|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
ALTERNATIVE NITRIDE DIFFUSION BARRIERS ON SILICON AND
GERMANIUM FOR COPPER METALLIZATION
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
Dedicated to my mother Rasilaben Rawal and my dear wife Purvi Rawal.
I would like to immensely thank my advisor Dr. David P. Norton. His enthusiasm
and knowledge on different subj ects of science are astounding. Dr. Norton' s exemplary
dedication towards his work and students along with his zeal to be the best in the world in
his area of research motivated me greatly to pursue and conduct superior quality research.
I would like to thank Prof. Rajiv K. Singh for providing me guidance and support during
my initial years in USA. I would like to thank Prof. Tim Anderson and Prof. Lisa
McElwee-White for their guidance and help with my research. They have helped me to
think and analyze my research critically.
I would like to thank my mother, Rasilaben Rawal, for her passion towards
education. Her undying zeal towards learning sowed the first seeds in me to respect and
acquire knowledge. Many thanks go to my high school chemistry teacher, Mr. Dubey and
physics teacher Mr. Parmar for providing support through my 12th standard. Special
thanks go to Mahendra Pandya and Geetaben Pandya, for supporting me throughout my
undergraduate years. I would like to thank my brother, Rakesh Rawal, and his wife,
Padmaj a Rawal, for their love and support. Most importantly, I would like to thank my
wife, Purvi Rawal, for her unconditional love and support. Her immense understanding
and perceptive nature have made my journey cherishing. I would like to thank all my
friends and well-wishers for their support and encouragement.
Last but not least, I thank GOD for giving me this life and opportunity to serve HIS
TABLE OF CONTENTS
ACKNOWLEDGMENT S .............. .................... iv
T ABLE S .............. .................... vii
LIST OF FIGURES ............ ....... ..............viii...
AB STRAC T ................ .............. xi
1 INTRODUCTION ................. ...............1.......... ......
2 LITERATURE REVIEW .............. ...............7.....
Refractory Metals as Diffusion Barrier .............. ...............7.....
Ti Diffusion Barrier............... ...............7.
Ta Diffusion Barrier ................. ...............8................
Cr Diffusion Barrier .............. ...............8.....
W Diffusion Barrier............... ...............8.
Binary Diffusion Barrier............... ...............9.
Refractory Intermetallics ................. ...............9.................
Refractory Carbides ................. ...............10.................
Refractory Nitrides ................. ...............12.......... .....
Ternary Diffusion Barriers .............. ...............15....
3 EXPERIMENTAL DETAILS AND CHARACTERIZATION ................. ...............18
Sputtering ............... ... .......... ...............18......
DC Magnetron Sputtering .............. ...............19....
RF Magnetron S puttering ................. ...._ ....___ ............2
Characterization ............ ..... .._ ...............21...
X-ray Diffraction ................. ...............21.................
Auger Electron Spectroscopy ................. ...............22........... ....
X-ray Photoelectron Spectroscopy ................... ............ ............... 23 .....
Scanning Electron Microscopy and Energy Dispersive Spectroscopy ................24
Atomic Force Microscopy ................. ......... ...............26......
Transmission Electron Microscopy ................. ...............26........... ....
Focused lon Beam ............... ... ............. ...............2
Van der Pauw Measurement and Four Point Probe ................. .....................28
4 PROPERTIES OF W-Ge-N AS A DIFFUSION BARRIER MATERIAL FOR
COPPER ................. ...............36.......... .....
Introducti on ................. ...............36.................
Experimental Details .............. ...............37....
Results and Discussion .............. ...............38....
Conclusion ................ ...............41.................
5 INVESTIGATION OF W-Ge-N DEPOSITED ON Ge AS A DIFFUSION
BARRIER FOR Cu METALLIZ ATION ....__ ....____ .......... .............4
Introducti on ............. ...... ._ ...............46...
Experimental Details .............. ...............47....
Results and Discussion .............. ...............48....
Conclusion ............. ...... ...............51...
6 COMPARATIVE STUDY OF Hf~Nx AND Hf-Ge-N DIFFUSION BARRIERS
ON Ge ................. ...............59.................
Introducti on ................. ...............59.................
Experimental Details .............. ...............60....
Results and Discussion .............. ...............62....
Conclusion ................ ...............66.................
7 EFFECT OF Ge OVERLAYER ON THE DIFFUSION BARRIER PROPERTIES
OF Hf~Nx .............. ...............75....
Introducti on ................. ...............75.................
Experimental Detail s .............. ...............76....
Results and Discussion .............. ...............78....
Conclusion ................. ...............80.......... .....
8 PROPERTIES OF Ta-Ge-N AS A DIFFUSION BARRIER FOR Cu ON Si ...........87
Introducti on ................. ...............87.____.......
Experimental Details .............. ...............88....
Results and Discussion .............. ...............89....
Conclusion ............. ...... __ ...............92....
9 CONCLUSIONS .............. ...............98....
LIST OF REFERENCES ............. ...... ._ ...............102...
BIOGRAPHICAL SKETCH ............. ......___ ...............110...
3-1 Sputtering yields of different ions ........................... ........._ ...... 3
LIST OF FIGURES
1-1 RC delay vs. technology nodes .............. ...............6.....
1-2 Void and Hillock formation in Al interconnects ................. .......... ................6
3-1 Schematic diagram of DC sputtering system with parallel plate discharge .............3 1
3-2 Schematic diagram of RF sputtering system with a capacitive, parallel plate
di schar ge ................. ...............3.. 1..............
3-3 Schematic diagram of X-ray diffraction set-up ................. .......... ................3 2
3-4 Schematic set-up principle of an atomic force microscope............... ...............3
3-5 Interaction output between a high-energy electron beam and thin specimen ..........33
3-6 Sample geometries for Van der pauw measurements .............. ....................3
3-7 Schematic diagram of Van der pauw configuration for measurement of RA and
RB TOSistances, respectively ................. ......... ...............34......
3-8 Schematic diagram of four point probe measurement ................. ............. .......3 5
4-1 X-ray diffraction patterns of as-deposited and annealed films at different
temperature of(a) WNx films (b) W-Ge-N fi1ms ........................... ...............42
4-2 AES depth profiles of Cu/WNx/SiO2/Si (a) as-deposited (b) annealed at 600 OC/1
hr (c) annealed at 800 OC/1 hr .............. ...............43....
4-3 AES depth profiles of Cu/W-Ge-N/SiO2/Si (a) as-deposited (b) annealed at 600
oC/1 hr (c) annealed at 800 OC/1 hr............... ...............44...
4-4 Resistivity vs. annealing temperature (a) for W-Ge-N and WNx films. Also
shown (b) is the resistivity as a function of sputter target power for the Ge and
W targets .............. ...............45....
5-1 X-ray diffraction patterns of as-deposited and annealed films at different
temperature of(a) WNx films (b) W-Ge-N films ........................... ...............53
5-2 AES depth profies of Cu/WNx/Ge (a) as-deposited (b) annealed at 400 OC/1 hr
(c) annealed at 500 OC/1 hr............... ...............54...
5-3 AES depth profies of Cu/W-Ge-N/Ge (a) as-deposited (b) annealed at 400 OC/1
hr (c) annealed at 500 OC/1 hr .............. ...............55....
5-4 ED S depth profie of Cu/WNx/Ge annealed at 500 OC/1 hr ................ ................. 56
5-5 ED S depth profie of Cu/W-Ge-N/Ge annealed at 500 OC/1 hr .............. .............57
5-6 X-TEM images of (a) Cu/WNx/Ge annealed at 500 OC/1 hr (b) Cu/W-Ge-N/Ge
annealed at 500 OC/1 hr ........._ ...... .___ ...............58...
6-1 X-ray diffraction patterns of Cu/Hf~Nx/Ge fi1ms in as-deposited and annealed
conditions for varying thickness of (a) 300 nm, (b) 50 nm, and (c) 15 nm .............68
6-2 X-ray diffraction patterns of Cu/Hf-Ge-N/Ge fi1ms in as-deposited and annealed
conditions for varying thickness of (a) 300 nm, (b) 50 nm, and (c) 15 nm .............69
6-3 FE-SEM images of 50 nm films annealed at 500 OC for 1 hr: (a) Cu/Hf~Nx/Ge
and (b) Cu/Hf-Ge-N/Ge .............. ...............70....
6-4 FE-SEM images of 50 nm films annealed at 600 OC for 1 hr: (a) Cu/Hf~Nx/Ge
and (b) Cu/Hf-Ge-N/Ge .............. ...............70....
6-5 X-TEM images of 50 nm films annealed at 600 OC for 1 hr: (a) Cu/HfNx/Ge and
(b) Cu/Hf-Ge-N/Ge .............. ...............71....
6-6 EDS depth profile of 50 nm thick Cu/Hf~Nx/Ge annealed at 600 OC for 1 hr...........72
6-7 EDS depth profile of 50 nm thick Cu/Hf-Ge-N/Ge annealed at 600 OC for 1 hr.....73
6-8 XPS chemical state data of Cu 2p peak at various sputtering times for 50 nm
thick film of Cu/Hf-Ge-N/Ge (a) as deposited and (b) annealed at 6000C for 1 hr.74
7-1 X-ray diffraction patterns of as-deposited films and annealed ones at the
temperature shown in the figure ....__ ......_____ .......___ ...........8
7-2 EDS depth profile of Cu/Ge(25nm)/Hf'Nx(7nm)/Si for a) as-deposited: b)
annealed at 500 OC for 1 hr: and c) annealed at 500 OC for 3 h............... ................84
7-3 AFM images of Cu/Ge(25nm)/Hf'Nx(7nm)/Si for a) as-deposited: b) annealed at
500 oC for 1 hr: and c) annealed at 500 OC for 3 hr .............. .....................8
7-4 HRTEM images of Cu/Ge(25nm)/Hf'Nx(7nm)/Si for a) as-deposited: b) annealed
at 500 oC for 1 hr: and c) annealed at 500 OC for 3 hr .............. .....................8
8-1 X-ray diffraction patterns of as-deposited and annealed at the temperature shown
in figure of a) Cu/TaN/Si (001) for 1 hr: b) Cu/Ta-Ge-N/Si (001) for 1 hr .............93
8-2 Field-emission SEM images after annealing at 400 oC for 1 hr for a) Cu/TaN/Si
(001) and b) Cu/Ta-Ge-N/Si (001)............... ...............94.
8-3 AES depth profile of Cu/TaN/Si (001) for a) as-deposited and b) annealed at 400
oC for 1 hr............... ...............95...
8-4 AES depth profile of Cu/Ta-Ge-N/Si (001) for a) as-deposited and b) annealed
at 400 oC for 1 hr............... ...............96...
8-5 Sheet resistance of Cu vs. annealing temperature for TaN and Ta-Ge-N ................97
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
ALTERNATIVE NITRIDE DIFFUSION BARRIERS ON SILICON AND
GERMANIUM FOR COPPER METALLIZATION
Chair: David P. Norton
Major Department: Materials Science and Engineering
As device dimensions shrink, there is an urgent need to replace conventionally used
Al interconnects to achieve increased current density requirements and better
performance at future technology node. Copper is a viable candidate due to its lower
resistivity and higher resistance to electromigration. However, Cu has its own problems
in its integration. It diffuses rapidly through SiO2, Si and Ge into the active regions of the
device, thereby deteriorating device performance. There is therefore a need to find a
diffusion barrier for Cu.
Amorphous ternary nitrides are investigated as a candidate diffusion barrier for Cu
metallization on single crystal Si and Ge substrates. Ge was chosen as a third element in
the binary matrix of WNx, HfNx and TaN due to its chemical similarity to Si and possible
integration in SiGe or Ge based devices. The addition of Ge helps in amorphization of the
binary matrix. It hinders grain boundary formation and increases the recrystallization
temperature compared to their respective binary nitrides. Cu was deposited in-situ after
nitride deposition. In the case of W-Ge-N on Si, the recrystallization temperature was
raised by 400 oC, while for Ta-Ge-N it was raised by 100 oC indicating better diffusion
barrier properties than their corresponding binaries.
A bilayer approach for diffusion barrier was also studied. Ge/Hf~Nx bilayer was
deposited on Si followed by in-situ deposition of Cu. Copper reacts with Ge and forms
Cu3Ge which has low resistivity, is less reactive with oxygen, and is a diffusion barrier
for Cu. By combining the material properties of Cu3Ge and HfNx, an excellent diffusion
barrier performance was demonstrated under stringent test conditions.
The integrated circuit industry doubles the number of transistors in a chip every
18 months. With more transistors built into a chip, minimum feature size decreases,
leading to faster devices. At present, we are moving towards the 45 nm node technology
and beyond which is guided by International Technology Roadmap for Semiconductors
(2005)2. For many years, gate length determined the device speed and was the bottleneck
for future generation devices. However, with decreasing dimensions, delay in
interconnects now plays a maj or role in influencing device speed. The interconnect delay,
also called resi stance-capacitance (RC) time constant delay, is mostly caused by a
increase in resistance at smaller dimensions. The RC delay is expressed by the following
RC = 2rDo
where p is resistivity, is interconnection length, t, is interconnect thickness, SED and
tlL are permittivity and thickness of interlevel dielectric (ILD). Also, the current density
increases as width of the interconnect decreases. Figure 1-1 shows the increase in RC
delay time in interconnects with decreasing feature size which dominates overall delay at
sub micron technology nodes.l
Aluminum has been the interconnect material of choice for many generations with
SiO2 (K = 3.9) as the dielectric material. This results in a tremendously high RC delay as
dimensions decrease. One way to reduce the RC delay is to decrease the dielectric
constant of SiO2. Fluorinated silicon dioxide (FSG) was used with a dielectric constant of
3.7 for the 180 nm technology node.2 Not until recently was low-k material (x = 2.7 -
3.0) being integrated in devices. At high current densities, electromigration performance
of Al is severely degraded leading to voids and hillock formation. Creation of voids leads
to discontinuous circuit causing failure of devices. Figure 1-2 shows void and hillock
formation in Al interconnect. Alloying Al with Cu showed an increase in
electromigration resistance, but it reached its usability limits as subsequent milestones in
IC industry were achieved.
