Evelyn Hu (UCSB)
Download
Report
Transcript Evelyn Hu (UCSB)
IC-DFN Workshop
Hangzhou, August 15, 2006
Designing for Uncertainty: Critical Issues for New
Nanoelectronic Technologies
Evelyn L. Hu
California NanoSystems Institute
UCSB
UNCERTAINTIES
A Few Candidate Technologies
• Technologies for post-CMOS?
• Architectures for new technologies
–Too early to extract device parameters?
Seeds of an architecture
Plan large-scale systems?
with defect tolerance
– Uncertainties in device variability, failure modes
• Where are the sources of errors in system operation,
in fabrication?
Beginnings….
Are We Finally at the Limits?
www.intel.com
Challenges to Further
Scaling (CMOS)
• Fundamental Physical Limits?
•
•
•
Intel 45 nm Shuttle Test Chip
– Leakage through dielectric
– Small current drive, electron
statistics
– Thermal noise
– Interconnect delay
High densities -> severe power
dissipation
Complexities in design &
verification
Economic Limits: the cost of
reliable fabrication
The Next Technology Generation
Continued faster, better (functional), cheaper
•
Performance
–
–
–
–
•
Faster access time
Low power
Operates over wide temperatures
Non-volatile
Manufacturability
–
–
–
Robust process latitude
Low-cost fabrication
Scalability
CHOICES? Benchmarks?
Selecting a post-CMOS Technology:
choices
• Limitations of present (scaled) technology
– Cross talk
– Leakage
– Charge modulation: statistical limitations of dopants
– Maintaining high noise margins at reasonable temperatures
• Desired scalability
– Scale down in size: fabrication at the nanoscale
– Scale up in complexity
Increased alignment accuracy
Issues of interconnect delay
Selecting a post-CMOS Technology:
Choices
Single Electron
• Address limitations
Transistors
– Leakage
– Cross talk
Quantum Cellular Automata
– Charge modulation: statistical
Spintronics:
limitations of dopants
alternative
– Maintaining high noise
‘state variables’
margins at reasonable temperatures
Simplify
• Desired scalability
manufacturability
– Scale down in size: fabrication at
Molecular Electronics
the nanoscale
– Scale up in complexity
Carbon Nanotube
Increased alignment accuracy
Issues of interconnect delay
transistors
No one technology addresses ALL the challenges, or provide ALL
the desirable features for the next-generation technology
Limiting Leakage: Single
Electron Transistors
electron
EC = Q2/2C;
Q = charge,
C = capacitance
ETh ~ kT; EC >> kT
For a small enough ‘island’ and very small
capacitance, C, and for Ec >> Eth,
THERE IS AN ENERGY COST TO ADDING
OR REMOVING CHARGE FROM THE ISLAND
(no leakage)
Limiting Leakage: Single
Electron Transistors
Ea = single-electron addition energy
electron
For a small enough ‘island’ and very small
capacitance, C, and for
Ec >> Eth, THERE IS AN ENERGY COST
TO ADDING OR REMOVING CHARGE
FROM THE ISLAND (no leakage)
Likharev, Electronics Below 10 nm
Room temperature (25 meV) operation only
possible for island diameter ~ few nanometers
Taking Advantage of Crosstalk:
Quantum Cellular Automata
QCA Cell:
Quantum dots with excess charge
Charge distribution
from electrostatic repulsion
Addition of charge -> changed
charge distribution, cellular ‘state’;
‘0’
‘1’
Majority Gate Device
Truth Table
Amlani et al., Science 284, 280 [1999]
(Notre Dame)
Nearest-neighbor interaction,
Fabrication & scale-up challenging, room temperature
operation unlikely
local computation
Operation at Room Temperature:
Magnetic Cellular Automata
Applied magnetic field
Cowburn & Welland,
Science 287, 1466 [2000]
Input dot = ‘0’
Input dot = ‘1’
Larger magnetic quantum dots
(110 nm diameter),
Material: Ni80Fe14Mo3X
Propagation of information through
exchange interaction between dots
Elongated dot, injector
Using Spin to Transfer
Information
GaMnAs Digital Alloys
David Awschalom,
Art Gossard
UCSB
Spintronic Technology
Spin-based Devices
Ferromagnetic spin filter
Semiconductor
Nanoscale-engineered
materials
Detector (Quantum well)
New device concepts
Powerful
new information
technologies
Experiments have shown long spin coherence lifetimes, but…
Need to understand best material systems and device configurations
Mechanism of control: Magnetic (e.g. MCA) or electronic?
