future trends in microelectronics – impact on detector readout final.pdfedwin hubble. detector...
TRANSCRIPT
Future Trends in Future Trends in Microelectronics Microelectronics –– Impact on Impact on
Detector ReadoutDetector ReadoutPaul O’Connor
Detector Development Symposium Paul O'Connor BNL April 5, 2006 2
OutlineOutline
• CMOS Technology Scaling
• Analog Circuits
• Radiation Effects
• Cost
Detector Development Symposium Paul O'Connor BNL April 5, 2006 3
Edwin HubbleEdwin Hubble
Detector Development Symposium Paul O'Connor BNL April 5, 2006 4
Gordon MooreGordon Moore
Detector Development Symposium Paul O'Connor BNL April 5, 2006 5
CMOS Logic ElementCMOS Logic Element• Pair of nearly-ideal switches in
series
• Complementary polarity ⇒common control voltage
• input impedance ~ ∞
• VDD-tolerant
• Zero static power
VDD
Detector Development Symposium Paul O'Connor BNL April 5, 2006 6
MOSFET ScalingMOSFET Scaling
• Voltages, dimensions reduced by α• Results:
αα
/ Speed
1 eCapacitanc
const. / eConductancconst.
CVI
VIE =r
const. density Power Density
1 Power/gate
1 energy Switching
2
22
32
αα
α
fCV
CV
20V 4V
Detector Development Symposium Paul O'Connor BNL April 5, 2006 7
Industry Scaling RoadmapIndustry Scaling Roadmap
• New generation every ~2 years with α = √2
• Lg (1970) 8 µm (2007) 18 nm
180
Detector Development Symposium Paul O'Connor BNL April 5, 2006 8
Transistor CountTransistor Count
Year
40048008
8080
8086
80286Intel386
Intel486Pentium
Pentium ProPentium II
Pentium IIIPentium 4
1970 1975 1980 1985 1990 1995 2000
MO
SF
ET
s
103
106
109
2.5e4
Detector Development Symposium Paul O'Connor BNL April 5, 2006 9
4004
8008
8080
8086
80286
Intel386
Intel486
Pentium
Pentium Pro/II/III
Pentium 4
Year
1970 1975 1980 1985 1990 1995 2000 2005
CPU Clock FrequencyCPU Clock FrequencyH
z
106
108
107
109
4e3
Detector Development Symposium Paul O'Connor BNL April 5, 2006 10
Transistor CostTransistor Cost$
100
10-3
10-6
Year
3e6
Detector Development Symposium Paul O'Connor BNL April 5, 2006 11
Digital State of the ArtDigital State of the Art
• 65 nm node: – SRAM test chip Nov. 2003
– production Nov. 2005
– “crossover” 3Q 2006
• 45 nm node: – 153MB SRAM Jan. 2006
– >109 MOSFETs
– production 3Q ‘07
Detector Development Symposium Paul O'Connor BNL April 5, 2006 12
Failures of Classical ScalingFailures of Classical Scaling
• ignores kT• ignores mobility degradation and velocity
saturation
• ignores fringing capacitance
• ignores tunneling
• ignores atomistic effects
• ignores industry economics
Detector Development Symposium Paul O'Connor BNL April 5, 2006 13
kTkT ≠≠ 00
• MOSFET in weak inversion when Vgs < Vth:
nkTqV
offnkT
VVq
D
THTHGS
eIIeII−−
=⇒= 0
)(
0
VDD
kT
• Can’t continue to scale VTH
• Tradeoff of speed for leakage current (static power)
• Multi-VTH processes
VTH
fast
low power
Detector Development Symposium Paul O'Connor BNL April 5, 2006 14
Gate tunneling currentGate tunneling current• Significant tunneling through
SiO2 when tox < 3nm
• Jtunnel increases 100X per generation
• Replace SiO2 with high-κdielectric (?)
