MAMI: Majority and Multi-Input Logic on Memristive Crossbar Array

Abstract

Memristive devices offer non-volatile storage capabilities, along with in-memory computation capabilities. Large scale passive crossbar arrays can be enabled by avoiding parasitic paths by means of a select device in series with a resistive switching device (1S1R). In this work, we utilize 1S1R Resistive RAM (ReRAM) arrays for realization of logic-in-memory operations, where each device is capable of performing the Boolean Majority operation with an input inverted. In contrast to the prior works, for the first time, we enable memristor-aided n-input OR operation in the same memristive array — some memristors initialized with data act as input and an additional memristor acts as output. By extending the reach of the underlying devices in terms of logic primitives, we boost the optimization of logical depth and device count. This improvement is demonstrated through a representative case study. This paves the way for new and improved memristive crossbar designs, compared to the state-of-the-art Majority only or NOR only memristive approaches.