Trends in Embedded Volatile and Non-Volatile Memories for Scalable Edge AI Applications

We present trends in embedded memory technologies, including both volatile and non-volatile memories, for edge AI applications. In edge AI applications, all input information from vision, voice and other real-time sensors is processed by embedded systems according to various inference tasks. Therefore, MPUs/MCUs for edge AI require a scalable products lineup to cost-effectively achieve real-time response, higher performance, and low power operation. To expand edge AI applications from MPUs/MCUs system perspective, we implement efficient memory access and memory size reduction for SRAM, as well as high-speed access and low power write for MRAM.