Regensburg 2019 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
SYCC: Symposium Identifying Optimal Physical Implementations for beyond von Neumann Computing Concepts
SYCC 1: Identifying optimal physical implementations of non-conventional computing
SYCC 1.4: Invited Talk
Friday, April 5, 2019, 11:15–11:45, H1
Artifcial Intelligence and beyond von Neumann architectures, a mutual opportunity — •Mirko Prezioso1,2, Farnood Merrikh Bayat1,2, and Dmitri Strukov2 — 1Mentium Technologies Inc, 3448 Elings Hall, University of California at Santa Barbara, Santa Barbara, California, 92106, USA — 2Department of Electrical and Computer Engineering, University of California at Santa Barbara, Santa Barbara, California, 93106, USA
The revolution in artificial intelligence was triggered not by any significant algorithm breakthrough, but by the availability of more powerful GPU hardware. Since then, more powerful dedicated digital systems have been developed but their speed and energy efficiency are still insufficient for more ambitious cognitive tasks. The main reason is that the use of conventional paradigm based on von Neumann architectures and digital mode operations for the implementation of neuromorphic networks, with their high parallelism and noise/variability tolerance, is inherently unnatural. Conversely, the performances can be dramatically improved using in-memory computing architectures based on analog computation. In this way, the key operation, the vector-by-matrix multiplication, is implemented on the physical level by utilization of the fundamental Ohm and Kirchhoff laws. I will discuss the recent progress of UC Santa Barbara group for the development of such analog and mixed-signal neuromorphic networks based on metal-oxide memristors. Additionally, I will discuss a more near-term solutions based on floating gate memory devices, which are being developed at Mentium Technologies, a spin-off of UCSB.