Heterogeneous memristive crossbar for in-memory computing

Abstract

It's been quite a while since scientists are seeking for the ancestor of von Neumann computing architecture. Among the most promising candidates, memristor demonstrates advantageous characteristics, which open new pathways for the exploration of advanced computing paradigms. In this work we propose the design of a novel crossbar geometry, which is heterogeneous in terms of its cross-point devices, allowing for the realization of true in-memory digital logic computations. More specifically, it is a combination of two stacked crossbar arrays with a shared intermediate nanowire layer. The variety of available cross-points types allows the execution of parallel memristive logic computations, where the logic state variable is voltage. Moreover, the utilization of insulating patterns in the crossbar arrays, at the expense of a small area-overhead, permits the simultaneous parallel read/write memory operation of two memory words. Memory/logic operation is determined through control signals driven from the peripheral CMOS-based driving circuitry, which also comprises row/column decoders, tri-state drivers, and summing/sense amplifiers to allow for the proper programming/reading of the memristive cross-points.