Professional Documents
Culture Documents
Ken Goodson
Alumni
Prof. Mehdi Asheghi Carnegie Mellon University (ME)
Prof. Dan Fletcher UC Berkeley (Bioengineering)
Prof. Bill King Georgia Tech (ME)
Prof. Katsuo Kurabayashi University of Michigan (ME)
Prof. Sungtaek Ju UCLA (ME)
Prof. Kaustav Banerjee UC Santa Barbara (EE)
Dr. Uma Srinivasan Xerox
Dr. Per Sverdrup Intel
Dr. Peng Zhou Cooligy
Dr. Maxat Touzelbaev AMD
Dr. Sanjiv Sinha Intel
Challenging (and Exciting) Time
for IC Thermal Management
Thermal issues are in the headlines:
Pumped liquid cooling in laptops and
desktops (Hitachi/Toshiba/Apple).
Motivation for multi-core processors.
0.3 oC/W
Rtotal (ITRS 2003)
0.2
Rheat sink
3DIC
Unprecedented startup
0.1 climate for thermal
technologies, ranging
Multicore
Rchip from microfluidics to
0 interfaces (>$100
2000 2005 2010 2015 2020 Million VC per year).
Thermal Resistance Hierarchy
Cinterconnect Device-Level SEM
Tinterconnect
qinterconnects metal
~ 10 ns ILD
Rinterconnect
Ctransistor
qtransistors
Ttransistors
~ 100 ps
Cchip
Rtransistor IBM
Tjunction
chip
~ 100 s
Rchip + TIM chip carrier
C heat sink
Tspreader Si chip
heat spreader
~1s Rheat sink heat sink
Tambient
Outline
1. On-Chip Thermal
Challenges
3. Nano Transistor
Simulations
On-Chip Thermal Challenges
40
Global Wires With Vias
Interconnect Tech. Node
22 nm
Temperature 30
32 nm
Field 45 nm
20 65 nm
T [ C]
90 nm
o
Student: Sungjun Im,
Trans Electron 10
Devices, in press 2005.
0
Silicon 0.0 0.2 0.4 0.6 0.8 1.0
Normalized Distance from Substrate
Interconnect
self heating
Rinterconnect
150
140
0.3 oC/W
50 W 130
120
Rtotal (ITRS 2003)
110
100
0.2
Rheat sink
90
80
100 W 70
60
3DIC
50
40
30
0.1
20
10
Multicore
0
Rchip
0
2000 2005 2010 2015 2020
Microprocessor hotspots &
power density (mm scale)
Rchip + TIM
Interconnects
0 0
2004 2006 2008 2010 2012 2014 2016 2018
Transistor hotspots, leakage,
mobility reduction
Rtransistor
Microprocessor hotspots
Interconnects
Source Drain
Bulk Silicon
Bulk FET
1
Channel Length (nm)
350 130 45 18 10
Source Drain
FinFET
Bulk Silicon
Carbon nanotube
transistor
Strained Si, Ge, SiGe HfO2
Silicide top gate (Al) CNT
SOI/SiGe Gate
S (Pd) D (Pd)
SiO2
Buried oxide
back gate
Silicon substrate (p++ Si)
IBM Stanford
3. Nano Transistor
Simulations
Phonon Overview
Responsible for heat conduction in semiconductors optical
p = hq
acoustic
E = h
Optical Acoustic
(slow) (fast)
Electrothermal Processes
High Electric
Field
Hot Electrons
225 nm (Energy E) E > 60 meV
~ 0.1ps
IBM E < 60 meV
~ 0.1ps Optical
OpticalPhonons
Phonons
lattice wave
~ 10ps
electron Acoustic
AcousticPhonons
Phonons
high
electron ~ 1 ms–1s
energy d
defect
phonon
E Heat Transfer
to Package
dopant
phonon
atom
Drain
Drain
Simple Drift-Diffusion q ' ' ' j E
10 nm
en n ep p
rr rg E Fn EFp eT Pp Pn T jn Pn jp Pp
Electron Monte Carlo (Jacoboni, 1983; Fischetti, 1988; Pop et al., 2005)
gen abs
1 d
q' ' '
t dV
Monte Carlo Simulation of Heat Generation
(Pop, Sinha, Goodson, Proc. IMECE 2002, Submitted to Journal of Applied Physics)
2 k x2 k y k z2
2
• Electrons treated as semi-classical
particles E 1 E
2 mx m y mz
• Electrons drift (free flight), scatter
2
M ( k ) g E k q
and select new state 2
(k )
• Yields full information about phonon
generation (Jacoboni, 1983)
optical
intervalley
g
f
intravalley
f
g
acoustic
k k q G
Monte Carlo of 1D Diode
Thesis work of Eric Pop
N+ N+
i-Si
qV
Impact of Ballistic Transport
Potential (V)
L=500 nm 100 nm 20 nm
Monte Carlo
Drift-Diffusion
d~
3. Nano Transistor
Simulations
Transistor Simulation Regime Map
Atomistic
Fundamental
BTE with
Research
Phonons
,
Wave models , Ju 0 0)
p 0
d r u n (2
er so
S v od r
Go da
BTE or m
M aj u
Monte Carlo i,
rk La 995)
wo a (1 )
h k
uc hut 9 9 0 5)
1 9
m ac )
u r ( ( 19
W 994 S h vi ch
Diffusion (1 o 0)
p an 1 97 ) )
83 ) 3)
2) A e r ( 985 9
(1 88 20
0 m
96 j a (1 ni (19
( tro 95)
(1 ek ani 6) o ad s
n oe r
t b ti st
e nd ( 1 9
tto Bl cca (19
8
a co het in u
L tta
ra J i sc W Da
St Ba dan F
Isothermal Ru
Drift Diffusion
Monte Carlo
Models
with Quantum
Moments
Monte Carlo
BTE
& BTE
Full Quantum
Electrons
Transistor Simulation Regime Map
Atomistic
Transistor Requirements
Available electrons phonons
BTE with
Software
Phonons
Wave models
~5 nm ~100 nm
MFP
BTE or
Monte Carlo ~5 nm ~1 nm
Diffusion
Currently
??
