Paper Title: Flash Enables Memory-Centric AI Computing in Computational Storage

Paper Abstract: With the advancements of artificial intelligence (AI), Internet of Things (IoT) and 5G, along with massive data generation and the demand of real-time, in-situ data processing is increasing exponentially. Memory-centric computing is emerging as a viable technology to complement traditional processor-centric computing. It minimizes data movement, thus reducing energy and processing time. New Flash memories are being developed rapidly for nonvolatile data and computing code storage with high-density, low-power, and small-form factors. Flash is a solid choice for accelerating some AI computation with low-power computing-in-memory approaches for a wide range of applications. From this presentation, attendees will glean an understanding of recent progress made by Flash-enabled, memory-centric computing solutions. Progress highlighted will multiply and accumulate (MAC) calculation for deep neural network, in-memory searching, non-volatile ternary content addressable memory (nvTCAM). Attendees will come away with a better grasp of the challenges, limitations and potential applications of memory-centric computing solutions.

Paper Author: Donald Huang, Product Marketing Director, Macronix

Author Bio: Donald Huang, PhD, is director of product marketing at Macronix International. Among his many responsibilities at Macronix he leads the marketing of the company's low- and ultra-low-power Flash memory products.