Impact of Non-Idealities on Computational Networks built from Low Energy-Barrier Magnet based Stochastic Neurons
Abstract
Emerging nano-materials technology, in particular low-energy barrier magnet based devices, promises to enable large scale energy-efficient neuromorphic and probabilistic computing hardware for ``modern'' applications such as optimization, machine learning, Bayesian decision networks, and noisy intermediate scale quantum (NISQ) devices. Simulation based analyses and initial hardware demonstrations have been highly encouraging. Therefore it is prudent to understand the impact of non-idealities that inevitably plague such nascent technologies. We focus the impact of non-idealities of such low-barrier magnets on circuit operations in terms of accuracy and reliability with energy-delay tradeoffs. It is hoped that this analysis helps in understanding the challenges ahead and help develop novel mitigation strategies.
- Publication:
-
APS March Meeting Abstracts
- Pub Date:
- 2021
- Bibcode:
- 2021APS..MARU71206G