3. Single Molecule Chips
• Chip constructed atom by atom of carefully
arranged atoms
• Missing atoms in lattice determine logic
• Some designs use the full molecule for
computation, some only surface level
• 1000 of these fit inside conventional circuit
• Much lower power per function
4. Nvidia 10 teraflops processor
• Graphics chip enabling exascale computing
– Very handy for brain emulation among other things
– Targeted for 2018 completion
• graphics core that can process a floating point
operation using just 10 picojoules of power
– Picojoules in current Fermi chips
– 1000 cores
5. Nanoscale light sensors
• Critical to optical interconnect of nanoscale
computational components
• Connection to atom scale components
– Single molecules, quantum dot interconnect
• Device < 4 nm wide
• Can be tuned electrically to different color
sensitivity
– Tie this to per neuron colored light signalling?
6. Racetrack memory
• Shock proof permanent storage 100,000 times
faster than hard disks
• Contains millions to billions of nanowires
• Racetrack-equipped computers would boot up
instantly
• Nearly 3 orders of magnitude less power use
than RAM, much less hard disk
8. Graphene Transistors
• Multi-transistor functionality possible per
transitor
• 10-10000 times faster than best silicon
transistors
• Behavior of graphene transistor can be
changed on the fly
– Triple mode N, P, and neutral
• Enables operation as amplifier changing frequency and
phase of signal
9. Quantum Dot Memory
• 1 TB per cm2
– Extremely high density
• Non-volatile
– Longer life (MTMF) than HD
• Up to 1 million years!!
• > 100 faster than DRAM
– < 10 ns
• Nanotube memory is contender
– But apparent manufacturing problems of high yield sufficiently
pure nanotubes
– Stable in vacuum and in presence of high ionizing radiation
• (great for space based systems)
10. Multicore -> Many Core
• 2 – 4 cores common
• 10 – 12 in pipeline
• Tilera has 100-core processor
– 546 gigabit per second performance
• Huge software problem
– New paradigm of programming?
– Automated conversion code?
– May need new operating system tech
11.
12. 3D chips
• Multiple layers of circuitry vertically interconnected into
single chip
– Vs 3D packaging which does not integrate into single device
• Heat dissipation issues
– requires forced fluid cooling in denser circuits
• Benefits
– Smaller footprint
– Speed due to short wire lengths
– 10 to 100 times less power
– Potentially cheaper fabrication
– Higher connectivity
– Heterogeneous integration
EETimes - Nvidia's chief scientist gave attendees at Supercomputing 2010 a sneak peak of a future graphics chip that will power an exascale computer.