Ultra-Low power neuromorphic computing with spin-torque devices

Abstract

Emerging spin transfer torque (ST) devices such as lateral spin valves and domain wall magnets may lead to ultra-low-voltage, current-mode, spin-torque switches that can offer attractive computing capabilities, beyond digital switches. This paper reviews our work on ST-based non-Boolean data-processing applications, like neural-networks, which involve analog processing. Integration of such spin-torque devices with charge-based devices like CMOS can lead to highly energy-efficient information processing hardware for applicatons like pattern-matching, neuromorphic-computing, image-processing and data-conversion. Simulation results for analog image processing and associative computing has shown the possibility of ~100X improvement in energy efficiency as compared to a 15nm CMOS ASIC.