Science & Engineering gg - Cornell...
-
Upload
nguyencong -
Category
Documents
-
view
218 -
download
0
Transcript of Science & Engineering gg - Cornell...
Nanoscale,,
Science & Engineering
& Our Place in the World
g g
Sandip Tiwari
Science -> Engineering -> Society
Science Engineering Society
16 1 00’…
16-1700’s
Gottfried Leibnitz Isaac Newton Thomas Savery
James Watt…
17-1800’s
Charles Darwin Dmitri Mandeleev
F it H bFritz Haber
18-1900’s
Albert Einstein Alexander Claude
Keck_Howard – Nov 6, 2007
19-2000’s??
Albert Einstein AlexanderFleming
Claude Shannon
Our Life Expectancy + Information Society
Information Revolution
TechnologyGiant Magnetoresistance – 1990s
Single-electron and Quantum-EffectMemories – 1990s
Q W ll L 1980
ComputingBusiness and scientific computers, personal computers, optical and magnetic storage,
Quantum-Well Lasers – 1980sLarge Scale Computation – 1970s
Hetero-Semiconductor Laser – 1970sLaser – 1960’s
Integrated Circuit – 1950’s
p , p g g ,liquid crystal displaysCommunicationsInformation superhighway, cellular phones, satellite communications, high capacity fib ti blIntegrated Circuit 1950 s
Transistor – 1947Vacuum Tubes – 1910’s
Telephony - 1876Science
fiber optic cablesSecurityAdvances in communications, command and control, sensors, weapon systems, mobile electronics
Correlated Electron States – 1980sScanning Tunneling Microscopy – 1980s
Single-electron effects – 1960sSemiconductor Tunneling – 1950s
BCS Th f S d ti it 1957
mobile electronicsEnergyPhotovoltaics, sensors, light-weight motors, high performance transformersTransportation
BCS Theory of Superconductivity – 1957Electron Microscopy – 1930
Electronic States in Crystals – 1920Wave Nature of the Electron – 1927
Quantum Mechanics – 1920s
pAutomotive electronics, sensors, avionics, air-traffic controlEntertainmentConsumer electronics
Keck_Howard – Nov 6, 2007
Quantum Mechanics 1920sX-Ray Diffraction – 1911Magnetoresistance - 1856
MedicineLasers, medical imaging, sensors
Based on National Academy Report
A Recent Timeline for Computing
?IBM 360
1964Micro/PC/Risc
1981?
Intel 4004
1971Cellphones, iPods, …IBM 360 c o/ C/ sc
IBM 801
Intel 40040.3 cm
0.4
cm
Cellphones, iPods, …Low power
Signal processingLeverage from integration
Software-hardware optimization
2300 Tx
0
Parallelized SupercomputersMulti-processor chips
Network switches
Semiconductors
Altair 8000 using intel 8080
2300 Tx10000 nm
~ENIAC(18000 tubes)
Network switchesReduced power-embedded memory
IntegrationSemiconductors
Reliability
Programming: Virtual machine and
IntegrationSystem Compaction
Integration
An Inflection Point(?) arising from
underlying science
Keck_Howard – Nov 6, 2007
Virtual machine and common software Reduced Complexity
in Programming-Hardware
eg a o
Scales, Energies & Small
Power Densities
102
A300
ConcordeF15
PhantomTurbines
)
cm2 )
10
747707
Phantom
cal
(kW
/kg
High end processors
ssin
g
(W/c
Air-Cooled Silicon
1 LibertyCyclone
or
Mec
han
ic
Low Power Processors nfo
. P
roce
s Limit
Propellers
10-1Wright
M b h
Den
sity
fo Low Power Processors
nsi
ty f
or
In
10-2
Otto
Maybach
PistonsPo
wer
Po
wer
Den
Keck_Howard – Nov 6, 2007
1880 1920 1960 200010-3
Source for mechanical engines: V. Smil (1991)
Scales
Physical LifeLarge
(Integration of many into a system)
Atoms, Molecules, Chemical Reactions,
Life forms,Weather, Environment, …
Fractal FormsAdaptation
…
Chaos??
