Spintronic Stochastic Dataflow Computing: Material Exploration
Abstract
The proposers, UCLA, are currently a performer under the DARPA ERI FRANC program. As part of this program, the UCLA team is developing spintronics devices such as voltage controlled magnetic tunneling junctions, CMOS circuits that employ these devices as non-volatile magnetic memory elements and true random number generators in novel computing architectures that improve the performance and energy efficiency of big data and other machine learning applications.Spintronics based magnetic memory promises to be non-volatile memory with lowest access energy, smallest cost, and more importantly, will allow integration with standard CMOS. The novel stochastic data computing architecture enables massively parallel compute that can greatly reduce the memory access requirements of modern big data applications. Together and individually, these approaches will have a considerable and beneficial impact on resolving the so called memory bottleneck that is the central challenge in modern computing systems.This proposal outlines a carefully crafted plan to collaborate with the Lawrence Berkeley National Laboratories (LBNL) with the goal of complementing and augmenting the efforts of the UCLA FRANC program. In particular, the proposal will take advantage of the excellent and varied capabilities of LBNL in modeling, simulating, and experimenting on new materials, device stacks, and in emulating computing systems on their high-performance computing (HPC) clusters based on CPUs and GPUs. The device and material modeling work will allow the UCLA team to identify and construct newer spintronic device stacks both for use in the FRANC program and a variety of applications of potential interest to DARPA and ONR. The HPC based modeling, simulation, andemulation work will allow the UCLA team to explore new neural network architecture based on stochastic computing, especially, in how to train them.The proposed work spans over two years, and budgets primarily for support of UCLA researcher support and materials costs. Costs that may be incurred by LBNL are not included as part of this proposal. The project will be supervised by Profs. Sudhakar Pamarti (PI), Kang Wang, and Puneet Gupta.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2020
- Source ID
- N000142012823
Entities
People
- Sudhakar Pamarti
Organizations
- Office of Naval Research
- United States Navy
- University of California, Los Angeles