HCMS: A Hybrid Conductance Modulation Scheme Based on Cell-to-Cell Z-Interference for 3D NAND Neuromorphic Computing

3D NAND flash memory can be selected as a promising candidate for the implementation of neuromorphic computing ascribe to high density, low write power and multi-level storage capability. Nevertheless, the broader range of application domains, such as advanced deep learning training and linear and partial differential equations solving, require a high degree of memory conductance precision. With the scaling of 3D NAND flash, the traditional modulation method of incremental step pulse programming (ISPP) suffers from accuracy and efficiency issues. In this letter, a novel hybrid conductance modulation scheme (HCMS) based on the cell-to-cell Z-interference effect is proposed to tackle this shortcoming by increasing Vpass bias. The presented scheme combines ISPP and Z-interference effect via a unique array programming sequence. The empirical results from the 3D NAND-based convolutional neural network (CNN) employed for MNIST dataset prediction reached an inference accuracy of 96.40%, which substantiates the superior accuracy and efficiency of the proposed modulation scheme.