Radiation Effects in Brain-Inspired Computing Systems
Abstract
This work explores the effects of radiation in digital and analog hardware implementations of brain-inspired computing systems. The IBM TrueNorth Neurosynaptic System was utilized as a baseline system to evaluate the effects of radiation on neural network applications emulated by von Neumann hardware. The IBM TrueNorth was irradiated using 4 MeV proton irradiation at the Vanderbilt Pelletron. The effects of Single-Event Upsets (SEUs) on network accuracy and performance were negligible, reducing the accuracy of an MNIST-trained Convolutional Neural Network (CNN) with trinary weights from98.82 to 98.78 after a 60 second exposure. Secondly, we also studied the effects of radiation on general analog in-mem
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 2021
- Accession Number
- AD1143545
Entities
People
- Brian D. Sierawski
- Jason Woo
- Michael Alles
- Subramanian Iyer
Organizations
- University of California, Los Angeles
- Vanderbilt University