Upload
jack-coughlin
View
213
Download
0
Embed Size (px)
Citation preview
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 1 -
Interconnect Working Interconnect Working GroupGroupInterconnect Working Interconnect Working GroupGroup
What’s up with wires?14 July 2010
San Francisco USAChristopher Case
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 2 -
ITWG Regional Chairs
US Christopher Case US Christopher Case
Japan Tomo Nakamura Hideki Shibata
Japan Tomo Nakamura Hideki Shibata
EuropeHans-Joachim BarthAlexis Farcy
EuropeHans-Joachim BarthAlexis Farcy
Korea Hyeon-Deok Lee Sibum Kim
Korea Hyeon-Deok Lee Sibum Kim
TaiwanDouglas CH Yu
TaiwanDouglas CH Yu
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 3 -
Partial List of Contributors
• Shuhei Amakawa• Nobuo Aoi• Sitaram Arkalgud• Lucile Arnaud• Koji Ban • Hans-Joachim Barth• Eric Beyne• Christopher Case • Chung-Liang Chang• Hsien-Wei Chen• Gilheyun Choi• Jinn-P. Chu• Mike Corbett • Alexis Farcy• Paul Feeney• Masayuki Hiroi• Harold Hosack
• Masayoshi Imai• Raymond Jao • Shin-Puu Jeng • Sibum Kim• Mauro Kobrinsky• Nohjung Kwak • Hyeon Deok Lee• Scott List• Anderson Liu• Didier Louis• JD Luttmer• Toshiro Maekawa• Akira Matsumoto• Azad Naeemi • NS Nagaraj• Tomoji Nakamura • Yuichi Nakao
• Akira Ouchi• Peter Ramm• Rick Reidy • Scott Pozder • Philip Pieters • Andy Rudack • Larry Smith• Mark Scannell• Hideki Shibata• Michele Stucchi• Wen-Chih Chiou• Weng Hong Teh• Thomas Toms • Manabu Tsujimura • Kazuyoshi Ueno• Detlef Weber• Osamu Yamazaki
0710
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 4 -
Interconnect scope• Conductors and dielectrics
– Starts at contacts– Metal 1 through global levels – Includes the pre-metal dielectric (PMD)
• Associated planarization • Necessary etch, strip and cleans• Embedded passives• Global and intermediate TSVs for 3D • Reliability and system and performance issues• “Needs” based replaced by – scaled,
equivalently scaled or functional diversity drivers.
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
MPU Cross-Section
Dielectric Capping
Layer
Copper Conductor with
Barrier / Nucleation
Layer
Pre-Metal Dielectric
Tungsten Contact Plug
Inter-Mediate(=M1x1)
Inter-Mediate(=M1x1)
Metal 1Metal 1
Passivation
Dielectric
Etch Stop Layer
ASIC Cross-Section
Semi-Global (=M1x2)
Semi-Global (=M1x2)
Metal 1 Pitch
Via
Wire
Via
Wire
Via
Wire
Metal 1 Pitch
Global (=IMx1.5~2µm)Global (=IMx1.5~2µm)
Inter-Mediate(=M1x1)
Inter-Mediate(=M1x1)
Metal 1Metal 1
Global (=IMx1.5~2µm)
Global (=IMx1.5~2µm)
1) MPU : Revised hierarchy2) ASIC : No drastic change, however semi-global should be kept at 2 x M1
Hierarchical Cross Sections
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 6 -
Technology RequirementsNow restated and organized as• General requirements
– Resistivity– Dielectric constant– Metal levels– Reliability metrics
• Level specific requirements (M1, intermediate, global)– Geometrical
• Via size and aspect ratio• Barrier/cladding thickness• Planarization specs
– Materials requirements• Conductor effective resistivity and scattering effects
– Electrical characteristics• Delay, capacitance, crosstalk, power index
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 8 -
Difficult challenges (1 of 2)Difficult challenges (1 of 2)• Meeting the requirements of scaled metal/dielectric systems
– Managing RC delay and power• New dielectrics (including air gap)• Controlling conductivity (liners and scattering)
– Filling small features• Barriers and nucleation layer• Conductor deposition
– Reliability• Electrical and thermo-mechanical
• Engineering a manufacturable interconnect stack compatible with new materials and processes – Defects– Metrology– Variability
