperature, and pressure. Gas phase species outside of these

boundary layers (where y > SM, 8c, or &r) move by both convection and diffusion.

Within the boundary layer (where y < 6M, 6c, or &r), velocity, concentration, and

temperature are non-uniform. As species leave the gas and deposit onto the substrate,

their gas phase concentration drops near the substrate surface. To compensate for this

drop in concentration near the surface, a net diffusive flux occurs bringing depositing

species from regions of higher concentration in the boundary layer to the near-surface

region. As the surface reaction generates reaction by-products, the concentration of

by-products near the substrate surface is higher than it is at locations farther from the

surface. This higher concentration of by-products near the substrate surface leads to a

net diffusive flux of by-product away from the substrate. Likewise, temperature varies

from the substrate into the gas phase. The temperature at the substrate is high, and this

drops steadily through the thermal boundary layer, until it reaches the bulk value.

Typical CVD temperature and concentration profiles are depicted in Figure 1-12.


TB PiB




-t----- ------X -----------_____ ;_________

T, P.
Substrate


Figure 1-12. Variation of temperature and precursor partial pressure near the substrate in
a CVD reactor. Note that while &r and 6c are shown as being equal, this is
not always true.








Transport to the substrate during CVD is assumed to occur by diffusion of

reactants through the concentration boundary layer at the surface. This boundary layer

arises due to consumption of the precursor at the surface, which causes a concentration

gradient to form between the surface and the bulk. Flux of reactant through this layer (J,

kgmol/m2-sec) is given by Equation 1-11 [Kel91]:

SDo (Ts TbXPb P)
cToRln(TT (1-11)


where Do is the diffusion coefficient (m2/sec) at temperature To (K), 8c is the thickness of

the concentration boundary layer (m), Pb and Ps are the bulk and surface partial pressure

of the reactant (Pa), Tb and Ts are the bulk and surface temperature (K) and R is the gas

constant (m3-Pa/kgmol-K). Once the reactant species diffuse across the boundary layer,

they may or may not be incorporated into the growing film surface. The mass transfer

flux (Jtr, kgmol/m2-sec), which is the amount of precursor incorporated at the film

surface, is defined in Equation 1-12:

k,(P -Peq)
Jti = (1-12)
RTs

where kd is the mass transfer coefficient (m/sec) and Peq is the equilibrium precursor

partial pressure (Pa) at the surface. When the deposition process reaches steady state, the

diffusion and mass transfer fluxes are equal, and a dimensionless parameter called the

CVD number, NCVD, can be defined [Kel91]:


(Pb -P) ka6cTln(T T
CVD (Pe-P ) DTs(Ts -Tb) (1-13)









The denominator on the left-hand side of the equation represents the degree of precursor

supersaturation at the substrate surface. When the amount of supersaturation is small,

Ps=Peq and NCVD >>1, hence all reactant reaching the substrate reacts immediately. This

is known as the mass transfer controlled regime. Conversely, if the supersaturation is

large, Ps approaches Pb and NCVD <<1, and the species reaching the substrate react very

slowly. This is known as the kinetically (or surface reaction) controlled regime. When

NCVD >>1, high points on a rough surface have a higher value of Ps, and the growth rate

at a high point is increased relative to a low point. This causes rough points on the

surface to be amplified by film growth when operating in the mass transfer controlled

regime. In contrast, when NCVD <<1, Ps approaches Pb everywhere along the surface, so

that the growth rate is similar at all points along the surface. This leads to a smoothing of

rough spots during film growth when operating in the kinetically controlled regime.

The type of carrier gas used in the reactor can have a significant impact on the

films grown during CVD. For fixed bulk and surface temperatures, gases with a lower

thermal conductivity will have a sharper temperature profile, meaning less upstream

heating of the precursor species as it approaches the substrate. Using a carrier gas with

higher thermal conductivity results in a smoother temperature profile, enabling more

upstream heating of the precursor. Increased upstream heating promotes pyrolysis, or

gas-phase thermal breakdown of the precursor into the intermediate or final reactant

species. Increasing the time available for precursors to undergo pyrolysis increases the

likelihood that the precursors have decomposed into the intermediate components (if

necessary) for final surface reaction. If the precursors undergo sudden pyrolysis very








near the substrate surface, they will not decompose as thoroughly, leaving large

molecular fragments that can be incorporated into and contaminate the growing film.

Ideally, transport of reactants to and products from the substrate surface should

occur in such a way to deposit a "clean" film containing the desired components. In

reality, this is unlikely, with atoms from reaction by-products and precursor ligand

fragments typically depositing to some degree in the films. Control of contamination by

ligand decomposition is extremely important to get films with desirable properties. An

example of single-source CVD, which has some contamination from precursor ligands,

is depicted in Figure 1-13.


S Atom desired in film
--- --- Atom not desired in film



Pyrolysis

S---- 0




S/ Surface
x easel a e
Substrate


Figure 1-13. Diagram of the CVD process for a single-source precursor.

1.6.3 Atomic Layer Deposition (ALD)

Standard CVD techniques introduce all required precursor gases to the reactor

simultaneously, making deposition rate and precise thickness control difficult during film









deposition. Although the ITRS roadmap suggests that CVD deposition of barriers will be

important in the near term, atomic layer deposition (ALD) methods will emerge as the

dominant solutions because of their superior conformality and improved thickness

control. As barrier thickness drops below 100 A, standard CVD techniques will approach

their usability limit due to difficulties controlling the deposition rate. Atomic layer

deposition (aka ALD, ALE or ALCVD) will be necessary to deposit extremely thin films

on the IC devices [Dan02], and its use in IC production at the 65 nm node is projected

[Bey02, Cha04]. The highly conformal, ultra-thin barriers afforded by ALD will be

essential in the future to minimize the barrier's impact on the resistance per unit length in

Cu interconnects [Kap02].

As a variant of CVD, ALD is well suited for deposition of ultra-thin, highly

conformal films over small device features. ALD enables monolayer addition with

precise thickness and composition control, irrespective of the underlying substrate's

topography [Dan02]. ALD has been used to deposit metal, semiconductor, dielectric and

seed layers [Dan02, Kla00a-b, Les02]. In particular, ALD has been used to deposit

high-k gate dielectrics at the front end and diffusion barriers at the back end of the

process. ALD is a "digital" process, involving the stepwise use of two or more gas phase

precursors, each of which is self-limiting on the substrate [RosOl]. A single precursor

gas is present in the reactor at any given time, so that a uniform layer of the precursor

may chemisorb to the substrate surface. Once this chemisorbed layer forms, the chamber

is evacuated or swept with inert gas [Sun92]. By minimizing chamber volume, rapid

deposition and quick purge/vacuum steps are promoted. This is important to minimize

the incorporation of background impurities into the film and also to prevent the process









from shifting to CVD mode, which can occur if additional reactant from the previous

pulse is in the reactor's atmosphere during the subsequent pulse. A second precursor gas

is then introduced to react with the chemisorbed layer from the previous step, forming a

new monolayer of material. The second precursor gas is also self-limiting, so that

reaction stops after the monolayer of chemisorbed material from the previous step has

been consumed. Film thickness is more difficult to control for growth with CVD, which

has been reported to close extremely narrow vias rather than depositing conformal films

in them [HauOO].

ALD was first reported in 1977 by Tuomo Suntola to deposit ZnS films for

electroluminescent displays [Goo86]. Since this first report, the utility of ALD to deposit

a variety of materials has been demonstrated. Materials such as metals (e.g., W, Cu, Ni,

Co, Ti, Ta, Ru, Pt, Al), metal oxides (e.g., A1203, ZrO2, HfO2) and metal nitrides (e.g.,

TiN, TaN, WNx) have been deposited using this technique [Les02]. In contrast to the

thin, flat films typically desired by ALD, the process has also been demonstrated to coat

porous, high surface area substrates for catalysis [Hau93].

ALD exploits the difference in energy between chemisorption and physisorption

to deposit film layers in a self-limiting fashion [Gat96, Goo86]. Temperature is selected

so that Echemisorption > kT > Ephysisorption. In other words, the temperature is high enough to

overcome physisorption forces to enable desorption of any physisorbed species, but it is

also low enough to prevent removal of chemisorbed layers. For an adequate pulse time,

complete surface reaction occurs and the film composition is determined by

thermodynamics (stable thermodynamic phase forms for the selected conditions).









The precursors must be volatile, thermally stable, self-limiting and highly

reactive (fast, complete reactions) with each other and with the substrate. They must also

have sufficient purity, have unreactive by-products, and not undergo self-decomposition

or cause etching of the film or substrate. The precursors should have a vapor pressure of

at least 0.1 Torr for delivery during ALD. Desirable ALD reactions should have large

negative AG values to ensure rapid surface reaction after the precursor pulse into the

reactor [Les02]. High reactivity enables rapid saturation of the film surface, which in

turn enables a good deposition rate. The need for high precursor reactivity is contrary to

CVD precursors, which require smaller negative AG values to prevent gas phase

nucleation and particle formation. The precursor dose must be high enough to saturate

the substrate surface, which must have reactive adsorption sites [HauOO]. In addition, the

reaction temperature must be chosen to enable reaction chemisorptionn) between the

precursor and reactive sites, while also avoiding gas phase decomposition or

condensation of the precursors [HauOO].

The advantages of ALD relative to CVD are excellent conformality (-100%),

inherent elimination of pinholes, good thickness uniformity, and elimination of potential

gas phase reaction/nucleation [Goo86, Rit03, Sun92, Sun93]. The disadvantages of ALD

include slow growth rate (0.06-0.6 pm/hr) [Goo86, Ros01], the possibility for substantial

incubation time to deposit the first monolayer of film, contaminant incorporation and the

potential for surface reconstructions to adversely affect deposition rate [RosOl, Sun93].

In the most aggressive applications, a diffusion barrier will need to be deposited on four

different surface materials simultaneously: two different insulator materials, an etch stop

layer (such as Si3N4) and Cu [Ele02]. Varying incubation times on the different materials








can cause significant deposition difficulty, so choice of precursor and deposition

conditions is essential to deal with situations like these. Precursor testing should

therefore be done on all possible underlying substrates, to ensure that a given material

can be reliably deposited with a given precursor. Another disadvantage is decreased film

purity relative to CVD grown films. Background contaminants have considerable time to

incorporate into the film during the pulse and purge steps, which can degrade film

structure and electrical properties. The purity of purge gases and control of out-gassing

from reactor walls and seals are critical, because inadequate control can lead to

considerable impurity incorporation causing modification of film structure and properties.

Typically, a faster growth rate means less sensitivity to contaminant incorporation

[Han03]. ALD must compete with the purity of PVD films, whose growth rates are 2 to

3 orders of magnitude higher in an environment that is 3 to 6 orders of magnitude cleaner

[Han03].

Several adjustable parameters are available for the ALD process. The first

parameter is the type of ALD reactor system used for deposition. The two main

categories for ALD systems are the open type (used with molecular beam epitaxy

(MBE)) and the closed type (used with LPCVD) [Sun93]. In the open type system,

pressure is fixed, as with an MBE system, which typically operates at ultra-high vacuum

(UHV). Precursors are introduced from two or more sources (e.g., Knudsen cell), which

are turned on and off to generate the pulsing action. Desorbed surface species collect on

the cold walls in the reactor system or by the vacuum pump. In the closed type system,

the atmosphere is turned over between reactant exposures to generate the pulsing action;

this is done either by evacuating the system (pressure modulation) or by introducing an









inert purge gas. In addition to the type of system used, the chamber volume may be

adjusted to accommodate more substrates (i.e., to increase throughput) or to decrease

purge/vacuum times.

