Self-Boosting Effect-based Electrochemical RAM with a-IGZO TFT for Highly Energy-Efficient Neural Network Computation

We present a self-boosting effect-based electrochemical random-access memory (SB-ECRAM), composed of an amorphous indium gallium zinc oxide thin-film transistor (a-IGZO TFT) and an electrochemical random-access memory (ECRAM) connected in series, which demonstrates significantly enhanced performance. With a 8-fold increase in programming characteristics and 483-fold improvement in update energy efficiency compared to standard ECRAM, the SB-ECRAM effectively reduces noise in vector-matrix multiplication (VMM) and achieves exceptional half-bias selectivity for the vector-vector outer product (VVOP) in neural networks. Moreover, training simulations on the MNIST dataset indicated a test accuracy of 97.39%, which is comparable to the floating-point baseline.