Temperature-Dependent Accuracy Analysis and Resistance Temperature Correction in RRAM-Based In-Memory Computing

In this article, a precise electro-thermal modeling framework of resistive random access memory (RRAM) fabricated using a commercial 40 nm CMOS back end of line (BEOL) process is established to perform thermal analysis of RRAM-based in-memory computing (IMC). The weight matrix is mapped into the model and simulation framework to calculate the temperature considering RRAM inner operating heat and various external working conditions such as operating temperatures and environmental temperatures. The thermal impact on chip performance is systematically studied with dedicated programming schemes for various application scenarios. A resistance-temperature compensation (RTC) scheme is proposed to offset thermal-induced weight deviation in RRAM-based IMC systems. RTC can significantly improve the performance of IMC systems working under extreme temperature conditions.