With increased demands on performance, an alternative interconnect material is
eventually desired. The logical choice is a transition to Cu interconnects. There are
several advantages in using Cu as an interconnect material. Bulk resistivity of Cu is 1.67
CLD-cm, which is almost 40% less than Al (p=2.65 CLD-cm).3 Resistance to
electromigration property of Cu is highly superior to Al, which leads to better device
reliability. Replacing the Al/SiO2 gate stack with Cu/low-K reduces the RC delay
However, Cu has its own share of problems. It diffuses rapidly through Si, Ge and
SiO2 and forms parasitic silicides (Cu3Si) and hinders the device performance. It forms
deep and shallow level traps in Si and Ge, respectively. Cu reacts with dopants and forms
Cu-D complexes (D is dopant atom) deteriorating device performance. Cu exhibits poor
adhesion on SiO2 and other low-K dielectrics. In addition to the above difficulties,
solutions also need to be found for anisotropic etching, poor corrosion and oxidation
properties of Cu. In order to get the desired benefits by switching from Al to Cu
interconnect, a viable diffusion barrier is required to prevent Cu diffusion.
The primary purpose of a diffusion barrier is to prevent intermixing of chemical
species with each other. This can be achieved via various mechanisms thereby classifying
the diffusion barriers based on the method. i.e., passive, sacrificial, stuffed and
amorphous diffusion barriers. A passive diffusion barrier is an ideal barrier which does
not react with any of the layers it separates. A sacrificial barrier would react with either
or both layers that it separates and get consumed. For a sacrificial barrier, the reaction
rate between the barrier and the layers is very important. It should be slow enough so that
the barrier can perform to acceptable levels for the useful lifetime of the device. For
polycrystalline or nanocrystalline thin films, rapid diffusion occurs via grain boundaries,
dislocations and surface defects.4 Diffusion is highly influenced by temperature. The
temperature dependence of diffusion coefficient (D) is given by the following equation:
whereDo is temperature-independent pre-exponential factor, Qd is the activation energy,
k is the Boltzmann constant and Tis the temperature.
The activation energy for grain boundary diffusion is approximately half the
activation energy required for lattice diffusion. As a result, grain boundary diffusion
dominates at lower temperatures.4 For polycrystalline thin films, diffusion through grain
boundaries and dislocations is the fastest. This mode of diffusion can be stopped using
two approaches. The grain boundaries and/or dislocations can be "stuffed" by impurity
atoms which would hinder diffusion through them. Another approach is to lower the
amount or eliminate the grain boundaries altogether. This can be achieved by low
temperature or room temperature deposition of nanocrystalline or a amorphous diffusion
barrier where the short-circuit pathways are lowered or eliminated. This is highly
desirable because of low/no addition to the thermal budget. This may not be the best
approach as films deposited at lower temperatures can have undesired modified
properties during subsequent high temperature processing steps.
Apart from the primary function of preventing diffusion of Cu, the diffusion
barrier should meet other stringent specifications. Some of them are:
The diffusion barrier should be thermodynamically stable with Cu and
underlying substrate under standard operating conditions. It should not
react with Cu or the substrate under thermal, mechanical or electrical
stresses encountered during other processing steps.
The density of diffusion barrier should be close to its bulk density in order
to avoid any defects, voids or dislocations which can compromise its
The diffusion barrier should have reasonable thermal and electrical
conductivity to avoid any parasitic capacitance effects and unwanted
The contact resistance of diffusion barrier with Cu and substrate should be
Diffusion barrier should behave well under the applied mechanical and
electrical stresses during subsequent processing.
Microstructure of the diffusion barrier should ideally be amorphous at
room temperature and remain amorphous after thermal treatment at higher
temperatures. In general, diffusion barriers with higher melting
temperatures have high recrystallization temperatures.
The barrier should have good conformality especially over high aspect
Diffusion barrier should adhere well with Cu and the surrounding
Deposition of diffusion barrier should be compatible with existing
processing facilities and infrastructures.
Chapter 2 will present background and a literature review of various diffusion
barriers, method of deposition and barrier properties. In Chapter 3, experimental methods
used to deposit diffusion barrier films for this study will be explained. Various
quantitative characterization techniques used to evaluate the diffusion barrier
performance under extreme thermal treatment will also be described. Recrystallization
properties and diffusion barrier performance of W-Ge-N deposited on Si/Sio2 aef
evaluated in Chapter 4. The behavior and performance of W-Ge-N diffusion barrier
deposited on Ge are investigated in Chapter 5. A comparison of barrier properties and
performance of HfNx and Hf-Ge-N deposited on Ge is described in Chapter 6. Chapter 7
will describe the excellent diffusion barrier properties of a Ge/Hf~Nx bi-layer diffusion
barrier deposited on Si. Properties of Ta-Ge-N diffusion barrier on Si will be evaluated in
Chapter 8 followed by conclusion in Chapter 9.
40~ nIM Cu 5m de~lly. Al nd 5502
r: r Umo de~lra Cu arid km ki2 Ij
35 O= Intezrronne t dly. alm 5102l~
SInterconned delay,.oc and lowr Ll.9.0
~~: ;F Undlow k
650 500 320 250 180j 130 100
Figure 1-1. RC delay vs. technology nodes.
Figure 1-2. Void and Hillock formation in Al interconnects.
Due to the unique diffusion property of Cu in Si, Ge, SiO2 and other low-K
dielectrics, there is an imminent need for a diffusion barrier for Cu. There has been a
plethora of research on different diffusion barriers. In general, metals or compounds with
high melting temperatures are suitable for the purpose because of less chance of having
grain boundaries, which are the fastest diffusion pathways. Therefore, refractory based
materials seem to be the best choice due to their high melting temperature and decent
electrical conductivity. A considerable amount of research has been conducted to find a
viable diffusion barrier for Cu based on refractory metals. Most of the research can be
classified in three categories.
Refractory metals as diffusion barriers
Binary diffusion barriers based on refractory metals
Ternary diffusion barriers based on refractory metals
This chapter will cover a comprehensive review of the above classifications.
Refractory Metals as Diffusion Barrier
As it became evident that switching from Al based interconnects to Cu based
interconnects was necessary, initial research focused on refractory metals as candidate
diffusion barriers. Primary interest was in Ti, Ta, Cr, and W as Cu diffusion barriers.
Ti Diffusion Barrier
Ti based diffusion barriers were extensively used for Al interconnects. A natural
choice was to extend its functionality in acting as a Cu diffusion barrier. However, a Ti
diffusion barrier fails at 350 oC by Ti-Cu compound formation.' A study conducted by
Ohta et al.6 also revealed failure of Ti diffusion barrier at 400 oC indicated by an increase
in resistivity after annealing. The resistivity increase was attributed to Cu diffusion
through Ti and subsequent reaction with Si to form Cu silicides.
Ta Diffusion Barrier
The two maj or advantages of using Ta as a diffusion barrier in comparison to Ti
are its very high melting temperature (3017 oC compared to 1668 oC for Ti) and
thermodynamical (interface and solubility) stability with Cu for very high temperatures.
The Cu-Ta phase diagram shows that both Cu and Ta are insoluble in each other even at
high temperature. Failure of Ta diffusion barrier occurs due to Cu diffusion through grain
boundaries formed in Ta after high temperature annealing followed by Cu3Si formation at
Ta/Si interface.' Ta also reacts with Si and forms silicides thus rendering itself unusable
as a Cu diffusion barrier.
Cr Diffusion Barrier
Cr was studied as a diffusion barrier for Cu because of its good corrosion
resistance properties and excellent adhesion promoter.8 However, Cu/Cr/Si multilayer
structure showed a huge increase in resistivity after annealing at temperatures higher than
400 oC. This was attributed to Cu-Cr binary phase formation. The formation of such a
phase compromises the integrity of the Cu/Cr/Si multilayer structure resulting in Cu
silicide formation indicating diffusion barrier failure.6
W Diffusion Barrier
W is a better diffusion barrier compared to Ti and Cr. It is chemically and
thermodynamically stable with Cu.9 Initial study showed that Cu/W/Si structures behaved
well even after annealing at 600 oC. However, it failed after annealing at 700 oC.6 CU
intermixes with W even at low temperatures of 260 oC, thus rendering it non-viable as a
solution to Cu diffusion.10 Preferentially oriented W(110) failed after annealing at 690 oC
for 1 hr due to consumption of W by a uniform silicidation reaction." Selective chemical
vapor deposited W layer on p -n junction diodes led to failure after annealing at
temperatures above 650 oC.12
Binary Diffusion Barrier
In order to overcome the drawbacks of lower recrystallization temperatures and
formation of grain boundaries in refractory metals used as diffusion barriers, their
respective binaries were explored. This class of diffusion barriers can generally be
classified in three categories.
These individual classifications will be explored further in the following sections.
Several of the refractory intermetallics were studied as Cu diffusion barriers. TiW
was used extensively in Al-based interconnects and so was also tried as a diffusion
barrier for Cu interconnects.13 The study showed that TiW (30 at. % Ti) not exposed to
air before deposition of Cu failed at 775 oC when RTA annealed in N2 gaS for 30 seconds.
The authors also summarized results of a variety of Cu/diffusion barrier/Si systems.
Table-1 in reference 13 shows a list of diffusion barrier investigated between Si and Cu.13
Amorphous Ni60N T40 diffusion barrier failed at 600 oC after annealing for 60 min. A
different composition of Ni57r\b42 failed at a lower temperature of 500 oC after 60 min.14
The authors also investigated the feasibility of Ni60Mo40 aS a candidate diffusion barrier
for copper. It failed at 500 oC after annealing for 60 min. Amorphous Ir-Ta alloy was
investigated for Cu diffusion barrier application." Sandwich structure of Si/a-Ir45Tass/Cu/
a-Ir45Tass failed after annealing at temperatures higher than 700 oC. Failure occurred at
750 oC by interdiffusion of Cu and Si as detected by Rutherford backscattering
spectrometry. It was also ob served that the recrystallization temperature of a-Ir45Tass was
lowered from 900 oC to 750 oC due to the presence of Cu. Alloyed Ta-Col6 was shown to
act as a Cu diffusion barrier deposited by e-beam evaporation. However, an intermetallic
phase of Co2Ta and metal-silicide phase of Co2Si formed after annealing at 500 oC.
Recently, a study conducted by J. S. Fang et al.17 explored the possibility of using
sputtered Ta-TM (TM = Fe, Co) as diffusion barriers for Cu. Tao~sFeo~s diffusion barriers
failed at 650 oC while Tao~sCoo~s failed at 700 oC after annealing, indicating better
performance of Tao~sCoo~s.
Interest in refractory carbides to serve the purpose of copper diffusion barriers
was due to two reasons. First, they have a high melting temperature, and second, they
have low resistivity indicating higher thermal and chemical stability. Among the
refractory carbides studied, primary interest has been in Ti, Ta and W based carbides. A
detailed study of TiGx as copper diffusion barriers was carried out by S.J. Wang et al.l
However, the Cu/TiCx/Si structure was unstable after annealing at temperatures higher
than 600 oC for 30 min. A more sensitive electrical characterization done by measuring
leakage current across Cu/TiCx/p+ n-Si diode structure revealed an early failure just after
annealing at 500 oC for 30 min. As many as 60% of the diodes measured for leakage
current after annealing at 530 oC had the leakage current in the range of 10-5 A/cm2,
which is three orders of magnitude higher than the leakage current measured at room
temperature. Approximately 80% of the diodes had a leakage current in the range of 10-3
A/cm2 after annealing at 550 oC, indicating a failure temperature somewhere around 500
TaC has a melting temperature of 3985 oC19 Suggesting good thermal stability,21
and room temperature resistivity of 27 CLD-cm,20 thus making it a potential candidate as a
Cu diffusion barrier. In this study, Junji Imahori et al.20 COmpared films with different
carbon concentrations, namely Ta53 47, Ta40 60, and Ta20 80. Of these, Ta53 47 f11ms
performed better than others. However, it too failed after annealing at temperatures
higher than 600 oC for 30 minutes. Films with 60 and 80% C concentrations failed
primarily due to diffusion of Cu through the amorphous C phase and lesser diffusion of
Cu occurred through grain boundaries of TaC. However, the prime reason for failure of
Ta53 47 WAS due to the Cu diffusion through the grain boundaries with Cu activation
energy of 0.9 eV. A study of the interfacial reactions in the Cu/TaC/Si before and after
annealing was conducted to determine the failure mechanism of the barrier layer.22 The
failure temperature was a function of thickness as 7, 3 5 and 70 nm TaC films failed at
600 oC, 650 oC and 750 oC respectively. The thicker TaC film integrity was compromised
at a lower temperature of 600 oC due to the formation of amorphous Ta(O,C)x layer at the
WCx was studied as a candidate diffusion barrier for Cu due to its high melting
temperature of around 2785 oC23 and low electrical resistivity.24 A hexagonal closed pack
W2C phase was achieved by chemical vapor deposition at a growth temperature of 600oC.