D i i i N d i i i
Anthropic
Deterministic Non-deterministic
Small
Keck_Howard – Nov 6, 2007
Small
(Nano; small materials, structures and devices)
Simple rules connect the scales
The Sinking of RMS Titanic!A Nanoscale Effect
Continuum
MesoScale
100
esoSca e
10-3
Atomistic
SemiEmpirical
10-6
Sca
le (
s)
Ab I iti
10-9
Tim
e S
Ab-Initio10-12
HMS Titanic
Keck_Howard – Nov 6, 2007
10-10 10-9 10-8 10-7 10-6 10-5 10-4 10-3 10-2 10-1 1 101 102 103
10-15
Length Scale (m)
Complex Oxides
Fe2O3 (ironstone) and Fe3O4 (magnetite)Found in cells and believed responsible for navigationMinimum units – domains stable – large sensitivity
La2/3Ca1/3MnO3, manganites, have large magnetoresistance
La1-xSrxCuO4, YBa2Cu3O7 or Bi2Sr2SaCu2O8 are high Tc
d
Magnetospirillum magnetotacticum
(Frankel)
superconductors
Some the strongest magnets are not metallic, but ceramic oxide insulators
NiO, CoO, MnO – antiferromagnetic oxides - insulators (naively expected to be conducting)
Va2O3 or Ti2O3 show a metal-insulator transition below characteristic temperature TMI
Even after multiple Nobel prizes in areas related to phase transitions, we can not predict these connections of scale well – how do single entities interact with ensembles.
Keck_Howard – Nov 6, 2007
Delocalized d-electrons of transition metals (e.g. Cu, Ni, Fe, Ru etc.) strongly coupling to oxygen p-orbitals. The enormous diversity of complex oxides arises from a delicate balance between de-localized electrons and effects of Coulomb interaction which localize electrons
Mott Hubbard transitions
Connections of Scale
Traveling particles: 2001 NASA satellite image of dust arriving in California from Asian deserts. SeaWiFS Project, NASA/Goddard Space Flight Center and
Traveling particles from recent (Oct-Nov , 2007) California fires leaving for China.
Keck_Howard – Nov 6, 2007
NASA/Goddard Space Flight Center and ORBIMAGE
2007) California fires leaving for China.
So, What is New in Nano?
We now have the ability to construct by synthesizing from atoms up (bottom-up) and pattern down from larger assemblies (top-down) We have sophisticated tools to characterize the nanoscaledown). We have sophisticated tools to characterize the nanoscale.
We can explore, characterize, and utilize phenomena across a breadth of the disciplines.
Nanoscale has brought together approaches of different disciplines - physical and life sciences, and engineering - at the length scale of the building blocks of the physical and living worldlength scale of the building blocks of the physical and living world.
POWERFUL INTERSECTION OF NATURE, SCIENCE & ENGINEERINGBut, So are the Difficulties with Complexities of Scale
The opportunities of understanding and in turn benefiting humanity by harnessing these capabilities is enormous
The scientific and engineering potential is vast
Keck_Howard – Nov 6, 2007
The scientific and engineering potential is vast.
The interdisciplinary research and applications are exciting.
Electronics
PhysicalBehavioral
Structural
ContinuumHi h Ab t ti (HDL)
100
Chip Floor Planning
PlacementTiming
C
High Abstraction (HDL)
10-3
Compartmentalization
DeviceBoltzmann
CompactASX, Spice, …
Lumped
CongestionRouting
VerificationTemperature
P
10-6
Sca
le (
s) A good way to lower cost and efficient and robust designs when within parameter band
HydrodynamicDrift-DiffusionMonte CarloAtomistic
InterconnectNoise
PowerElectromagnetic10-9
Tim
e S parameter band
What happens when at “allowed” limits?
QuantumMaster Eq.NEG
10-12
Keck_Howard – Nov 6, 2007
10-10 10-9 10-8 10-7 10-6 10-5 10-4 10-3 10-2 10-1 1 101 102 103
Length Scale (m)
10-15
Power & Energy in Computing
Keck_Howard – Nov 6, 2007
Source: Roger Schmidt, IBM??
