TechCapsules

AI-summarised brief · reviewed before publication

EdgeCortix, a Japanese fabless semiconductor firm, announced RAIDEN, a scalable AI chiplet platform designed for “Physical AI” workloads that run at the edge rather than in data centers. The architecture can be configured as a single‑die (X1), two‑die (X2) or four‑die (X4) system, all sharing the company’s DNA‑X accelerator and MERA software stack. At full X4 scale, RAIDEN delivers up to 3.36 PFLOPS of FP4 compute, 256 GB of memory, 548 GB/s memory bandwidth, 1.54 TB/s die‑to‑die bandwidth and up to 6.4 Tb/s chip‑to‑chip connectivity for further scaling. The platform targets demanding edge applications such as aerospace, defense, robotics and high‑performance edge servers, and already has design wins in those sectors. By integrating compute, memory, bandwidth and software in a modular package, RAIDEN aims to eliminate the system‑level compromises that have limited current edge AI deployments.