This growth temperature is fairly high keeping in mind future technology requirements
where incorporation of low-K materials will put an even more stringent temperature
constraint.25 Low temperature CVD process to deposit W2C WAS demonstrated by Y. M.
Sun et.al.26 TEM studies revealed 5-6 nm W2C CryStallites in an amorphous matrix with a
W/C ratio of 2:1 until 450 oC. The Cu/W2C/SiO2/Si stack remained intact after 400 oC
annealing for 8-9 hours. Unfortunately, higher temperature annealing was not done in the
study which would have revealed capacity of the diffusion barrier to withstand higher
thermal stress. Room temperature sputter deposited WCx was tested as a candidate
diffusion barrier for Cu on Si.24 Fifty nanometer WCx failed according to analytical
results after annealing at temperatures higher than 650 oC for 30 min. Electrical
measurements, however, lowered the failure temperature to 550 oC. At 700 oC, the
metallurgical stability of the WCx/Si interface is compromised. W reacts with Si to form
W5Si3 at the WCx/Si interface. Cu diffuses through the barrier layer and possibly through
defects in the barrier layer into the Si substrate to form Cu3Si phase. In as-deposited
condition, 60% of the measured diodes fail with a leakage current of 10-9 A/cm2. Also,
60% of the diodes fail after annealing at a temperature of 550 oC with a leakage current
of 10-' A/cm2.
The activation energy for Cu to diffuse through grain boundaries is low making
this the fastest diffusion pathway and killer of devices. A method to stop Cu diffusion is
to block these super highways by "stuffing" them. Oxygen was used as a stuffing agent in
TiN diffusion barriers used for Al metallization. Al reacts with 02 and forms an
aluminum oxide phase that hinders Cu from further diffusion. However, the same concept
could not be applied to Cu.3 Nitrogen is a good stuffing agent that can solve the purpose.
Excess nitrogen in the film moves out to the grain boundaries and stuffs them. Cu
diffusing through the grain boundaries experiences a repulsive force from the nitrogen
thus stopping it from diffusing through the barrier film.27 Refractory nitrides are
interesting candidates for Cu diffusion barrier application due to their lower resistivity,28
high melting temperatures, lower heat of formation indicating better stability and the
ability to block Cu diffusion by stuffing the grain boundaries by nitrogen. Some of the
binary refractory nitrides investigated for Cu diffusion barrier applications are Ti, Ta, W
and Hf based which will be discussed further in detail.
TiN was extensively used as a diffusion barrier for Al based metallization.
Considerable effort was made to utilize the existing knowledge and processing methods
for Cu based metallization as well. However, TiN deposited by both PVD and CVD
methods resulted in a columnar grain structure in which grain boundaries run through the
entire thickness of the barrier film.29-32 These columnar grain boundaries provide easy
pathways for Cu diffusion and subsequent reaction with Si to form silicides. The
properties of TiN films greatly depend on the deposition conditions which affect the
microstructure, density and other relevant barrier properties of the film. The resistivity
values range from 20 2000 CLD-cm and density ranges from 3.2 5.0 g/cm3 33 A
comparative study of different deposition methods and conditions to deposit TiN films
was done by Park et al.33 Porous films have lower densities making them susceptible for
impurities like oxygen which alter the diffusion barrier properties. The TiN films failed in
the temperature range of 500 750 oC depending on the film properties, 500 oC being for
CVD deposited films having lowest density and 750 oC being for sputter deposited films
having highest density. Atomic layer deposition techniques were also tried for TiN
barrier deposition with focus on the influence of microstructure, resistivity and impurity
content on film barrier properties.34 However, the barrier film failed after annealing at
500 oC for 1 hr. The maj or drawbacks of CVD based techniques are unwanted C and O
impurities, particulate generation, high deposition temperatures, and uncontrolled
thickness variations across the surface.
WNx is also an attractive candidate for Cu diffusion barrier applications due to its
low resistivity and the ability to be deposited as an amorphous phase. Due to the absence
of grain boundaries, Cu diffusion is hindered or slowed. However, the phase of WNx
mainly governs the barrier properties. A tungsten rich WNx (x < 0.5) phase tends to
dissociate at temperatures as low as 450 oC into W and W2N leading to barrier failure by
diffusion of Cu through the grain boundaries due to lower recrystallization
temperatures.35 A nitrogen rich WNx phase (x = > 1) tends to be amorphous and remain
amorphous for higher temperatures, but the gain is traded off with higher resistivity.
Uekubo et al demonstrated the feasibility of W2N as a diffusion barrier, as compared to
W and WN, and showed that 8 nm W2N was able to stop Cu diffusion in Si until 600 oC
for 30 min.36 Failure of the diffusion barrier was due to recrystallization and grain
boundary formation. An array of deposition methods have been tried to deposit WNx like
sputtering,37 inOrganic CVD,38 plasma-enhanced CVD,39 metal-organic CVD,40 and
TaN is being currently used for Cu diffusion barrier applications due to its high
melting temperature and thermal stability. Among the various phases of Ta-N,
stoichiometric TaN has a melting temperature of 3087 oC and heat of formation
(A~Hf= -120 kJ/mol)43 making it more stable than Ta2N which has a melting temperature
of 2050 oC and heat of formation of AHf = -98 kJ/mol. Chemical vapor deposition of
tantalum nitride results in an insulating tetragonal Ta3N5 phase which is not suitable for
diffusion barrier applications.44,45 A bi-layer structure of Ta/TaN is currently being used
to overcome the obstacle of adhesion problem of TaN. TaN does not adhere well to Cu46
but it does adhere with SiO2.47 In case of Ta, it adheres well to TaN but not with SiO2 and
is not an effective diffusion barrier by itself.47 Also, TaN helps nucleate Ta in the a-Ta
phase, which has a BCC structure and a resistivity of ~15-30 CLD-cm, as compared to
when Ta is directly deposited on Sio2 where it nucleates in the P-Ta phase, with a
resistivity of ~150-220 CpD-cm.47 However, with device dimensions shrinking,
conformality of high aspect ratio is very critical. PVD48 has been used until now for
deposition of liner and Cu seed layer, with ionized PVD (I-PVD)49 being the latest in the
technology that has been able to extend the functionality to lower dimensions. However,
with future technology constraints, the liner thickness should be less than 10 nm.5o Since
the overall liner-Cu seed layer thickness requirement decreases, either some maj or
modifications has to be done to the I-PVD process or a switch to a better conformal
process like atomic layer deposition (ALD) has to be adopted.
Ternary Diffusion Barriers
The activation energy required for Cu diffusion through the grain boundaries is
low. Binary diffusion barriers have been investigated in which one of the elements, like
nitrogen, "stuffs" the grain boundaries and disallows Cu diffusion by blocking. However,
at higher temperatures the binary compound recrystallizes, forming parasitic grain
boundaries that should be avoided. Another way to solve the problem of creating a
diffusion barrier for Cu is to form an amorphous matrix that acts as a diffusion barrier
and remains amorphous (not recrystallized) when annealed at higher temperatures. This
can be achieved by adding a third element into the binary matrix. This addition frustrates
the binary lattice structure and delays or avoids the recrystallization process when it is
annealed at higher temperatures. Considerable research has been done on amorphous
ternary nitride diffusion barriers for Cu metallization. Most of the research work has been
done by incorporating Si, B or C as the third element in the binary matrix of refractory
nitrides. Some of the work will be discussed below.
TiSixN, were initially studied as Ti based liners and used as diffusion barriers for
Al metallization. Amorphous TiSixN, is attractive due to the absence of grain boundaries.
The properties of diffusion barriers depend on the method of deposition and its chemical
composition. Chemical vapor deposited TiSixN, films (25 nm thick) were studied as
candidate diffusion barriers for Cu.51 The barrier failed after annealing at temperatures
higher than 700 oC for 30 min. Unfortunately, the barrier resistivity is too high (800 CLa-
cm) for applications. MOCVD deposited TiSiN films were also investigated as diffusion
barriers for Cu.52 The deposition steps involved a H2 N2 plaSma treatment to increase the
density of the film. Comparison of films that were plasma treated and not plasma treated
was done. Films were characterized by secondary ion mass spectroscopy (SIMS) and
Secco etch method for diffusion barrier failure. Secco etch revealed that failure occurred
in films that were not plasma treated after annealing at 550 oC for 1 hr. For plasma treated
films, failure occurred after annealing at 600 oC for 1 hr. SIMS detected Cu diffusion in
the sample that was not plasma treated at 450 oC, whereas Secco etch revealed etch pits
in the same sample after annealing at 550 oC.
TaSixN, (x=1.4, y=2.5) films with different thickness (t= 5-40 nm) were sputter
deposited on Si by Lin et al53 to evaluate their diffusion barrier characteristics. Electrical
measurements on a pn diode structure revealed failure of 5 nm TaSixN, after annealing
at 500 oC for 30 min. The TaSixN, structure remained amorphous even after annealing at
800 oC, which is a good indicator that addition of Si to the TaN matrix was effective in
keeping the lattice structure amorphous. Failure of the diffusion barrier was due to Cu
diffusion through localized defects into Si. The nitrogen content plays an important role
in determining the diffusion barrier characteristics. With the increase in the nitrogen
content, the diffusion barrier properties are enhanced as the nitrogen helps in preventing
formation of TaSi2 phase after annealing at higher temperatures in Ta-Si-N films.54
There has been some research on WSixN, diffusion barriers for Cu metallization.
LPCVD deposited WSixN, shows promise as a diffusion barrier compared to Ta and Ti
counterparts.55,56 Efforts are also made to take advantage of good conformal coverage
obtained using the CVD deposition process while achieving superior diffusion barrier
properties similar to those achieved by PVD methods.56 Researchers have worked on
using boron or carbon as a third element in the binary matrix." W-B-N seems to be a
better diffusion barrier compared to W-Si-N due to its lower resistivity while having the
same performance as W-Si-N.5
EXPERIMENTAL DETAILS AND CHARACTERIZATION
Sputtering is a physical vapor deposition technique where incident ions remove
atoms (sputter) from the target surface by momentum transfer process. Sputtering was
discovered in the 19th century by Grove and Pulker and reported by Wright in 1877. The
number of atoms removed from the target surface by a single incident ion is called the
sputtering yield. The sputtering yield depends on the mass of the atoms (target) and ions,
bombardment energy of the incident ions, angle of incidence and binding energy of the
atoms in the target material. Table 3-1 shows the sputter yields of different ions when
bombarded on the target materials as indicated. Different classifications are used to
define a particular sputtering technique depending on the type of sputtering configuration
and also on the type of a reactive species inside the deposition chamber. As a result, there
are different sputtering techniques like DC sputtering, RF sputtering, triode sputtering,
magnetron sputtering and "unbalanced" magnetron sputtering. Depending on the absence
or presence of a reactive species, it can be called non-reactive or reactive sputtering,
respectively. The energy transfer taking place due to collision between two hard spheres
can be given as:
E, 4M~,MCos 6
E, (M, +M,)2
whereE,, E, are the energy of the target and incident particle respectively, M~,,M~, are
the mass of the target and incident particle and B is the angle of incidence with respect to
the surface normal of the target.
Currently, almost any target material can be deposited by sputtering due to the
latest advancement in technology. Co-sputtering can be carried out to deposit a
compound on the substrate by sputtering two targets simultaneously or by sputtering a
single target in a reactive atmosphere or by using a compound target in an inert
atmosphere. Some of the advantages of the sputtering deposition process are given below.
Ability to coat large areas with uniform thickness
Low-temperature deposition process
Flexibility of target materials choice including insulators and semiconductors
Target composition is replicated in the composition of deposited film.
Good adhesive property obtained in the deposited films
Sputtering can be up, down or sideways
Reactive sputtering possible
Environmentally friendly process technology
Can be easily scale up for commercial production
The two maj or kind of sputtering configurations conventionally used are DC and RF
magnetron sputtering. They are described in detail below.
DC Magnetron Sputtering
In a DC sputtering process, the target is the cathode and either the substrate or the
chamber walls are the anode. A very high negative DC voltage (several kV) is applied to
the cathode. Ar gas is supplied in the chamber to create plasma. Ionized Ar' ions are
accelerated to the cathode, bombarding the surface and sputtering atoms from the cathode
surface. They also generate secondary electrons that ionize the gas atoms creating
increased amounts of Ar+ ions. This increases the probability of collisions at the cathode
surface. A circular magnet (magnetron) below the target traps the electrons emitted from
the cathode surface to remain near the cathode. These electrons hop in a cycloid fashion
on the target surface due to (E x B) forces acting on it. As the electrons are trapped near
the surface region, the probability of creating more Ar+ ions increases which increases the
sputtering yield. Another advantage of using the magnetron source is that low gas
pressures can be used inside the chamber to maintain a stable plasma. Since the gas
pressure is low, internal gas collisions decrease leading to higher yields of sputtering
target material. As there is less chance of internal gas collisions, the sputtered materials
impact the anode surface with a high kinetic energy. Figure 3-1 shows a schematic DC
sputtering chamber set-up. The disadvantage of DC sputtering is that the cathode material
needs to be conductive. Insulating materials cause a charge build-up near the surface
quenching the plasma.
RF Magnetron Sputtering
The maj or difference between DC and RF magnetron sputtering is the source. In
radio frequency (RF) sputtering, an RF source is used, typically 13.56 IVHz. A matching
network is used to optimize power transfer from the RF source to the discharge and a
blocking capacitor is used in the circuit to create a DC bias. With the use of an RF source,
sputtering of insulating targets has been made possible. Figure 3-2 shows a schematic set-
up of a RF sputtering. The sputtering process is similar to DC sputtering. Low gas
pressures can be used for RF sputtering. One consideration during RF sputtering is to use
target materials with high thermal conductivity. Low thermally conductive materials
develop a high enough thermal gradient due to the sputtering process to cause brittle
fracture of the target.