The second adjustable parameter is the type of precursors used for the deposition

process. The precursors must be self-limiting, very reactive with each other and with

substrate, and generate the desired film structure/stoichiometry. Many precursors have

been tested in the literature for various materials, with halide chemistry often being used

due to easy removal of the halide ligands (with H2, NH3, etc.). In addition to precursor

selection, pre-treatment of one or more of the precursors before introduction to the

reactor is another adjustable option. For example, Ta films were deposited using

plasma-enhanced ALD with sequential pulses of TaCl5 and H2 [Kim02b]. The H2 was

cracked to atomic H in an rf plasma system before being introduced to the reactor.

The third parameter is substrate temperature, which should be selected to enable

chemisorption of first layer and desorption of outer, physisorbed layers [Gat96]. A fourth

parameter is the reactant pulse time. A longer pulse time enables more complete surface

reaction, and closer approach to thermodynamic equilibrium at the film surface. The last

parameter, which is unique to closed type systems, is reactor pressure. ALD process

pressure in closed type systems typically ranges from 1 to 10 Torr [Les02], which can be

modulated to evacuate the chamber if an inert gas purge is not desired.

The process is depicted in Figure 1-14. The result of running in ALD mode is a

linear growth rate with the number of cycles. The number of monolayers grown per

cycle (normally <1), multiplied by the number of cycles, gives the film thickness, where

one cycle includes one pulse of each of the precursor gases. Several growth cycles are








typically necessary to deposit a monolayer of material, because steric crowding on the

film surface prevents 100% surface coverage for each pulse. In practice, monolayer by

monolayer growth is unlikely, and there may be two or more monolayer levels growing

together during deposition. Film growth is dependent on several factors, including

reactivity of the precursors, the number of reactive sites available on the substrate/film

surface, and the size of the precursor molecule (larger molecules typically yield lower

growth rates due to steric hindrance during the reaction cycle) [HauOO]. ALD cycle time

typically ranges from 0.5 to 5 seconds.


A-X A-X X X
A-X A-X X AX A







B--Y X-Y X-Y
B-Y
BB X-Y
B-Y X-Y
S X B B
SI
B B
X X A A
A A I
A




Figure 1-14. Diagram of the ALD process. (a) Introduction of first precursor. (b)
Absorption of first precursor. (c) Introduction of second precursor. (d)
Complete monolayer deposition of A-B.

The growth rate for an ALD process should be linear with the number of growth

cycles. If this is not the case, the process may not be self-limiting. To check this,

reactant pulse time can be changed to determine if the experiment is self-limiting, or if









longer pulse times result in thicker films. ALD growth reportedly occurs by formation of

islands during nucleation, hence the barrier should have thickness at least equal to that

required to close the surface of the substrate [Bey02]. This closure thickness depends on

the deposition conditions and the barrier materials being used. Electrical measurements

(such as 4-point probe) can be used to determine if the deposited film is continuous.

Films may have high resistivity until enough cycles have passed that the film is

continuous; the film's resistivity will then drop substantially. This is one way to

determine if the first monolayer has been completely filled in, if depositing a conductive

layer on Si, for example. Extrapolating from a plot of resistivity vs. the number of

deposition cycles, one can estimate the deposition cycle after which a continuous film is

finally deposited. Ion scattering spectroscopy has also been used to determine when the

surface has closed.

The appeal of ALD to industry lies in its ability to manage line resistance and to

provide better stress migration performance than PVD films [Pet03]. A thin ALD barrier

film at the via bottom is instrumental in lowering via resistance. In addition, Cu seed

layers must be ultra-thin, continuous, and have excellent conformality to ensure

consistent Cu deposition during ECD, so ALD will be essential for seed deposition. ALD

is also excellent for varying the composition of the deposited film with thickness

(composition grading), which can be important to ensure a film's compatibility with

underlying layers.

1.7 Copper Deposition Methods

Electrochemical deposition (ECD), also known as electrodeposition or

electroplating, is the technique currently used in industry to deposit void-free bulk Cu









layers on ICs with aggressive topographies [ITR02]. This technique has the advantages

of low deposition temperature, high deposition rate, and low manufacturing cost [Chi98].

The ECD process begins with the deposition of a Cu "seed layer" on the barrier surface,

because the barrier layer's resistivity is typically too high to enable uniform

electrodeposition. Seed layer deposition is currently done by PVD, which is projected to

be used down to the 45 nm device node [Pet03]. Once the seed layer is deposited, the

sample can be immersed in the electrochemical bath and the ECD process can begin.

During Cu ECD, an electrolyte containing Cu cations and sulfate anions (Cu2+ and SO42,

respectively) is put into contact with the desired deposition surface. Electrons are

introduced to the deposition surface (called the cathode) and reduce the Cu2+ ions in the

ECD solution, causing these ions to plate out as metallic Cu. A Cu seed layer deposited

by PVD (prior to the ECD process) is typically used as the cathode, on which Cu2+ ions

plate out to form Cu(0). The plating reaction is:

Cu2+ + 2e- Cu(0) (1-14)

To complete the electrical circuit, an anode is placed in the electrolyte solution.

The anode introduces current to the solution, which replaces positive charges lost by the

Cu plating reaction, enabling the electrolyte to remain electrically neutral [ReiOO]. Since

Cu2+ ions are present in the plating solutions for ECD, backside protection of the Si wafer

from the bath is essential during deposition. Failure to do this can result in massive Cu

in-diffusion through the backside of the wafer, which will occur rapidly based on the

diffusivity of Cu in Si (as discussed above). At the industrial level, the ECD plating tools

have an o-ring around the edge of the wafer, which allows the device side of the wafer to

contact the plating solution while preventing the solution from touching the backside. At








the research lab level, waxes or glues may be used to protect the backside of the

substrate.

The use of special additives enables the ECD technique to fill small features from

the bottom up with Cu (superfilling). To ensure deposition of a microvoid-free, uniform

thickness Cu layer by ECD, the plating current (and therefore the seed layer thickness)

must be essentially constant from edge to center across the wafer [Sin02a]. Hence, high

uniformity of the Cu seed layer is essential. The seed layer is usually deposited on the

barrier surface by sputtering, however, which is characterized by poor step coverage in

small features [And99, Wan03]. A Cu seed layer with poor step coverage leads to poor

Cu coverage during ECD. To overcome the conformality challenges associated with

sputtering, several different approaches are being pursued.

The first approach is to deposit Cu seed layers by CVD [NorOl]. Generally, Cu

CVD precursors are classified into two categories: Cu(I) and Cu(II) compounds. Cu(I)

compounds, such as Cu(hfac)(TMVS), where hfac is hexafluoroacetylacetonate and

TMVS is trimethylvinylsilane, are typically more volatile, and can be used without a

carrier gas or a reducing agent [NorOl]. These compounds tend to be liquids and are

capable of deposition below 2000C, but are therefore more reactive and less thermally

stable, making control of the deposition rate more difficult. Copper(H) compounds, such

as Cu(hfac)2, are more thermally stable, but typically require a reducing agent (such as

H2) to remove the ligands during deposition [Kod94]. In addition, these compounds tend

to be solids, requiring a deposition temperature higher than 2500C [NorOl]. The major

issue for Cu CVD from both Cu(I) and Cu(II) precursors is incorporation of halides and

carbon from the precursor ligands into the Cu film, which can have detrimental effects on








film properties such as adhesion and resistivity. More detail on Cu CVD precursors and

deposition is available elsewhere [Kod94].

Deposition of a Cu seed layer by CVD followed by PVD and reflow of Cu has

also been reported [Fri99a, Jai99]. Cu reflow typically involves the deposition of Cu by

PVD methods, especially sputtering, followed by subsequent annealing of the device.

The annealing process enhances surface diffusion of Cu, which enables Cu transport into

and filling of small features [Fri99a]. The driving force for this diffusion is the chemical

potential gradient associated with differences in surface curvature along the surface of the

deposited Cu. To minimize the total free energy on the Cu surface, Cu atoms diffuse

along the surface from convex regions to concave ones [Fri99a]. The net result is

thinning of convex overhang regions at the tops of features and filling of concave parts of

the features, such as via/trench sidewalls and bottoms. This diffusion-mediated process

is good for movement of Cu to fill submicron feature sizes. Anneal time can be modified

by changing the anneal temperature, but the anneal temperature cannot be higher than the

temperature stability limits for the device, and must be low enough to prevent bulk Cu

diffusion, which can lead to delamination and a change in Cu texture. Typically, Cu is

annealed for 13-14 minutes at 450C during the reflow process [Fri99a].

Another approach is to deposit Cu seed layers by ALD. Several different

precursors have been tested for ALD Cu seed deposition, including Cu(hfac)(TMVS),

Cu(hfac)2, Cu(thd)2, [Cu(C3H7)NC(CH3)N(C3H7)]2, and CuCl, where thd is tetramethyl

heptanedionate [Kod94, Lim03, Nor01]. The first four compounds are bulky,

metalorganic sources, which have a very low growth rate per cycle due to steric

hindrance. In addition, the first two compounds contain F, making the possibility of F








incorporation into the deposited Cu film an issue. The last compound, CuCl, is a solid

with very low vapor pressure, hence transport to the reactor is very challenging. [Huo02,

Jup97, NorO1].

Electroless Cu plating, which deposits Cu without use of a sputtered Cu seed

layer, may also be used [Wan03]. This technique involves deposition of Cu from an

ionic solution, where the deposition surface catalyzes a redox reaction, without any

external electrodes [Sha95]. A specific component from the solution (e.g., glyoxylic

acid) serves as a reducing agent, and is oxidized on the catalytic surface, releasing one or

more electrons [Wan03]. These electrons reduce Cu' ions from the solution, causing the

Cu to plate out as a film on the deposition surface. While this technique has been used to

deposit Cu seed layers with good conformality and low resistivity (1.7 jiQ-cm), and has

also been used to do Cu filling [Lee98a], control of the deposition rate is not as precise as

sputtering. More detail on this deposition technique is available elsewhere [Sha95,

Sha97].

The last approach is seedless ECD [ITR02, Sin02a], where Cu is directly

electroplated onto the underlying barrier material using an external electrode, but without

any seed layer. This technique is the most technologically and economically promising,

because it eliminates the seed layer deposition step from the current IC metallization

process and does not require new equipment expenditures. Selection of a diffusion

barrier material that enables seedless Cu electrodeposition could result in significant cost

savings to the metallization process.








1.8 Statement of Problem

The three main challenges of shifting to Cu interconnects on IC devices are: a) Cu

deposition technique b) patterning method for the Cu layer, and c) which barrier material

and deposition method will prevent Cu-Si interdiffusion while fostering adhesion to

neighboring layers [And99].

The first challenge has been addressed by use of ECD. Since Cu is resistant to

chemical etching, the application of standard reactive ion etching (RIE) techniques used

to pattern Al is not feasible for Cu. The second challenge, patterning of Cu

interconnects, has therefore been addressed by the dual-damascene deposition process.

