Collaborative Training for Compensation of Inference Errors in NOR Flash Computing in memory Chips
Tong Zhang, Weiping Pan, Hao Li · 2024
We propose a co-training error compensation method applied to NOR Flash computing in memory chips. The method utilizes a co-training technique to learn the computation error and data conversion error of NOR Flash computing in memory chips. Unlike traditional error compensation methods, this method does not need to analyze the error causes and error data, but uses the error compensation capability of deep learning to learn the computation error and data conversion error of NOR Flash computing in memory chips. Through experiments, it is proved that the method can compensate the analog computation error and AD conversion error inside the chip. Compared with the method without co-training, the accuracy of the method on the MNIST dataset is improved by 9.91%.