Characterization is an important part of research. It helps in identifying physical,
electrical, magnetic, optical and chemical properties of the samples that helps in
understanding the fundamental behavior of the material being studied. Some of the
characterization techniques used in this study are discussed below.
X-ray diffraction (XRD) is one of the most versatile non-destructive technique
used todate to identify the crystalline phases and crystallinity of the sample. Advances in
X-ray diffraction and understanding the basic physics between the interaction of X-ray's
and the sample have given access to a plethora of information. Currently, information
regarding strain, in-plane epitaxy, defect structure, film thickness etc. can easily be
obtained from the constructive and destructive interference of X-rays with the sample.
Figure 3-3 shows a basic X-ray diffraction set up. Cu Ku radiation is generated
and impinged on the sample. This radiation undergoes constructive or destructive
interference after reflecting from the sample depending on the path difference between
the incident and reflected X-rays. Constructive interference will cause a peak at a
particular 26 angle according to Bragg's rule as given below:
nAl = 2dSine
where Ai is the wavelength of X-ray (Cu Ku radiation 1.54 A+), d is the distance between
two consecutive (hkl) planes and O is the angle between the incident X-ray and the
sample surface. In a cubic system the d-spacing is given as:
hkih +k +1~
where ao is the lattice constant.
For a single crystal, there are only specific orientations that satisfy the Bragg's
law. Diffraction peaks from these planes appear in the diffraction pattern. However, for a
polycrystalline film having differently oriented grains, diffraction peaks appear when
those grains meet the diffraction conditions.
The full width at half maximum of an X-ray diffraction peak gives information
about the grain size. The relationship is called the Scherrer equation:
where t is the grain size, ii is the wavelength of x-ray, B is the full width at half maximum
and O is the Bragg angle.
Auger Electron Spectroscopy
Auger electron spectroscopy is an excellent technique for surface and sub-surface
analysis. It is a highly sensitive technique as it probes only from few angstroms to few
nanometers in depth from the surface. Usually, an electron beam is used to probe and
excite electrons in the atoms of the sample. As the core shell electron leaves the atom, it
is in an excited state. A higher shell electron fills the core shell vacancy and the energy
emitted by this transition is transferred in exciting another electron to emit from the atom
which is called the Auger electron. Depending on the kinetic energy of the electron
measured, the binding energy of the emitted Auger electron is calculated by the following
where K.E is the kinetic energy of the Auger electron, and Ex- EL and EL are the energies
of the K and L shells, respectively.
A total of three electrons are needed in order to fulfill an Auger transition. As a
result, elements like H and He are undetectable with AES. Suppose an electron from the
K-shell of an element is removed by photons or electrons. The electron from the L shell
fills up this vacancy with enough energy left to emit an electron from the L shell. This is
called a KLL transition. Similarly, now higher shells would fill up the L shell and would
result in a LMM type of transition. The Auger equipment can be combined with an ion
gun, which would sputter the sample and expose a new surface to be examined. A group
of all those data points would give the depth profie characteristics of the sample studied.
This is therefore a very good method to study interfaces and diffusion profies of
elements inside the sample. The following information can be obtained from an Auger
Type of elements
Amount percentage of the elements
Chemical state of elements
Valence band density of states
X-ray Photoelectron Spectroscopy
As the name suggests, a high energy photon source (X-ray) is bombarded on the
surface of the sample. It ej ects a core shell electron in the atom transferring it to an
excited state. The atom returns back to ground state as the higher shell electron fills up
the core shell vacancy, emitting the excess energy in the form of photon or by emitting an
Auger electron. The kinetic energy of the emitted core shell electron is detected which
gives a plethora of information about the sample. The expression of kinetic energy
measured is as below:
K.E = hv B.E
X-ray photoelectron spectroscopy is a very sensitive technique for surface
analysis as it probes only few angstroms from the surface. This is because only few
electrons near the surface are able to escape the sample while most of the electron excited
by the x-ray lose their energy by collisions with atoms and are not able to exit the surface.
The flux of electrons able to exit the surface of the sample without being scattered (Id) is
given by the following expression:
where Io is the original flux of electrons generated at depth d, 3Ae is the inelastic mean free
path of electrons, 9 is the angle of electron emission.
As the binding energy of an electron is a characteristic of an atom, elemental
information can by obtained by XPS. Also, a general trend is that if the charge on an
atom increases, so does the binding energy of the electron. As a result, the valence state
of the atom can be known. Amount percentage of the element present can be determined
by peak intensity. The valence electrons participate in the bonding process and thus
compound information can also be known from XPS data analysis. A depth profile is
possible. However, it is very time consuming.
Scanning Electron Microscopy and Energy Dispersive Spectroscopy
Scanning electron microscopy is used for surface and cross-section imaging,
topographical information and compositional information of a binary alloy system. An
electron beam strikes the sample surface and various kinds of elastic or inelastic
interactions occur resulting in emission of electrons. The electrons emitted out of the
sample are of generally two types
When the incident primary beam of electrons impinge on the sample surface, they
can undergo inelastic scattering by colliding with electrons of the sample, transferring
some of its energy to them which exit the surface of the sample depending on the amount
of energy transferred. If sufficient energy is transferred, then electrons from the sample
exit the surface and are collected by the detector. Electrons with energy of less than 50
eV are classified as secondary electrons. The primary beam can also undergo elastic
scattering with the nucleus of the atom where transfer of energy is null or very small.
These electrons have high energy and can exit the surface easily. The amount of back-
scattered electrons depends on the atomic number of the atom. A higher atomic number
(Z) yields a higher amount of back-scattered electrons which results in increased
brightness of the image. However, for a constant Z, the back scattering yield remains
unchanged if the primary beam energy is above 5 keV. The secondary electron yield does
not depend much on the Z. The backscattering and secondary electron images can be
used to complement each other in seeking information.
Since the primary beam has high energy, it can remove core electrons from the
atom thus exciting it. The atom goes back into the ground state by filling the core
vacancy by an electron from a higher orbital. The excess energy can either be emitted in
the form of a photon or an Auger electron. The emitted photon can be detected and used
for further study depending on its energy. This method is called as energy dispersive
spectroscopy (EDS). Since the emitted photon is characteristic of the atom from which it
emits, elemental information of the sample can be known. An elemental map of the
surface can be created with use of today's instruments. Quantifieation of the amount of a
particular element can be carried out depending on the peak intensity and using ZAF
correctional factors, where Z is atomic number, A is absorption and Z is secondary x-ray
Atomic Force Microscopy
The topography and roughness of thin films can be determined using atomic force
microscope. The probe in an AFM is a sharp tip created at the end of a cantilever beam
having a spring constant of 0. 1-1.0 N/m. As the tip is brought near the surface of the
sample, van der Waals forces become active between the tip and sample surface atoms
resulting in the deflection of the tip. This deflection is monitored and measured by a laser
striking the back of the cantilever beam which plots the surface of the sample as the tip is
moved across the sample. A general schematic of the AFM set up is show in Figure 3-4.
The tip can be rastered over the sample surface and a 3-D plot can be obtained with
nanometer spatial resolution. Lower spatial resolution can be obtained if the scanned area
is reduced and slow scans are used. AFM is operated in two modes namely contact mode,
where the tip is in contact with the sample surface, and tapping mode, where the tip is
vibrated over the sample surface. The tapping mode is particularly useful for analyzing
soft samples like polymers.
Transmission Electron Microscopy
The wave nature of electron was first theorized by Louis de Broglie in 1925. The
term "electron microscope" was first used by Knoll and Ruska in 1932 when they build
the first electron microscope and obtained images. TEM is an excellent tool to obtain
high resolution lattice images and diffraction patterns of electron transparent samples.
The high spatial resolution is due to the fact that high energy electrons are used to probe
the sample which have extremely small wavelength (on the order of A+). The wavelength
of the electron depends on the energy as per the expression:
where h is wavelength of electrons in nm, E is energy of electrons in eV.
A typical TEM operates at 200 to 400 keV. A highly coherent beam of
monochromatic electrons is focused on the sample using a series of electromagnetic
lenses. Some of the possible interactions between sample and the electrons are shown in
Figure 3-5. For transmission electron microscopy, information is collected below the
sample from the deflected and un-deflected electrons on a fluorescent screen to form
images or diffraction pattern depending on where it' s focused below the sample. Unlike
scanning electron microscope, the entire sample is in focus all the time as along as its
electron transparent. Some of the drawbacks of TEM are as follows
Limited sampling volume
Complex data interpretation
Electron diffraction is a very important quantitative characterizing technique to
analyze a sample. Information from a particular area of a sample can be obtained by
using selected area aperture (SAD) which focuses only on the area of interests. Other
information, like defects (line and point defects), burger's vectors, 3-D loops, crystal
orientation, orientation relationship between film and substrate or multilayer films and
sample crystallinity can be obtained.
Focused lon Beam
The samples required for TEM in this study were prepared by focused ion beam
(FIB) technique. Focused ion beam uses an electron beam for imaging and a gallium ion
source for imaging as well as milling. The reason for using Ga+ ions for milling is
because the mass of Ga is approx. 127000 times that of electron and so provides a huge
momentum transfer. The Ga source is heated to a liquid on a tip and Ga' ions are
extracted and focused on the sample by applying a bias. The ion source is also used for
milling purposes to prepare an electron transparent sample. Platinum is used to avoid
milling the area of interest. Once the sample is electron transparent it is set free and
transferred using glass rods to a grid for TEM analysis. Some of the advantages of using
FIB as compared to conventional TEM sample preparations are
Selected feature can be prepared for analysis
Excellent control over the cross-section to be prepared
Insulating samples can also be used
No mechanical damage in the area of interest
Van der Pauw Measurement and Four Point Probe
Van der Pauw technique is used to measure the resistivity of the sample in the
semiconductor industry due to its ease and simplicity. The technique doesn't depend on
the shape of sample. Four ohmic contacts are prepared on the four corners or periphery of
a sample. An acceptable version of sample geometry for this measurement is shown in
Figure 3-6. As shown in the Figure3-7, a DC current is applied between contacts 1 and 2
(I12) and voltage is measured along contacts 3 and 4 (V43). This resistance RA is measured
This is followed by another measurement by applying DC current between
contacts 1 and 4 (114) and voltage is measured between contacts 2 and 3 (V23). This gives
resistance RB aS given by the following equation:
The sheet resistance Rs is related to RA and RB through the following equation:
expR + expr =1
The bulk resistivity can be calculated as per the expression:
p = Rsd
where d is thickness of the film
The sheet resistance of copper before and after annealing was measured by four
point probe. A schematic set-up of four point probe is as shown in Figure 3-8. The DC
current (I) is supplied through the extreme probes and corresponding voltage (V) is
measure by the inside probes. The sheet resistivity p of a thin film of thickness (d) is
calculated by using the following equation:
In 2 Il
And the sheet resistance (Rs) is given by
where k is the geometric factor and its value is 4.53 for semi-infinite thin film.
Table 3-1. Sputtering yields of different ions.
Al (27) Si (28)
by 500 eV ions
Cu (64) Ag (106)
He+ (4 amu)
Ne+ (20 amu)
Ar+ (40 amu)
Kr+ (84 amu)
Xe+ (131 amu)
Figure 3-1. Schematic diagram of DC sputtering system with parallel plate discharge.
I Ivacurum Chambrer
Figure 3-2. Schematic diagram of RF sputtering system with a capacitive, parallel plate
Figure 3-3. Schematic diagram of X-ray diffraction set-up.
Sample Surface Canileer T
Figure 3-4. Schematic set-up principle of an atomic force microscope.
Figure 3-5. Interaction output between a high-energy electron beam and thin specimen.
Square1 or arcrtanlgle:
contracts at thze edges
or inside! the
rec ta ngle:
4 the earne~rs
Nort Reco~mmended ~
Figure 3-6. Sample geometries for Van der Pauw measurements.
RA = V43 / In2
Rr,= V, /I, 3,
Figure 3-7. Schematic diagram of Van der Pauw configuration for measurement of RA
and RB TOSistances, respectively.
-I+ ~I + I
Figure 3-8. Schematic diagram of four point probe measurement.
PROPERTIES OF W-Ge-N AS A DIFFUSION BARRIER MATERIAL FOR COPPER
For present and future integrated electronic technologies, the use of barrier
materials to enable materials integration is becoming increasingly important. In current Si
technology, the push for higher circuit density and low RC time delays has made copper
the material of choice for interconnects due to its higher resistance to electromigration
and lower resistivity as compared to Al. Unfortunately, Cu is known to show poor
adhesion to most dielectric materials and rapidly diffuses into Sio2 and Si. This
obviously degrades the electrical properties of devices ", creating the need for
intermediate layers that provide a barrier to Cu diffusion.
When considering the role of microstructure in diffusion processes, amorphous
materials are generally better suited than polycrystalline phases, as grain boundaries
provide high diffusivity pathways for Cu diffusion through the barrier material. Among
the material systems currently being developed, binary nitrides, such as TaN59,60 and
WNx61,62 are receiving significant attention. For Ta-based barriers, the Ta-Cu phase
diagram indicates that Ta and Cu are effectively immiscible even at their melting
temperature. Unfortunately, recrystallization of TaN films occurs at approximately 600
oC, which is relatively low for diffusion barrier applications. WNx is also an interesting
candidate as it is relatively easy to synthesize as an amorphous film. In this case, the
nitrogen content in WNx films has a significant influence on its diffusion barrier
properties. For WNx films with low nitrogen content, the recrystallization temperature
can be on the order of 450 oC.3 Higher nitrogen content yields higher recrystallization
temperature. WT\x films grown by physical vapor deposition have been reported to
exhibit a recrystallization temperature as high as 600 oC. Yet, the primary mode of failure
remains diffusion through grain boundaries that form during heat treatments.