Variability
30% in frequency
5x in leakage
Keck_Howard – Nov 6, 2007Borkar (2007)
Process Variability – Energy at Nanoscale
Keck_Howard – Nov 6, 2007
Source: Dennard (2005)
Energy per Operation
10-12
Logic Operation
10-14J)
Memory Read 10000 e-
10-16
Ope
ratio
n (J
1000 e-
100 e-
10
These energy numbers are related to error rates and acceptable speeds
10-18
10 kT ln2nerg
y pe
r O 10 e- acceptable speeds
for cmos static
10-20
kT ln2
E
Limit in dissipative information processingBits deleted
inverter
Keck_Howard – Nov 6, 2007
10-22
1 10 100 Feature Size (nm)
Bits deleted
Stochastic Variance of Nanoscale
10 nm 5 nm 3nm 1 nm
3D B d Th h ld J ti C it3D: Bandgap, Threshold, Junctions, Capacitances, …
n ~28,000 ~3,500 ~750 ~28
σ(n) / n 0.6% 1.7% 3.7% 19%( )
2D: Inversion Charge, Interface Defects, Contacts, …
n 920 230 83 9
σ(n) / n 3.3% 6.6% 11% 33%
1D: Wire Resistance, Wire Defects, Molecular Contacts, …
n 30 15 9 3
Yield of chain: Y = (1 – pf)m where pf is failure property of each element
n 30 15 9 3
σ(n) / n 18% 26% 33% 57%
Keck_Howard – Nov 6, 2007
f f
From DOE Reports
Nearly 13% of electricity is used for computers and peripherals
39% of primary energy input goes towards generating
peripherals
goes towards generating electricity
69% of this electrical energy is69% of this electrical energy is lost in transmission and inefficiencies
72% of this electrical energy is generated by greenhouse emission processes
Keck_Howard – Nov 6, 2007
emission processesWe should care!We are responsible.
Edvard Munch
Energy, Heat & Work
STHG Δ−Δ≡ΔGibbs Free Energy Entropy
Enthalpypy
TH
Reversible isothermal expansion/isothermal heat addition
A B
Carnot Cycle
TC
H
Reversible adiabatic expansion
CD
Reversible adiabatic compression
SA SB
CDReversible isothermal compression
No engine operating between two heat reservoirs can be more efficient than a Carnot engine operating between the same reservoirs
Work Efficiency for Combustion Engine < 30% (typically ~25%)
Keck_Howard – Nov 6, 2007
y g ( yp y )
Work Efficiency for Information Engine too has Constraints
Energy: A Fundamental Constraint at Nanoscale
In any approach based on charge transport and change in electromagnetic fieldsPower /Energy dissipation and heat removal set the
αU
S. Tiwari et al., IEEE Conf. Em. Tech. (2006)
Power /Energy dissipation and heat removal set the fundamental constraints of system operationA
Q energy per operation
activity factor
Individual devicesLarge collection of devices
x-section area of heat removal
density of heat removal
time constant
Individual devicesQ =105 W/cm2
t ~ 5 ps
Large collection of devicesQ = 100 W/cm2
t ~ 5 ns
Error rates are related to energy
Minimum energy: U ~ 1000kT
Error rates are related to energyCMOS Inverter: pf ~ exp (-2VDD/σ(VT))Synthesis: variance ~ exp (-ΔE/kT)
ΔE ~ 0.1 eV => pf ~ 2x10-2
Keck_Howard – Nov 6, 2007
Energy/Power Dissipation Complexity/Adaptation
Energy Scales of Processing
Koyama et al. Nuzzo et al.Self Assembly: Oxide Growth Self Assembly: Molecular
ΔE = 0.1 to 0.5 eV (self-assembly)
Vladiviroma et al.
ΔE = 1.5 to 2.5 eV (oxide growth)
Probabilities proportional to exp (-ΔE/kT)Smaller activation energies lead to larger varianceShort range and long range order
Energy Error Rates
Keck_Howard – Nov 6, 2007
5 o 5 e (o de g o )ΔE > 2 eV for dopants used
Energy Error Rates0.1 eV ~2x10-2
0.5 eV ~5x10-9
1.5 eV ~1x10-26
Electronics at Massive Integration Using Nanoscale
Terascale integration capacityg p y
Billions will be unusable due to variations
General cores
Special cores
f ti l bl k
Many will fail in time
Intermittent failures
Interconnect fabric
functional blocks
TB/s memory bw Intermittent failures
A desired performance at power and cost still needed
low power buses
configurability
fi bilit
hierarchy
TB/s memory bw
Dynamically self-test, detect errors, reconfigure, & adapt
sensors for temp, time
adaptation
reconfigurability
Keck_Howard – Nov 6, 2007
Configurability: Defects and Testing
Defects have non-local effect
Enormous penalty arising from non-linear rate of connectivity with defects; affects congestion, power, area, ….
