MIPS, a subsidiary of GlobalFoundries (GF), has introduced new capabilities for its MIPS Actus reference platform, targeting the computing and control requirements of humanoid and industrial robots. Built on GF’s FDX FD-SOI technology, Actus combines real-time motor control, deterministic networking and edge processing on a single chip designed for compact robotic joints. The platform aims to connect AI inference with precise physical movement while addressing the power, thermal and safety constraints of production robotics.
As humanoid and industrial robots move from research environments toward commercial deployment, their computing requirements are expanding beyond the processors that run AI models. Robots must translate decisions into precise physical actions, coordinate multiple joints and process sensor data with predictable timing. MIPS is targeting this challenge with advances to its Actus reference platform, which brings motor control, networking and edge intelligence into the robotic joint.
Announced at electronica 2026, the updated MIPS Actus platform is designed to address what the company describes as the final stage between AI decisions and physical action. This is particularly important in humanoid arms, where joints from the elbow to the fingertips must coordinate smooth movements within strict limits on size, weight, power consumption and heat.
Real-Time Control and Edge Processing on One Chip
Actus combines two MIPS Atlas M8500 real-time RISC-V processors with an ARC VPX vector processing unit. The architecture is designed to handle deterministic motor control and local intelligence without requiring every processing task to be routed through a central computing system.
The two Atlas M8500 processors use hardware multithreading to reduce context-switching delays during interrupt servicing. MIPS also incorporates dedicated trigonometric acceleration, which the company says delivers more than 16 times the performance of standard trigonometric libraries for field-oriented motor control algorithms.
Field-oriented control is widely used to regulate electric motors by managing current and magnetic fields. In robotics, its execution affects torque, positioning and movement smoothness. Faster mathematical operations and predictable processing can help controllers respond to changing motor conditions with tighter timing requirements.
The platform also includes high-resolution pulse-width modulation (PWM) and sigma-delta analog-to-digital converters (ADCs). MIPS specifies 100-picosecond-class PWM resolution, targeting precise control of high-torque, high-current motors.
Alongside motor control, the ARC VPX 512-bit vector processor enables sensor fusion and predictive maintenance models to run locally on the same chip. This allows information from sensors to be processed close to the actuator, potentially reducing dependence on centralized computing and communication between separate processing components.
The announcement does not disclose independent benchmarks for complete robotic systems, including end-to-end inference latency, power consumption or motion-control accuracy. Its performance figures should therefore be understood as vendor-reported specifications rather than demonstrated improvements across production robots.
Deterministic Networking Connects Robotic Joints
Coordinating multiple robotic joints requires reliable communication between local controllers and a central system. Delays or inconsistent message delivery can affect synchronization, particularly in robots performing complex movements.
Actus integrates gigabit Ethernet with three-port Time-Sensitive Networking (TSN), alongside CAN-FD and serial interfaces. TSN is designed to provide predictable communication timing over Ethernet, making it relevant to systems that need coordinated data exchange rather than best-effort network delivery.
The architecture allows joint controllers to exchange position and torque information with a central controller. In a humanoid robot, this could support coordinated movement across arms, hands and other motor-driven components.
Security is another part of the platform design. MIPS says Actus includes a hardware security module with post-quantum cryptography capabilities, while functional-safety support is intended to address requirements in production robotics.
These features reflect a wider industry challenge: robots operating in factories, warehouses and other commercial environments need computing systems that combine performance with predictable behavior, secure communications and safety-oriented design.
MIPS Builds a Broader Physical AI Architecture
Actus completes MIPS’ physical AI platform portfolio, joining Acies and Aegis.
Acies is the company’s inference-native AI platform, while Aegis is designed as a safety-native CPU platform. Together, the three offerings are intended to cover different stages of robotic computing, from AI model execution and safety-critical processing to deterministic control and actuation.
The architecture is based on RISC-V, an open instruction set architecture that allows semiconductor developers to build customized processor implementations. For robotics manufacturers, this approach can provide flexibility in tailoring computing systems to specific workloads and product requirements.
MIPS is positioning Actus as a way to bring more processing functions into compact actuator designs. Integrating control, networking and intelligence on one chip could reduce the need for multiple components and help lower board area and system complexity.
The platform is built using GlobalFoundries’ FDX FD-SOI technology, with AutoPro150 and on-chip MRAM. MIPS identifies the manufacturing technology as a foundation for low-power operation, integration and compact designs. The company also points to reduced board area and system cost as potential benefits, although it has not provided comparative cost or power figures.
Software Development and Production Readiness
MIPS is also addressing the software development process. Developers can begin working with a virtual platform before silicon becomes available, then move to the Actus-1 evaluation board and software development kit for hardware prototyping.
Green Hills Software is among the ecosystem partners supporting the software stack for safety-certified products. This matters because robotics manufacturers need more than functional silicon: development tools, operating environments and safety-related software support are necessary to move designs toward commercial deployment.
Omdia Senior Research Director Edward Wilford described the combination of real-time motor control and deterministic networking on a single chip as addressing an engineering gap in emerging robotic designs.
The broader opportunity is tied to the evolution of physical AI. As robots become capable of more complex autonomous tasks, the industry will need computing architectures that connect high-level inference with low-level control. Central AI processors remain important, but the ability to execute time-sensitive workloads close to motors and sensors is becoming a distinct part of the system architecture.
MIPS’ Actus platform is designed to address that layer. Its commercial relevance will depend on silicon availability, software ecosystem maturity, integration with robot manufacturers’ designs and independently validated performance under real-world operating conditions.
Market Landscape
Physical AI is expanding the semiconductor market beyond processors dedicated to training and inference. Humanoid robots, industrial automation systems and autonomous machines require a combination of AI acceleration, real-time control, sensor processing, networking and safety mechanisms.
This creates demand for edge architectures that can execute time-sensitive workloads close to physical components. RISC-V-based platforms offer an opportunity for customization, while technologies such as TSN support coordinated communication across distributed control systems.
MIPS is positioning Actus within this emerging category by combining motor control, vector processing and networking in a compact platform. Its differentiation will depend on how effectively manufacturers can integrate the technology into production robots and validate performance, reliability, power efficiency and safety.
Top Insights
- Physical AI focus: Actus targets the computing layer connecting AI decisions with precise movement in humanoid and industrial robots.
- Integrated processing: Two real-time RISC-V processors and vector processing combine motor control with local sensor intelligence.
- Deterministic networking: TSN Ethernet, CAN-FD and serial interfaces support coordinated communication between robotic joints and central controllers.
- Security and safety: Hardware security with post-quantum cryptography and functional-safety support address requirements for production robotics.
- Development ecosystem: Virtual platforms, evaluation hardware, an SDK and software partners support the path from design to prototyping.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
