Electrical Engineering Seminar Series

Speaker Name: 
Dmitri Strukov
Speaker Title: 
Associate Professor ECE
Speaker Organization: 
UC Santa Barbara
Start Time: 
Monday, November 13, 2017 - 10:40am
Location: 
E2 - 192

TALK TITLE: Emerging Memories for Neurocomputing

Abstract: The present-day revolution in deep learning was triggered not by any significant algorithm breakthrough, but by the use of more powerful GPU hardware. Though this revolution has stimulated the development of even more powerful dedicated digital circuits, their speed and energy efficiency are still inadequate for more ambitious cognitive tasks and/or systems with severely limited power budget. On the other hand, the network performance may be dramatically improved using mixed-signal integrated circuits based on emerging nonvolatile memories, where the key inference-stage operation, the vector-by- matrix multiplication, is implemented on the physical level by utilization of the fundamental physical laws.

In my talk I will review the recent progress of such mixed-signal neuromorphic networks based on two types of memory technologies – floating-gate memories and memristive arrays (also known as ReRAM). In particular, a minor modification of a highly optimized embedded NOR flash memory has already enabled a successful demonstration of the first 180-nm medium-scale network for pattern classification. The experimentally measured delay and energy dissipation per inference were at least three orders of magnitude better than those reported for digital implementation of the same task, with a similar fidelity, using the 28-nm IBM’s TrueNorth chip. The detailed estimates show that the transfer to the similar 55 nm technology will allow the implementation of much larger networks, keeping a similar performance lead over the most prospective digital networks. Another way toward further scaling down of the mixed-signal neuromorphic networks is provided by novel nonvolatile two-terminal devices - memristors, which may have a sub-10- nm footprint, and are suitable for 3D integration. My group has developed a new technology of fabrication of these devices, sufficiently reproducible to demonstrate the first simple integrated neuromorphic networks.

Bio: Dmitri (“Dima”) Strukov is a Professor of Electrical and Computer Engineering at University of California at Santa Barbara. Prior to joining UCSB, Dmitri worked as a postdoctoral associate at Hewlett Packard Laboratories (Jan. 2007 – Jun. 2009) on various aspects of nanoelectronic systems. He received a MS in applied physics and mathematics from the Moscow Institute of Physics and Technology in 1999 and a PhD in electrical engineering from Stony Brook University in New York in 2006. He is a member of ACM, MRS, and IEEE societies. Dmitri’s research broadly concerns different aspects of computation, in particular addressing questions on how to efficiently perform computation on various levels of abstraction. His current research focus is on hardware implementations of artificial neural networks with emerging memory devices.