To PiM or Not to PiM: The case for in-memory inferencing of quantized CNNs at the edge

Gabriel Falcão, João Dinis Ferreira

Abstract: As artificial intelligence becomes a pervasive tool for the billions of IoT (Internet of things) devices at the edge, the data movement bottleneck imposes severe limitations on the performance and autonomy of these systems. PiM (processing-in-memory) is emerging as a way of mitigating the data movement bottleneck while satisfying the stringent performance, energy efficiency, and accuracy requirements of edge imaging applications that rely on CNNs (convolutional neural networks).