Transcript Principles of Computer Architecture Dr. Mike Frank

Mesoscale Bulk Electronics
Beyond the MOSFET
• Mesoscale:
–
–
–
–
An intermediate scale, on the order of ~10 nm,
Materials have some properties of bulk material,
But surface effects are important,
And more quantum phenomena become important
• Bulk:
– Materials & structures fabricated using bulk
processes, w/o atomic precision
• Electronics:
– Electron states are used for primary informationprocessing operations
• not photons (optical), or whole atoms (mechanical)
What happens @ mesoscale?
• MOSFET scaling hampered by quantization of:
– charge:
• becomes important @ L  10 nm in all materials
– energy levels:
• important in semiconductors @ L  10 nm
• Can alternative device operating principles
exploit these quantization effects rather than be
hampered by them?
• Some approaches:
– Single-electron transistors
– Quantum wells / wires / dots, quantum-dot CAs
– Resonant tunneling diodes / transistors
“Coulomb blockade effect” SET
• Based on charge quantization
• # of energy levels may not be noticeably
quantized
• Comprised of an island of (typically) metal
surrounded by insulator.
• Narrow tunnel junctions to transistor
source/drain -  5-10 nm typical
• Gate controls # of electrons that may occupy
island
– within precision of 1, out of millions
Quantum Wells/Wires/Dots
• Usually use semiconductor material.
• Electron position is narrowly confined in 1, 2,
or 3 dimensions, respectively.
• E between distinct momentum states becomes
large
• In quantum dots, total # of mobile electrons
may be as small as 1!
• Non-transistorlike quantum-dot logics:
– Most notably, quantum dot “cellular automata” by
Notre Dame group
Quantized Energy Levels
• The narrower the space,
– the smaller the  gap between normal
modes #n and #n+1
– the larger the frequency & energy gap
between those modes
• More confined
spaces have wider /3, 3E
energy gaps between
their distinct
/2, 2E
momentum states.
, E
Resonant Tunneling Diodes
• Usually based on quantum wells or wires
– 1-2 effectively “classical” degrees of freedom
Source
Drain
Island (narrow bandgap)
Tunnel barriers (wide bandgap)
Electron tunnels
through barrier
Quantized momentum state
Electron flow
Occupied states in
conduction band
Energy
Unoccupied states
Resonant Tunneling Transistors
• Like RTDs, but an adjacent gate electrode helps
adjust the energy levels in the island
Gate
Source
Drain
Future Semiconductor Structures
•
•
•
•
•
•
•
•
SOI (Silicon-on-Insulator)
Band-engineered transistors
Vertical transistors
FinFETs (Chenming Hu group @ Berkeley)
Double-gate transistors (e.g. Philip Wong IBM)
Multi-layer chips (Lee @ Stanford)
“Quantum FET” analysis (Merkle ‘93)
atom-width wires (need ref)
Go through ITRS presentation
Nanoelectronics Technologies
• Scaled MOSFET structures - prev. slide
• Quantum wells/wires/dots - covered last time
– quantum dot cellular automata - go thru website
• Various “single-electron” devices - today
• “Spintronics” - electron (&/or nuclear?) spin
based electronics- today
• Molecular electronics - today or Friday
Quantum Dot Cellular Automata
• Wires: x vs +, fan-out, wire-crossing
• Speed: 2 ps/cell
– compare: light can go 0.6 mm in 2 ps
– ordinary electronic signals ~0.3 mm
– MOSFET gate delay according to ITRS ‘99:
• 11 ps in ‘99, 5.7 in ‘05, 2.4 in ‘14
• Gates: inverters, majority gates, full adder
• Paradigms:
– ground state computing
– clocked QCA pipelining (adiabatic, reversible)
• Molecular version: 20 fs/cell (100x smaller)
Spintronics
UF contacts:
Arthur Hebard,
Jeff Krause
• Cf. Das Sarma group at UF
• Info written into spin orientation of electrons
– persists for nanoseconds in conduction e’s
– compare ~10 fs lifetime for momentum decay
• Spin control, propagation along wires,
selection, & detection
• Datta-Das and Johnson spin-based transistors
• Potential medium for quantum computation
Molecular Electronics
• Tour wires
• Molecular switches
• Carbon nanotube devices
Helical Logic
• Proposal by Merkle & Drexler ‘96
• Do w. conductors & insulators only!
See plastic
transparencies,
readings
for details
– no fancy semiconductors, superconductors, or
tunnel junctions needed...
• The wires are the devices!
– Uses simple Coulombic repulsion
between electrons to do logic
• Scalable to single electrons & atom-wide wires!
• Externally clocked...
– by rotation of CPU within a fixed electrostatic field
• Can be used reversibly… 10-27 J, 1K, 10 GHz!
HL: Overall Physical Structure
• Consider a cylinder of sparse (high-permissivity) insulating material
(e.g., air), containing embedded helical coils of cold conductive
or semiconductive wire, rotating on its axis in a static, flat electric
field (or, unmoving in a rotating field).
• An excess of conduction electrons
will be attracted to regions on
wire closest to + field
direction.
• These electron packets
follow the field along as it
rotates relative to the
cylinder.
• Next slide: Logic!
Switch gate operation: 1 of 3
Data
wire
Condition
wire
Switch gate operation: 2 of 3
Data
wire
Coulombic
repulsion
Condition
wire
Switch gate operation: 3 of 3
Data
wire
Condition
wire
Nano-mechanical logics
See plastic
transparencies,
readings
for details
• First proposed by Drexler, 1992 (& earlier)
– Typically, very low leakage!
• due to high energy barriers (mechanical rigidity) in
interactions involving bonded atoms, vs. just electrons
– Pretty fast due to small size, but probably...
• ~1000’s × slower than molecular electronics might be
– basically, because atoms are ~1000’s × heavier than electrons
• Drexler’s logic of rods, cams, springs
– Molecular scale components
• Covalently bonded, atomically precise
• Merkle’s (1993) “buckling” logic
– No sliding-contact interfaces
– Scalable from macroscale to mesoscale
Also see
Smith’s
planar
mechanical
logics
Molecular Electronics
• Tour wires
• Various molecular switches
• Various carbon nanotube devices
See plastic
transparencies,
readings
for details
• Potential problem areas:
– High resistance of existing molecular devices.
– Maintaining thermal reliability in face of low node
capacitances and voltages.
– High leakage currents, due to tunneling or thermal
excitation over small, narrow barriers.
Biochemical computing
• Selected points on DNA computing:
– Adleman’s experiment
– Cyclic Mixture Mutagenesis
• Reversible DNA Turing Machines
See
readings
for details
– Seeman’s self-assembling structures
– Winfree’s tile self-assembly logics
• DNA computing has many disadvantages:
– High cost of materials
– Slowness of diffusive molecular interactions
– Slowness/cost/unreliability of lab steps
• Prob. won’t ever be a cost-effective computing
paradigm (except maybe for in vivo apps)
Optical computing
• Not viable at the nanoscale anytime soon!
– Due to entropy density issues mentioned earlier:
• High enough info. flux requires extremely energetic
photons, with too-high effective temperatures
• Or, waveguides considerably smaller than photon
wavelengths - EMF theory suggests: Impossible!
• All-optical computing requires nonlinear
interactions, between photons & materials.
• Optics (or more generally, EMF waves) will
remain useful for communications, but only:
– in contexts where extreme bandwidth density is not
required (or extreme temperatures can be tolerated)