MIPS Releases Physical AI and Robotics Reference Designs
The new suite pairs real-time RISC-V cores with AI acceleration for robotics and other physical AI applications.
Updated on Nov. 12, 2026 in Semiconductors

MIPS released a suite of reference designs for physical AI and robotics applications, including the MIPS Actus-1, MIPS Acies-1 and MIPS Aegis-1. The designs incorporate RISC-V profiles RMV32 and RVA23 and are delivered as modules, M.2 cards and motherboards.
Why it matters
Physical AI and robotics systems need both fast real-time control and enough compute for AI workloads in a small power budget. Ready-made reference platforms can shorten the path for developers building such systems on the RISC-V architecture.
The Acies-1 M.2 module, built on the MIPS Atlas S8200 RVA23 platform, delivers 50 TOPS of AI performance at 10 W across eight cores with 50 GB/s of bandwidth. The Actus-1 module uses 100-ps-class PWM resolution for field-oriented control, and MIPS has not published independent benchmarks for these figures.
The players
MIPS
MIPS is a processor IP company that offers an array of RISC-V intellectual property, from hard real-time cores to cores supporting AI large language models.
GlobalFoundries
GlobalFoundries is a semiconductor foundry whose FDX FD-SOI process technology was used to build the MIPS Actus-1 module.
The details
The Actus-1 pairs two MIPS Atlas M8500 real-time RISC-V cores, is built on GlobalFoundries' FDX FD-SOI technology, uses MRAM for on-chip non-volatile storage and includes a gigabit Ethernet port tied to a three-port time-sensitive networking switch. AI acceleration is supported by the Atlas SDK, a Graph Performance Simulator and an IREE-based MLIR compiler that handles TensorFlow, PyTorch and ONNX models.
Timeline
- Reference designs released on November 12, 2026.
The Tech Race
The release applies the RVA23 RISC-V profile to physical AI hardware, extending a standard that RISC-V processor vendors are adopting for AI-capable designs. It positions MIPS to compete with other RISC-V and Arm-based suppliers targeting robotics and edge AI.
Developers building robots or other physical AI products can start from ready-made modules and M.2 cards rather than designing real-time and AI hardware from scratch. The reference platforms include AI software tooling, so teams can test TensorFlow, PyTorch and ONNX models on the hardware.
The takeaway
Developers evaluating hardware for robotics should compare real-time control performance and AI throughput per watt across reference platforms before committing to a design.
Further reading
Explore more semiconductor developments shaping AI hardware and embedded computing.
Source note: This article includes information reported by Electronic Design.