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 9 -
Difficult challenges (2 of 2)• Meeting the requirements with equivalent scaling
– Interconnect design and architecture (includes multi-core benefits)
– Alternative metal/dielectric assemblies• 3D with TSV
– Interconnects beyond metal/dielectrics• 3D• Optical wiring• CNT/Graphene
– Reliability• Electrical and thermo-mechanical
• Engineering a CMOS-compatible manufacturable interconnect system– Non-traditional materials (for optical, CNT etc.)– Unique metrology (alignment, chirality measurements,
turning radius etc)
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 10 -
Eff
ecti
ve D
iele
ctr
ic C
on
sta
nt;
keff
Year of 1st Shipment
ITRS1999
ITRS2001
ITRS2005
ITRS2003
Before 2001,unreasonable RM
without logical basis
Before 2001,unreasonable RM
without logical basis
ITRS2007-2010
Historical Transition of ITRS Low-k Roadmap
2009 decreased max bulk by 0.1 - no significant change on eff in 2009
2010 no changes
2009 decreased max bulk by 0.1 - no significant change on eff in 2009
2010 no changes
Since 2003, based on wiring capacitance calculation of three kinds of dielectric structures and validated against publicationsSince 2003, based on wiring capacitance calculation of three kinds of dielectric structures and validated against publications
2009 Summer Conference
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 11 -
2010 Low k or nothing?
Air gap architectures will be required for bulk <2.0•No viable materials expected to be available.•Mechanical requirements easier to achieve with air-gaps.
•End of the material solution and the beginning of an architecture solution.
Year of Production 2010 2011 2012 2013 2014 2015
MPU/ASIC Metal 1 ½ Pitch (nm)(contacted)
45 38 32 27 24 21
Interlevel metal insulator – bulk dielectric constant (κ)
2.3-2.5 2.3-2.5 2.3-2.5 2.1-2.3 2.1-2.3 2.1-2.3
Year of Production 2016 2017 2018 2019 2020 2021 2022 2023 2024
MPU/ASIC Metal 1 ½ Pitch (nm)(contacted)
18.9 16.9 15.0 13.4 11.9 10.6 9.5 8.4 7.5
Interlevel metal insulator – bulk dielectric constant (κ)
1.9–2.1 1.9–2.1 1.9–2.1 1.7–1.9 1.7–1.9 1.7–1.9 1.5–1.7 1.5–1.7 1.5–1.7
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 12 -
Air Gap
Pictures (top left, clockwise): NXP, IBM, Panasonic, TSMC
Approaches•Creation of air gaps with non-conformal deposition
•Removal of sacrificial materials after multi-level interconnects
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 13 -
2010 Barrier/Nucleation/Resistivity
• ALD barrier processes and metal capping layers for Cu are lagging in introduction – key challenge
• Resistivity increases due to scattering and impact of liners
•No known practical solutions
Year of Production 2010 2011 2012 2013 2014 2015
MPU/ASIC Metal 1 ½ Pitch (nm)(contacted) 45 38 32 27 24 21
Barrier cladding thickness Metal 1 (nm)
3.3 2.9 2.6 2.4 2.1 1.9
Conductor effective resistivity (µΩ‑cm) Cu Metal 1
4.08 4.30 4.53 4.83 5.2 5.58
Year of Production 2016 2017 2018 2019 2020 2021 2022 2023 2024
MPU/ASIC Metal 1 ½ Pitch (nm)(contacted)
18.9 16.9 15.0 13.4 11.9 10.6 9.5 8.4 7.5
Barrier cladding thickness Metal 1 (nm)
1.7 1.5 1.3` 1.2 1.1 1.0 0.9 0.8 0.7
Conductor effective resistivity (µΩ‑cm) Cu Metal 1
6.01 6.33 6.7 7.34 8.19 8.51 9.84 11.30 12.91
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
Cu Contact Transition Prospective 5% of total parasitic resistance for contact resistance
(agreed with PIDS, FEP and INTC in 2008)
Estimation of contact resistance used to forecast timing for Cu
TargetW-plugCu-plug
45/40nm 32/28nm 22/20nm
marginal
W-plug will be applicable until 22/20nm node at 2013.
Cu-plug will be able to satisfy the requirement beyond 15/14nm node after 2016.
