All-magnetic analog associative memory

Associative memories and neurocomputing systems represent a massively parallel computing paradigm. Although extremely efficient for certain applications such as pattern recognition and image processing, such systems are not efficiently implemented using traditional CMOS design technology due to the required circuit complexity and corresponding power consumption to perform neural computations and represent programmable connections between artificial neurons. In this paper we propose to use newly-proposed mCell devices, which act as switchable non-volatile resistances, to implement an all-magnetic analog associative memory. We describe our building blocks for both artificial neurons and synapses using such magnetic devices. We verify the functionality of our proposed architecture using circuit simulation models of the mCell devices based on the first-order device physics.