The Sensible Machine concept started with an OSTP RFI response by Stan Williams and grew over six months into a IEEE-led community activity advocating the advancement of computing in the direction of learning machines. The talk will discuss some of this history to motivate a technical discussion.
The talk will explain emerging ideas that could facilitate the Sensible Machine direction, with this talk concentrating on the architectural end of the technology stack rather than materials and physics. Through Moore's Law, the semiconductor industry focused on line width reduction for microprocessors and memory. However, a Sensible Machine should have logic and memory integrated down to the device level for speed and energy efficiency. While numerical computing will remain important, a Sensible Machine will make extraordinary use of computational primitives for learning and should be optimized for those. Sensible Machines could possibly drive a new growth path for the computer industry, yet the advancement may be in applications becoming increasingly sophisticated essentially due to the computer auto-programming software rather than just more efficient Boolean logic. There is also a path to improving traditional supercomputing through learning and neural networks.
NSCI Committee
1:00 p.m. - 2:00 p.m. (Gaithersburg, Bldg. 221, Rm. B145)
11:00 a.m. - 12:00 p.m. (Boulder, VTC in 1 1107)
Slides from the presentation are now available.
Erik P. DeBenedictis (BIO)
Sandia National Laboratory