Characterizing and Modeling Radiation Effects in Neuromorphic Computing Paradigm
Abstract
Neuromorphic computing is emerging as a competitive computing paradigm for processing complex data-intensive tasks such as image classification and object recognition. A combination factors such as the end of Moore s law, the bottleneck between processor and memory (particularly for data intensive applications), and the maturity of new device technologies are leading to a resurgence of interest in neuromorphic hardware throughout the semiconductor industry. IBM s recent development of TrueNorth[1], a SRAM-based neuromorphic chip, exemplifies this type of paradigm shift. Advances in resistive memory (RRAM) technologies that can be integrated back-end-of-line (BEOL) in CMOS fabrication are also a driving force behind the resurgence. Hybrid neuromorphic architectures consisting of CMOS spiking neurons communicating with each other in massively parallel RRAM-based synaptic arrays offer the promise of reduced power consumption, increased density, as well as the potential for increased radiation hardness. In this program we propose to conduct fundamental research investigating radiation effects in the neuromorphic computing paradigm. We will characterize and model radiation effects on two neuromorphic platforms (CMOS SRAM based and beyond-CMOS RRAM based) using a hierarchical device-circuit-architecture co-design methodology
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 10, 2017
- Source ID
- HDTRA11710038
Entities
People
- Hugh Barnaby
Organizations
- Arizona State University
- Defense Threat Reduction Agency