Selecting a post-CMOS Technology:
Choices
Single Electron
• Address limitations
Transistors
– Leakage
– Cross talk
Quantum Cellular Automata
– Charge modulation: statistical
Spintronics:
limitations of dopants
alternative
– Maintaining high noise
‘state variables’
margins at reasonable temperatures
• Desired scalability
– Scale down in size: fabrication at
the nanoscale
Challenges in device
– Scale up in complexity
Increased alignment accuracy
Issues of interconnect delay
fabrication profound
Limited architectures for
large scale systems
Incorporating natural nanoscale building blocks:
Carbon Nanotubes
Beautiful structural order in carbon
nanotubes
Exceptional electrical properties
A single carbon nanotube can be made into a
transistor
Chen et al., Science 311, 1735 [2006]
or a ring oscillator
Can dope single carbon nanotube
both n-type and p-type
Enhanced compactness, multifunctionality
Challenges: control of conductivity,
doping, assembly
Incorporating natural nanoscale building blocks:
molecular switches
wire
Apply electric
signal
wire
Stoddart & Heath, UCLA
With very dense nanowires (20 nm)
in cross-bar geometry
HP has taken these concepts to
larger-scale arrays, considered
architectures and defect tolerances
Developing the Molecular
Crossbar Platform
Wu et al., Applied Physics A
80, 1173 [2005]
Cross-bar configuration, with
molecular interlayer
34 x 34 cross-bar memory,
30 nm half-pitch, Ti/Pt wires
I-V of single device:
HYSTERETIC SWITCHING
On= 1.5 positive bias, top
electrode; Off = negative bias
ON/OFF ~ 10
HP has used this architecture
for memory (a) and logic (b)
(a) Chen et al., Nanotechnology14, 462 [2003]; (b) G. Snider, Applied Phys.A 80, 1165 [2005]
Sample of ‘defect-tolerant’ nano-architecture
Lay out demultiplexer
circuit on crossbar
geometry
Kuekes et al,
Appl. Phys.A 80, 1161 [2005]
address
signal
Defect
‘stuck open’
Demux circuit
(not defect-tolerant)
Defect-tolerant circuit
2-bit address passes through
CMOS encoder -> 3 bit encoded address
6-bit signal vector u -> redundant input
address
Sample of ‘defect-tolerant’ nano-architecture
Calculated percentage of
usable nanowires, versus
defect probability for
different levels of
redundancy, d (address
bit)
Summary and Beginnings
Technologies for post-CMOS?
A wide variety of candidates, at different levels of maturity
No one technology addresses ALL the challenges, or provide
ALL the desirable features for the next-generation
technology
Architectures for new technologies
Initial work on cross-bar geometry with molecular switches
Simple architectures, error-tolerant schemes provide
important benchmarks
Consideration of appropriate architectures CRITICAL (even
in the face of uncertainty) to help sort and direct progress
of technology
Selecting a post-CMOS Technology:
Choices
Single Electron
• Address limitations
Transistors
– Leakage
– Cross talk
Quantum Cellular Automata
– Charge modulation: statistical
Spintronics:
limitations of dopants
alternative
– Maintaining high noise
‘state variables’
margins at reasonable temperatures
Simplify
• Desired scalability
manufacturability
– Scale down in size: fabrication at
Molecular Electronics
the nanoscale
– Scale up in complexity
Carbon Nanotube
Increased alignment accuracy
Issues of interconnect delay
New Opportunities in Resilient, Manufacturable
Information Systems?
The Emergent Integrated Circuit of the Cell
Hanahan & Weinberg, Cell [2000]
transistors