Resistive Neural Hardware Accelerators

Deep neural networks (DNNs), as a subset of machine learning (ML) techniques, entail that real-world data can be learned, and decisions made in real time. However, their wide adoption is hindered by number software hardware limitations. The existing general-purpose platforms used to accelerate DNNs are facing new challenges associated with the growing amount exponentially increasing complexity computations. Emerging nonvolatile memory (NVM) devices compute-in-memory (CIM) paradigm creating architecture generation increased computing storage capabilities. In particular, shift toward resistive random access (ReRAM)-based in-memory has great potential implementation area- power-efficient inference training large-scale network architectures. These process IoT-enabled AI technologies entering our daily lives. this survey, we review state-of-the-art ReRAM-based DNN many-core accelerators, superiority compared CMOS counterparts was shown. covers different aspects realization present limitations, prospects. comparison accelerators shows need for introduction performance metrics benchmarking standards. addition, major concerns regarding efficient design include lack accuracy simulation tools codesign.