Home / Technology / EdgeCortix Unveils Scalable Chiplet for Physical AI
EdgeCortix Unveils Scalable Chiplet for Physical AI
24 Sep
Summary
- EdgeCortix launched RAIDEN, a scalable AI chiplet platform for Physical AI.
- RAIDEN scales from one die to four, offering up to 3.36 PFLOPS.
- Customer design wins include aerospace, defense, and robotics applications.
EdgeCortix, a Japanese semiconductor company, has launched its next-generation scalable AI chiplet platform, RAIDEN, designed to power Physical AI applications. This platform addresses the need for compute, memory, and connectivity to scale together as AI models evolve.
RAIDEN features a unified, energy-efficient architecture that scales from a single compute die to a four-die flagship. The maximum configuration targets up to 3.36 PFLOPS of AI compute, 256 GB of memory, and substantial memory and inter-die bandwidth. This modular approach allows customers to adapt to diverse performance and power requirements without changing their core architecture or software investments.
The RAIDEN platform is purpose-built for "thick edge" deployments, bringing high-performance AI closer to real-world operations. It is engineered to execute full Physical AI workloads, including perception, reasoning, and action, utilizing EdgeCortix's DNA-X accelerator architecture and MERA software stack. Customer design wins already secured include next-generation aerospace and defense systems, and robotic platforms.
Customer sampling for RAIDEN is anticipated in early 2027, with volume production slated for the latter half of 2027. The RAIDEN Early Access Program is currently open for qualified customers and partners.