One approach to achieve higher recrystallization temperature is to consider
ternary compositions. The additional element added to the refractory metal-nitride
composition frustrates the recrystallization behavior, rendering a stable amorphous
mixture at high temperatures and thus minimizing grain boundary diffusion. While failure
of ternary diffusion barriers still occurs through grain boundaries formed due to
decomposition and recrystallization of the film, this process generally takes place at
higher processing temperature. For this reason, there is significant interest in ternary
nitride alloys, such as Ta-Si-N,63-65 W-Si-N,56,66 W-B-N,67 and Ta-W-N68 due to their
high recrystallization temperature as compared to the binaries. In this study, we report on
the diffusion barrier properties of W-Ge-N thin films for Cu metallization. The W-Ge-N
alloy is chemically similar to W-Si-N, should be more resistant to recrystallization than
WNx, and may prove attractive for integration with SiGe or Ge devices.
The W-Ge-N films were deposited at the rate of 8.64 nm/min on thermally grown
SiO2 (630 A+)/n-type (100) Si substrates by reactive sputter deposition. For comparison,
WNx films were also deposited under similar conditions. The substrates were sequentially
cleaned with trichloroethylene, acetone, and methanol for 5 min each in an ultrasonic
bath. The substrates were then loaded into the multi-target R.F. sputter deposition system
via a load-lock. The base pressure of the sputtering chamber was 7 x 10-6 Torr. Typical
forward sputtering power for the W and Ge targets was 200 W and 100 W, respectively.
Nitrogen was incorporated into the films by leaking a mixture of Ar and N2 at the ratio of
1 : 0.9 into the chamber at a fixed chamber pressure of 1 1.5 mTorr. The thickness of the
films was measured using a stylus profilometer. In the experiments reported here, film
thickness was maintained in the range 300 to 360 nm. All targets were pre-sputtered
before deposition to remove any contaminant present on the target surface.
To assess the compatibility and diffusion properties of W-Ge-N with respect to
Cu metallization, the nitride layer deposition was followed by in situ deposition of a Cu
film 90 nm thick. During Cu deposition, Ar gas was used as the sputter deposition gas at
a fixed chamber pressure of 15 mTorr. After deposition, the individual samples were
annealed in a separate vacuum chamber with a base pressure of 4 x 105 Torr at 400, 600
and 800 oC for 1 hr to study and compare the diffusion barrier properties of the films. The
crystallinity of films and formation of any intermetallic compounds by annealing were
characterized by X-ray diffraction (XRD) measurements. Resistivity was measured by
the Van der Pauw method, while Auger electron spectroscopy was used to characterize
the Cu diffusion profile in the nitride films. Energy dispersive spectrometry (EDS) was
used to determine the composition of the films.
Results and Discussion
Initial studies focused on the crystallinity of the W-Ge-N films, both as-deposited
and after high temperature annealing. Figure 4-1 shows XRD spectra of WNx and W-Ge-
N films, both as-deposited and annealed at 400, 600, and 800 oC. The data show that
films of both compositions are amorphous in the as-deposited condition. For the WNx
films [Fig.4-1(a)], recrystallization is clearly evident in the XRD pattern for annealing
temperature of 400 oC and higher. The onset of grain structure will provide undesired
diffusion paths via the grain boundaries. In contrast, the W-Ge-N film shows no evidence
of recrystallization upon annealing at 400 or 600 oC. The XRD data [Fig.4-1(b)] for W-
Ge-N films annealed at 400 and 600 oC show no peaks related to the nitride material.
Only at 800 oC do polycrystalline peaks appear. The addition of Ge to the W-N solid
presumably frustrates crystallization, thus rendering the films amorphous for more severe
annealing conditions relative to WNx. It should also be noted that the W-Ge-N film was
far less susceptible to oxidation via ambient atmosphere exposure as compared to WVNx.
This may prove advantageous in terms of device processing.
To assess the behavior of these films as diffusion barriers to Cu, the chemical
profile of the annealed structures was determined by Auger electron spectroscopy. Figure
4-2 shows the depth profiles for the WNx film (a) as-deposited and upon annealing at (b)
600 oC, and (c) 800 oC. For the as-deposited diffusion barrier layer in Figure 4-2(a), there
is a well-defined interface between Cu and WNx and the SiO2 buffer layer is evident. In
contrast, after annealing at 600 oC, the Cu signal is seen throughout the barrier layer and
into the SiO2/Si. This is in agreement with the XRD data, which show the formation of
grain structure upon annealing at a temperature of 600 oC. Annealing at 800 oC [Fig. 4-
2(c)] yields a Cu diffusion profile similar to that for the 600 oC anneal. There is also some
apparent intermixing of W and Sio2 Seen at the WNx/SiO2 interface in the Auger depth
profiles of the annealed samples, which is perhaps related to a change in surface
roughness. The intensity of the nitrogen profile decreases slightly as we increase the
annealing temperature to 800 oC. This may reflect the decomposition of WNx and
liberation of N2 fOm the films under those conditions. This is not unexpected as Affolter
et at. 69 have shown that nitrogen is liberated from W-N alloy thin films when annealed at
temperatures above 700 oC.
The chemical composition of Cu/W-Ge-N/SiO2/Si structures was also examined
with Auger electron spectrometry. Figure 4-3 shows the chemical depth profiles in (a) as-
deposited W-Ge-N and after annealing at (b) 600 oC and (c) 800 oC. In Figure 3(a),
distinct interfaces at the Cu/W-Ge-N and W-Ge-N/SiO2 boundaries show that there is no
Cu diffusion during growth. At an annealing temperature of 600 oC, the interfaces and
layers remain distinct, consistent with no or minimal diffusion of Cu through the barrier
layer. These results suggest that W-Ge-N films possess superior diffusion barrier
properties as compared to WNx. At an annealing temperature of 800 oC, Cu is observed in
the diffusion barrier as seen for the WNx film. Again, Cu diffusion correlates with the
appearance of grain structure in the film (i.e., (1 11) reflection of P-W2N in the XRD
pattern). This is also consistent with N loss suggested by the AES sputter profile, and
thus its ability to stuff the diffusion pathways.
The resistivity of the W-Ge-N films was measured using the Van der Pauw
method. In general, the resistance of the as-deposited W-Ge-N films is higher than WNx.
As shown in Figure 4-4(a), the resistivity of WNx and W-Ge-N decreases as the
annealing temperature increases. While the resistivity of WNx decreases upon annealing
at 400 oC, the resistivity for W-Ge-N remains relatively unchanged after annealing at 400
oC. There is a progressive decrease in resistivity for both materials with further increase
in annealing temperature. The change in resistivity correlates with onset of grain
structure, suggesting electron transport across grain boundaries. The decrease may also
be due to the loss of nitrogen from the films. The resistivity of WNx is two orders of
magnitude lower than W-Ge-N at 800 oC reflecting the robustness of W-Ge-N film, for
which increased decomposition temperature slows nitrogen evolution. Figure 4-4(b)
shows the resistivity of W-Ge-N films vs. sputtering power to the Ge target. Clearly, the
resistivity scales with Ge content in the film.
In conclusion, the diffusion barrier properties of W-Ge-N thin films have been
investigated. X-ray diffraction shows recrystallization of WNx films at an annealing
temperature of 400 oC and higher, while W-Ge-N films show recrystallization peaks only
at an annealing temperature of 800 oC, suggesting that the addition of Ge frustrates the
recrystallization behavior of WNx. The AES data show complete Cu diffusion across the
WNx layer for an annealing temperature of 600 oC, while for W-Ge-N films the Cu/W-
Ge-N and W-Ge-N/SiO2 interfaces remain distinct at that temperature, indicating that W-
Ge-N has better diffusion barrier properties. The resistivity of both films decreases with
increasing anneal temperature, with the resistivity of WNx two orders of magnitude lower
than W-Ge-N after annealing at 800 oC. This behavior is consistent with enhanced
stability of W-Ge-N with respect to film decomposition and subsequent nitrogen
P-W,N (ll )
SCu (111) 80
40 45 50
5'5 60 65
Figure 4-1. X-ray diffraction patterns of as-deposited and annealed films at different
temperature of (a) WNx films (b) W-Ge-N films.
5 10 15
20 25 30 35
O 1 1'5 20 25 30 35 40 45
Figure 4-2. AES depth profiles of Cu/WNx/SiO2/Si (a) as-deposited (b) annealed at 600
oC/1 hr (c) annealed at 800 OC/1 hr.
5 10 15 20 25 30 35
O 5 10 15 20 25 30 35 40 45 50
0 5 10 15 20 25 30 35
Figure 4-3. AES depth profiles of Cu/W-Ge-N/SiO2/Si (a) as-deposited (b) annealed at
600 oC/1 hr (c) annealed at 800 OC/1 hr.
(a) 10 -
I I I I I I I I
0.50 0.75 1.00 1.25 1.50 1.75
H target RF: power I \'ars)
Figure 4-4. Resistivity vs. annealing temperature (a) for W-Ge-N and WNx films. Also
shown (b) is the resistivity as a function of sputter target power for the Ge and
a a -5- W-Ge-N\
0 100 200 300 400 500 600 700 800 900
Anniealing temp. (oC)
INVESTIGATION OF W-Ge-N DEPOSITED ON Ge AS A DIFFUSION BARRIER
FOR Cu METALLIZATION
The progression to ever-decreasing semiconductor device dimensions brings with
it new challenges for material integration. As SiGe-based microelectronic technology
moves toward a 45 nm technology node, there is an imminent need to replace Al
interconnects. Cu is an attractive candidate because of its low resistivity and high
resistance to electromigration as compared to Al.70,71 Replacing Al/SiO2 interconnect
technology with Cu/low-k dielectrics can yield a large reduction in RC time constant
delay, prOViding for a significant motivation to carry out the replacement search for Al.
Unfortunately, Cu has known problems with adhesion to low-k materials. It also has a
high diffusion rate in silicon and germanium creating deep level traps in Si and a shallow
level in Ge located at 0.04 eV near the valence band.72-74
Diffusion barriers are needed to achieve integration of Cu with Si-Ge and Ge. In
diffusion barrier materials the diffusion pathways are interstices, vacancies and grain
boundaries. Diffusion through grain boundaries is fastest and more prominent. By
"stuffing" the grain boundaries with selected dopants, diffusion can often be hindered. In
general, amorphous materials are highly preferred due to the absence of grain boundaries.
Common diffusion barrier materials studied for Cu metallization are refractory metal-
nitrides, which include TaN, TiN, HfN and WNx.42,59,75-79 However, these have limited
utility as viable diffusion barriers because of their relatively low recrystallization
temperatures. Recrystallization of amorphous diffusion barriers can often be inhibited by
the addition of a third element to the matrix. Previous results show higher
recrystallization temperature for W-Ge-N as compared to WNx as the introduction of Ge
effectively frustrates the matrix.so
SiGe devices have higher mobility as compared to Si devices and the flexibility of
band-gap engineering, thus being applicable in high speed electronics. The properties of
the metal/SixGel-x contact layer is important for semiconductor device applications.
Considerable research has been done on the chemical reactivity of metals, such as Co,
Ti,72,82 Pt,83 Pd,84 and CusS on SiGe. In most cases, it results in the formation of metal-Si,
metal-Ge, or metal (SixGel-x) alloys in the temperature range of 400-600 oC. In particular,
direct deposition of Cu on Sil-xGex results in the formation of unstable Cu3(Sil-xGex) in
the temperature range of 250-400 oC.8 One solution could be to grow a metal-rich Cu3Si
or Cu3Ge phase directly on Sil-xGex. Unfortunately the Cu3Si phase reacts with oxygen
when exposed to air. The Cu3Ge phase is more stable and less reactive with oxygen.86 In
general, the need to identify viable barriers of Cu integration with Ge-based structures
persists. In this work, we report on the diffusion barrier properties of W-Ge-N thin films
deposited on Ge for Cu metallization. The properties are compared to WNx deposited on
Ge under similar conditions to evaluate its suitability and properties. Due to the ternary
nature of this diffusion barrier, W-Ge-N is expected to have better recrystallization
properties as compared to WVNx.
W-Ge-N films were deposited on p-type Ge (001) by reactive sputtering. The Ge
substrates were cleaned by a standard procedure reported elsewhere" to remove any
organic residue or impurity on the surface. The substrates were loaded in the reactive
sputter chamber with a base pressure of 3 x 10-6 Torr via a load-lock. Nitrogen was
incorporated in the film by flowing a mixture of N2 and Ar at a ratio of 1:3 at a fixed
pressure of 10 mTorr. Prior to deposition, all targets were cleaned by pre-sputtering by
flowing Ar+N2 in the chamber. Forward sputtering power for the W and Ge targets were
200 W and 100 W respectively. The deposition rate for W-Ge-N film on Ge was 10.2
nm/min under the above mentioned conditions. Thickness was measured by a stylus
profilometer and was kept in the range of 50 to 300 nm.
Cu metallization was carried out in-situ after depositing W-Ge-N thin films to
determine suitability as a diffusion barrier. Sputter deposition of Cu was carried out by
flowing Ar gas at a fixed chamber pressure of 5 mTorr. After depositing the film stack,
individual samples were separately annealed in the range of 400 oC 700 oC in a tube
furnace. Ar gas was flowed through the tube furnace at 65 sccm for at least 10 hours
before starting the annealing to remove any residual air mixture. Typical annealing
experiment was carried out for 1 hr to study the diffusion barrier properties of the film.