In this process, Cu is deposited across a pre-patterned wafer by one of the

aforementioned deposition techniques. Chemical mechanical polishing (CMP) uses a

grinding pad and slurry to remove excess Cu, leaving only the desired spots (e.g.,

trenches and vias) filled with metal. Since the dual-damascene process deposits Cu in

both trenches and vias, the need for tungsten (W) plugs at the upper levels of the IC,

which are used to fill vias on Al based devices, is eliminated. Removing this W plug

lowers the electrical resistance of the device, both because Cu has a lower resistivity than

W and because the Cu-W interface is eliminated.

The third challenge, barrier material and deposition technique, is an ongoing issue

as device features continue to shrink. Ti/TiN barriers, deposited by sputtering, are still

used at the contact level with W plugs, where TiN protects the contact from F in the WF6

precursor and improves W adhesion [ITR02, Kal00]. Ti reacts with Cu to form cuprides,

making Ti thermodynamically unstable as a diffusion barrier for Cu [RamOO]. Moreover,

if TiN films are Ti rich, the excess Ti can react with Cu and lead to failure of the TiN








barrier. In addition, TiN barriers with thickness below 20 nm reportedly fail to prevent

Cu diffusion due to grain boundary diffusion [Kal00].

A Ta/TaN dual layer barrier structure is currently used as the Cu diffusion barrier

at the intermediate and upper wiring levels on IC devices. Cu has good adhesion to Ta,

which encourages deposition of low-resistivity Cu (111) on its surface [Pet03], while

TaN has good adhesion to the dielectric layer [ITR02, Sin02a]. The Ta/TaN dual layer

barriers are deposited by modified sputtering methods [ITRO2, Sin02a], however, which

have a finite lifetime due to aforementioned step shadowing issues. Barrier deposition

takes place on the industrial scale in a cluster tool, which contains several chambers

connected by a common vacuum transfer chamber. The barrier is deposited in one

chamber, and the wafer is then moved in-vacuo to a separate Cu seed layer deposition

chamber. Once the Cu seed layer is deposited, the wafer is pulled out of the cluster tool

and put into a separate system for Cu ECD.

The "holy grail" of diffusion barrier research is a robust material / deposition

method couple that meets all of the above listed barrier requirements and is extendable to

future device generations. In addition, a material/deposition pair which would enable Cu

metallization to extend from the upper levels of the device down to the contact level

would eliminate the need for Al metallization at the local level and the use of W contact

plugs, yielding greater device speed. The interconnect structure's features (trenches and

vias) become increasingly more aggressive (narrower diameter and higher aspect ratio) as

the wiring levels get closer to the contact level. Research should therefore focus on

meeting coverage requirements at device level feature dimensions to enable future

extension of Cu throughout the IC.









1.9 Hypothesis

It is evident that a variant of CVD will be required to meet future demands for

conformal diffusion barrier deposition on aggressive IC device topographies. New

precursors will be required to meet the increasingly stringent demands on film properties.

In an effort to avoid halide incorporation, synthesis and testing of novel WNx and WNxCy

MOCVD precursors were pursued in this work, for reasons that will become clear in

Chapter 2.

Depositing WNx and WNxCy films from novel precursors by MOCVD should

minimize halide content, have low temperature deposition, enable good adhesion to

neighboring layers, and enable a single-step CMP process. In addition, these materials

appear to be useful for direct, seedless electroplating, and have been reported to resist Cu

diffusion. WNx is also reported to deposit in amorphous form more easily than TaNx

[RamOO], and the addition of C to form WNxCy should enhance the ability to deposit

amorphous films even further.

Eventually, a shift from CVD to ALD will be necessary to meet conformality

requirements in small device features. Hence, our strategy is to initially pursue WNx and

WNxCy films by MOCVD and to eventually transition to precursor development and

testing for ALD.

The film growth and procedure and characterization techniques used to analyze

the films will be discussed in Chapter 4. Data, analysis and conclusions for each

precursor will be presented in a separate chapter, while the thermodynamic analysis for

the W-N-C-H-Cl system is presented in Chapter 3. Cu diffusion results for barriers






52

deposited with an isopropylimido-based W complex will be discussed in Chapter 8, and

future work will be discussed in Chapter 9.















CHAPTER 2
REVIEW OF THE LITERATURE

From a thermodynamic standpoint, a barrier material must be chosen so that the

chemical potential (the thermodynamic description of reactivity) at each interface ensures

little or no reaction. A major thermochemical difference between Al and Cu is their

reactivity with silicon. Al can exist in equilibrium with Si, while Cu cannot, readily

reacting to form silicides [RamOO]. This non-equilibrium suggests that the

thermodynamic driving force for diffusion in the Cu-Si system is larger than that in the

Al-Si one. An ideal barrier separating Cu from Si must therefore be stable to and

non-reactive with both Cu and Si. Moreover, the barrier should be stable with other

potential materials in the IC, such as low-k and high-k dielectrics.

Many materials have been examined to determine their potential as diffusion barriers

for Cu metallization schemes. A comprehensive list of various materials tested as

diffusion barriers can be found in previous review articles [Jai99, Kal00, Kim03b].

2.1 Cu-Si Interconnects without a Barrier

Ideally, the easiest route to IC device fabrication would be to deposit Cu directly

onto any neighboring layers. Hymes et al. [1998] formed copper silicide compounds by

sputtering Cu onto thin films of Si, with the goal of using these compounds as passivation

layers to prevent further Cu-Si interdiffusion and reaction. At room temperature, Cu3Si

and CusSi formed, while all Cu3Si decomposed to CusSi after annealing at only 3000C.

The CusSi compound disappeared after annealing at 5000C, leaving only pure Cu and








dissolving the Si into the Cu layer. The inability to form a thermally stable Cu-Si

passivation layer highlights the instability of the Cu-Si interface and underscores the

need for a diffusion barrier to separate these materials.

2.2 Refractory Metal-Based Barriers

The term refractory refers to a material with a melting point above 1800C

[Pie96]. Due to their low self-diffusion coefficients, refractory metal-based materials

have been frequently studied for use as diffusion barriers [Liu00], with current refractory

metals of interest for diffusion barrier applications including Ti, Ta, W and Ru.

2.2.1 Unary Refractory Metal Barriers

Unary refractory metals have been explored as potential barrier materials, due to

their low self-diffusivity and low electrical resistivity. Some drawbacks associated with

these materials include their tendency to crystallize (enabling grain boundary diffusion)

and to react with Si at temperatures greater than 4500C [JohOOa]. The applicability of

several unary refractory metals as barriers will be discussed in the following sections.

2.2.1.1 Ti Barriers

While Group V and VI transition metals (e.g., Ta and W) are stable and

non-reactive with Cu, Group IV transition metals such as Ti are unstable, forming

several cuprides, including Cu4Ti, Cu4Ti3, CuTi and CuTi2 [RamOO]. For this reason, Ti

is not a viable choice for a Cu diffusion barrier, as its reaction with Cu would degrade the

barrier structure and lead to rapid device failure.

2.2.1.2 Ta Barriers

Tantalum (Ta) does not react with Cu [Wan94], and has demonstrated good

oxidation resistance and enhanced Cu (111) texturing during Cu ECD [Chi02]. Ta








generally fails to suppress Cu diffusion, however, because grain boundaries readily form

during deposition and subsequent thermal cycling of the Ta layer. In addition, Ta reacts

with Si to form silicides, making the Ta-Si interface unstable, and disqualifying Ta metal

as a potential single barrier layer material to separate Cu and Si.

2.2.1.3 W Barriers

Tungsten (W) is another transition metal that does not react with Cu [RamOO,

Wan94]. It does, however, react with Si [RamOO]. W reportedly failed as a Cu barrier

due both to a silicidation reaction above 6700C [Tak97b] and crystallization, which

causes grain boundary formation [Cha97c]. Sputtered W barriers have also been reported

with columnar grains, which extended across the entire barrier thickness and caused

barrier failure above 7000C [Mer97]. While PECVD W has been reported with very low

resistivity (10 p-cm) [Cha97b], films deposited at 3500C had an open grain boundary

structure, which led to a low activation energy (0.46 eV) for Cu diffusion [Gup95].

Metallic W has also been deposited by ALD, using WF6 and Si2H6 precursors over a

temperature range of 30 to 3500C [Ela01]. Below 1000C, some Si surface species were

left on the surface and remained unconsumed by the WF6 precursor exposure.

2.2.1.4 Ru Barriers

Ru, like Ta and W, is non-reactive with Cu [Chy03], and has been tested as a Cu

diffusion barrier material. The tested Ru barriers were deposited by PVD methods,

however [Jos03], which have a limited lifespan, as mentioned in Chapter 1. In addition,

Ru silicides are reported to form after annealing at 5000C [Jel03]. A sputtered Ru layer

sandwiched between two TiN layers reportedly did not enhance the barrier's ability to

prevent Cu diffusion [Kim03c]. Ru deposited by CVD has been studied for use as the








metal electrode for DRAM capacitors [Aoy99, LeeOOa, Mat02], and Ru has also been

deposited by ALD [Aal03]; reports of CVD and ALD Ru as Cu diffusion barriers,

however, have not been given.

2.2.2 Binary Refractory Metal-Based Barriers

To overcome the limitations associated with unary refractory metal barriers,

binary compounds containing Ti, Ta and W have been widely studied. The addition of a

second, nonmetal element to the refractory metal tends to increase the likelihood of

depositing amorphous films, because the nonmetal element disrupts the crystallization

process. The non-metal also generally increases film resistivity, however. The

properties and applicability of several refractory metal carbides and nitrides will be

discussed in the following sections.

2.2.2.1 Refractory Metal Nitrides (M-N)

Adding nitrogen to refractory metals gives rise to refractory metal nitrides, where

the term nitride refers to compounds formed between nitrogen and other elements with

equal or lower electronegativity [Pie96]. Three characteristics play a role in the

formation of metal nitrides: the size ratio of nitrogen to the other element, the

electronegativity difference between nitrogen and the metal, and the electronic bonding

characteristics between nitrogen and the metal [Pie96]. Below a N/metal radius ratio of

0.59, interstitial nitrides form, while above 0.59, covalent nitrides (such as Si3N4)

typically form [Pie96]. The early transition metals (Groups IV, V and VI) have a large

enough host lattice to enable formation of interstitial nitrides, where the nitrogen atoms

sit on interstitial sites in the metal lattice. The atomic radius of N is 0.74 A, while that for

W, for example, is 1.394 A, yielding a N/W radius ratio of 0.53. The nitrogen and metal








atoms have a large electronegativity difference, so the N atoms "nest" in the interstitial

sites of the metal lattice [Pie96]. Interstitial nitrides have bonding with a combination

of metallic, covalent and ionic character, high hardness and strength, and high thermal

and electrical conductivity. These nitrides tolerate nonmetal vacancies, making them

susceptible to the presence of interstitial impurities such as oxygen. Most early transition

metals have a BCC structure, which cannot accommodate significant nitrogen levels at

the interstices. To form an interstitial nitride, the metals switch to a close-packed

structure, such as FCC or HCP, to ensure adequately sized interstitial sites that can

accommodate N. Close-packed FCC transition metals have both tetrahedral and

octahedral interstitial sites, but since the tetrahedral sites are too small to accommodate

the N atoms, N only occupies the octahedral sites [Pie96].

