Kioxia Unveils 5TB, 64GB/s Flash Memory Module for AI and Edge Computing Advancements

Kioxia has developed a prototype flash memory module with a capacity of 5 terabytes (TB) and a bandwidth of 64 gigabytes per second (GB/s), designed to support large-scale artificial intelligence (AI) models and edge computing applications. The development was part of the "Post-5G Information and Communication Systems Infrastructure Enhancement R&D Project (JPNP20017)," commissioned by Japan’s New Energy and Industrial Technology Development Organization (NEDO).

The module addresses the challenge of balancing capacity and bandwidth in conventional DRAM-based memory by utilizing daisy-chain connections with flash memory beads, maintaining high bandwidth even as capacity increases. Kioxia implemented a 128 gigabits per second (Gbps) PAM4 (Pulse Amplitude Modulation with 4 Levels) high-speed, low-power transceiver to reduce connection numbers while achieving higher bandwidth. Additionally, flash prefetch technology was developed to minimize read latency during sequential data accesses, and low amplitude signaling with distortion correction enhanced memory bandwidth to 4.0 Gbps.

The prototype, which uses a PCIe 6.0 (64 Gbps, 8 lanes) host interface, achieves its 5TB capacity and 64GB/s bandwidth with less than 40 watts of power consumption. This advancement supports the growing demand for Multi-Access Edge Computing (MEC) servers in the post-5G/6G era, wh...