Resistive Memory

Resistive memory (ReRAM) is a promising technology for accelerating deep neural networks (DNNs) by integrating computation and memory, thus overcoming the von Neumann bottleneck. Current research focuses on optimizing ReRAM-based in-memory computing (CIM) architectures for DNN inference and training, exploring techniques like basis combination for reprogramming-free operation and hardware-aware quantization to mitigate device variability and noise. These advancements aim to improve energy efficiency, speed, and accuracy of DNNs, particularly for resource-constrained applications like edge AI and neuromorphic computing, impacting fields such as medical imaging and generative AI.