Nitrogen addition suppresses crystallization in the films, and helps to repel the

advance of Cu through the grain boundaries, due to a repulsive interaction between Cu

and N [EksOl, Tak97a]. Excess nitrogen in these films migrates from the bulk

polycrystals to the grain boundaries, where repulsive interactions between Cu and

nitrogen "stuff" the boundary and halt Cu diffusion [EksOl, Sha89]. Formation of copper

nitride (Cu3N) by reactive sputtering has been reported, but this compound is unstable,

decomposing to Cu and N under vacuum at 1000C [Liu98, Mar95].

Refractory metal nitrides tend to have high hardness, good chemical stability and

high conductivity [Lev98, Nag93]. Three nitrides of current interest are FCC titanium

nitride (TiN), FCC tantalum nitride (TaN), and FCC tungsten nitride (WNx). While FCC

TiN and TaN have melting points of 2950 and 3093C, respectively [Pie96], FCC WNx

does not melt, but instead decomposes to BCC W and N2 gas under vacuum at elevated








temperature (850C) [Suh01]. Since this decomposition temperature is substantially

higher than the typically processing window for ICs, FCC WNx is a viable diffusion

barrier candidate material. The barrier properties of TiN, TaN and WNx are discussed in

more detail below.

2.2.2.1.1 TiN

Titanium nitride (TiN) was widely used as a diffusion barrier for the Al-Si

system, and its reported resistivity ranges from 20 to 2000 pL2-cm [Par96]. PVD has

dominated TiN barrier deposition, resulting in non-uniform coverage of high aspect ratio

structures and a columnar grain structure [Gal99, Kal00]. TiN performs poorly with Cu

metallization due to this columnar microstructure [Nic95], which enables rapid Cu

diffusion through the barrier [Gal99]. Plasma treatment and exposure to SiH4, however,

have been reported to improve TiN's resistance to Cu diffusion [Jos02]. Another

difficulty involves the reactivity of Ti rich TiN films with Si and Cu. For TiN films with

a Ti/N ratio > 1 (i.e., film is unsaturated with nitrogen), the excess Ti will react with Si or

Cu, leaving the barrier permeable to Cu and the device vulnerable to rapid failure

[Kim92, Lee94a, RamOO].

To avoid the problems associated with PVD TiN, several different CVD

techniques have been tested. Lu et al. [2000] used LPCVD of TiC14 + NH3 at high

deposition temperature (6300C), but the resulting films had a columnar grain structure

and suffered from halide incorporation. MOCVD TiN has shown only 40-50%

conformality in small contact structures [Fal98], has high impurity incorporation, and

requires an additional plasma-processing step to stabilize the films in low resistivity

form [Gal99]. MOCVD TiN films deposited below 4500C were very porous, with low








density and high resistivity, and had poor resistance to Cu diffusion [Par95b]. Capping

the film with an ultra-thin Si3N4 layer has also enhanced performance of porous TiN

films deposited by MOCVD of tetrakisdimethylamido tantalum (TDMAT). This thin

capping layer improves barrier capability without significantly increasing the film

resistivity, but adds another step to the barrier deposition process [Lu98a].

While oxygen stuffing of the TiN barrier (due to air exposure) helps it to resist Al

in-diffusion (due to aluminum oxide formation), it does not prevent Cu in-diffusion, due

to the lack of a stable, passivating Cu oxide layer [Par95a]. In addition, TiN was not

suitable for Cu electroless plating, as its redox potential was higher than that of the Cu

electrode, preventing the initial displacement reaction from occurring at the TiN surface

[Wan03].

2.2.2.1.2 TaN

TaN is a proven Cu diffusion barrier, and is reportedly stable against Cu-Si

interdiffusion up to 7200C [Wan94]. Copper with a PVD Ta/TaN dual layer barrier is

currently used as the interconnect scheme for intermediate and upper level wiring on IC

devices. The dual layer barrier is required to overcome adhesion difficulties: Cu adheres

well to Ta, but not to TaN, while SiO2 and most dielectrics adhere well to TaN, but not

Ta. [ITR02, Sin02a]. In addition, the TaN layer promotes deposition of the low

resistivity (20-30 pQ-cm) a-Ta phase, whereas the high resistivity (180 p.2-cm) 3-Ta

phase forms when Ta deposits directly on SiO2 [Tra03]. But, since PVD TaN has much

higher resistivity (-200 to 250 iQ-cm [Sun98a, Tra03]) than Ta, the TaN layer thickness

should be held to a minimum. A 7 A thick TaN film was found to foster a-Ta deposition

and to prevent Cu interdiffusion [Tra03]. The lifespan of this bilayer deposition








technique is limited, however, due to the inability of PVD processes to deposit conformal

films at ever shrinking device dimensions.

LPCVD routes to TaN deposition have been examined as well, as this deposition

method provides superior conformality to sputtering techniques, which are projected to

reach their usability limit at the 65 nm node in 2007 [Han03]. CVD of TaN was reported

by reacting halide precursors, including TaBrs, TaF5 and TaC15, with NH3 [Kal99, Hil00].

Resistivities ranged from 395 to 5000 p.Q-cm depending on the precursor used, with the

lowest value corresponding to TaCl5 [Kal99, Che99b, Hil00]. Halide incorporation,

however, ranged from 0.5 to 4.5 at. %, with a value of 1.0 at. % for the TaC15 precursor

[Kal99, Hil00, Che99b], While plasma assisted CVD reduced the resistivity of films

deposited by TaBrs + NH3 to 150 pQ-cm, an increase in Br content up to 3 at.% occurred

[Che99a].

Attempts to grow TaN by MOCVD resulted in films with high resistivity and

some carbon contamination. When pentakis(dimethylamido)tantalum (PDMATa) was

used for MOCVD of TaN, the resulting films contained the insulating Ta3N5 phase

[Fix93]. Pentakis(diethylamido)tantalum (PDEATa) was also tested, but the films also

had very high resistivity (7000-60000 iQ-cm) [Cho98, Cho99]. MOCVD using

tert-butylimido-tris(diethylamido)tantalum (TDEATa) has also been tested, resulting in

good step coverage (-100%) but high deposition temperature (up to 6500C) and

resistivity (900-2000 pgQ-cm) [Tsa95]. Low film density and grain boundary formation

were the typical causes of barrier failure in MOCVD TaN [Kal00].

One study of an ALD TaN/PVD Ta barrier has also been given [Ho04]. The

precursors used in this study were not disclosed, but the barrier scheme was shown to








work with a 65 nm Cu back-end-of-line (BEOL) interconnect structure. The ALD TaN

barrier had good yield and reliability, and a 16% reduction in RC time delay over a PVD

TaN/PVD Ta barrier scheme was demonstrated.

2.2.2.1.3 WNx

As mentioned above, N has been added to W to deposit WNx films. These films

show promise as thermodynamically stable, stuffed barriers for separation of Cu and Si

and have been demonstrated as excellent glue layers for W/Si and W/SiO2 interfaces

[Gal97, Kim92]. Many deposition techniques, including ion implantation, sputtering,

LPCVD, PECVD, MOCVD and ALD have been used to deposit WNx films. A detailed

discussion of WNx films deposited by these techniques will be deferred until Section 2.7

below, as we have chosen to focus on this material for our work.

2.2.2.2 Refractory Metal Carbides (M-C)

Unlike the nitrides, refractory metal carbides from all three Groups (IV, V and

VI) tend to be hard, wear-resistant materials with high melting points and good chemical

resistance [Pie96]. The Group IV, V and VI refractory metals form interstitial carbides,

with crystal structures similar to those for the nitrides. The electrical resistivity of

carbides is typically lower than that for nitrides with equivalent crystal structure, due to

weaker bonding between the metal and C relative to bonding between metal and N

[Nak87]. While these properties make the carbides interesting diffusion barrier

candidates, C, unlike N, does not exhibit a repulsive interaction with Cu, making the

carbides less resistant to Cu diffusion. The barrier properties of TiC, TaC and WCx are

discussed below.








2.2.2.2.1 TiC

One report of TiC (deposited by sputtering) as a Cu diffusion barrier has been

given [WanOla]. While the TiC resisted metallurgical failure at temperatures up to

6500C, the same films suffered electrical failure after annealing at temperatures just

above 5000C.

2.2.2.2.2 TaC

Sputter deposited TaC films have also been tested as Cu diffusion barriers

[Ima97, Lau02, Mor98]. While these films had low resistivity (210 p.Q-cm for 100 nm

thick film), carbon-rich TaC films contained low-density carbon regions surrounding

TaC grains [Ima97]. These low-density regions are facile paths for Cu penetration

through the barrier, with 7 nm TaC films failing to prevent Cu diffusion above 550C

[Lau02]. This indicates that C, unlike N, lacks the ability to chemically repel Cu

diffusion through the film, and is reflected in TaC's lower activation energy for Cu

diffusion as compared to W2N or TaN [Mor98].

2.2.2.2.3 WCx

The bulk resistivity of I-WCx is slightly lower than that for P-WNx, with

reported values of ~ 41 jQ-cm and -50 dQ-cm, respectively [Lee93, Nic78]. MOCVD

WCx from W(CO)6 + C2H4 has been tested as a Cu diffusion barrier [Sun01b, Vel00].

Deposition temperatures ranged from 250 to 5100C, with 50 nm thick films deposited at

2900C having resistivity of 250 pQ-cm. A 70 nm thick WCx film annealed for 8 hours at

4000C resisted Cu diffusion. Sputtered WCx with resistivities ranging from 200 to 1000

p.2-cm were also tested as Cu diffusion barriers, with the incorporation of C resulting in

smaller grain sizes relative to pure W films [Vel00, Wan01b]. The onset of silicide








formation above 7000C, however, indicated the instability of the WCx/Si interface and

also compromised barrier integrity against Cu diffusion. Kim [2003a] reported

deposition of tungsten carbide (WCx) films by plasma assisted ALD (PAALD) using the

bis(tert-butylimido)bis(dimethylamido)tungsten [(tBuN)2(Me2N)2]W precursor with H2

and N2 carrier gas. The films were deposited on SiO2/Si substrates at 2500C, and had

growth rates ranging from 0.4 to 0.7 A/cycle, 100 % conformality in 0.15 pm features

with a 15:1 aspect ratio, and resistivities ranging from 295 to 22000 pQ-cm. The films

contained 63 at.% W, 28 to 42 at.% C, 2 to 7 at.% N, and 1 to 6 at.% O. Cu barrier

integrity tests were not reported, however.

2.2.3 Ternary Refractory Metal-Based Barriers

The addition of a third element to a binary refractory metal nitride matrix tends to

further disrupt the microstructure, increasing the likelihood of nanocrystalline or

amorphous phase formation [IstOO, Jai99, RamOO]. C, Si and B are the most examined

third elements added to metal nitride films.

2.2.3.1 Refractory Metal Carbonitrides (M-C-N)

The addition of C to refractory nitrides has two general purposes: to increase the

likelihood of amorphous film deposition, and to decrease film resistivity relative to the

binary nitride. The C and N intermix on the interstitial sublattice of the host metal, and

lattice structures are similar to those for the nitrides and carbides. The advantages of

adding C to the N on the non-metal sublattice are tempered by the decreased ability of C

(relative to N) to chemically repel Cu diffusion. The composition and microstructure of

the films must therefore be carefully controlled to deposit low resistivity films with high

stability against Cu diffusion. Free carbon (C not bound to W), in particular, should be








minimized, as it has minimal impact on Cu diffusion and detrimentally affects film

resistivity.

