STMicro Unveils STM32H7P AI Microcontroller
The new chip pairs a 520 MHz Arm Cortex-M7 CPU with a hardware signal processor and an 85 GOPS neural processing unit for industrial devices.
Updated on Nov. 10, 2026 in Semiconductors
STMicroelectronics announced the STM32H7P microcontroller on November 10, 2026, adding a dedicated computation accelerator to its existing STM32H7 series. The chip is currently supplying samples to OEM customers.
Why it matters
The product performs AI inference and signal processing in industrial devices without a separate high-performance computing unit, offloading matrix and digital signal processing tasks from the CPU. This approach targets real-time control applications where processing load would otherwise strain the main processor.
The NPU delivers up to 85 GOPS of AI processing, with a 520 MHz Arm Cortex-M7 CPU, up to 2MB of flash, and rated operation up to 125 degrees Celsius. Performance figures come from ST's own tests, and results beyond the cited applications have not been independently verified.
The players
STMicroelectronics
A Europe-headquartered semiconductor manufacturer known for its STM32 family of microcontrollers used across industrial, automotive, and consumer electronics.
The details
The STM32H7P adds a hardware signal processor that performs matrix and vector operations and digital signal processing in place of the CPU, alongside a neural processing unit for AI workloads. The LQFP100 package with 1MB flash starts at $4.43 per unit for orders of 10,000 units, and mass production is scheduled to begin in January 2027.
Timeline
November 10, 2026: STMicroelectronics announces the STM32H7P microcontroller.
January 2027: Scheduled start of mass production and availability of evaluation boards and discovery kits for developers.
The Tech Race
The STM32H7P extends the existing STM32H7 series by adding a dedicated computation accelerator to a microcontroller platform already used in industrial designs. It positions STMicroelectronics to add on-chip AI inference to real-time controllers without requiring a separate high-performance processor.
Developers building industrial controllers can add AI inference to real-time designs using a single microcontroller instead of a separate computing unit. The LQFP100 variant with 1MB flash starts at $4.43 per unit for 10,000-unit orders, and evaluation boards arrive from January 2027.
The takeaway
Integrating AI processing directly into microcontrollers lets engineers run inference on the same chip that handles real-time control, reducing system complexity. Developers evaluating new industrial designs should plan around the January 2027 availability window for development hardware.
What happens next
Mass production of the STM32H7P series and availability of evaluation boards and discovery kits are scheduled to begin in January 2027.
Further reading
Explore more semiconductor coverage to see how chipmakers are integrating AI processing into embedded hardware.