X-ray diffraction was used to identify any intermetallic phases and access the crystallinity
of the films after annealing. Cu diffusion profile through the film was determined by
Auger electron spectroscopy (AES) and Energy dispersive spectrometry (EDS). Interface
properties were determined by cross-section transmission electron microscopy (X-TEM).
Results and Discussion
Crystallinity and high temperature phase formation of the Cu/W-Ge-N/Ge film
structure before and after annealing was determined by XRD. The XRD spectra for
Cu/WNx/Ge and Cu/W-Ge-N/Cu film structure is shown in Figure 5-1, both as-deposited
and annealed in the temperature range of 400 700 oC. The WNx diffusion barrier shows
little crystallization in as-deposited condition as evident by (111) peak, whereas W-Ge-N
fi1m is amorphous. The (111) peak intensity increases at 500 oC indicating further
crystallization as evident from Figure 5-1(a). This leads to increase in the formation of
undesired grain boundaries that provide fast diffusion paths. Cu diffuses through the
grain boundaries and reacts with the underlying Ge substrate resulting in the formation of
a Cu3Ge phase that is clearly evident from Figure. 5-1(a). In comparison, there is no
recrystallization of W-Ge-N fi1ms at 500 and 600 oC. However, at 600 oC, Cu reacts with
Ge in the W-Ge-N layer, and subsequently with Ge substrate to form the Cu3Ge phase
which is evident from Figure 5-1(b). This results in depletion of Ge from the W-Ge-N
fi1m. Upon further high temperature annealing, recrystallization of the Ge-depleted W-
Ge-N takes place leading to barrier failure. This result indicates that adding Ge to W-N
alloy effectively hampers the recrystallization process even at high annealing
temperatures as compared to WVNx.
The Cu diffusion profile through the diffusion barrier in as-deposited and
annealed structures was determined by Auger electron spectroscopy. The fi1m thickness
for WNx and W-Ge-N was 50 nm for the AES chemical profile study. Figure 5-2 shows
the chemical profile of Cu for Cu/WNx/Ge structure (a) as-deposited and after annealing
at (b) 400 oC and (c) 500 oC. The Auger profile shows negligible Cu diffusion in WNx at
400 oC. However, upon annealing to 500 oC, Cu is seen to rapidly diffuse through the
barrier film and to the substrate. This is in agreement with XRD data at 500 oC where it
shows increased crystallization of WNx, thus increasing the grain boundaries, i.e.,
diffusion pathways, resulting in the formation of Cu3Ge phase. The nitrogen signal
intensity steadily decreases at subsequent high temperature annealing compared to the as-
deposited sample indicating decomposition of WNx by liberating N2 at higher
The chemical profile of Cu/W-Ge-N/Ge structures was also measured by Auger
electron spectroscopy as shown in Figure 5-3(a) as-deposited and after annealing at (b)
400 oC and (c) 500 oC. As is evident, there is little or no Cu diffusion through the barrier
upon annealing at temperatures less than 500 oC. The Cu/W-Ge-N and W-Ge-N/Ge
interfaces are distinct in as-deposited and 400 oC annealed samples. At 500 oC, Cu starts
consuming the Ge in the film. This is also supported by XRD data [Fig. 5-1(b)], where a
small intensity (-111) Cu3Ge peak appears for annealing temperature of 500 oC. Upon
annealing at higher temperature, Cu completely consumes the Ge in the film, thereby
depleting the W-Ge-N matrix. This results in the recrystallization of WNx as is evident
from Figure 5-1(b) at 700 oC resulting in rapid Cu diffusion. The result above suggests
that W-Ge-N hinders recrystallization by frustrating recrystallization of the matrix and is
a better diffusion barrier as compared to WNx. The high oxygen content noticed in the
chemical profiles is due to background oxygen during sputtering. The negative enthalpies
of formation of W-O and Ge-O bond are 672 kJ/mol and 659.4 kJ/mol respectively
indicating the stability of the compound after formation. Longer pre-sputtering time of
target could help in reducing the oxygen content in the deposited film.
Figure 5-4 and Figure 5-5 shows the Cu diffusion profile in WNx and W-Ge-N
film annealed at 500 oC, respectively, which was measured by Energy dispersive
spectroscopy (EDS) attached to a cross-section transmission electron microscope
(XTEM) system. As seen in Figure 5-4, Cu signal is seen throughout the WNx film and
into the Ge substrate. However, very little (half counts as compared to WNx) Cu signal is
seen coming from the W-Ge-N (Fig. 5-5) diffusion barrier. This result corroborates the
above mentioned XRD and AES data suggesting excellent diffusion barrier properties for
W-Ge-N as compared to WNx. The interface properties were determined by XTEM.
Figure 5-6 shows XTEM images of (a) Cu/WNx/Ge structure and (b) Cu/W-Ge-N/Ge
structure both annealed at 500 oC. The WNx/Ge and W-Ge-N/Ge interfaces are abrupt,
indicating no intermixing or reactions between them even after annealing. However, the
Cu/WNx and Cu/W-Ge-N interface is rough, suggesting some intermixing and Cu
diffusion in WNx films. XTEM images shows well-defined (111) grain structure for Cu
films deposited on W-Ge-N [Fig. 6(b)] as compared to WNx films [Fig. 6(a)]. This may
be important as (111) oriented Cu films has high resistance to electromigration.
In conclusion, the diffusion barrier properties of W-Ge-N thin films deposited on
p-Ge (001) substrates were investigated. W-Ge-N films have a higher recrystallization
temperature (700 oC) as compared to WNx films as shown by X-ray diffraction. The
failure of W-Ge-N diffusion barrier films at high temperatures occurs by Cu diffusing
through the grain boundaries formed by recrystallization of Ge depleted W-Ge-N. This
Ge depletion is caused by the consumption of Ge in the barrier film by Cu at high
temperatures, thereby forming the Cu3Ge phase. Removal of Ge from the film makes the
W-Ge-N barrier more likely to recrystallize. In contrast, failure of WNx diffusion barrier
at high temperature takes place by diffusion of Cu through WNx grain boundaries.
Nitrogen evolution from the film at high temperature causes decomposition of the
diffusion barrier, thus enhancing crystallization of WNx film. The AES data clearly
shows complete Cu diffusion throughout the WNx layer at 500 oC annealing temperature,
whereas there is little or negligible Cu diffusion through W-Ge-N films. This suggests
that W-Ge-N is a better diffusion barrier. This is substantiated by the Cu profile measured
by EDS. The Cu films deposited on W-Ge-N have better orientation as compared to WVNx
after annealing at high temperatures, thereby increasing its resistance to electromigration.
35 40 45 50 55 60
B-W N (111)
Cu Ge (020)
35 40 45 50 55 60
Figure 5-1. X-ray diffraction patterns of as-deposited and annealed films at different
temperature of (a) WNx films (b) W-Ge-N films.
2 4 6 8 10 12
400 "C annealed
O 3 6 9) 12
15 18 21
500 oC annealed
0 10 20 30
Figure 5-2. AES depth profiles of Cu/WNx/Ge (a) as-deposited (b) annealed at 400 OC/1
hr (c) annealed at 500 OC/1 hr.
(a) 100 --,------- Ge
0 5 10 15 20 25 30 35
400 of annealed
0 5 10 15 20 25 30 35 40
500 of annealed
0 5 10 IS 20 25 30
Figure 5-3. AES depth profiles of Cu/W-Ge-N/Ge (a) as-deposited (b) annealed at 400
oC/1 hr (c) annealed at 500 OC/1 hr.
SGe x 5
-A Cu x 5
0 50 100 150 200
Figure 5-4. EDS depth profile of Cu/WNx/Ge annealed at 500 OC/1 hr.
O 20 40 60 80 1. 120 140 160 180
Figure 5-5. EDS depth profile of Cu/W-Ge-N/Ge annealed at 500 OC/1 hr.
Figure 5-6. X-TEM images of (a) Cu/WNx/Ge annealed at 500 OC/1 hr (b) Cu/W-Ge-
N/Ge annealed at 500 OC/1 hr.
COMPARATIVE STUDY OF Hf~Nx AND Hf-Ge-N DIFFUSION BARRIERS ON Ge
For many years, aluminum has been the primary interconnect metal for Si-based
integrated circuits. However, with device dimensions shrinking to sub-45 nm and
demands for high current density increasing, the conductivity and electromigration
properties of Al become limitations to performance. In response, Cu is beginning to
replace conventional Al interconnects given its better electromigration resistance and
lower electrical resistance.70,71,s? The use of low resistivity Cu compared to Al
significantly reduces the circuit time constant delay to make the circuit faster. As the
need for high speed electronics grows, there is also a renewed interest in Ge and SiGe
based devices because of inherent advantages of Ge over Si, i.e., smaller E,, higher
mobility of charge carriers and lower dopant activation energy.88-90 Sil-xGex -based
devices are also of interest because of the innate flexibility to tailor the bandgap through
the alloy composition.91-93 These factors provide sufficient impetus to investigate barrier
layer materials needed in incorporating Cu interconnects in Sil-xGex and Ge-based
For interconnect applications, copper cannot be deposited directly on Si-Ge since
it diffuses rapidly in Si and Ge creating deep level traps.73,94 For the case of Si, it forms
copper silicides at saturation. It also passivates dopants by forming Cu-D (D is dopant
atom) covalent pairs thus altering the intended doping levels.95 COpper is also known to
diffuse rapidly in Ge with an average diffusion coefficient of 3 x 105 c2S- nte70t
900 oC temperature range.96 COpper introduces three acceptor levels in Ge, two at
Ev+0.04 and Ev+0.32 eV and another at Ec-0.26 eV respectively.74 Direct deposition of
Cu on Sil-xGex results in the formation of Cu3(Sil-xGex) and passivation of the dopants.97
In addition to the above issues, Cu also exhibits poor adhesion to dielectrics commonly
used in Si device structures.'
Considerable work has focused on identifying viable Cu diffusion barrier
materials on Si. Since amorphous materials lack grain boundaries that are fast diffusion
pathways, they are ideally suited for application as a diffusion barrier. Recent material
systems that have been studied as possible Cu diffusion barriers for Si include refractory
metal nitrides, such as TaN, TiN, WNx.59,61,75,97 These binary nitrides, however, tend to
recrystallize at moderate temperature, thus becoming susceptible to rapid Cu diffusion.
There is significant interest in identifying diffusion barrier materials that remain
amorphous at high processing temperature and effectively block Cu diffusion. Increasing
the temperature necessary for crystallization can often be achieved by the addition of a
third element to a binary matrix material. Some of the ternary materials systems that have
been studied include Ta-Si-N, W-Si-N, W-Ge-N.98-100 In this paper, we report on the
recrystallization of HfNx and Hf-Ge-N thin films deposited on Ge and their diffusion
barrier properties for Cu metallization.
HfNx and Hf-Ge-N thin films with varying thickness (15, 50 and 300 nm) were
deposited on p-Ge (001) single crystal substrates by reactive sputtering at room
temperature. Prior to deposition, the substrates were cleaned with trichloroethylene,
acetone and methanol in an ultra-sonic bath for 5 min each to remove any organic residue
from the surface. The substrates were introduced in a reactive sputter deposition chamber
with a base pressure of 3 x 107 Torr via a load-lock. Nitrogen was incorporated in the
film by flowing Ar and N2 in the chamber at a ratio of 3:1. The total chamber pressure
during deposition was 10 mTorr. Prior to deposition, the targets were cleaned in-situ by
pre-sputtering with Ar+N2 at a fixed chamber pressure of 15 mTorr. The forward
sputtering power for Hf and Ge was 200 and 100 W, respectively. The typical deposition
rate for HfNx and Hf-Ge-N films was 1.8 and 6.23 nm/min, respectively. Identical
thickness was achieved for both films by varying the deposition time. Film thickness was
measured by a stylus profilometer.
Nitride film deposition was followed by in-situ deposition of Cu films. The
forward power used for Cu deposition was 200 W. The Cu thickness was maintained
constant at 300 nm for all films. The deposition was carried out by flowing Ar inside the
chamber at a fixed chamber pressure of 5 mTorr. Individual film stacks were then
separately annealed in a tube furnace in the temperature range of 400 to 700 oC for 1 hr.
Before starting the annealing process, the tube was purged by flowing Ar gas at 65 sccm
for at least 10 hrs. The film crystallinity before and after annealing was determined by X-
ray diffraction (XRD) while the film surface morphology and roughness after annealing
were determined by field emission-scanning electron microscopy (FE-SEM). The
chemical depth profile of Cu diffusion through the diffusion barrier was determined by
energy dispersive spectroscopy (EDS). The chemical state analysis of Cu and
intermetallic compound formation with Ge was investigated by X-ray photoelectron
spectroscopy (XPS). Interface reactions and properties were determined by cross-section
transmission electron microscopy (X-TEM).