2.2.3.1.1 TiCN

One report of a TiCN Cu barrier, deposited by MOCVD of

tetrakis(dimethylamino)titanium (TDMAT), has been given [Eiz94a]. While the films

resisted Cu in-diffusion after annealing at 6000C, they had high resistivities, ranging

from 3000 to 20000 Q2-cm for 200 A films.

2.2.3.1.2 TaCxNy

TaCxNy films deposited by sputtering of a TaC target in an Ar/N2 atmosphere

were also tested as Cu diffusion barriers [SunOla]. TaCxNy was found to have superior

thermal stability to the respective binary phases, and its resistance to Cu diffusion was

greater than TaC due to stuffing of the grain boundaries with N. The films had relatively

low resistivity (-300 uQ-cm), and prevented Cu diffusion after a 30 min anneal at

6000C. Jun [1996] used pentakis(diethylamido)tantalum to deposit TaCxNy films by

MOCVD, but resistivity was high (2 6000 .Q-cm) and the barrier failed after annealing

for 1 hour at 5000C. MOCVD of TaCxNy films was also done using a mixture of

pentakis(dimethylamido)tantalum and pentakis(diethylamido)tantalum [HosOO]. These

films prevented Cu diffusion after a 30 min anneal at 5000C, but had high resistivity (>

4000 aQ-cm).

2.2.3.13 WNxCy

The bulk resistivity of 3-WCx is somewhat lower than P-WNx, and as expected,

the resistivity of P-WNxCy is reportedly lower than that for P-WNx. Carbon and

nitrogen in these films can deposit at interstitial lattice sites (i.e., bind to W) or as free C








or N. Given a choice, free N is more desirable, because although it increases film

resistivity (as does free C), it also repels Cu diffusion. Free C also increases resistivity,

but is less effective at preventing Cu diffusion. The difficulty, then, is to minimize or

eliminate deposition of free C, while encouraging some N to deposit interstitially and

allowing some to deposit as free N. This should give minimum film resistivity and

maximum Cu resistance.

WNxCy films were deposited by non-reactive sputtering of W-C and W-N

targets, with the resulting films being amorphous, but maintaining some short-range

order [Vie02]. These films were not tested for Cu barrier reliability, however.

MOCVD was used to deposit WNxCy as a barrier for Cu metallization in a patent

application [FukOO]. Amorphous WNxCy films were deposited by reacting a gas

containing W, such as WF6, W[N(CH3)]6, or W[N(C2Hs)]6, with a hydrocarbon gas, such

as CH4, and a nitrogen supply source, such as a nitrogen plasma, at 360C. X-ray

diffraction indicated a peak between 36 and 38 200 and a second position between 42 and

44 20, both of which are indicative of FCC WNxCy deposition. While resistivities down

to 275 pQ2-cm were reported, details of Cu barrier testing were not included.

Several studies have reported ALD deposition of WNxCy. The deposition used

sequential reactions of WF6, NH3 and Et3B [Ele03, Kim03d, Li02, Li03, Pet02, Smi02].

The lowest reported resistivity was 210 piQ-cm for a film deposited at 3500C [Ele03,

Li02]. One group reported that a 120 A thick WNxCy film was stable for a 30 min anneal

up to 7000C [Kim03d]. These ternary films had high density and excellent adhesion to

copper, but contained some amount of F (0.5-1 at.%) [Ele03, Li03].








2.2.3.2 Other Ternary Refractory Metal Compounds

Several ternary silicide systems, including TiSixNy [Bai96, ChiOl, EisOOb, Jos02],

TaSixNy [Lee99, Som97], and WSixNy [Bla97, Hir1O], have been examined for diffusion

barrier applications. While the addition of Si promotes amorphous film growth and

improves Cu adhesion [HarO1], it also increases film resistivity and can decrease failure

temperature [Kal00]. The ratio of metal/Si must be greater than 1.67 to ensure stability

with Cu [Nic95]. The failure mechanism in these films tends to be grain boundary

diffusion after barrier crystallization [Nic95], and the stability of these films is

questionable due to the potential for Si out-diffusion into and reaction with the

neighboring Cu layer. These barriers are also reported to have poor adhesion to low-k

dielectrics [Pet03], and the majority of research on ternary silicide barriers has relied on

PVD methods, which will have limited use in the future.

2.2.3.2.1 TiSiN

TiSiN films were deposited by reactively sputtering a Ta-Si target in an Ar/N2

gas mixture [Iij95, Rei94]. Reid et al. [1994] deposited amorphous films with a

stoichiometry of Ti0.34Si0.23N0.43 and resistivity of 680 pQ-cm. These films resisted Cu

diffusion after a 30 min anneal at 6500C. lijima et al. [1995] deposited amorphous films

with a stoichiometry of Tio.31Si0.19N0.50 and resistivity of 500 jQ-cm. These films

resisted Cu diffusion after a 30 min anneal at 6000C.

LPCVD of TiCl4+SiH4+NH3+H2/Ar at 5000C was also reported [Bla97].

As-deposited films were microcrystalline, with a composition of Tio.47Sio.o7No.46.

Chlorine and oxygen contamination levels ranged from 3 to 5 at.%. This film prevented

Cu diffusion after a 6000C anneal for 1 minute.


_~__ *








MOCVD of TiSiN films has been reported with multiple precursor schemes,

including TDMAT + SiH4 + NH3 [ChiOl, Jos02] and TDEAT +SiH4 + NH3 [Bai96]. A

50 A thick TiSiN film had good wettability to Cu and low resistivity (350 LQ-cm)

[Chi01]. The Si was suggested to improve Cu adhesion when thin oxidized layers are

present on the barrier, and also to reduce Cu agglomeration [Pet03]. TiSiN films

deposited by a combination of MOCVD and plasma assisted CVD, however, allowed Cu

to diffuse through after annealing at 5000C [Jos02]. Amorphous TiSiN films deposited at

4000C contained ~3 at.% C and 0, and had high resistivity (>2000 jIQ-cm) [Bai96].

Metalorganic ALD of TiSiN films was reported using TDMAT, NH3 and SiH4 at

1800C [MinOO]. The film contained small crystallites in an amorphous phase, and film

stoichiometry was Ti0.32Si0.18N0.50, but barrier film resistivity and Cu testing were not

reported.

2.2.3.2.2 TaSiN

Several reports of TaSiN thin films as Cu diffusion barriers have been given.

TaSiN films were deposited by reactive sputtering of a Ta-Si target in N2 [HarOl] and

Ar/N2 [Kim97a, Kol91, Lee99] gases. This ternary layer had better barrier performance

and better adhesion to Cu than a Ta/TaN dual layer barrier, but details of the barrier's

performance and electrical properties were not given. Lee et al. [1999] deposited films

with Tao.43Sio.o4No.53 stoichiometry, with a failure temperature of 8250C (decomposing to

Cu3Si +TaNx + TiSi2 above this temperature) but the film had high resistivity (1419

pQ-cm). Kolawa et al. [1991] deposited 100 nm thick amorphous films with a

stoichiometry of Ta0.36Si0.14N0.50. The films had high thermal stability, preventing

interdiffusion between neighboring Cu and TiSi2 layers up to 9000C, had O content








below 3 at.%, and had resistivity of 625 ji2-cm. Kim et al. [1997a] varied the N content

of the TaSiN films to determine the impact on barrier performance. Films with N content

greater than 40 at. % resisted Cu diffusion after an 8000C anneal, while those with lower

N levels failed after a 7000C anneal. Information on film resistivity was not provided,

however.

LPCVD of TaCl5+SiH4+NH3+H2/Ar at 500C was also used to deposit TaSiN

[Bla97]. As-deposited films were microcrystalline, with a stoichiometry of

Tao.35Si.11N0.54, and contained Cl and O levels ranging from 3 to 5 at.%. The films failed

to prevent Cu diffusion after a 1 minute, 6000C anneal, however.

2.2.3.2.3 WSixNy

Several reports of WSixNy thin films as Cu diffusion barriers have been given.

Reactive sputtering of a W-Si target in N2/Ar gas was used to deposit amorphous

WSio.6N films [Min96, Shi97], which crystallized after annealing at 8500C [Shi97]. Film

resistivity was 430-450 2Q-cm, and the WSio.6N film blocked Cu diffusion after a 30

min anneal at temperatures up to 6000C.

WSixNy films were also deposited by N2 plasma nitridation of sputtered WSix thin

films [Hir98, Hir99, HirOl]. As-deposited films were amorphous, and their effectiveness

decreased with increasing film crystallinity. A WSiN(6 nm)/WSix(14 nm) bilayer

resisted Cu diffusion after annealing for 1 hour at 400C [Hir99]. Electrical properties of

the barrier film were not reported, however. WSixNy thin films were inadvertently

formed by annealing a W/WNx/poly-Si structure at 800C [Nak97]. The resulting

W/WSixNy/poly-Si structure was stable up to 950C, above which silicidation of the W








layer occurred. The ternary film was not tested as a Cu diffusion barrier, however, and

bulk resistivity of this film was not reported.

WSixNy films were also deposited by LPCVD of WC16+SiH4+NH3+H2/Ar at

500C [Bla97]. As deposited films had a stoichiometry of Wo.54Si0.12No.34, contamination

levels of Cl and O ranging from 3 to 5 at.%, and high resistivity (1000 SgQ-cm). These

layers prevented Cu diffusion after a 6000C anneal for 1 minute, and crystallized at

9000C.

Amorphous WSixNy thin films were also deposited by PECVD of WF6+ N2 + H2

+ SiH4 at 380 to 4000C [Eck03]. While film resistivity was low (308 pQ-cm), layer

uniformity and thermal stability was poor, as films crystallized after a 1 hour anneal at

6000C. Amorphous WSixNy thin films were also deposited by PECVD of

WF6+Si2H6+NH3 at 3500C [Gok99]. Films with compositions of WL.21Si0.14N and

Wo.96Sio.24N were deposited, which crystallized after annealing at 800C for 30 min.

Film resistivities below 200 p2-cm were reported after annealing at 4500C, but film

performance with Cu was not reported.

2.2.3.2.4 WBxNy

Several reports of WBxNy thin films as Cu diffusion barriers have been given.

WBxNy films with a variety of stoichiometries were deposited by reactive sputtering of a

W-B target in N2/Ar gas [Rei95]. A W0.64B0.20No.16 film showed the best resistance to

Cu diffusion, preventing it after a 30 min anneal at 8000C. This stoichiometry also had a

low resistivity of 220 gi2-cm. WBxNy films were also deposited by reactive sputtering

of W and W2B5 targets in N2/Ar gas, with substrate temperature ranging from 25 to

5000C [LeeOlb, ParOO]. Film stoichiometries ranged from Wo.9oBo.o5No.o5 to









W0.57Bo.o5No.38, with resistivities as low as 140 pC2-cm obtained for the stoichiometry

with the lowest N content. As-deposited films were amorphous, crystallizing after a

7000C anneal [ParOO], and resisted Cu diffusion after a 30 min anneal at 8000C [LeeOlb].

WBxNy films deposited by reactive sputtering of a W2B5 target in N2/Ar gas at 350C were

used to form Schottky contacts to GaAs [Kim98b].

WBxNy films were formed by ion implantation of BF2+ into WNx thin films,

which were deposited by PECVD of WF6 + NH3 + H2 on Si (100) at 3000C [Kim97b,

Kim99a]. The ternary film was amorphous, and resisted N out-diffusion after annealing

at 8000C. As-deposited ternary bulk film resistivity was 200 p.Q-cm, which was slightly

higher than the value for the binary WNx film, and the ternary film resisted Cu diffusion

after annealing at temperatures up to 750C.