Keck_Howard – Nov 6, 2007
Digital Circuits: Configurability
A. Kumar et al., DFT, 280(2004)
Defects limit testability and limit the 80
100
ule
s te
ste
ddINT=10-2
N=1x106
p=0.5
yusable devices and interconnects
High yields, similar to current electronics are required
20
40
60
nta
ge
of m
odu dINT 10
dINT=10-3
dINT=10-4
dINT=10-5
electronics, are required
0 500 1000 1500 20000
Per
cen
Number of iterations
dINT=0
ion
60
8
12dINT = 0.001
ower
dis
sipa
ti
p=0.5 p=0.6 p=0.7
40
60
dINT = 0.001
e in
are
a
p=0.5 p=0.6 p=0.7
4
8
ncre
ase
in p
o
20
acto
r in
crea
se
Keck_Howard – Nov 6, 2007
104 105 106
0
Fac
tor
in
Number of modules104 105 106
0
Fa
Number of modules
Simulated Performance of Defect Tolerant Architectures
R-fold modular redundancyNAND multiplexingReconfiguration (using knowledge of faulty
uir
ed
knowledge of faulty devices)
ncy
Req
F 90%
Red
un
da For 90%
chipyield
R
Keck_Howard – Nov 6, 2007 Nikolic, Sadek and Forshaw
Allowable probability of failure1012 devices assumed
Adaptation
5.8 nm
Front Gate
Front OxideGateSiO2 SiO2GateSiO2 SiO2GateSiO2 SiO2
Self-aligned Back-Gate:
A structure where the
5.3 nm
7 0 nm
5.8 nmFront Oxide
Silicon
Back Oxide
Source DrainSi Channel
Back Gate
SiO2
Source DrainSi Channel
Back Gate
Source DrainSi Channel
Back Gate
SiO2
threshold voltage can be changed so that the device, circuit &
system can be 7.0 nm
Back Gate
Back Oxide
V 0 3 i
VBG
= 0 to -3 in
yadapted to needs.
1E-8
1E-6
Increasing VBG
VBG
= 0 to -3 in
steps of -0.5 V
nt, I
DS (
A)
1E-8
1E-6
Increasing VBG
ent,
I DS (
A)
BG
steps of -0.5 V
1E-12
1E-10
VDS
= 1.0 V
Dra
in C
urre
n
1E-12
1E-10
VDS
= 0.1V
Dra
in C
urre
Keck_Howard – Nov 6, 2007
-1.5 -1.0 -0.5 0.0 0.5 1.0
Front Gate Voltage, VFG
(V)
-1.5 -1.0 -0.5 0.0 0.5 1.0
Front Gate Voltage, VFG
(V)
Sub-Threshold Swing
L~ 1 μm
Keck_Howard – Nov 6, 2007
W. Chan et al., (2007)
Adaptive Imagers
computational sensors
tunableoptics
low latencyVladimir BrajovicIntrigue Technologies
processingsensing
“soft”computation
on-chipprocessing
computational sensors
sensoryadaptation
imageformationadaptation
algorithmicadaptation
LIi
Ij
Adaptive Pixel-to-Pixel “Switch”
top down adaptationLi
Li < Lj
Li Ii
top-down adaptation Lj
Li > Lj Lj
1Ω
1Ω
Keck_Howard – Nov 6, 2007
jΩ Ω
Vladimir Brajovic
Keck_Howard – Nov 6, 2007
Vladimir BrajovicIntrigue Technologies Local gain, offset adaptive compensation
Geometric Leverage: Operational Amplifier
VDD
CC
CLVIN- VIN+
VOUT
VSS
Supply 1.5V 0.6V
Threshold voltage change through buried bias
Supp y 5 0 6
Gain (dB) 108.9 99.8
Gain Bandwidth (MHz) 32.1 29.5
Open-Loop Gain Voltage
Keck_Howard – Nov 6, 2007
Open-Loop Gain Voltage Gain (V/V)
4.3x104 3.9x104
Phase Margin (degrees) 84 89.8 A. Kumar et al. (2006)
Environment of the Challenge
System-level OptimizationHierarchy
P i f T h l
Power In: ~100 W/cm2
Energy Error Rates, S/NProperties of Technology
AdaptationEnergy Partitioning
Network/Communication
Energy Error Rates, S/N
10 nm F 4x1013 F2 in 2.5 inch& more in multilayer 3D
Network/CommunicationDesign Tools
Hardware-Software
(reproducibility, resilience, robustness, routing, reconfiguration, …)
Keck_Howard – Nov 6, 2007
So, the Biggest Problems
Today’s challenge: Power and DesignAdapt
