January 03, 2019: Seminar by Avik Ghosh

Information processing in the post Moore’s Law era

Avik Ghosh
Dept of Electrical and Computer Engineering
Dept of Physics
University of Virginia
Date   : January 03, 2019
Time   : 15:00 - 16:00 HRS
Venue  : ECSU Seminar Room, 9th Floor, S. N. Bose Bhavan
The 20th century belonged rightfully to the silicon transistor, whose dimensional reduction over several decades offered sustained economic growth in the high tech industry. As Moore’s Law started its inevitable slow-down, there was increased focus on energy efficient hardware, exploring novel materials and devices such as layered two-dimensional systems, complex oxides, tunnel devices and spintronics. Having finally reached the end of device scaling, the emphasis has now shifted to new computing paradigms such as offered by quantum information processing, machine learning and artificial intelligence. In other words, novel software, as well as software inspired application-specific hardware.
On the hardware end, we still have some room to push against fundamental physical restrictions using clever design. I will briefly introduce some of these efforts – bypassing the Boltzmann limit on switching voltage with energy filtering and feedback coupled networks, bypassing the super-paramagnetic limit on memory with topological excitations, bypassing single photon and single molecule sensing limits using innovative band engineering. On the software end, however, the field is wide open.  One example is brain inspired in-memory computing and signal processing, such as using stochastic neurons. I will argue that noisy binary probabilistic bits (‘p’-bits), fabricable with present day technology, can act as a poor man’s qubit for solving NP-hard problems. Furthermore, introducing an explicit time variable with analog stochastic neurons can allow real time learning, tracking and prediction. Thus through a judicious design of software inspired hardware, we may still be able to sustain the broad economic core of Moore’s law, namely scaling the number of functions per dollar.
Biosketch: Avik Ghosh is Professor at the Charles Brown Dept of Electrical and Computing Engineering and the Dept of Physics at the University of Virginia. He did his PhD in condensed matter theory at the Ohio State University, and a postdoctoral fellowship in Electrical Engineering at Purdue University.  He is the UVA site-director of the NSF-Industry University Cooperative Center on Multifunctional Integrated Systems Technology (MIST). Ghosh has authord 125+ refereed papers and a book (“Nanoelectronics – a Molecular View”, World Scientific) in the area of computational nano-materials and devices. He has given over 125 invited lectures worldwide. He is Fellow of the Institute of Physics (IOP), senior member of the IEEE, and has received the IBM Faculty Award, the NSF CAREER Award, a 2006 best paper award from the Army Research Office, and UVA’s All University Teaching Award. His group’s work with Columbia University on negative index behavior in graphene was voted by the editors of Physics World as one of the top10 breakthroughs of 2016.
All are cordially invited.
Dipti Prasad Mukherjee
Electronics and Communication Sciences Unit