Building on decades of innovation in the Arm Cortex-M domain, Arm and STMicroelectronics are joining forces to deeply integrate intelligence into scenarios where machines interact with the physical world.
Key Highlights:
· At last week's STM32 Summit, STMicroelectronics announced the launch of the new Arm Cortex-M-based STM32A series of microcontrollers, designed to ensure the safe and reliable operation of autonomous systems.
· Arm and STMicroelectronics are deepening their long-standing collaboration to jointly drive the development of next-generation Cortex-M-based microcontrollers, meeting the growing demands of intelligent machines for perception, processing, and real-time control.
· As intelligence extends toward sensors and actuators, Cortex-M provides precise, predictable, and continuous processing power, translating AI decisions into physical actions.
Embedding artificial intelligence (AI) into industrial machines, robots, and vehicles is one of the most complex engineering challenges facing the industry today. These physical AI systems need to perceive their surroundings, decide on actions, coordinate movements, and respond quickly and safely, all while meeting strict constraints on power consumption, weight, heat dissipation, and cost.
Choosing the right approach to address this challenge is key to unlocking the immense economic and productivity value of the physical industries. Arm estimates that by the 2030s, these industries will generate a $200 billion annual compute market opportunity.
Physical AI Requires Intelligence at Every Compute Layer
The engineering challenge of physical AI lies not only in building more AI models but also in creating complete systems that translate intelligence into reliable physical actions.
Take a robot picking up a glass as an example. The robot must identify the object, determine its position, plan how to grasp it, and calculate the appropriate gripping force to avoid crushing the glass. Actuators and sensors continuously monitor position, pressure, and motion status, while motors and control systems adjust in real time as conditions change.
From Arm's perspective, physical AI systems comprise three interconnected "compute layers":
1. Perception-Driven Intelligence: Understands the surrounding environment to enable autonomous navigation, obstacle avoidance, and real-time motion.
2. Interaction-Driven Intelligence: Understands and responds to human commands through voice, gestures, and other interaction methods, enabling natural and seamless interaction and information exchange between humans and machines.
3. Actuation Control: Coordinates and completes precise physical movements through real-time control of motors, joints, and other actuators.
Additionally, the cloud or data centers provide a fourth compute layer for device fleet management, simulation, and training, offering complementary support to the first three compute layers. It is this complete compute architecture that makes physical AI possible.
Across these compute layers, functional safety and cybersecurity must serve as the foundation, not only to protect systems and data but also to ensure that systems operate in a predictable and reliable manner, which is crucial for building trust in autonomous machines.
Three synergistic compute layers collectively enable physical AI systems to perceive, interact, and act, supported by cloud computing, with functional safety and cybersecurity贯穿ing all compute layers.
These compute layers must also collaboratively handle a range of complex AI and traditional software processes. These processes operate at different speeds and granularities while needing to interact and coordinate with each other.
At the perception-driven intelligence and interaction-driven intelligence layers, Arm Zena Compute Subsystems (CSS) provide the foundation for building complete compute systems, supporting the integration of processors, application CPUs, AI accelerators, memory, and interconnects. Meanwhile, by collaborating with industry leaders such as Cadence, Arm helps partners simplify the integration and verification of related technologies, accelerating these systems to market.
Components at the actuation control layer need to respond quickly and predictably, translating AI decisions into physical actions. Embedded processors play a key role in this process.
At the actuation control layer, real-time perception, decision-making, and precise control work together to ensure reliable execution of physical actions.
Building Next-Generation Embedded Computing for Physical AI
For decades, Arm and STMicroelectronics have driven the widespread adoption of embedded computing in industrial, automotive, and consumer markets. Today, their partnership is helping shape the next-generation Cortex-M roadmap to meet the increasing demands of smarter machines for perception, processing, and control capabilities.
At last week's STM32 Summit, STMicroelectronics announced a new strategic direction: creating a converged STM32 platform for physical AI and launching the new STM32A series of microcontrollers (MCUs). Built on Cortex-M, this MCU series is designed for autonomous systems with high requirements for predictability and redundancy, ensuring safe and reliable operation even if some components fail.
Drew Henry, Executive Vice President of Physical AI at Arm, stated: "The next major frontier in computing is integrating intelligence into real-world machines, enabling them to move, produce, and perform tasks more intelligently. Building on over twenty years of deep collaboration, Arm and STMicroelectronics are working together to drive the continuous evolution of Cortex-M to meet the growing demand for edge intelligence, allowing machines to perceive the physical world and take action, thereby unlocking the potential of next-generation physical AI systems."
The Cortex-M portfolio is already powering billions of devices, including industrial controllers and automotive systems. Meanwhile, Cortex-M continues to expand the capabilities of MCU-level systems, bringing advanced digital signal processing and machine learning capabilities to power- and cost-constrained applications through processors like Arm Cortex-M55 and Cortex-M85. As intelligence extends toward sensors and actuators, Cortex-M provides the precise, predictable, and continuous processing power needed to translate AI decisions into physical actions. For more details on how Cortex-M and Cortex-R processors support physical AI systems, please refer to this blog post.
Driving Embedded Computing Toward Greater Intelligence
The future of physical AI cannot rely solely on increasingly intelligent models; it depends on whether machines can effectively translate this intelligence into precise, safe, and reliable physical actions.
Through continued investment in its technology roadmap and deeper ecosystem collaboration, Arm is injecting more intelligence into every compute layer, from perception and interaction to real-time sensing and actuation control.
This is a crucial foundation for unlocking the potential of next-generation physical AI systems and seizing the significant economic and productivity opportunities they bring.
– End –