Reference SoC platform targets physical AI robotics control

Robotics platforms increasingly require tightly integrated compute, connectivity and control architectures to support physical AI workloads and real-time motion control. In this context, MIPS, a GlobalFoundries company, and Inova Semiconductors GmbH introduced a robotics control reference platform for humanoid and edge robotics applications.

A hardware architecture for sense-think-act control chains
The reference platform is designed as a hardware-based system-on-chip (SoC) architecture supporting the full robotics signal chain covering sensing, processing, actuation and communication. The approach targets applications such as humanoid robots, robotic arms and other physical AI systems where size, weight and power constraints influence system design.

The architecture supports mixed-criticality computing, enabling real-time control loops alongside secure AI workloads. This type of integration is intended to support robotics platforms requiring deterministic response times together with local AI processing.

The platform is manufactured on GlobalFoundries’ FDX process platform, including the 22FDX^® process technology, to support low-power operation and predictable performance characteristics for embedded robotics workloads.

Combining RISC-V compute with zonal communication design
The platform builds on Inova Semiconductors’ experience with zonal architectures from the automotive data ecosystem, applying similar communication principles to robotics control design. The architecture integrates INOVA’s APXpress high-speed interface with several MIPS compute elements.

These include the MIPS Atlas M8500 RISC-V high-performance microcontroller processor IP, the MIPS Atlas S8200 RISC-V AI processor IP, and MIPS Atlas mixed-signal technologies. Together, these components support multi-axis motion control and high-performance data exchange across multiple network topologies.

The design also aims to reduce software overhead for data communication by shifting more functionality into hardware, which can shorten development cycles and simplify platform scaling across different robot configurations.

Reference design targeting faster robotics development cycles
The reference platform is positioned as a development blueprint for robotics manufacturers building humanoid and advanced robotic systems. The zonal architecture approach is intended to support modular design, allowing reuse of architectural components across multiple product variants.

The system is also intended to reduce bill-of-materials complexity by consolidating compute and communication functions into a custom SoC platform capable of supporting mission-critical robotics workloads and secure on-device AI processing.

Simulation environment supports early software development
Early access to the platform is supported through the MIPS Atlas Explorer environment, a simulation-based hardware/software co-design platform. This environment provides virtual models of the compute architecture, allowing software teams to begin optimisation of vision-language-action models and related robotics control software before physical hardware becomes available.

This co-design approach enables earlier validation of robotics software stacks and AI models aligned with the intended control architecture.

www.inova-semiconductors.com

Edited by industrial journalist, Aishwarya Mambet — AI-powered.