Arm and STMicroelectronics target embedded compute for physical AI
Arm and STMicroelectronics are working on the next generation of embedded computing for physical AI, with ST announcing a converged STM32 platform and Cortex-M-based STM32A microcontrollers. The effort is aimed at bringing more intelligence closer to the sensors, actuators and machines that interact with the physical world.
- Arm and STMicroelectronics are shaping the next Cortex-M roadmap for physical AI.
- ST announced a converged STM32 platform and Cortex-M-based STM32A microcontrollers.
- Physical AI systems combine perception, interaction and actuation computing planes.
- Arm says safety, security, predictability and redundancy are essential for autonomous machines.
- Arm estimates a $200 billion annual compute opportunity in physical industries into the 2030s.
Physical AI must turn decisions into action
Arm describes physical AI as a systems challenge that extends beyond creating more capable AI models. Industrial machines, robots and vehicles must perceive their surroundings, decide what to do, coordinate movement and respond quickly and safely while operating within limits on power, weight, heat and cost.
The company uses a robot picking up a glass as an example. The system must recognize the object, determine its position, plan a route and calculate the force needed to grasp it without breaking it. Sensors and actuators then monitor pressure, position and movement as motors and control systems adjust in real time.
Arm identifies three interconnected computing planes in these systems: perception-driven intelligence, interaction-driven intelligence and actuation execution. Cloud or data center computing provides an additional layer for fleet management, simulation and training. Safety and security underpin all of the planes, according to Arm.
Arm and ST outline a Cortex-M direction
At the STM32 Summit this week, STMicroelectronics announced a strategic direction toward a converged STM32 platform for physical AI. It also introduced a series of Cortex-M-based STM32A microcontrollers for autonomous systems that require high levels of predictability and redundancy.
Arm says these systems are designed to maintain safe and reliable operation even when individual components fail. The companies’ collaboration combines Arm’s processor architecture and software ecosystem with STMicroelectronics’ expertise in embedded applications.
The work builds on decades of Arm Cortex-M development across industrial, automotive and consumer products. Arm says the Cortex-M portfolio powers billions of devices, while the Cortex-M55 and Cortex-M85 bring signal processing and machine learning capabilities to microcontroller-class systems constrained by power and cost.
Why embedded control matters for physical AI
Perception and interaction workloads can use Arm Zena Compute Subsystems, which provide a foundation for integrating processors, application CPUs, AI accelerators, memory and interconnect. Arm also points to collaboration with Cadence as a way for partners to simplify integration and verification.
The actuation execution plane has different requirements. Its components need to deliver fast, predictable responses so AI decisions can become precise physical movements. Embedded processors handle real-time sensing, decision-making and control, keeping intelligence close to the sensors and actuators.
For users and the wider market, the announcement signals that physical AI will depend on coordinated computing across models, embedded processors and control systems. Arm estimates that physical industries represent a $200 billion annual compute opportunity into the 2030s. The company says continued investment in its technology roadmap and ecosystem partnerships is intended to help unlock that opportunity through more efficient, reliable and responsive machines.
