EdgeCortix Unveiled Scalable RAIDEN AI Chiplet Platform
The new hardware architecture aims to resolve performance bottlenecks for Physical AI applications in data centers.
Updated on Sept. 23, 2026 in Semiconductors
Live Poll
Do you believe new AI chip technologies will lead to better performance for physical machines?
EdgeCortix has introduced its RAIDEN AI chiplet platform, designed to offer scalable computing power from a single die to a four-die configuration. The system utilizes the company’s proprietary DNA-X architecture to address memory and connectivity limitations in advanced AI tasks.
Why it matters
Current AI accelerators struggle to balance memory, compute, and connectivity for Physical AI, often forcing trade-offs that limit system-level performance. This modular design seeks to eliminate those compromises through a unified software environment.
The flagship RAIDEN X4 configuration delivers 3.36 PFLOPS of FP4 compute and supports 256 GB of memory. It features 548 GB/s of memory bandwidth and provides up to 6.4 Tb/s of chip-to-chip connectivity.
The players
EdgeCortix
A semiconductor company that specializes in AI-focused energy-efficient hardware and software stacks.
Kawasaki Heavy Industries
A Japanese industrial conglomerate that manufactures heavy equipment, aerospace components, and defense systems.
Unigen Corporation
A specialized firm that focuses on the design and manufacturing of custom server and storage solutions.
New Energy and Industrial Technology Development Organization
A Japanese national research and development agency that provides funding for innovative energy and industrial projects.
The details
The platform utilizes a modular chiplet architecture that allows developers to scale workloads across one, two, or four dies using the consistent MERA software stack. Kawasaki Heavy Industries has already selected the technology for its upcoming aerospace and defense projects.
Timeline
EdgeCortix unveiled the RAIDEN platform on September 23, 2026.
Customer sampling is scheduled to begin in early 2027.
Volume production is planned for the second half of 2027.
The Tech Race
The RAIDEN platform builds on the MERA software stack to ensure interoperability across various chiplet configurations. This modular approach marks a departure from static single-die accelerators, signaling a broader industry shift toward flexible, heterogeneous hardware designs.
Developers and server engineers can expect improved flexibility when designing Physical AI systems, reducing the need for custom hardware compromises. While the hardware is not for individual consumers, these advancements will likely influence the efficiency and costs of AI-driven enterprise services.
The takeaway
The shift toward modular chiplet architectures allows companies to scale computing power dynamically rather than relying on fixed-design processors. This evolution is critical for supporting the next generation of high-demand AI applications in aerospace and defense.
What happens next
Customer sampling for the RAIDEN platform is scheduled to begin in early 2027, with volume production expected to commence in the second half of 2027.
Further reading
For more on the latest advancements in chip architecture, explore our Semiconductors section.
More information
Interested developers can sign up for the RAIDEN early access program registration to begin evaluating the hardware.
Live Poll
Do you believe new AI chip technologies will lead to better performance for physical machines?
