Flash-Based Computing-in-Memory (CiM) Architectures for High-Accuracy Classification

1. Abstract Computing-in-memory (CiM) has been widely studied as an efficient solution to eliminate frequent data transfer, and it can provide high-speed real-time data processing. In this work, as a hardware-accelerated denoising approach, Flash-based CiM is designed to improve the accuracy and efficiency of in-suit clinical diagnosis. By adopting NOR arrays to proceed parallel computation of linear convection-diffusion equations, the quality and classification accuracy of ultrasound images has been successfully enhanced. For denoised images, as high as 99.6% classification accuracy can been achieved, indicating that Flash-based CiM is promising as an efficient approach to assist clinical diagnosis.