STM32U3B5/C5 Ultra-Low-Power MCU Features 640 KB RAM, 2 MB Flash, and HSP Accelerator to Run AI without Batteries

STM32U3B5/C5 Ultra-Low-Power MCU Features 640 KB RAM, 2 MB Flash, and HSP Accelerator to Run AI without Batteries

CNX Software – Embedded Systems News
CNX Software – Embedded Systems NewsMar 5, 2026

Key Takeaways

  • 640 KB SRAM, 2 MB flash, 96 MHz Cortex‑M33
  • HSP accelerator enables AI on energy‑harvesting devices
  • STM32U3C5 adds cryptographic core and security features
  • Up to 14 I/Os operate at 1.08 V, 5 V‑tolerant
  • Power consumption as low as 3 µA in Stop mode

Summary

STMicroelectronics introduced the STM32U3B5 and STM32U3C5 microcontrollers, expanding its ultra‑low‑power Cortex‑M33 family. Both devices combine up to 640 KB SRAM, 2 MB flash, and a new hardware signal processor (HSP) accelerator that can run AI/ML workloads on energy‑harvesting power sources. The STM32U3C5 adds a dedicated cryptographic core and advanced security features, while both chips support a rich peripheral set including CAN‑FD, I³C, and multiple UARTs. Pricing starts at roughly $3 per unit for the B5 and $4.7 for the C5 in 10 K volumes.

Pulse Analysis

The demand for ultra‑low‑power microcontrollers has surged as edge AI moves beyond battery‑powered gateways into truly autonomous sensors. STMicroelectronics’ STM32U3 series targets this niche by marrying a Cortex‑M33 core with near‑threshold operation down to 0.65 V, delivering a CoreMark efficiency of 4.12 CoreMark/MHz. This efficiency, combined with a modest 96 MHz clock, positions the U3B5/C5 as compelling alternatives to higher‑power MCU families while keeping silicon costs competitive.

At the heart of the new devices is the hardware signal processor (HSP) accelerator, which ST claims offers up to 13× the performance of a standard Cortex‑M33 and up to 9× better power efficiency than the STM32U5 line. Benchmarks show the HSP outperforms even Cortex‑M55 designs with M‑VE extensions, delivering real‑time inference for keyword spotting, image classification, and visual wake‑word detection on a power budget that fits energy‑harvesting sources such as solar or kinetic modules. This leap enables developers to embed AI directly into wearables, environmental monitors, and industrial IoT nodes without relying on bulky batteries.

For engineers, the STM32U3C5’s integrated cryptographic core and comprehensive TrustZone security stack simplify compliance with emerging data‑privacy regulations. Coupled with the familiar STM32CubeMX tooling and the STM32 AI ecosystem, the migration path from prototype to production is streamlined. While the per‑unit price—approximately $3 for the B5 and $4.7 for the C5 at 10 K volumes—is higher than legacy low‑end MCUs, the added AI capability and security justify the premium, especially as competitors scramble to match ST’s battery‑less AI proposition. The U3 series therefore sets a new benchmark for power‑constrained intelligent devices.

STM32U3B5/C5 ultra-low-power MCU features 640 KB RAM, 2 MB Flash, and HSP accelerator to run AI without batteries

Comments

Want to join the conversation?