Neuromorphic Computing Enabled by Spin-Transfer Torque Devices

Abhronil Sengupta, Priyadarshini Panda, Anand Raghunathan, Kaushik Roy

Abstract: Neuromorphic computing offers immense possibilities in the development of self-learning, fault-tolerant, adaptive cognitive systems. However, the computing models are in complete contrast to the present sequential von-Neumann model of computation. Even custom analog/digital CMOS implementations of neural networks have been unable to achieve the ultra-low power and compact computing abilities of the human brain. In this tutorial, we review some of the neuromorphic computing models and demonstrate the manner in which spin-transfer torque effects in emerging spintronic devices can offer a direct mapping to such underlying neural computations.