EdgeCortix
About
About EdgeCortix
EdgeCortix combines reconfigurable AI-accelerator silicon with a software toolchain for compiling and deploying neural networks at the edge. SAKURA accelerator chips and modules provide local inference capacity, while the MERA compiler maps models across heterogeneous hardware and deployment targets.
Founded in 2019, the company operates from Japan with additional teams in India, Singapore and the United States. Its robotics relevance is as an embedded-compute and development layer: customers can use the platform for perception and decision workloads where latency, power consumption and disconnected operation limit cloud processing.
Timeline
Events
- 2026-09-24 - Product Launch - EdgeCortix unveils RAIDEN chiplet platform for Physical AI
The four-die X4 design targets up to 3.36 PFLOPS FP4 compute, 256 GB of memory and 1.54 TB/s die-to-die bandwidth. EdgeCortix says RAIDEN has design wins, including Kawasaki Heavy Industries aerospace and defense products and a Unigen server platform. The X1, X2 and X4 configurations share the DNA-X accelerator architecture and MERA software stack; volume production is planned for late 2027.
Source - 2026-09-08 - Contract Award - Kawasaki Heavy Industries signs multi-year EdgeCortix development agreement
The initial programme is valued at several million dollars. Work covers edge-AI development, system integration, feasibility studies and prototype platforms. The agreement targets low-power onboard AI for future aerial systems rather than a currently fielded production configuration.
Source
Relations
Relationships
- Kawasaki Heavy Industries
Technology Partner - EdgeCortix and Kawasaki Heavy Industries are working under a 2026–2028 edge-AI development and integration programme.
Commercial
- Kawasaki Heavy Industries signs multi-year EdgeCortix development agreement
2026-09-08 · Order / contract · Kawasaki Heavy Industries awarded EdgeCortix a 2026–2028 development and integration programme for future AI-enabled aerial defence and autonomy systems.
Source