STMicroelectronics launches three STM32A safety MCUs for automotive

By Nathan Brooks

STMicroelectronics launched three STM32A variants at STM32 Summit 2026, extending the STM32 line into automotive, robotics, energy and other safety-critical control systems. The A4, A5 and A9 span entry-level local control through multi-domain real-time operation, with different memory, safety and security levels.

What does the unified STM32 platform change?

ST has combined Arm Cortex-M cores, the STM32Cube ecosystem, shared hardware IP, architectures and manufacturing methods under a unified platform for robotics, industrial, automotive and physical-AI workloads. ST expects common software and validated configurations to reduce firmware and toolchain redesign across product families.

ST says its shared manufacturing technologies make it easier to “reuse software and designs while products scale.” The stated goal is to keep core software and tools stable as device families broaden.

The process portfolio includes 40-nm flash, 18-nm FD-SOI with embedded phase-change memory, and TSMC embedded nonvolatile memory technologies. For developers, the stated benefit is architectural continuity between MCU families; for ST, it is a wider set of products sharing the same development environment.

How do the A4, A5 and A9 divide safety-critical work?

The STM32A4 targets entry automotive local-control nodes, local sensors and safety functions. It offers up to 0.5 MB of memory and lists QM/ASIL B and cryptography support.

The A5 addresses mid-range safety-critical industrial work, including robotics, energy systems and industrial automation. It provides up to 4 MB of memory, ASIL B support and a hardware security module.

The STM32A9 handles multi-domain, safety-critical control with up to 12 MB of memory, ASIL D support, a hardware security module, split-lock architecture and Ethernet-oriented connectivity with RCP support. Its split-lock design supports ASIL D operation, while ST says it can be configured for high-performance operation when full ASIL D support is unnecessary.

What does the G5 bring to power and motor control?

ST is launching the STM32G5 as a mixed-signal MCU family for digital power, high-end motor control and precision real-time control. The family is already in production on ST's 40-nm process, pairing an Arm Cortex-M33 core with TrustZone security. It runs at up to 280 MHz and delivers 1,136 CoreMark.

Its timing hardware includes a high-resolution timer with 111-ps resolution for advanced PWM and control, plus a programmable logic array. The analog set includes up to three 12-bit ADCs running at 5 MSPS, eight comparators, eight DAC channels, including two buffered external channels, and up to three operational amplifiers. Multiple motor-control PWM and general-purpose timers complete the control feature set.

What does the H7P add for edge inference?

The STM32H7P targets industrial edge AI, digital power, motor control and condition monitoring. It uses a Cortex-M7 CPU and a hardware signal processor for DSP functions.

Some variants add the Neural-ART NPU, which reaches up to 85 GOPS and moves AI inference away from the main processor. The dedicated inference lane leaves the Cortex-M7 available for control work.

The processor runs at up to 520 MHz and delivers 2,773 CoreMark. The family includes DSP instructions, a memory protection unit, 32 kB of instruction cache and 32 kB of data cache.

Up to 2 MB of flash and 736 kB of RAM sit alongside USB, three FDCAN interfaces, four 12-bit ADCs and eight DAC channels, including two external 12-bit outputs. Three advanced motor-control timers and two high-resolution timers cover precision timing. The specification concentrates communications, analog measurement and motor timing in one MCU.

How much AI work moves into the toolchain?

ST updated its Edge AI Suite with four tools covering model selection through deployment on STM32 hardware. Model Explorer provides a web interface for finding candidate models, while Dataset Factory streamlines data preparation.

Model Factory handles training, customization, quantization and optimization for STM32 MCUs, and Model Bench runs remote benchmarks on real ST boards. Together, the tools let teams evaluate, prepare, optimize and validate embedded models before deployment.

Can coding agents work without inventing hardware?

The new STM32Cube for Agentic AI framework gives coding agents access to maintained STM32 documentation, validated configurations, reference examples and ST development resources. It works with verified reference code and STM32CubeMX, ST's deterministic code-generation tool. Its code-review functions examine API calls, hardware interactions and deployment flexibility, while developers retain control of familiar STM32 workflows.

The framework supports GitHub Copilot, Claude and Cursor. ST says the combination accelerates code generation while preserving access to validated materials, thereby “reducing hallucination risk.” ST positions the agents as assistants within familiar STM32 development workflows.

What does Sidekick do?

STM32 Sidekick is an AI-powered assistant that accepts an STM32 series, part number, tool or application as context. It answers technical questions, finds documentation, explains concepts, compares devices or features and assists with debugging. Its knowledge comes from reference manuals, datasheets, user manuals, wikis, application notes and community-knowledge-base articles.

Sidekick adds citations and links to its answers so developers can verify the source material. Users can reach it through st.com, the ST Community, STM32CubeProgrammer 2.23 and later, Developer Zones, support pages and STM32 product pages. The citation trail gives engineers a way to check the answer rather than trust the tone.

ST disclosed no pricing, capacity, revenue or shipment schedule for the new families. The near-term competitive test is whether the A-series breadth, unified toolchain and 85-GOPS inference option convert this launch into production sockets.