• Multi-tox processes
κκ ε
ε
−−=
hi
SiOhitEOT 2
Detector Development Symposium Paul O'Connor BNL April 5, 2006 15
Atomistic EffectsAtomistic Effectsrandom dopantdistribution
oxide thickness variation
line edge roughness
σVTH, RTS σVTH, σJtunnel, σµ σVTH
A. Asenov, IEEE Trans. Electron Dev. 50(9), 1837 (2003)
Detector Development Symposium Paul O'Connor BNL April 5, 2006 16
Scaling in PracticeScaling in Practice• Until 180nm node:
– follow classical scaling with α = √2– 2.8X performance per generation
• Now:– continue (super) scaling Lg
– sub-scaling of tox
– VDD, VTH have stopped scaling• Gate density and speed continue
to scale• Increase of E, conductance• Switching energy decreases only
by 1/α not 1/α3
• Power density increase ~ α• Static power from leakage, gate
tunneling make power problem worse
SiO2 latticeconstant
3 kT/q
Detector Development Symposium Paul O'Connor BNL April 5, 2006 17
Power is biggest impediment to Power is biggest impediment to further scalingfurther scaling
LG-1
LG-4
E. Nowak, IBM J. Res. & Dev. 45(2), 169 (2002)
Detector Development Symposium Paul O'Connor BNL April 5, 2006 18
M. Bohr, Intel, 2003 IEDM
E. Gusev et al., IBM, 2004 IEDM
F.-L. Yang et al., IMEC, 200 VLSI Symp.
S. Thompson, Intel,IEEE T-ED 51(11), 1790 (2004)
NextNext--generation Transistorsgeneration Transistors
M. Ieong, IBM, Sol. State and IC Tech. 2004
Strain Engineering• Reduce mobility degradation
UTSOI, DG-FET• Suppress SCE through improved electrostatics
High-k gate dielectric• Suppress Jtunnel
Detector Development Symposium Paul O'Connor BNL April 5, 2006 19
SiGeSiGe HeterojunctionHeterojunction Bipolar Bipolar Transistor (HBT)Transistor (HBT)
• Power-performance ~ 2X standard CMOS at same feature size
• Cost also ~ 2X
• Provides 3 – 4 year “head start”over CMOS
• Eventually CMOS provides better cost-performance
Detector Development Symposium Paul O'Connor BNL April 5, 2006 20
Analog: Speed, Gain & MatchingAnalog: Speed, Gain & Matching• Speed/intrinsic gain tradeoff
– fT ~ 1/LG for min-LG devices– Output conductance gds sensitive to geometry
and bias– Gain gm/gds severely degraded in ultra-scaled
devices:• failure to adhere to classical constant-E scaling
• Threshold Mismatch– Dopant fluctuations:– Overcome by increasing W*L– Power penalty
ασ ~min, −∆ LGVTH
Pelgrom, Philips, 1998 IEDM
0
50
100
150
200
250
250 180 130 90 65
Technology Node, nm
f T, G
Hz
0
5
10
15
20
fT
gm/gds
g m/g
ds
f T, H
z
P, W10-6 10-4 10-2
107
108
109
1010
1011
250nm
65nm
130m
Detector Development Symposium Paul O'Connor BNL April 5, 2006 21
Analog LowAnalog Low--VVDDDD ChallengesChallenges• Increasing ratio of VTH/VDD rules out use of many classical analog
design toplogies, e.g. cascodes.• To achieve same performance at lower VDD, analog circuits need
increased power.
