In-Data vs. Near-Data Processing: The Case for Processing in Resistive CAM

Near-data in-storage processing research has been gaining momentum in recent years. Typical processing-in-storage architecture places a single or several processing cores inside the storage and allows data processing without transferring it to the host CPU. Since this approach replicates von Neumann architecture inside storage, it is exposed to the problems faced by von Neumann architecture, especially the bandwidth wall. We present a novel processing-in-storage system based on Resistive Content Addressable Memory (RCAM). RCAM functions simultaneously as a storage and a massively parallel associative processor. RCAM processing-in-storage resolves the bandwidth wall faced by conventional processing-in-storage architectures by keeping the computing inside the storage arrays, thus implementing indata, rather than near-data, processing. We show that RCAM based processing-in-storage architecture may outperform existing in-storage designs and accelerator based designs. RCAM processing-in-storage implementation of K-means achieves speedup of 4.6—68 relative to CPU, GPU and FPGA based solutions. For K-Nearest Neighbors, RCAM processing-in-storage achieves speedup of 17.9—17,470 and for Smith-Waterman sequence alignment it reaches speedup of almost 5 over a GPU cluster based solution.