Results and Discussion
The HfNx films were amorphous in the as-deposited condition and showed no
signs of recrystallization for any film thickness even after high temperature annealing. A
lack of crystallization upon annealing is desirable as formation of grain boundaries leads
to rapid Cu diffusion. The HfNx diffusion barrier properties are expected to be attractive
based on its high melting temperature (3330 oC).101 Materials that have a high melting
temperature also generally show a have high recrystallization temperature since both
process involve bond breaking. HfN films have been shown to be stable to thermal
decomposition up to 1000 oC.102 Fig. 1 shows the X-ray diffraction patterns for
Cu/Hf~Nx/Ge as-deposited and after high temperature annealing in the range of 400 to 700
oC in an Ar atmosphere. For 300 and 50 nm thick Hf~Nx diffusion barrier films that were
annealed at a temperature of 600 oC or greater, the Cu films exhibit a shift in the Cu (1 11)
peak towards smaller 26 values. This may indicate a reaction with the HfNx film. It is
noted, however, that for the 300 nm thick HfNx barrier, no Cu3Ge phase was formed even
after annealing at 700 oC. For the 50 nm thick HfNx film (Fig lb), as aforementioned,
there is a definitive shift of Cu (1 11) peak at 700 oC annealing temperature which may be
due to reaction of Cu with the underlying HfNx layer. Also evident for the 50 nm thick
HfNx sample is the formation of non-stoichiometric Cu3-xGe phase after annealing at 700
oC. This indicates Cu diffusion through the HfNx diffusion barrier and reaction with the
underlying Ge substrate. Cu3-xGe phase could also have been formed by possible Ge out-
diffusion through the diffusion barrier and subsequent reaction with Cu. For the ultra-thin
HfNx diffusion barrier film (15 nm), the barrier fails at lower temperature as evident from
Fig. Ic that shows Cu3Ge phase formation after annealing at 500 oC and above. The 400
oC anneal pattern, however, does not reveal evidence of barrier failure.
The properties of Cu/Hf-Ge-N/Ge multilayers were then examined and compared
to the Cu/Hf~Nx/Ge samples. Fig. 2 shows X-ray diffraction patterns for Cu/Hf-Ge-N/Ge
as-deposited and after high temperature annealing in the range of 400 to 700 oC in Ar
atmosphere. Based on the behavior of the thickest film, the Hf-Ge-N films remained
amorphous after annealing at a temperature as high as 7000C for all film thicknesses. For
the 300 nm thick Hf-Ge-N film (Fig. 2a), there is little or no shift in the Cu (111) peak
even after annealing at 700 oC. For the 50 nm thick Hf-Ge-N film thickness (Fig 2b), Cu3-
xGe phase formation is evident after annealing at 600 oC indicating that Cu has diffused
through the barrier film to react with the underlying Ge substrate. It is also noted for the
600 oC annealed sample that the Cu (200) and (111) peaks are no longer present,
suggesting significant loss of Cu to the underlying material. At 700 oC annealing
temperature, sufficient Cu diffuses through the barrier layer to form stoichiometric Cu3Ge
phase, again indicating barrier failure.
These data suggests that, while Hf-Ge-N and HfNx have similar recrystallization
behavior, the diffusion barrier properties of HfNx are superior. In particular, the absence
of the Cu (200) peak for the film on 50 nm thick Hf-Ge-N annealed at 600 OC suggests
significant diffusion as compared to HfNx.
One possible factor in determining the properties of the two materials relates to
the relative percentages of the elements present in the diffusion barrier. As mentioned
before, HfNx and Hf-Ge-N films were deposited at the rate of 1.8 and 6.23 nm/min,
respectively. As the forwarding power to Hf was kept constant during deposition of both
films, the total Hf content in the HfNx sample is about 3.5 times greater than that for the
corresponding Hf-Ge-N film of the same thickness. This results in the deposition of a Ge-
rich Hf-Ge-N film. Cu is known to react readily with Ge. For example, at room
temperature, a 20 nm Cu3Ge reaction layer will form at a Cu/Ge interface in 24 hrs in a
binary reaction couplel03 and the reaction rate should increase with increased anneal
temperature. The atomic percentages of each element present in Hf-Ge-N film as
measured by Auger electron spectroscopy were 3 at. % nitrogen, 41.5 at. % oxygen, 28.8
at. % germanium and 26.7 at. % hafnium, respectively, in as-deposited condition.
The surface morphology of the barrier materials was revealed by field emission-
scanning electron microscopy. Fig. 3 shows a comparison of FE-SEM micrographs for
(a) Cu/Hf~Nx/Ge and (b) Cu/Hf-Ge-N/Ge annealed at 500 oC; samples that retained barrier
integrity as evidenced by the XRD patterns shown in Fig. lb and 2b. The thickness of
each HfNx and Hf-Ge-N layers was 50 nm. The grain structure observed in the
micrographs is that of the Cu. Note that there is no evidence of delamination. Fig. 4
shows the FE-SEM micrographs for 50 nm thick (a) Cu/Hf~Nx/Ge and (b) Cu/Hf-Ge-N/Ge
films annealed at 600 oC. After annealing at 600 oC, the surface morphology is
significantly different for the copper films on HfNx as compared to that on Hf-Ge-N
films. For Cu on the HfNx, the Cu films are continuous with a roughness similar to that
seen for the 500 oC anneal. For the Cu on Hf-Ge-N, however, significant Cu segregation
is observed. This is consistent with the suppression of the (002) Cu peak for this
structure and annealing temperature. Reaction with the Hf-Ge-N and possible Cu
diffusion through the Hf-Ge-N film leads to depletion of Cu from the surface and
segregation of Cu islands. This apparent Cu loss on the surface is in agreement with the
XRD data showing the appearance of Cu3Ge peaks for Cu films on Hf-Ge-N barriers that
are annealed at 600oC. No such peaks are detected for the comparison HfNx film
suggesting improved diffusion barrier quality of the latter film.
The interface properties and reactions were examined by cross-section
transmission electron microscopy. Fig 5 shows the X-TEM images of the 50 nm thick (a)
Cu/Hf~Nx/Ge and (b) Cu/Hf-Ge-N/Ge films after annealing at 600 oC for 1 hr. Cu
diffusion is clearly seen in the Hf-Ge-N film with the formation of Cu3Ge phase formed
below the diffusion barrier film. The image of the HfNx film, however, shows a
negligible amount of Cu diffusion as indicated by a continuous Cu film on the surface
and no indication of formation of the Cu3Ge phase. The discontinuous layer at the
Hf~Nx/Ge interface is due to delamination of Hf~Tx. This could be due to TEM sample
preparation prepared by focused ion beam (FIB). The chemical diffusion profile of Cu
was determined by energy dispersive spectroscopy attached to the cross-section TEM.
Figs. 6 and Fig. 7 show the chemical diffusion profile of Cu after annealing at 600 oC for
1 hr for the HfNx and Hf-Ge-N films respectively. A Cu signal is present in the Hf-Ge-N
barrier layer and Cu3Ge has clearly formed by transport through the barrier film to the Ge
substrate. In contrast, HfNx shows little Cu signal is seen from the Ge substrate indicating
that the HfNx barrier layer prevented Cu diffusion to the Ge substrate. The EDS peak
positions of Cu and Hf overlap each other. As a result, a bump in the Cu intensity profile
is observed in the HfNx layer. Also, a strong Hf EDS intensity is seen from the entire
region of the Cu layer due to the EDS detector' s inability to differentiate between Cu and
The chemical state of Cu and intermetallic phase formation after annealing was
determined by X-ray photoelectron spectroscopy (XPS). Fig. 8 compares the Cu 2p3/2
peak shifts in Cu/Hf-Ge-N/Ge fi1m for different sputtering times in the (a) as deposited
material (b) after annealing the fi1m at 600 oC for 1 hr and Ge 2p3/2 peak shifts in (c)
Cu/Hf-Ge-N/Ge for different sputtering times after annealing at 600 oC. The Cu surface
of the as-deposited film is clearly oxidized forming a CuO layer. This is evident from the
characteristics satellite peaks formed for Cu+2. After sputtering, however, the peak shifts
and matches with pure Cu (932.8 eV). As seen in Fig. 8b, after annealing the 50nm
Cu/Hf-Ge-N fi1m at 600 oC, the Cu 2p3/2 peak in the as-received condition forms at 934.8
eV indicating its reaction with Ge and formation of Cu3-xGe. This is also consistent with
the XRD and X-TEM data which show the formation of Cu3-xGe. The peak intensity
increases with sputtering time as more Cu participation in the Cu-Ge bond is revealed.
Apparently, there is slight shift in the Cu 2p3/2 peak to 933.3 eV after 60 minutes
sputtering indicating that Cu might react with oxygen and form a Cu-O compound. The
Ge 2p3/2 peak appears at 1221.9 eV after annealing at 600 oC. The Ge 2p3/2 peak intensity
increases with sputtering time indicating increased Ge participation in the Cu-Ge bond
formation. There is a shift in the Ge 2p3/2 peak position to 1221 eV after sputtering for 60
minutes. This might be due to reaction with oxygen and formation of Ge-O bond.
In conclusion, a comparative study of the diffusion barrier properties of Hf-Ge-N
and HfNx deposited on (001) Ge single crystal wafers was conducted. The FE-SEM
images show almost identical surface morphology of Cu fi1ms after annealing at 500 oC.
Annealing at 600 oC, however, results in considerable extent of diffusion across the Hf-
Ge-N films leaving a discontinuous Cu film on the surface. Furthermore, sufficient Cu
transport occurs to form Cu3Ge which is evident from XRD data. In contrast, little or no
diffusion takes place for HfNx films of the same 50 nm thickness and annealing condition
leaving Cu films continuous and smoother. This is also substantiated by cross-sectional
TEM images which clearly show the formation of a Cu3Ge below the Hf-Ge-N diffusion
barrier, but no such phase is formed in the comparison HfNx film after annealing at 600
oC. The chemical valence state was determined by XPS and the results point to Cu-Ge
bond formation after high temperature annealing. The chemical diffusion profile
measured by EDS shows Cu signal emanating from the Hf-Ge-N diffusion barrier and the
underlying Cu3Ge phase formed after annealing at 600 oC. There is little or no Cu signal ,
however, observed in the HfNx diffusion barrier and underlying Ge substrate, indicating
that the HfNx diffusion barrier was successful in preventing Cu diffusion to the substrate.
It is thus concluded that HfNx is an attractive diffusion barrier for Cu on Ge, while Hf-
Ge-N demonstrates limited utility.
I I I I I I I I I I
29 (de; g.
Figure 6-1. X-ray diffraction patterns of Cu/HfNx/Ge films in as-deposited and annealed
conditions for varying thickness of (a) 300 nm, (b) 50 nm, and (c) 15 nm.
31 5 4 4 5 5
Figure 6-2. X-ray diffraction patterns of Cu/Hf-Ge-N/Ge films in as-deposited and
annealed conditions for varying thickness of (a) 300 nm, (b) 50 nm, and (c) 15
Figure 6-3. FE-SEM images of 50 nm films annealed at 500 OC for 1 hr: (a) Cu/Hf~Nx/Ge
and (b) Cu/Hf-Ge-N/Ge.
Figure 6-4. FE-SEM images of 50 nm films annealed at 600 OC for 1 hr: (a) Cu/Hf~Nx/Ge
and (b) Cu/Hf-Ge-N/Ge.
Figure 6-5. X-TEM images of 50 nm films annealed at 600 OC for 1 hr: (a) Cu/Hf~Nx/Ge
and (b) Cu/Hf-Ge-N/Ge.
11 lf i
Figure 6-6. EDS depth profile of 50 nm thick Cu/HfNx/Ge annealed at 600 OC for 1 hr.
-#*--Ater .C -1 so.Te~rr n
Binding Ener~g NY.:
Bindine Energy (sk `
Bindine EngErs (sY:
Figure 6-8. XPS chemical state data of Cu 2p peak at various sputtering times for 50 nm
thick film of Cu/Hf-Ge-N/Ge (a) as deposited and (b) annealed at 6000C for 1
EFFECT OF Ge OVERLAYER ON THE DIFFUSION BARRIER PROPERTIES OF
Cu is gradually replacing Al as an interconnect material in integrated circuits due
to its lower resistivity and higher electromigration resistance.104 The limitations on
materials properties will become more stringent with shrinking device dimensions as we
move towards the 45 nm device node technology.'os For example, the current density and
the associated heat generation increase rapidly as more devices are packaged on a single
chip. The properties and behavior of interconnects become the limiting factor in
determining the circuit speed under these extreme circumstances. Although Cu is
preferred in metallization schemes for resistivity reasons, it rapidly diffuses in Si creating
deep level traps and reacts with dopant atoms, deteriorating device performance.106-10
There is thus a need to find a diffusion barrier material for Cu metallization. Ideally, a
single crystal diffusion barrier with no defects is highly desirable due to lack of grain
boundaries. Growth of single crystal barriers, however, is not practical due to the process
restrictions (e.g. thermal budget) and material properties (e.g. lattice parameter and
thermal expansion coefficient mismatches between Cu and the underlying substrate (Si or
low x material). An amorphous system is then preferred due to lack of grain boundaries,
which provide facile pathways for diffusion of Cu. Avoiding recrystallization and the
formation of grain boundaries is thus necessary for the diffusion barrier to function.
Refractory metal nitrides are being studied intensely as diffusion barriers for Cu because
of their high melting temperatures. Some of the refractory metal nitrides being considered
are WNx,42,109,110 TaN, 111-113 and TiN.114-117 Diffusion barriers of these materials fail at
relatively low temperature (500 to 600 oC) due to recrystallization. Ternary diffusion
barriers such as W-Si-N,55 Ti-Si-N,118-119 W-B-N,120 and W-Ge-Nso have better diffusion
barrier properties owing to higher recrystallization temperatures compared to their
respective binary systems.