Deposition of WBxNy thin films by PECVD of WF6 + NH3 + BloH14 + H2 on Si

(100) at 3500C has also been reported [Kim97b, Kim98a, Kim02a]. As-deposited film

stoichiometries ranged from Wo.90B0.05No.05 to Wo.38B0.42N0.20, depending on the B1oH4/

NH3 ratio, with the optimal film stoichiometry being W0.46B0.25No.29. Resistivity ranged

from 100 to 844 jLQ-cm, with a value -700 pLQ-cm for a 2000 A thick film with optimal

stoichiometry. Films with optimal stoichiometry remained amorphous even after

annealing at 8000C, and resisted Cu-Si interdiffusion up to this temperature. Tungsten

rich as-deposited films were polycrystalline, and B atoms reportedly out-diffused even

faster during annealing that did the N atoms. Both the Wo.9oBo.o5No.o5 and Wo.80Bo.15No.05

stoichiometries failed to resist Cu diffusion after annealing above 6000C.









2.3 Justification for WNx (and WNxCy) as the Barrier Material

The microelectronics industry already uses Ti, Ta and W in one form or another

(such as TiN barriers, Ta/TaN barriers and Ta205 capacitors, and W plugs) for memory

and processor devices [Jai99]. TiN is used as a barrier for Al based metallization, but

TiN is unsuitable for Cu metallization, as mentioned earlier. Cu diffusion barrier

research has focused on TaN and WNx, with WNx appearing to have technological

advantages over TaN as a Cu diffusion barrier. WNx is known to be an effective

diffusion barrier against copper penetration at temperatures up to 750 *C [Pok91], and

adhesion of CVD Cu to WNx is stronger than that to TaN [Iva99]. In addition, WNx

outperforms TaN as a liner material for seedless electrochemical deposition (ECD), as

ECD Cu shows stronger adhesion to WNx film layers [ITR02, ShaOl, Sin02a]. A suitable

liner material would enable seedless ECD of Cu. Elimination of the Cu seed layer,

typically deposited by sputtering, would increase throughput by removing a process step,

as well as remove the complications associated with sputtering in ever-shrinking device

features. The combination of these steps would result in cost savings per wafer, higher

wafer throughput, and higher quality devices for a WNx based system as compared to its

TaN counterpart [Gal99].

In addition, to enable electroless plating of Cu onto the barrier materials, TaN

barrier layers require a pre-deposition HF etch, while WNx barriers do not. Tungsten

oxide, present on the WNx surface due to air exposure, is readily dissolved in the

electroless plating solution, hence the pre-deposition etch step is unnecessary [Wan03].

There are also processing advantages for WNx. During chemical mechanical

polishing (CMP), TaN is removed at a rate 18 times slower than Cu, while WNx is









removed 1.5 times faster [Gal99]. One report gives WNx removal rates of -100 A/min

for CMP pressures below 0.5 psi, while comparable removal rates for TaN require

polishing pressures close to 2 psi [Tak02]. The slow removal rate of TaN means that the

pad must be in contact with the Cu/TaN surface for an extended period of time to

planarize the wafer. This extended contact time leads to dishing of the copper, where

excessive copper is removed from the device. Dishing of both the Cu and dielectric

layers impacts line resistance and planarity of the device [Jai99]. A time-consuming

two-step CMP process, which includes a slurry and pad changeout, is required to

minimize copper dishing during planarization of TaN.

From the above discussion, it is evident that WNx films are the most promising

candidates for diffusion barrier materials. The resistivities for these films are reasonably

low, and WNx has shown good adhesion to Cu and other potential neighboring materials.

CVD and ALD routes to low-resistivity WNx have been established, although the

precursors must be optimized to minimize halide contamination and to control carbon

levels. Recent studies have examined WNxCy for diffusion barrier applications, due to its

ability to deposit in amorphous form, its lower resistivity relative to WNx, and its

excellent adhesion to Cu. A review of the properties of WNx, along with a brief

discussion of WNxCy films, follows.

2.4 WNx Film Properties

The four commonly seen phases for tungsten nitride are BCC WNx, FCC 0-WNx,

HCP WNx, and SHP-WN [Gui93, Nak87, Wri89]. The structure and properties of these

phases are discussed in more detail in Chapter 3. FCC I-WNx is the desired phase for

Cu diffusion barriers because it has the lowest bulk resistivity (-50 pIQ-cm) of the








various tungsten nitride phases [Lee93]. The structure of FCC WNx is NaCI type, with

W atoms at FCC sites and N atoms at octahedral interstitial sites. The FCC P-WNx

phase, with x = 0.5 (also called P-WNo.5 or 1-W2N) has lattice constant a=4.126 A,

hence it has a lattice mismatch of -24% with Si (a=5.431 A) and -27 % with GaAs

(a=5.653 A) [JCP88]. The P-WNo.5 stoichiometry is described as a defective NaCl-type

structure, with half of the octahedral interstitial sites filled with N and half of them vacant

[HonOO]. Bonding in metal nitrides is complex, with both metallic (valence electrons

delocalized; non-directional bonding) and covalent (valence electrons shared; highly

directional bonding) characteristics, due to the combination of metal-metal and

metal-nonmetal interactions [Tot71]. Hones et al. [2000] described metal-nonmetal

bonding in NaCI type structures as having 7n-like and a-like bands due to overlap of the

d and 2p orbitals from the metal and nonmetal, respectively. These bands (called pdn

bands) have ionic character, with bonds having more ionic character when this band is

populated and more covalent-metallic character when the band is depopulated [HonOO].

Shen et al. [2000a], using transmission electron diffraction (TED), found W-N and

W-W nearest neighbor distances of 2.08 and 2.92 A, respectively, in sputter deposited

P-W2N films. The P-W2N lattice parameter (4.126 A) represents the distance between

centers of W atoms at the covers of an FCC unit cell, while the 2.92 A W-W correlation

represents the distance between centers of the corer and face centered W atoms in the

FCC unit cell (Figure 2-1). The theoretical density of FCC P-WNo.5, based on 4 tungsten

and 2 nitrogen atoms per unit cell, is 18.0 g/cm3, while experimentally determined

densities range from 8.0-17.9 g/cm3 [Bos91, Gal97, Hec02, Mar93, Sam80, Sot03].














4.126 A N

2.92 A




Figure 2-1. Distances between W atoms in the FCC face of the 3-W2N phase.

Shen and Mai [2000b] have reported that binding energy shifts (per XPS) for W

and N in polycrystalline P-WNo.5 films indicate an ionic bonding character in the films

(consistent with W4-NI-). The apparent binding energy difference between the metal

and N atom, from XPS, gives an idea of the degree of charge transfer. A large binding

energy difference infers greater bond ionicity [Pri95]. Moreover, adding more nitrogen

to the films has been suggested to cause a decrease in the number of free electrons

provided by W in the solid film [Sot03], which is consistent with increasing film

resistivity with nitrogen content. At lower N levels, the metal's band structure is

retained, so that metal nitrides show metallic properties [Muk93]. Matsuhashi and

Nishikawa [1994] reported FCC P-WNx films to have a work function before annealing

of 5.0 eV, where work function is a measure of the energy required to remove an electron

from the material to a state of rest outside the material (i.e., at the vacuum level).

Transmission electron diffraction (TED) intensities were also used to determine the

concentration of W-N neighboring pairs in the P-WNx films. Results showed that the

concentration of W-N neighboring pairs did not increase when bulk N content was

increased above x-0.5 [ShenOOb]. This indicates that excess N in the films migrated to








the polycrystalline grain boundaries rather than filling the remaining vacant interstitial

sites in the FCC lattice. Shen et al. [2000c] reported no SHP 6-WN formation, even for

WNx with x=1.22, which shows that excess N migrated to the grain boundaries in the

polycrystalline FCC films.

Other reported properties for the binary nitride include a compressive stress of 4.3

+ 0.5 GPa, nanohardness of 30 GPa, and a Poisson's ratio assumed to be 0.25 for films

with the WNo.6 stoichiometry [HonOO]. In addition, depending on the W/N ratio, thick

WNx films were reported with refractive indices and absorption coefficients ranging from

3.20 to 4.00 and 1.20 to 3.95, respectively, as determined by ellipsometry [Boh90].

Oxygen impurities tend to form solid solutions with metal carbides and nitrides

[Tot71], and oxygen is reported to oxidize the surfaces of polycrystals [Par96].

Impurities such as O, N and C are reported to enhance the stability of diffusion barrier

films [Cha97c, Cha99, Lee94a]. Various stoichiometries of tungsten oxynitride have

been reported. The JCPDS powder diffraction standard indicates two phases of W(N,O)x,

one with x=0.62 and a=4.138 A, and the other with x= 0.57 and a=4.126 A [JCP88]. A

WN1.3400.42 phase with a=4.153 A has also been reported [Sel95].

2.5 WNxCy Film Properties

Little information is available in the literature on the structure and properties of

WNxCy films. In going from nitrides to carbides, electron density increases around the

metal atomic sphere and decreases around the non-metal atomic sphere, consistent with

more ionic bond character for the nitrides and more covalent bond character for the

carbides [Gub94]. This is also consistent with the higher resistivity of 3-WNx relative to

P-WCx [Kim03a, Nak87]. Adding carbon to the nitride phase therefore leads to a








decrease in film resistivity. The FCC WNxCy phase, with N and C intermixing on the

interstitial sublattice of FCC W, has been predicted by thermodynamic analysis [Fri99b,

Hua97, Jon93], and has been studied for diffusion barrier applications. A detailed

experimental analysis of the composition, structure and electrical properties of these films

has not been reported, however.

WNxCy film deposition was reported using WF6+NH3+B(C2H5)3 at 3500C [Ele03,

Li03]. The film composition was Wo.ssNo.15sC.30, and films had a nanocrystalline P-WCx

or P -WNx cubic structure in an amorphous matrix along with low resistivity (210 to 400

pQ-cm). The films had F, O and B impurities below 0.5 at.%, H content < 4 at.%,

density of 14 g/cm3 and resistivity of 300 to 400 JiQ-cm [Li03]. Another report using

the same chemistry gave a film composition of Wo.57No.13C0.30 and resistivity ranging

from 600 to 900 pQ-cm, but deposition temperature was not given [Smi02].

Kim et al. [2003d] also reported deposition of WNxCy films using WF6+NH3+

B(C2H5)3 at 3500C. Film composition from RBS was 48 at.% W, 32 at.% C and 20 at.%

N, and film density was 15.37 g/cm3. The film had low resistivity (350 p-Q-cm), and had

an electron diffraction pattern that closely matched those for P-WNo.5 and P -WCo.6.

One HR-TEM lattice fringe spacing was 2.39 A, which was between the interplanar

spacings for (111) P-WNo.5 (2.38 A) and P -WCo.6 (2.43 A). The other lattice fringe

spacing was 2.08 A, which was between the interplanar spacings for (200) P-WNo.5 (2.06

A) and P -WC0.6 (2.11 A). These values support the presence of a ternary WNxCy solid

solution. Moreover, a 12 nm WNxCy film prevented Cu diffusion during annealing up to

7000C.








