Skip to main content
U.S. flag

An official website of the United States government

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

A System for Validating Resistive Neural Network Prototypes

Published

Author(s)

Brian Hoskins, Mitchell Fream, Matthew Daniels, Jonathan Goodwill, Advait Madhavan, Jabez J. McClelland, Osama Yousuf, Gina C. Adam, Wen Ma, Muqing Liu, Rasmus Madsen, Martin Lueker-Boden

Abstract

Building prototypes of heterogeneous hardware systems based on emerging electronic, magnetic, and photonic devices is an increasingly important area of research. On the face of it, the novel implementation of these systems, especially for online learning in artificial intelligence, poses new and unforeseen challenges in mixed signal data acquisition, hyperparameter optimization, and hardware co-processing. Depending on the type of device, unpredictable and stochastic behavior can also be significant obstacles, as well as poorly repeatable hysteretic effects or performance degradation. Dealing with these emerging device challenges on top of more traditional computer hardware problems, like quantization errors, timing constraints, and even simple hardware and software bugs is a enterprise fraught with many potential pitfalls and mistakes on the road to the development of reliable heterogeneous systems. Equally important to the construction of the physical prototype is the co-development and integration of a design verification framework that can extensibly allow for predictable and deterministic behavior of not only the entire system but, also, all of its constituent parts in a modular way, allowing for simulation and implementation to be seamlessly integrated. This work discusses Daffodil-lib, a Python based prototyping framework which, from hardware to software, enables everything from a simple script-based simulation to a compiled hardware-timed experiment, to everything in between with no syntactical changes for the end user. This system is designed to simulate and control a 20,000 device resistive memory array for neural network simulations and beyond while maintaining isomorphism to a realizable integrated circuit.
Proceedings Title
International Conference on Neuromorphic Systems 2021 (ICONS 2021)
Conference Dates
July 27-29, 2021
Conference Location
Knoxville, TN, US
Conference Title
International Conference on Neuromorphic Systems 2021

Keywords

Design Verification, Artificial Intelligence, Hardware for Machine Learning, Machine Learning, FPGA

Citation

Hoskins, B. , Fream, M. , Daniels, M. , Goodwill, J. , Madhavan, A. , McClelland, J. , Yousuf, O. , Adam, G. , Ma, W. , Liu, M. , Madsen, R. and Lueker-Boden, M. (2021), A System for Validating Resistive Neural Network Prototypes, International Conference on Neuromorphic Systems 2021 (ICONS 2021), Knoxville, TN, US, [online], https://doi.org/10.1145/3477145.3477260, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=932369 (Accessed December 11, 2024)

Issues

If you have any questions about this publication or are having problems accessing it, please contact reflib@nist.gov.

Created July 27, 2021, Updated November 29, 2022