Recently Hf~Nx78,121 has received significant interest as a diffusion barrier material
due to its high melting temperature (3330 oC),101 which translates to higher
recrystallization temperatures., Failure of diffusion barriers typically takes place due to
recrystallization of the barrier film giving rise to parasitic grain boundaries. The integrity
of the diffusion barrier can be enhanced in several ways, for example, by adding a third
element into the binary matrix,122 choosing a refractory nitride system having very high
melting temperature, or by combination of materials that collectively hinder Cu diffusion.
In this paper it is demonstrated that the third approach is a viable one to formulating a
diffusion barrier for Cu metallization. Specifically, a novel bilayer diffusion barrier of Ge
(25 nm)/Hf~Nx (7 nm thick) on Si is tested as a barrier for Cu. The candidate diffusion
barrier is compared with a simple HfNx layer deposited under identical conditions. The
results indicate that the integrity of the diffusion barrier was maintained even under
strenuous test conditions.
Ge/Hf~Nx diffusion barrier films were deposited on p-Si (001) single crystal wafers
by a reactive sputtering process. The native oxide on Si was first etched by dipping the
substrate in 7:1 buffered oxide etchant and rinsing with D.I. water. After drying the
substrates with ultra high purity N2 gaS, the substrates were loaded into the deposition
chamber via a load-lock. The chamber base pressure was maintained at 3 x 107 Torr. The
sputtering targets were pre-cleaned before deposition by flowing Ar gas inside the
chamber at a fixed chamber pressure of 15 mTorr. The forward sputtering power used for
Hf and Ge targets was 200 and 100 W, respectively. HfNx diffusion barrier films were
deposited by flowing Ar + N2 gaS inside the chamber at a fixed pressure of 10 mTorr.
Once the desired thickness of HfNx was achieved, the N2 flOw was stopped and a Ge over
layer was deposited via Ar ion sputtering. The thicknesses of the HfNx and Ge films
were maintained at 7 nm and 25 nm, respectively, by varying deposition time.
Cu thin films were deposited in-situ on the Ge layer at a fixed pressure of 5 mTorr
without breaking vacuum. Forward sputtering power used for Cu was 200 W. The
thickness for the Cu film was 300 nm for all samples. The substrate was rotated at 20 rpm
throughout each deposition process to ensure film uniformity. Subsequent to Cu
metallization, individual diffusion barrier samples were annealed separately in a tube
furnace in the temperature range 400 to 700 oC for 1 hr to test the integrity of the
diffusion barriers. Ultra high purity Ar gas was used to purge the tube furnace at a flow
rate of 65 sccm for at least 10 hr. Individual samples were also annealed at 500 oC for 1
to 3 hr. The film crystallinity and intermetallic phase formation were determined using a
Phillips APD 3720 X-ray diffractometer (XRD) while the Cu depth profile was
determined using a JEOL Superprobe 733 energy-dispersive spectrometer (EDS). The
integrity of the interface was assessed by examining cross-sections on a JEOL 2010F
high-resolution transmission electron microscope (HRTEM). Samples were prepared
using Dual beam strata DB23 5 focused ion beam (FIB) and the surface roughness of Cu
was determined using atomic force microscopy (AFM) on a SPM/AFM Dimension 3100.
Results and Discussion
Thick HfNx films (thickness 300 nm) without a Ge layer were first grown with an
apparent amorphous structure in the as-deposited condition since no diffraction peak was
observed that was assignable to Hf`Nx and the underlying Si (400) reflection at 69. 15 50
was evident (not shown). The HfNx (thickness 300 nm) remained amorphous even after
annealing at high temperature (400 700 oC) (not shown). Figure 1 shows the X-ray
diffraction patterns of thinner HfNx (7 nm) diffusion barrier films with (Fig. la) and
without Ge (Fig. lb). The sample with only a 7 nm thick HfNx film, however, failed after
annealing at 400 oC for 1 hr as evident from the formation of Cu3Si (Figure lb). By
providing an over layer of Ge, the efficacy of HfNx was significantly enhanced (Figure
la). The integrity of the Ge/Hf~Nx stack remains intact up to 600 oC annealing temperature
indicating a significant increase in its performance as a diffusion barrier. Only after
annealing at 700 oC are copper silicide peaks evident, indicating improved performance
of the bilayer diffusion barrier until 600 oC. For Ge/Hf~Nx films annealed at 500 oC in the
range of 1 to 3 hrs (Figure Ic), the diffusion barrier stack does not show any indication of
Cu transport across it. The integrity of the Ge/Hf~Nx stack at these severe test conditions is
This excellent performance can be attributed to a combined effect of both Ge and
HfNx. Ge was deposited as an over layer because of its rapid reactivity with Cu to form
Cu3Ge, which also has a high electrical conductivity.23, 103,123 Moreover, the Cu3Ge is less
reactive with oxygen86 than Cu3Si and itself is a good diffusion barrier for Cu.95 Using Ge
as a stand alone diffusion barrier for Cu, however, fails at 500 oC as evident by formation
of copper silicide peaks (Fig. Id). When Ge is used as an over layer, it reacts with Cu to
form Cu3Ge, which hinders Cu diffusion. After passing through the Cu3Ge layer, Cu
encounters the amorphous HfNx layer, which further limits its diffusion. This is an
excellent example of a synergistic effect of two different materials in achieving a
common goal. The Cu intensity peak ratio between Cu (111) and (200) increases as the
temperature is increased. It also increases when annealed at constant temperature but with
increasing anneal time. This apparent recrystallization of Cu to yield a preferred
orientation of Cu along the [1 11] direction is advantageous as Cu has superior resistance
to electromigration in the  direction.
Figure 2 shows the atomic depth profiles in Cu/Ge/HfNx/Si film stacks before and
after annealing at 500 oC. As clearly seen in Figure 2a for the as-deposited stack, there is
a noticeable overlap of the Cu intensity profile with the Ge over layer. This supports the
hypothesis that Cu reacts with Ge even at room temperature. The EDS peak positions of
Cu and Hf overlap each other. As a result, a bump in the Cu intensity profile is observed
in the HfNx layer. Also, a strong Hf EDS intensity is seen from the entire region of the Cu
layer due to the EDS detector' s inability to differentiate between Cu and Hf. The Cu
intensity profile declines sharply as it moves inside the Si substrate indicating no Cu
diffusion into the Si at room temperature. For samples annealed at 500 oC for 1 hr (Fig.
2b) and 3 hr (Fig. 2c), the Cu intensity profile again decreases sharply as the scan moves
into the Si substrate. This is consistent with the XRD data, which also show no sign of Cu
diffusion and formation of copper silicides. The oxygen content as measured by auger
electron spectroscopy in Hf~Nx films was approx. 20 at. %. This can be rectified by using
lower base pressure and longer target pre-sputtering times.
The surface roughness was quantitatively measured by atomic force microscopy.
Figure 3 shows AFM plots of the same Cu/Ge/Hf~Nx/Si stacks profiled in Figure 2. As
evident in these images, there is no significant change in the rms roughness values
between the as-deposited samples (5.182 nm) and samples annealed at 500 oC for 1 hr
(9.373 nm) and 3 hr (6.476 nm). The data suggest that the Cu layer is fairly smooth with
the small amount of roughness likely produced by the reaction of the Cu with Ge.
Subsequent annealing does not promote further reaction or surface atom migration.
The film interface integrity and abruptness were determined by high-resolution
TEM for the same 3 samples analyzed by EDS depth profiling and AFM. As shown in
Figure 4 the HfNx/Si film interface appears to be very abrupt for all three samples, which
suggests no intermixing between the layers. This is consistent with the XRD data, which
show no evidence of copper silicide formation for annealing at 600 oC for 1 hr and at 500
oC for 3 hr.
In summary, the effectiveness of a novel Ge(25 nm)/Hf~Nx(7 nm) bilayer to serve
as a diffusion barrier for Cu was investigated. XRD data suggests that the barrier stack
does not fail even after annealing at 600 oC for 1 hr and shows no signs of copper silicide
formation even after annealing at 500 oC for 3 hr, indicating excellent diffusion barrier
quality of the Ge/Hf~Nx bilayer structure. The as-deposited Cu/Ge/HfNx/Si (001) stack as
well as ones annealed at 500 oC for 1 and 3 hr were further examined by EDS, AFM, and
HRTEM. Elemental EDS profiles show a sharp decline in the Cu intensity profile at the
HfNx/Si interface, indicating no Cu diffusion into the Si substrate. The results of these
measurements also suggest that Cu reacts at room temperature with the Ge to form
Cu3Ge. The Cu film surface was smooth in the as-deposited condition with no significant
changes after annealing at 500 oC for 3 hr. This is beneficial as smoother films exhibit
lower contact resistance. Lattice images of the Hf~Nx/Si interface reveal atomic
abruptness, indicating excellent stability of the diffusion barrier and showing no signs of
intermixing. Annealing of a single HfNx layer barrier, however, showed Cu silicide
formation at 400 oC. Overall, the 32 nm Ge/Hf~Nx bilayer stack has excellent diffusion
barrier properties for Cu metallization due to synergistic behavior of two different
35 40 45 50 55 60 65
Si Kpr (-100)
35 40 45 50 55 60 65
Figure 7-1. X-ray diffraction patterns of as-deposited films and annealed ones at the
temperature shown in the figure: a) Cu/Ge(25nm)/Hf'Nx(7nm)/Si (001) for 1
hr: b) Cu/Hf~Nx(7nm)/Si (001) for 1 hr: c) Cu/Ge(2 5nm)/Hf~Nx(7nm)/Si
annealed at 500 oC for increasing time durations and d) Cu/Ge(25nm)/Si
(001) for 1 hr.
(d)Si KS; (400)
o00C, I hr
600 "C, I h~r
oC0", I hr
Cu!11~ 1/ 40oC, I hr
Cu (20)A deposited
5 40 45 50 55 60 65
35 40 45 50
500 oC, 3 hr
55 60 65
Figure 7-1. (cotd.).
0 1 20 3~0
500 "C, I bour
0 30 60 90
0 2 40 60
Figure 7-2. EDS depth profile of Cu/Ge(25nm)/Hf'Nx(7nm)/Si for a) as-deposited: b)
annealed at 500 OC for 1 hr: and c) annealed at 500 OC for 3 hr.
1 5 0.d nmdi
U m 7 .41000 pr isd iv
O 50 ,00 nm/di
Fgre -.AMiae fC/G(5m/fx7m/ifo )a-eoie:b
aneldat50o or1h:ad )anaeda 0 C o r
Figure 7-4. HRTEM images of Cu/Ge(25nm)/Hf'Nx(7nm)/Si for a) as-deposited: b)
annealed at 500 OC for 1 hr: and c) annealed at 500 OC for 3 hr.
PROPERTIES OF Ta-Ge-N AS A DIFFUSION BARRIER FOR Cu ON Si
As the minimum feature size in integrated circuits continues to decrease, there is a
need to replace Al in interconnects due to the limited conductivity and poor
electromigration performance of Al under high current densities. Cu is a viable
alternative for replacing Al due to its 1.7 times lower resistivity compared to Al and
better electromigration properties under high current densities.71,124 In addition, the RC
time delay achieved by an Al/SiO2 Stack needs to be lowered significantly to increase the
device speed. The RC time constant can be significantly lowered by combining Cu with a
low-k material thus enhancing the performance of the device.3
Although the aforementioned properties of Cu are advantageous for device
manufacture, Cu diffuses very effectively through Si, SiO2 and Ge, degrading the
electrical properties of the device.104 It is an acceptor in Ge introducing traps at E, + 0.04
eV, E, + 0.32 eV and E, 0.26 eV.74 CU is a donor in Si and creates traps from 0.2 to 0.5
eV above the valence band.125 It also reacts with Si to form parasitic copper silicides at
the interface. In addition, Cu reacts with dopants in Si to form complexes that affect the
device characteristics.95 Thus, use of Cu as an interconnect for future technology nodes
will require a viable diffusion barrier. There has been a significant interest in refractory
nitrides such as WNx, TiN, HfNx, and TaN61,110,114-1 16,121,126-128 as diffusion barrier
candidates for Cu metallization. These diffusion barriers, however, typically fail at
moderate temperature (400-600 oC) limiting their use. The primary mode of failure for
these diffusion barriers is by Cu diffusion through grain boundaries formed by
recrystallization of the barrier material upon annealing. By increasing the
recrystallization temperature, grain boundary formation can be delayed, thereby
increasing the robustness of the diffusion barrier. Addition of a third element to the
binary matrix induces amorphization at room temperature and also delays the
recrystallization process, making ternary solutions interesting as diffusion barrier
candidates. Some of the ternary nitride diffusion barriers being studied include W-Si-N,
Ta-Si-N, and W-Ge-N.99,81,129
In this paper, we report on the barrier layer properties when Ge is added to TaN.
Ge is of interest because it displays chemical behavior similar to that of its congener Si
and might be compatible with future Ge and SiGe based devices. Diffusion barrier
properties of Ta-Ge-N were compared with TaN deposited under identical conditions.
The results indicate that a Ta-Ge-N diffusion barrier fails at a higher temperature than
TaN, suggesting superior diffusion barrier properties.
Ta-Ge-N diffusion barriers were deposited on p-Si (001) wafers by reactive
sputtering process at room temperature. Prior to deposition, the wafer was etched in 7:1
buffered oxide etch to remove its native oxide and then rinsed with deionized water. The
substrate was loaded in the sputtering chamber that is maintained at 3 x 107 Torr base
pressure. Sputtering targets were pre-sputtered before deposition at an Ar pressure of 15
mTorr to remove any contamination on the surface. The forward power used for Ta and
Ge was 200 W and 100 W, respectively. The diffusion barrier films were then deposited
by flowing Ar and N2 at a chamber pressure of 10 mTorr. For comparison, TaN diffusion