Answer the questions an experienced engineer would ask you, and get a shortlist of families worth evaluating — plus, more usefully, every family that was ruled out and the single constraint that did it. 42 families surveyed, at family granularity, from a dataset kept separately so it can be corrected as it ages.
Runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere. Survey last updated 2026-09-23, 42 families.
There are several thousand microcontroller and SoC part numbers in production, and no selection tool can pick one for you honestly. What it can do is get you from every family on the market down to the five or six worth an afternoon each, and tell you exactly why it discarded the rest — which is the more useful half, because the reasons are where you discover that one of your requirements is the expensive one.
So this works at family granularity, never at part granularity. When it says a family offers CAN-FD and PCIe, that means parts in the family commonly offer those; it does not promise a single part offers both. Once you have a shortlist, the vendor's own parametric search is what picks the part, and nothing here substitutes for it.
The table of ruled-out families is sorted so that anything eliminated by a single constraint appears first. Those are the near misses, and they usually prompt the most productive conversation on a project: a family that fails only on temperature grade might have an automotive variant; one that fails only on longevity might be fine if the product is a three-year consumer device; one that fails only on Linux capability might work perfectly if the display could move to a separate module.
If nothing survives, that is a real result too, and almost always means two requirements are fighting. ASIL-D with Linux on one chip, a coin-cell power budget with Ethernet, or 150 °C with an edge-AI NPU are all combinations that push you towards splitting the design across two devices — which is what the empty result is telling you.
The dataset behind this tool is kept as a separate, versioned artifact rather than being buried in the page, with its own provenance notes, per-family confidence marks and an explicit list of its own limitations. That is deliberate: a dataset this size will be wrong somewhere and will go stale everywhere, and it should be correctable without touching the tool that reads it.
Entries marked medium confidence are ones where the family is broad or moving quickly — new parts add PCIe, TSN or an NPU, and older ones quietly move to not-recommended-for-new-design. Treat those as prompts to check rather than as facts. The figures throughout are bands and typical family characteristics, not datasheet values for a specific device.
Where a family is marked ASIL-D or SIL3, that means the vendor offers a safety package — a safety manual, an FMEDA, a safety element out of context, and usually a certified compiler and diagnostic library. It does not mean the device is "ASIL-D" on its own, because no component is: the integrity level belongs to your item, and the silicon is one input to that argument. Choosing a part with a mature safety package saves a great deal of work, but it does not remove the obligation to do the FMEDA on your design.
If that is the road you are on, the ASIL determination calculator establishes the target, FIT and PMHF shows what each element has to reach, the safety mechanism selector covers what to put around the parts that cannot reach it alone, and decomposition is the lever when a single device cannot carry the goal. If the shortlist has pushed you towards an external DRAM interface or a high-speed serial link, the memory bandwidth and bus bandwidth calculators size those, and the stackup planner turns the resulting BGA into a board.
First compiled 2026-09-23 and revised since. Every entry records the date its vendor link was last checked, so a stale row is visible in the dataset rather than only in hindsight.
| Date | Change |
|---|---|
| 2026-09-23 | Added Puya PY32F0, Artery AT32F4, Allwinner T113 / D1, widening coverage of the low-cost Chinese vendors. The list already had WCH and GigaDevice but nothing at the true cost floor (Puya is roughly eight cents in volume, below many 8-bit parts), nothing at the STM32-alternative performance tier, and nothing Linux-capable that was not a Western part at a Western price. Each entry states the cost of choosing it as well as the saving -- thin English documentation, weak errata discipline, short or absent longevity commitments, and community rather than vendor kernel support. |
| 2026-09-23 | Added MCX N, MSPM0, STM32N6, