A Resource-Efficient Weight Quantization and Mapping Method for Crossbar Arrays in ReRAM-based Computing-in-Memory Systems

Resistance Random Access Memory (ReRAM) crossbar arrays have been used in compute in-memory (CIM) application owing to its high bit-density, non-volatility, and capability to perform multiply-accumulate (MAC) calculations efficiently. The expansion of the size of the crossbars has led to the emerging challenge of high IR voltage drop and more complex logic control devices. In this paper, we propose two complementary weight quantization-mapping methods to solve this problem at high and low bits, respectively. The proposed method utilizes group quantization for clustering to merge weights in higher bits and leverages differential properties to conduct spectral clustering for merging weights in lower bits. Experimental results demonstrate significant savings in crossbar resources (45%) with minimal precision loss (less than 2%). Morover, We designed a carrier board-FPGA testing platform and deployed a neural network on a $32 \times 32$ size ReRAM crossbar. The results show that the proposed algorithm saves $\mathbf{4 2 \%}$ of units, and the recognition accuracy of the MNIST dataset is within an acceptable range (91.5% to 88.3%).