BTI Aging Influence and Mitigation in Neural Networks Oriented In-Memory Computing SRAMs

Static random access memories (SRAMs) are commonly used in In-Memory Computing (IMC) architectures that perform neural network related computations directly in the memory unit, overcoming the data transfer constrains of traditional Von-Neumann architectures. However, Bias Temperature Instability (BTI) aging is a growing reliability concern for nanometer technology SRAMs. In this paper, we examine the BTI aging influence on SRAM based IMC architectures and propose an aging mitigation method to enhance their reliability and lifetime. According to our simulation results, BTI aging can affect the computation reliability of IMC cells. Aging mitigation is achieved by periodically alternating active with inactive transistors and vice-versa in these cells.