CISE-ANR: FET: Small: Hybrid Stochastic Tunnel Junction Circuits for Optimization and Inference

University Of Maryland, College Park, College Park MD

Investigators

Abstract

Neuroscience research shows that pervasive randomness in brains is fundamental to their stability and computational ability. This observation inspires probabilistic models that are useful for a variety of learning and optimization tasks. Conventional computers are not well suited to solving such problems because they are fundamentally deterministic. In this work, the researchers propose to develop probabilistic unit cells by augmenting commercial computer chips with thermally unstable magnetic devices that naturally exhibit probabilistic behavior. Distributed networks of such devices will enable emulating and accelerating powerful stochastic computational models. Reverse engineering the brain is one of the major challenges of the 21st century. Such an endeavor will undoubtedly affect the way computation is understood. This proposal, inspired by a probabilistic interpretation of neural activity will develop a hybrid probabilistic technology as a prototype to efficiently transfer this insight into a tangible technology and then to the broader community. The research will require contributions of a diverse international team from a variety of fields such as material science, device physics, electrical engineering, and computer science. The results of this research will be disseminated in the form of publications, presentations, short pedagogical YouTube videos in various languages, and lab tours for the general public. Neuroscience research shows that pervasive randomness in brains is fundamental to their stability and computational ability. This observation inspires probabilistic models that are useful for a variety of learning and optimization tasks. Conventional computers are not well suited to solving such problems because they are fundamentally deterministic. In this work, the researchers propose to develop probabilistic unit cells by augmenting commercial computer chips with thermally unstable magnetic devices that naturally exhibit probabilistic behavior. Distributed networks of such devices will enable emulating and accelerating powerful stochastic computational models. Reverse engineering the brain is one of the major challenges of the 21st century. Such an endeavor will undoubtedly affect the way computation is understood. This proposal, inspired by a probabilistic interpretation of neural activity will develop a hybrid probabilistic technology as a prototype to efficiently transfer this insight into a tangible technology and then to the broader community. The research will require contributions of a diverse international team from a variety of fields such as material science, device physics, electrical engineering, and computer science. The results of this research will be disseminated in the form of publications, presentations, short pedagogical YouTube videos in various languages, and lab tours for the general public. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.