Paper Title: ReRAM for Implementing Neural Network Synapses

Paper Abstract: ReRAM technology is widely considered the best candidate for implementing artificial synapses for neuromorphic computing. The movement of ionic species in ReRAM cells closely resembles mechanisms typical of biological synapses in the human brain. Furthermore, neuromorphic circuits demonstrate good resilience to the intrinsic device variability and noise typical of ReRAM. A CEA/Leti test vehicle uses SiOx resistive memories from Weebit in a neuromorphic architecture to demonstrate a spiking neural network (SNN) composed of analog neurons and resistive synapses. The chosen topology is a fully-connected single-layer SNN, which can handle up to 144-dimensional input vectors and 10 output classes. The designed-integrate and fires neurons allow the use of existing supervised offline learning tools. Digit recognition using this SNN on the MNIST database will be demonstrated.

Paper Author: Amir Regev, CTO, Weebit-nano

Author Bio: Amir regev is the Chief Technology Officer (CTO) of Weebit-nano and one of its technology founders and leader. Amir career exceeds two decades of experience in device and technology for the semiconductor industry, with particular expertise in Emerging memory and Flash memory development. Prior to establishing Weebit-Nano, Amir took senior Engineering roles leading technology companies in the semiconductor and memory field, including Intel, SanDisk Micron and Marvell accumulating broad knowledge and expertise in multiple engineering fields including Device, Technology Development, Quality & Reliability, and ASIC R&D. Amir was involved in developing the most advanced 45nm NOR Flash technology to date. Amir holds a M.Sc. in Electrical Engineering from Tel-Aviv University and a B.Sc. in Material Science and Engineering from Ben-Gurion University. In 2015 Amir graduated with a B.A. honors in Psychology from the Open University of Israel.