An IMPLY-based Memristive Multiplier for Computing-in-Memory Systems with Weight-Stationary CNN Acceleration

Abstract

Adders and multipliers based on memristive Material Implication (IMPLY) logic are widely used in primary building blocks of Arithmetic Logic Unit (ALU). To solve the issue that the existing IMPLY-based multipliers cannot protect the input operands, this paper presents a novel data non-destructive memristive IMPLY-based semi-parallel multiplier for Computing-in-Memory (CIM) systems, by assigning function-specific memristors for data-protection and introducing additional switches for higher parallelism. Simulation results show that the proposed multiplier can achieve 30% faster than conventional semi-parallel design and 9.1 % less memristors against the state-of-art semi-serial design for 4-bit multiplication, while preventing the input weight from destruction as required by CNN weight reuse.