Vlakhov et al. [1995] deposited W, WC and WNxCy films to study their electrical

resistivities, and stated that bulk tungsten nitrides, carbides, and carbonitrides were good

superconductors [Vla95]. The W films were deposited from both W(CO)6 and WCl6, the

WC films were formed by annealing the W films in a carbon-containing atmosphere, and

the WNxCy films were deposited by co-reacting W(CO)6 +NH3 + CH3COCH3 [Vla95].

The room temperature (300 K) resistivities of W, WC and WNxCy films with thickness of

0.4 gLm were compared. The W and WC films had similar resistivities, ranging from 109

to 191 tQZ-cm, while the WNxCy films had a much higher resistivity of 651 pQ-cm.

2.6 Amorphous WNx Film Deposition

Incorporation of smaller solute atoms into a metal matrix will cause a crystalline

solid solution to become unstable above a certain solute concentration, fostering

amorphous film growth [E1190]. To enable amorphous film deposition in binary systems,

the atomic radii of the two elements must differ by more than 10 % [E1190]. The atomic

radius of N is 0.74 A, while that for W is 1.394 A, hence N is 47% smaller than W, and

so WNx should be able to deposit in amorphous form. Moreover, nitrogen can act as a

roadblock to trap diffusing W on the film surface, or can serve as a nucleation site for

lattice defects. The trapped W and permanent defects caused by N incorporation prevent

crystallization, and amorphous growth occurs [SheOOb].

Several reports of amorphous WNx deposition by a variety of deposition

techniques have been given. These include reactive sputtering, LPCVD, PECVD,

MOCVD, annealing W in NH3, PLD and ion implantation. FCC f3-WNx films with x <

0.5 were typically amorphous, while those with x = 0.5 contained polycrystalline








P-WNo.5, and those with x > 0.5 contained P-WNo.5 and additional N at the grain

boundaries [She00e,f].

Depending on their composition, amorphous WNx films have been reported to

crystallize at temperatures ranging from 480-6200C [SuhOl]. Additional contamination

from C will likely increase the crystallization temperature of WNxCy films.

2.7 Demonstrated Uses of WNx

WNx as a Diffusion Barrier

Kilbane and Habig [1975] first reported growth of WNx thin films by reactively

sputtering a W target in a mixture of Ar/N2 gas. Reichelt and Bergmann [1975] also

studied deposition of WNx thin films by RF sputtering. Ten years later, two reports were

published describing the use of sputter deposited WNx films as diffusion barriers [Aff85,

Kat85]. Since then, many studies of WNx as a diffusion barrier separating Cu and Si

have been reported. References for some of these reports are listed in several review

articles [Jai99, Kal00, Kim03b].

The use of WNx as a diffusion barrier on low-k materials has also been examined.

The interaction between WNx and the low-k material hydrogen silsesquioxane (HSQ)

has been studied [ZenOOa]. Retention of H in the HSQ film is very important, as the

dielectric constant (k) of the film is a strong function of H content (as Si-H). A W2N

barrier film deposited by PECVD of WF6+N2+H2 prevented the release of H (as H2) from

HSQ more effectively than its TaN counterpart deposited by PVD [ZenOOa]. Adhesion

strength between WNx and two other low-k dielectrics, an aromatic hydrocarbon

(SiLKT, from Dow Chemical) and a polyarylene ether-based polymer (Flare 2.0T,

from Allied Signal), has also been studied [LanOO]. TiN, TaN and WNx all had similar








adhesion strengths to these low-k materials, suggesting that a similar bonding

mechanism exists between the barriers and these low-k films [LanOO]. These barriers are

expected to have stronger adhesion than Cu to the polymer layers, because Cu, unlike the

refractory metals, has fewer unfilled d-orbitals available for bonding to the surface

[LanOO].

WNx films have also been tested to separate Cu from the low-k material fluorine

doped silicon oxide (SiOF) [Lee98c]. WNx had poor adhesion to SiOF layers which had

not been pre-treated with 02 plasma at 3000C, and failed to prevent Cu-SiOF

intermixing. When the SiOF layer was pre-treated, however, WNx had good adhesion to

and prevented Cu diffusion into SiOF for a 30 second anneal at 9000C. Surface oxidation

and densification of the SiOF, caused by the plasma pre-treatment, are believed to

enhance barrier capability on pre-treated SiOF [Lee98c].

WNx films have also been tested as diffusion barriers between the Ni and W

layers in an InxGal-xAs/Ni/W contact structure for use with GaAs [Uch97]. These

reactively sputtered WNx films prevented out-diffusion of In from the InxGalxAs layer,

and also lowered the contact resistance from (4 to 1 -cm).

WNx formed by nitridation of an amorphous CVD-W layer using N2 plasma was

used as a barrier in the multilayer structure Cu/amorphous-WNx/amorphous-W/p+n-Si

diode [Cha99]. The structure maintained its integrity after annealing at 725C. The same

group determined the upper limit for a Cu/amorphous-WNx/amorphous-W/Si multilayer

structure to be 7500C [Cha97a, Cha97c].








WNx as Gate Electrode in MOSFET Devices

Historically, metal oxide semiconductor field-effect transistor (MOSFET)

devices have used SiO2 as the gate dielectric. Shrinking device sizes are forcing a

decrease in SiO2 layer thickness. Devices formed with SiO2 layers less than 3 nm thick

suffer from high leakage current due to direct tunneling of electrons through the oxide,

hence an alternative material with higher permittivity is required to prevent tunneling

[LeeO1c]. Ta20O is a potential alternative due to its high dielectric constant (20-25)

relative to SiO2 (3.9) [May90, Par98a]. New contact materials, including WNx, are being

investigated for devices with Ta205 gates. WNx/Ta205 structures had lower leakage

current than TiN/Ta205 structures after annealing at 9000C [LeeOlc]. In addition, the

WNx/Ta205 structure had no interfacial reaction, while the TiN/Ta20s interface degraded

after annealing. The superior performance of WNx is reportedly due to diffusion of N

from WNx into the Ta2O5 layer, which suppresses Ta and O out-diffusion [LeeOlc].

Diffusion of N into Ta205 did not occur for the TiN/Ta20O structure.

WNx has also been used as a diffusion barrier and W source in poly-Si gate stack

structures [Gal00, Yan02]. Without a barrier, W reacts readily with Si, forming a highly

resistive WSix phase [Kas94]. Kasai et al. [1994] formed a W/WNx/Poly-Si gate

structure with a 50 A thick layer of WNx. The sheet resistance of W in this gate structure

was low (1.6 Q/ ) compared to one without the barrier layer (18 Q/ ), which suffered

from silicide formation. First, PECVD WNx was deposited on poly-Si and annealed to

form a W-Si-N layer. Then, the structure was annealed at temperatures up to 10000C in

either an Ar or N2 atmosphere to convert some of the remaining WNx layer to W. The

newly formed layer of W can then be used as the contact metal for the gate. While








as-deposited WNx films had sheet resistance of 19.2 Q/ the W layer produced by

annealing had sheet resistances as low as 1.28 Q/ [Gal00]. Sputtered WNx was tested in

a similar manner; formation of WSi2 was suppressed due to formation of a Si3N4 layer at

the WNx-Si interface after a rapid thermal anneal (RTA) at 10000C [Yan02]. Kang et al.

[2001,2002] used a sputtered, 100 A thick WNx film as a barrier layer in a W/WNx/poly

Sil-xGex gate structure for use in a CMOS-FET. Takagi et al. [1996] used a W/WNx

polysilicon gate electrode to produce a CMOS device. The WNx layer suppressed silicide

formation after a 30 second anneal at 950C, and the specific contact resistance for a

W/WNx/poly-Si structure was 10-7 Q-cm2 [Tak96]. In addition, the sheet resistance of

WNx deposited on poly-Si was reported to be an order of magnitude lower than that for

WSix [Cho02].

WNx as a Gate Electrode in MESFET Devices

Rectifying (Schottky) contacts for GaAs self-aligned gate field-effect transistors

(SAGFETs) must have good thermal stability, good adhesion to GaAs, and high Schottky

barrier height [Lee95, Pac91]. WNx Schottky contacts formed by PECVD of

WF6+NH3+H2 on GaAs maintained their interface integrity after a 30 second, 1000C

RTA [Lee95]. Yu et al. [1988] sputter deposited thin WNx films for use as a Schottky

contact to GaAs, while Boher et al. [1990] deposited thick WNx films. Kim [1994] used

PECVD of WF6 + NH3 + H2 at 3500C to deposit WNx Schottky contacts to GaAs. WNx

nucleated easily on GaAs and had higher thermal stability than W films, maintaining the

W2N stoichiometry and blocking As out-diffusion during a RTA at 10000C for 30

seconds [Kim94]. Grain boundaries in the PECVD WNx films were stuffed by excess N,

preventing Ga and As out-diffusion, and this Schottky contact structure was stable up to








850C. Paccagnella et al. [1991] examined the effect of GaAs pre-treatment on the

performance of WNx/GaAs Schottky diodes. Schottky barrier height was highest for

WNx/GaAs diodes annealed at 800C, where GaAs was pretreated by H2 plasma [Pac91].

Sputter deposited WNx has also been used as a gate metal in GaAs SAGFETs [Nag94,

Gei86, Uch86]. In particular, WNo.o4 was found to produce Schottky barriers with

excellent stability after annealing for 20 min at 8100C [Gei86]. WNx films have also

been deposited onto 4H-SiC to form Schottky contacts with high temperature stability

[Pec97, Kak99]. The contacts, deposited by room temperature magnetron sputtering of a

W target in Ar/N2 gas, remained rectifying up to 1200"C despite formation of the W5Si3

and W2C phases [Pec97].

WNx as an Ohmic Contact in HBT Devices

Thermally stable, low resistance ohmic contacts are critical to produce high

speed, high frequency devices such as heterojunction bipolar transistors (HBTs). Park et

al. [1998b] deposited the Au/Pt/Ti/WNx metal contact structure onto n-InGaAs, which is

a cap layer used for AlGaAs/GaAs HBTs. The minimum contact resistivity for the

Au/Pt/Ti/WNx/n-InGaAs structure was 9.5 x 10-8 L-cm2, which was obtained after

annealing at 4000C [Par98b]. In addition, the morphology of the contact remained

smooth over a wide annealing range.

WNx as a Top Electrode in DRAM Devices

WNx films have also been investigated as top electrodes for Ta205 dynamic

random access memory (DRAM) capacitors [ChoOl, Mat94]. Cho et al. [2001] deposited

both TiN and WNx by reactive sputtering in Ar/N2 ambient to test their stability as a

diffusion barrier at the interface in a W/Ta205 gate structure. The structures with WNx








barriers had higher thermal stability, as they were more resistant to Ta and O

out-diffusion than TiN. The increased stability was presumably due to migration of N

from the WNx film to the WNx/Ta205 interface, which prevented Ta and O from diffusing

out of the Ta20s. Suzuki et al. [1998] used WF6 with two different N sources, NH3 and

NF3, to deposit WNx thin films by PECVD. When tested as top electrodes in DRAM

capacitors, these films had leakage currents that were an order of magnitude lower than

PVD TiN top electrodes. Matsuhashi and Nishikawa [1994] tested sputtered WNx, TiN

and TaN as top electrodes in Ta205 DRAM capacitors. After annealing for 30 min at

8000C, the WNx/Ta205 structure exhibited lower leakage current than structures with TiN

or TaN electrodes [Mat94]. Kim et al. [2001] used a W/WNx/Poly-Si gate stack to

produce polysilicon based DRAM devices.

