Memristor-based computing-in-memory accelerator speeds up compressed sensing at the edge by overcoming traditional von Neumann bottleneck

https://www.eurekalert.org/news-releases/1109311

"CMOS inefficient for compressed sensing reconstruction due to high energy/ latency caused by data shuttling between processor/ memory)... overcome using memCS: 128Kb memristor chip with CIM architecture executing core CS calculations directly within memory... 31.11 dB peak signal-to-noise ratio, 94.2% classification accuracy, closely matching pure software implementations... 11.22X faster, 30.46X more energy-efficient... excels in power-constrained edge computing scenarios"