Tinne-Techniques for Innovative Nanoscale Neuromorphic Electronics

Mircea Stan

Electrical & Computer Engineering Department, School of Engineering and Applied Science, UVa

 William Levy

Department of Neurological Surgery, School of Medicine UVa

This interdisciplinary project will investigate several novel techniques for implementing nanoscale Neuromorphic electronic circuits and systems. The initial focus will be on high-level feasibility questions with more details and implementation aspects planned for future follow-up project proposals that will likely include other PIs as well. The motivation for the project comes from several directions: (1) the recent renewed interest in Neuromorphic systems at various funding agencies (such as the DARPA SyNAPSE program [Synapse], or the China-Brain program [China-Brain], etc.), (2) the impasse that regular systems have in achieving significant AI progress, (3) the promise of extreme low-power operation (such as pulse signaling and adiabatic operation) for some biologically-inspired methods, and (4) the interesting possible mapping of several novel nanoscale technologies (such as the newly “rediscovered” memristor by HP Labs) to Neuromorphic concepts.

The main ideas of this proposal are to:

• model Magnetic Tunnel Junctions (MTJs) as memristors and optimize the resulting figures of merit and explore the applicability to Neuromorphic applications using the new models,
• explore the possibility of nanoscale adiabatic circuit schemes for memristors in general, and fro MTJs in particular,
• explore Interpulse Interval Coding (ICC) schemes for extreme low power communication,
• combine the above concepts into a coherent Neuromorphic system concept that is scalable and requires low power consumption.