WNx as an X-ray Absorber Mask

The use of amorphous WNx films in absorber masks for X-ray lithography has

also been demonstrated [Lee97, Lee98b]. WNx has good stress controllability, strongly

attenuates x-rays, and has a coefficient of thermal expansion (CTE) very close to

common x-ray membrane materials such as SiC, BN, SiNx, and CN, making it a

desirable x-ray mask material [Lee98b]. WNx masks were grown by reactive sputtering

of a W target in an Ar/N2 atmosphere on room temperature indium tin oxide (ITO) coated

Si substrates. Film microstructure depended on the nitrogen content of the films, with 20

at. % N being the upper limit for the amorphous phase. As N levels increased above this,

film microstructure shifted to polycrystalline P-WNo.s.








WN, as a Liner for Cu Deposition

Sputtered WNx films were also shown to be good liner materials for seedless

electrochemical deposition, due to strong Cu-WNx adhesion and high Cu nucleation

density on the WNx surface [Sha01]. ECD Cu was deposited on air-exposed WN, films,

and the composition of the ECD bath was controlled to strip off surface oxides [ShaOl].

Use of TiN as a seed layer for electroless Cu deposition resulted in only sporadic Cu

coverage on the barrier layer [Mur95], while electroless Cu deposition occurred readily

on WNx films [Wan03]. For the same bath temperature (700C), electroless Cu deposition

occurred more rapidly on WNx than on TaN [Wan03]. Moreover, Cu deposited by CVD

had low via resistance on WNx, but high via resistance with Ta based barriers [Jai99].

Other WNx Uses

P-WNo.5 has found use in bulk form (i.e., powder) as a catalyst for quinoline

hydrodenitrogenation [Abe93], n-heptane isomerization [Sel95], deamination of

2-octylamine and alcohol dehydration [Lee92, Luc96]. Reactively sputtered amorphous

WNx films have also been deposited and used as a means to form equiaxed, low

resistivity W films in W/poly-Si gate structures [Lee03]. Amorphous WNx films

annealed at 1273 K released nitrogen to form BCC W with a resistivity of 12 pQ-cm,

which was similar to the value for pure, sputtered W. The presence of a small amount of

residual N in the films at 1273 K suppressed silicide (WSix) formation, enabling the W

film to retain its low resistivity.

WN, Etching

Several etching studies have been done on WNx films. Lee [98b] first reported

use of SF6 + Ar + N2 in an inductively coupled plasma (ICP) etching system to








anisotropically etch WNx thin films. Etch rates ranged from 4000 to 14000 A/min.

Vijayendran et al. [1999] used an NF3/N2 reactive ion etch (RIE) to remove WNx films.

Reyes-Betanzo et al. [2002] used SF6 + Ar to anisotropically etch W and WNx films.

2.8 WNx Deposition Techniques

A variety of techniques to deposit WNx have been tested in order to optimize key

film properties, which include film resisitivity and deposition temperature. Ideal film

resistivity was reported to be 500 p.-cm, [Ele03], while a temperature ceiling of 400C

has been given for IC production by numerous reports [EisOOb, Ele02, HauOO, Kim03b,

Les02, Sun01b]. Depositing the barrier at the absolute minimum temperature, however,

may not be the best solution, as other processes occurring during IC processing will

typically take the device temperature up to ~4000C. If the barrier is deposited at much

lower temperature than the processing temperature ceiling, significant shift in barrier

structure can occur during subsequent high temperature processing, leading to Cu

diffusion and potential barrier failure.

The reported deposition techniques include reactive sputtering, annealing of W in

NH3, plasma nitridation of W, LPCVD, PECVD, MOCVD, and ALD. Reports of each of

these techniques to deposit WNx are discussed below.

2.8.1 Annealing W in NH3

Deneuville et al. [1989] reported formation of WNx by annealing sputtered W

films on Si in an NH3 atmosphere from 500 to 11000C. The N/W ratio increased with

temperature, going from 0.37 to 1.85 across the aforementioned temperature range.

While several polycrystalline structures were postulated based on film composition, XRD

results were not given to support them. High temperature and pressure, along with long








anneal times, are required to form WNx in this manner, making this an undesirable

deposition technique for IC manufacturing.

2.8.2 Plasma Nitridation / Ion Implantation of W

W films, deposited by CVD of WF6 + SiH4 + H2 were nitrided to form thin WN,

films using an in-situ N2 plasma treatment [Yeh96]. This nitride layer prevented WAll2

formation after annealing at 550C, where WA112 formation can degrade contact

structures using W plugs between Al and Si. WNx films were also formed by N+ ion

implantation into W metal substrates [Zha99]. Barrier testing with Cu was not done for

either of these films, however. N+ ion implantation was also done on W films deposited

by PECVD of WF6 + H2 to form amorphous WNx films [Kwo95]. These films resisted

Cu penetration after a 30 min anneal at 8000C, while polycrystalline PECVD WNx films

failed. The nitridation technique is directional in nature, however, meaning that good

conformality will be difficult in highly aggressive future device topographies. Chang

[1997a-c, 1999] reported N+ ion implantation into W films deposited by CVD of WF6 +

SiH4. The films had resistivity of 198 pQ-cm and resisted Cu diffusion for annealing

temperatures below 7000C, but had poor adhesion to the underlying Si substrate.

2.8.3 Pulsed Laser Deposition

Soto et al. [2003] first reported pulsed laser deposition of tungsten nitride films

from a W target in the presence of N2. Films were deposited on n-type Si (111) and on

Coming glass slide substrates, at N2 pressures ranging from lxl0-8 to lxl0-1 Torr. Films

contained W, N and some O, believed to originate from W target contamination, and had

higher density than films produced by DC magnetron sputtering. This technique, like








plasma nitridation, is also directional, making its ability to deposit highly conformal films

in future device generations questionable.

2.8.4 Reactively Sputtered WN, Deposition

Typically, sputtered WNx films are formed by sputtering a W target in a N2/Ar

atmosphere. A variant of this, known as nitrogen ion beam sputter deposition (IBSD),

has also been reported [Eiz94b, Gal93], where a nitrogen ion beam, rather than an Ar+

ion beam, is used to liberate W atoms from the W target. Nitrogen ions or radicals

backscattered from the target are then incorporated with W at the substrate to deposit

WNx. Both variants enable amorphous film deposition due to the decreased mobility of

W in the presence of N on the substrate surface. N can inhibit W mobility on the

substrate surface by blocking W diffusion or by trapping W. Decreased W diffusivity

leads to defect formation, failure of crystal growth and amorphization of the growing film

[SheOOb]. Sputtering can be done at low deposition temperature (at or near ambient),

which protects temperature sensitive components from thermal damage during the barrier

deposition process.

Many reports of sputtered WNx deposition have been given in the literature. The

key properties of these films are summarized in Table 2-1. While sputter deposited WNx

films are generally contaminant free, the major drawback of sputtering, as mentioned in

Chapter 1, is poor conformality in small, high aspect ratio device features.

2.8.5 LPCVD WNx Deposition

LPCVD of WNx has been researched as a possible alternative to sputtering for

smaller device features with high aspect ratios. Many reports of LPCVD WNx films have












0% e 00000
0%~0
88 a m

M~ dl~ =0 ? 3s
Q,- 00
kNn 06db~




00
o $~





cs~cq














'00
-B










bo






Wz







0 E U N






+++
cuoo
IS 8 S






























in 0
E- U

1^^ ? ^ ^

g^ ^5 ^

^
1^ ^
|^^
o
0

U5
._ in
.9 *-- Y o





.2 ,'-
1 t^AS (
(S~ A b ^





1-a a i .9 a
Ev l ,.P ,Z


been given in the literature. The key properties of these films are summarized in Table 2-








As evident from Table 2-2, LPCVD reactions are based on halide chemistry, and

include the reduction of WF6, WO3, or WC16 by NH3 or NF3 in a H2 atmosphere [Chi93,

Gon02, Mar93, Nag93, Nak87, Sak96, Suz98, Vol85]. Nagai and Kishida [1993]

reported that use of WCl6 as the tungsten precursor for reaction with NH3 and H2 was

more thermodynamically favorable than using WF6, as the Gibbs free energy change was

larger for the chloride precursor reaction. Marcus et al. [1993] have reported high

LPCVD deposition rates, ranging from 1800-4500 A/min.

One report (not included in the table) of the "indirect" deposition of J3-WNx has

been given, where this phase formed after annealing a metastable W3Ns phase [Zha97].

Reacting WNC13 with ZnN2 produced the W3N5 phase at 4000C, and subsequent

annealing at 6000C resulted in formation of the f3-WNx phase.

The halide precursors used for LPCVD can create difficulties during barrier

deposition and subsequent device processing. WF6 is reported to consume Si during the

reaction, forming SiF4, which leaves vacancies on the Si substrate [Kim91, Kim92,

Lai98a]. In addition, there are concerns that residual F and Cl in the barrier contribute to

corrosion of metal interconnects [Fal98, Kel99, Raa93]. Contamination levels of

0.1-0.9% F have been reported for LPCVD films [Mar93]. Moreover, reactive

by-products (e.g., hydrogen halides) resulting from LPCVD are a material handling

concern [Cur92, Kim91].

Adduct formation is another difficulty associated with LPCVD. Adducts are

compounds produced by chemical addition of two or more reactants. These adducts can

settle onto the substrate and barrier film during deposition, resulting in dislocations

and/or pinholes in the film that can lead to barrier failure. Gas phase adducts such as






90


WF6-4NH3, NH4F2, and NH4F have been reported during LPCVD reactions [Nak87,

Suz98, Tsa96].

I .


o E 2





a




0 0
a o < a


00

o












0 + + +







uder, rather than film, produced in this report
0 0






bulk -W2N powder, rather than film, produced in this report
E ^ ^^ <>
5~ ^' 2 2X 2 '^
c~Y 1/^ 9
cx s 3(
au_________Q___


(j te lulk B-2







2.8.6 PECVD WN, Deposition

The same halide chemistries used for LPCVD are also used for PECVD, whereby

the plasma assists fragmentation of the reactants, lowering reaction temperatures

considerably. After fragmentation by the plasma, the precursor fragments travel to the

heated substrate, react on its surface and deposit a film. Many reports of PECVD WNx

films have been given in the literature. The key properties of these films are summarized

in Table 2-3.

Films deposited by PECVD show excellent adhesion, with resistivities as low as

70 ~2-cm. Growth rates are moderate, ranging from 300-400 /min [Gal97], and

conformality up to 70 % has been reported for trench holes with diameter > 0.30 lm

[Kim93].

PECVD films suffer from reduced conformality in small diameter, high aspect

ratio trench features, however, due to the directional nature of the plasma [JohOOa,

Tsa96]. Despite the low resistivity and deposition temperatures that PECVD offers, its

inability to deposit highly conformal films in high aspect ratio features makes its use in

future barrier deposition questionable. Moreover, PECVD films, like LPCVD films,

suffer from halide impurities. Concerns about halides contaminating the barrier films

also exist for PECVD, since the same reactants are often used for both LPCVD and

PECVD. Contamination levels up to 4 at.% F, which are higher than those for LPCVD,

have been reported for PECVD films [Lee93]. Another difficulty associated with

PECVD is gas phase adduct formation; adducts such as (WF6)x(NH3)x and NH4F have

been reported during WNx PECVD [Lu98b, Suz98]. These adducts can contaminate

films grown by PECVD and deposit particles on the film surface, both of which can