(performance limited by thermal noise)
A. Annema et al., IEEE JSSS 40(1), 132 (2005)
0.1 1Min. Feature Size, um
103
104
Dyn
am
ic R
an
ge
10 uW100 uW1000 uW
SNR, BW held constant Power held constant
100X Power4X SNR
Detector Development Symposium Paul O'Connor BNL April 5, 2006 22
Noise in scaled CMOSNoise in scaled CMOS• For CSA, consider noise in relation to gate capacitance• Thermal noise improvement from higher fT:
– ENCth ~ √1/α• 1/f noise depends on interface trap density, device area, inversion charge density –
– no consistent trend with scaling• Choice of L,W for minimum noise needs to account for moderate inversion operation• Gate leakage current shot noise (parallel)
G. Anelli 2005
0.01 0.1 1100
200
300
400 TECHNOLOGY SUPPLY COST/RUN
0.35µm 3.3V 14k$ 0.25µm 2.5V 19k$ 0.18µm 1.8V 32k$
En
erg
y R
esol
utio
n [
EN
C]
Preamplifier Power [mW]
P O'Connor and G. De Geronimo, Prospects for charge sensitive amplifiers in scaled CMOS, NIM A480, 713 (Mar. 2002)
EN
C (
rms
e- )
Detector Development Symposium Paul O'Connor BNL April 5, 2006 23
Radiation Tolerant InverterRadiation Tolerant Inverter
metal 1 polysilicon
p+ guard ring
NMOS Enclosed Layout Transistor (ELT)
VDD
n+ diffusion p+ diffusion
PMOS
contact to diffusion
DRAIN
DRAIN
GATE
SOURCESOURCEGATE IN
OUT
Detector Development Symposium Paul O'Connor BNL April 5, 2006 24
Radiation Effects below 0.25Radiation Effects below 0.25µµmm• Thinner gate oxides more
tolerance to Total Ionizing Dose
• Thick field oxide charging – edge leakage in NMOS– loss of isolation NWELL-
NWELL• 0.13µm IBM technology harder
than 0.25µm:– non-ELT devices OK except
narrow/thick-oxide NMOS– to 70 Mrad+– lateral isolation OK, no guard
ring needed– SEU cross section higher,
lower critical charge– No radiation-induced SEL
ELT
non-ELT
F. Faccio, CERN, FEE Workshop Snowmass 2003
Detector Development Symposium Paul O'Connor BNL April 5, 2006 25
FabFab Line CostLine Cost
http://www.tf.uni-kiel.de/matwis/amat/elmat_en/index.htmlElectronic Materials© Prof. Dr. Helmut FöllUniversity of Kiel; Faculty of Engineering
$B
Detector Development Symposium Paul O'Connor BNL April 5, 2006 26
Mask Set and Fabrication CostMask Set and Fabrication Cost
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.5um 0.35um 0.25um 0.18um 0.13um
Technology
Cost/run ($M)
Cost/mm^2 ($)
Technology
0
0.2
0.4
0.6
0.8
MASK SET ENGINEERING RUN
$M
Detector Development Symposium Paul O'Connor BNL April 5, 2006 27
Example Example -- BNL ASIC Design BNL ASIC Design GroupGroup
• Develops Low Noise FEE for gas and semiconductor detectors– accelerator experiments– medical imaging– astrophysics– homeland security
• 70% Federal, 30% commercial• Staff
– 5 ASIC professionals– 1 DAQ engineer– 1 Technician
• CAD suite and test labs• 7 – 8 runs/year through MOSIS
– 0.35, 0.25, and 0.18um mixed-signal CMOS– 5K – 600K MOSFETs per chip– $20K – $80K per MPW run
• Development cost per design ~ $200K x no. revisions needed
Detector Development Symposium Paul O'Connor BNL April 5, 2006 28
SummarySummary• Power dissipation is the major obstacle to further CMOS
scaling.
• Foundry and mask costs going up as process options (multi-Vth, multi-tox, HBT, passives, etc.) added for SOC.
• Analog design is compromised by the low supply rail.
• Except for high radiation resistance and packing density, ultra-scaled technologies offer few advantages to justify their enormous cost.
• Plenty of opportunity to innovate in (n-4)-generation CMOS.
Detector Development Symposium Paul O'Connor BNL April 5, 2006 29
Bon AppBon Appéétit!tit!