STMicroelectronics used its STM32 Summit 2026 to reorganise its microcontroller business. The company says its industrial, automotive and robotics MCUs will now sit on a single STM32 platform, and it introduced a new family, STM32A, aimed at automotive and high-safety designs. Unite.AI reports that the keynote took place on November 3, 2026. The official press release followed from Geneva on November 4 and calls the change a "major strategic evolution."
ST is presenting this as continuity: Arm Cortex-M, the STM32Cube software ecosystem and a shared set of process technologies, now applied across market segments that used to be handled separately. The new STM32A parts are the first products built under that framing.
The three pillars, as ST describes them
The press release rests the unified platform on three things:
- Arm Cortex-M continuity. The platform is built on Arm Cortex-M. ST also describes itself as a lead partner shaping Arm's future Cortex-M portfolio. Drew Henry of Arm is quoted in the release, which describes Cortex-M as the architecture of choice for more than 100 billion devices. Unite.AI adds that Cadence is participating as a key design partner in the Cortex-M roadmap work.
- The STM32Cube ecosystem. ST says STM32Cube and the broader Arm ecosystem provide a consistent development environment across the platform and help customers retain established workflows.
- Shared process technology. ST lists 40nm Flash, 18nm FD-SOI with embedded phase-change memory (PCM), and TSMC embedded non-volatile memory (eNVM).
The process list matters to anyone tracking where future STM32 parts will come from. ST is saying the same small set of silicon technologies will serve everything from general-purpose industrial parts to automotive safety controllers, and that TSMC eNVM is one of them alongside its own 40nm and 18nm FD-SOI nodes. The release does not say which process each STM32A part uses, and neither do we.
The release quotes ST's Remi El-Ouazzane alongside Arm's Henry. The quotes add context on the strategy but no specifications.
Meet the STM32A family
ST says the first three STM32A microcontrollers are sampling to customers now, with broader availability planned for 2027. Here is how the press release positions each one:
- STM32A4: entry-level parts for automotive local control nodes and safety systems.
- STM32A5: mid-range parts for safety-critical work in industrial automation, robotics, energy and automotive.
- STM32A9: multi-domain, safety-critical real-time control with Ethernet.
ST's release does not include memory sizes or functional-safety ratings. Unite.AI, covering the keynote, reports these figures:
- STM32A4: up to 0.5 MB of memory, QM/ASIL-B targets, with cryptography.
- STM32A5: up to 4 MB of memory, ASIL-B.
- STM32A9: up to 12 MB of memory, ASIL-D, Ethernet-oriented. Unite.AI adds that a split-lock configuration can deliver twice the ASIL-B performance.
Treat those figures as secondary reporting until ST publishes datasheets. Unite.AI notes that its article was written by an AI research agent and reviewed by its editors. We have not seen core counts, clock speeds, pin counts or package options for any STM32A part, so we are not printing any.
The split-lock point is worth explaining because it reflects a real design choice. In lockstep mode, two cores run the same code and compare results, which supports the highest integrity level. In split mode, the same cores run independently for more throughput at a lower safety level. Unite.AI says the STM32A9's split-lock architecture can prioritise ASIL-D or, per ST, provide twice the ASIL-B performance when ASIL-D is unnecessary. We have not seen ST's own documentation of how this works.
Availability, precisely
ST's official position is that the three parts are sampling, with broader availability planned for 2027. Unite.AI is more specific: it says broader availability starts with the STM32A9 in the first half of 2027. No pricing has been announced for any STM32A part.
Edge AI and the robotics thread
The "physical AI" part of the announcement comes mostly through Unite.AI's keynote coverage. It reports that ST's Neural-ART accelerator, available in the STM32N6, is planned for the STM32H7P in the first half of 2027 and is included in the STM32A9. That would make the STM32A9 more than a safety controller: it would also be a safety-rated part with an on-chip NPU.
Unite.AI also reports a proof of concept with NVIDIA built on the Holoscan Sensor Bridge and running on STM32H7. ST says it will ship as an STM32Cube component in the coming months. None of these items has a published part number, spec sheet or ship date beyond the timing above.
What It Means for Makers
The STM32A family is not a hobbyist product. ASIL ratings, automotive qualification and "sampling to customers" all mean these chips go to Tier 1s and industrial OEMs first. Nothing has been announced about distributor stock, development boards or pricing, and you should not expect an STM32A board on a maker's bench before the 2027 availability window at the earliest.
The platform message does matter on the bench, though. Circuit.rocks, writing for hobbyists on November 5, puts it simply: skills you learn on a cheap STM32 board carry over to the industrial, robotics and automotive parts. If ST holds to Cortex-M continuity and one STM32Cube ecosystem, the STM32Cube tools and code you use on a cheap STM32 board should carry over to the industrial and automotive parts, though ST has not yet shown that for STM32A. For students and for engineers moving from prototype to product, that portability is the practical benefit.
Circuit.rocks also confirms that the STM32A4, A5 and A9 are sampling now, that wider availability is expected next year, and that no price has been given. For something you can buy today, it points to the STM32G5, built on a 280 MHz Cortex-M33 and orderable from US$1.034 each in 10,000-unit quantities, as reported on November 5, 2026. That is a volume price, not a single-unit or dev-board price.
Status summary: the unified STM32 platform is announced. The STM32A4, STM32A5 and STM32A9 are sampling to customers, with broader availability planned for 2027, the STM32A9 first in H1 2027 per Unite.AI. Neural-ART on STM32H7P is planned for H1 2027. The Holoscan Sensor Bridge component is expected within months. None of the new parts has a public price yet.
