In-memory Computing on Memristor Crossbar: Architecture and Algorithms

Abstract

Frequent data transfers between memory and the processor in typical von Neumann computers lead to increased power consumption and performance degradation. These problems are addressed by non-von Neumann architecture that uses in-memory computing to reduce the need for data transfer by carrying out computations directly within the memory. Memristorbased non-volatile memory, which enables effective data processing and storage, is a key component in this evolution. This article discusses instruction-driven in-memory set operations, inmemory comparisons of integers and floating point numbers, and the impact of sneak-path on the memristor crossbar. It is observed that in-memory operations require less energy than CPU-based operations at a cost of minimal delay.