OK
0
50
100
150
200
2007 2010 2013 2016Year
Re
sis
tan
ce
(
/c
on
tac
t)
15nm
NG
65~54nm 45~32nm 27~21nm 19nm~ITRSCommercial
Assumption•Aspect ratio=5.5•Barrier for W-plug
PVD-Ti 10nm@btm, 2nm@sideCVD-TiN5nm@btm/side
•Barrier for Cu-plugPVD-Ta(X) 5nm@btm, 1nm@sideCVD-Ru(X) 2nm@btm/side
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
0.1
1.0
10.0
2009 2012 2015 2018 2021 2024Year
Jmax
JEM
- 15 -
Wire current limit – width dependence
• The color boundaries may actually be width-dependent.– Hu et al., Microelectronics Reliability, vol.46, pp.213-231, 2006.
• Jmax will increase with frequency and reducing cross-section, while JEM will scale with the product w*h according to EM lifetime dependence on wiring width
0.1
1.0
10.0
100.0
2005 2010 2015 2020 2025
Year
Jmax
(M
A/c
m2)
Jmax 2008
Jmax 2007
2008 Update 2010 To be revised
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 16 -
H. Shibata added published data based on Yokogawa et. Al. IEEE Trans on ED.2008
EM
Lif
etim
e Im
pro
vem
ent
Rat
io
Normalized Resistance Increase Ratio
CuSiN
CuAl
CoWP
CuSiN ( 1 )
CuSiN ( 2 )
CuGeN
CuSi(N)+ Ti-BM
< CoWP-Cap >
Si
Si
SiSi
SiSi Si
SiSi
Si
Si
SiSi
Al
Al Al
Al
Al
Al
Al
Al
AlAl
AlAl
AlAlAl Al
< CuAl-Alloy >
< CuSiN-Cap >
< CuGeN-Cap >
Ge GeGeGe
< CuSiN-Cap+Ti>
Metal CappingVarious lifetime improvement approaches against the resistivity increase
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 17 -
High Density TSV Roadmap or“enabling terabits/sec at femtojoules/bit”
• The Interconnect perspective - examples:– High bandwidth/low energy interfaces between memory and logic– Heterogeneous integration with minimal parasitics (analog/digital, mixed substrate materials, etc.)– “Re-architect” chip by placing macros (functional units) on multiple tiers (wafers) and connect using HD
TSVs• Defined a 3D interconnect hierarchy
• TSV dimensions• Minimum contact pitches• Overlay accuracy
– Described process modules
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 18 -
Emerging Interconnect Changes
OUT:• Air gaps Process Module Section
•Increasing maturity of integrated air gap solutions
• 3D Process Module and Architecture Sections•3D with TSVs nearing production
IN:• Focus on Cu Replacements
• Section on Native Device Interconnects
• Identified need to jointly consider switch and interconnect properties of new switch options
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
All roads lead to C?
- 19 -
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 20 -
Native Device InterconnectsNanowires GNRs
CNTs Spin Transport
• Doped NWs require silicidation for lengths > 1µm
•Multi-fanouts are very difficult
•Serial, multi-input AND gates are easy
•Multi-fanouts are easy•Multi-layer routability incurs quantum resistance
•Diffusion and spin wave transport are 1000 times slower than electron transport
•Spin relaxation lengths are ~ 1µm
1k/µm
1 M
12k
mfp
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
Potential interconnect state variables
GA Tech is using a Benchmarking Approach:
Define possible transportation mechanisms in general
Model delay and energy per bit for various transport mechanisms.
Benchmark new state variables against their conventional counterpart as a function interconnect length.
From Naeemi et al.
Transport Mode
Information Token
Diffusion Direct ExcitonsIndirect ExcitonsSpinPseudoSpin
Drift Indirect ExitonsSpinPseudospin
Ballistic Transport (Fermi Velocity)
SpinPseudospin
Spin Wave Spin
EM Waves Photons Plasmons
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
Delay versus Length
Communications of new state variables are all slow compared to conventional CMOS interconnects.
Interconnect Length (Gate Pitch)
Del
ay (
ps)
CMOS (Min-Size Driver)
Diffusio
n
Drift Ballistic
CMOS (5x Driver)
Spin Waves (104 m/s)
Spin Waves (105 m/s)Two spin wave phase velocities of 104 (realistic) and 105 m/s (optimistic) are considered.