H d S ft d iHardware-Software co-design
Future challenge: Reliable Systems from Unreliable ComponentsVariations and reliability
Resiliency
CostNew technologies that are efficient at hardware-software co-designg g
Non-compartmentalized education and practice
Keck_Howard – Nov 6, 2007
Major Problems of Society
Efficient low cost sustainable photovoltaicsNumerous “nano” ideas – Gretzel cell, new organics, inexpensive silicon approaches multi bandgapssilicon approaches, multi-bandgaps, …
Efficient low energy sustainable technologiesNumerous “nano” ideas – Efficient catalysis, self-assembly techniques in fabrication, novel material synthesis, low power electronics, low power displays, …
HealthcareNumerous “nano” ideas – health diagnostics and repair, local treatments neural probes new characterization methods smallertreatments, neural probes, new characterization methods, smaller portable instruments, …
Keck_Howard – Nov 6, 2007
Hydrogen Use: Fuel Cells
H2 Electricity
Elimination of combustion and heat from energy conversion
Hydrogen provides up to 60% efficiency
Storage for 300 mile driving(capacity higher than liq. H2!
-
Pt
metal
better catalystshigher activity
Hydrogen provides up to 60% efficiency
++
+
e
H2
O2
metal
core-shell nanoparticle medium
high surface area+
+
++ +
H2O
O
shell : dissociation H2 ⇔ 2 H
H2 gas
OOSO3
-H+
SO3-
H+
OOSO3
-
H+SO3
-
H+ core: high density storage
nanoscale ⇒ fast transit time
~ 10 nm
Keck_Howard – Nov 6, 2007
better membraneshigher temperature
higher ion conductivity
designed nanoscale materials
World Map (Normalized to Population)
Keck_Howard – Nov 6, 2007
R. Webb, Nature 439, 800(2006)
A Narrow View
The Russians play Chess, plotting their moves with a strategy that looks decades into the future.strategy that looks decades into the future.
The Japanese play Go, systematically surrounding each technological territory with their pieces until they make ittechnological territory with their pieces until they make it their own.
The Europeans play Bridge, kicking a lot under the table while presenting a smooth performance above its surface.
USA: We play Monopoly.
The world does not know how China and India play
Keck_Howard – Nov 6, 2007
The world does not know how China and India play.
From Prof. Les Eastman’s Bulletin Board
From the 1980’s
This is more true now than everPursuit of science, engineering and technology
is more central to our world’s well being now gmore than ever.
We all need to do our part and contribute to collective
Keck_Howard – Nov 6, 2007
pprogress
NanoExpress
Mobile Nanotechnology Demonstration Laboratory
Developed at Howard UniversityDeveloped at Howard University
>4000 Visitors in ~ 6 months
S h l S t J dSchools, Scouts, Judges, Community …
Hands-On Equipment:a Photolithographya. Photolithographyb. AFMc. SEM/TEMd. Probe Statione FTIRe. FTIRf. Optical Microscope
Keck_Howard – Nov 6, 2007
Conclusions
We all need to work together
We need to think deeper
Small has large effects; there are connections of scaleWe are all individually an important contributor and cause of successes and failure of our society
Education
Research and Technological Innovation working together
Keck_Howard – Nov 6, 2007