Available
Isothermal
Drift Diffusion
Monte Carlo
Models
with Quantum
Moments
Monte Carlo
BTE
& BTE
Full Quantum
Electrons
Monte Carlo of 18 nm Thin-Body
SOI Transistor
ITRS Specs:
Current
Atomistic
models split flux phonon BTE http://nanoheat.stanford.edu
BTE or
Electron MONET
Monte Carlo Monte
Carlo
with
ESD phonon
Diffusion
Studies dispersion
& branch
accuracy
Isothermal
Drift Diffusion
Monte Carlo
Models
with Quantum
Moments
Monte Carlo
BTE
& BTE
Full Quantum
Electrons
Monte Carlo Implementation: MONET
E. Pop et al., J. Appl. Phys. 2004
optical
2 k2
k 2
k
2
E 1 E
y
x
z
2 m m mz 50 meV
x y
Analytic band
q 0 vs q cq 2
Full band
OK to use 20 meV
acoustic
Sanjiv Sinha
Hotspot Temperature
ULTRA-THINBODY
Temperature (oC)
Actual Junction
Temperature
Year1 1
Based on ITRS 2003
Molecular-Dynamics Simulations of Phonons†
0.8
Sinha et al., J. Appl. Phys., 97, 23702 (2005)
0.6 LO phonon wave packet
0.4
(longitudinal displacements)
0.2
uz/a0 x 103
0
-0.2
-0.4
-0.6
-0.8
0 200 400 600 800 1000
z (nm)
1 m 0.35
0.2
†
In collaboration with Drs. Phillpot & 0
0 200 400 600 800 1000
Schelling at Argonne National Lab z (nm)
Atomistic Simulation of Phonon
Hotspot Decay
Sinha, Schelling, Phillpot, and Goodson, Journal of Applied Physics 2005
Alumni
Prof. Mehdi Asheghi Carnegie Mellon University (ME)
Prof. Dan Fletcher UC Berkeley (Bioengineering)
Prof. Bill King Georgia Tech (ME)
Prof. Katsuo Kurabayashi University of Michigan (ME)
Prof. Sungtaek Ju UCLA (ME)
Prof. Kaustav Banerjee UC Santa Barbara (EE)
Dr. Uma Srinivasan Xerox
Dr. Per Sverdrup Intel
Dr. Peng Zhou Cooligy
Dr. Maxat Touzelbaev AMD
Dr. Sanjiv Sinha Intel
EXTRA SLIDES
CMOS Transistor Technologies
Strained-Si
channel Tbody
Gate Gate Gate
Gate
Tbody Drain
Gate
Source Drain
Gate
Source
PillarFET
(UT Austin)
T E NE
C Q J NE KT
t t
90 nm bulk
Sanjiv Sinha
Fully-Coupled Simulations – Monte Carlo
Effect of Source/drain
resistance is amplified: Uncoupled
Voltage [V]
Coupled
Position [nm]
A Multiscale, Multi-Carrier Approach
Sponsorship: IBM, SRC, and Intel. Students: Eric Pop, Sanjiv Sinha
MONET
nq nq nq
vq .nq
t q
fast Monte Carlo for electron and
phonon distributions in nanodevices
Gate Gate
450 430 390 410
6
5 CDM
4
IESD (A)
3
2 HBM
1
Metal/via damage
K. Banerjee et al., IRPS 1996 -1
MM
0 20 40 60 80 100 120ns
Electrostatic Discharge
3
level cooling
2 HBM Governs peak transistor
1
temperature & mobility
Governs relative
-1 reduction, leakage
MM importance of
augmentation
0 20 40 60 80 100 120ns failure modes for ESD
gate
Stress switching ESD / EOS Phenomena
Timescales clock
period