Driver delay is not included for new state variables (upper bound performance)
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
Required Area Down-Scaling
Important implications at the Device and architecture levels.
Novel interconnects can compete with their conventional counterparts only if the new circuits occupy smaller areas.
To have smaller circuits switches must be smaller or same functions can be obtained with fewer switches.
For longer interconnects larger circuit area scaling is needed.
For a given area scaling, maximum circuit size with no signal conversion can be obtained.
Diffusion
Drift, 0.8V
Ballistic
Spin Waves (105 m/s)
Interconnect Length (Gate Pitch)
Nee
ded
Are
a S
calin
g A
CM
OS/
AP
ost-
CM
OS
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
Energy per bit
ESWB 0M .HextV
0 410-7H / m
M magnetization change A/m
Hext External magnetic field Oe
V= Volume of the interconnectm3
For a Co30Fe70 SWB, energy is taken to be 1aJ/um*
* A. Khitun et al, Nanotechnology, vol. 18, no. 46, 2007.
CMOS
DIFFUSION
DRIFT,0.8V
SWB
The excitation energy has been optimistically assumed to be negligible.
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 25 -
Emerging Interconnect Summary Table• Interconnect options include Cu Extensions, Cu
Replacements and Native Device Interconnects
Option Potential Advantages Primary ConcernsOther metals (W, Ag, silicides)
Potential lower resistance in fine geometries
Grain boundary scattering, integration issues, reliability
Nanowires Ballistic conduction in narrow lines
Quantum contact resistance, controlled placement, low density, substrate interactions
Carbon Nanotubes
Ballistic conduction in narrow lines
Quantum contact resistance, controlled placement, low density
Graphene Nanoribbons
Ballistic conduction in narrow films, planar growth
Quantum contact resistance, control of edges, deposition and stacking
Optical (interchip)
High bandwidth, low power and latency, noise immunity
Connection and alignment between die and package, optical /electrical conversions
Optical (intrachip)
Latency and power reduction for long lines, high bandwidth with WDM
Benefits only for long lines, need compact components, integration issues, need WDM
Wireless Available with current technology, wireless
Very limited bandwidth, intra-die communication difficult, large area and power overhead
Superconductors Zero resistance interconnect, high Q passives
Cryogenic cooling, frequency dependent resistance, defects, low critical current density
~Example of Approaches for Cu Replacements ~
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
Formidable Task: Fabricating Dense Bundles of SWNTs
Fabricating Bundles of Densely Packed SWNTs for interconnects has proven very challenging.
Making contacts to horizontal bundles of SWNTs is very difficult.
Promising progress in creating aligned isolated SWNTs by transferring SWNTs grown on sapphire to other substrates.
Single SWNTs are too resistive for general purpose CMOS circuits.
Y. Nishi and H.-S. Philip Wong (Stanford) 26
Young Lae Kim et al. (RPI, RICE & NorthEastern Univ
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
Graphene – a promising interconnect option beyond Cu/Low k?
Graphene Potentials•Device/interconnect synergism•Tunable bandgap•Ballistic transport for > 10m•Planar technology•Excellent Jmax
•Potential CVD deposition process
Graphene Areas for Research•Layer to layer coupling/scattering•Low T CVD deposition•Effective bandgap control
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA
PrognosisInterconnects can impose major limits on both charge and non-charge based devices.
Most transport mechanisms for novel state-variables are relatively slow.
For the same speed, the new circuits must be smaller in terms of physical area.
The new devices must offer better delay-energy trade-offs compared with conventional CMOS devices.
Interconnects have important implications at the device, circuit and architecture levels.
ITRS 2010 Summer Conference – 14 July 2010 San Francisco, CA - 29 -
Summary 2010• 3D and air gaps moved out of emerging sections • Low-k – unchanged – second time in 10 years
– Air gaps expected to be solution for kbulk <2.0• Jmax current limits are width dependent - a new concern• Barriers and nucleation layers are a critical challenge
– ALD barrier process is lagging in introduction – sub 1 nm specs– Ru hybrid approaches proliferating– Capping metal for reliability improvement
• New Interconnect 3D TSV roadmap tables• Cu contact need expected >2013• Emerging Interconnect Properties being developed• First principle consideration of interconnects properties for new switches – CNT, graphene, nanowires etc.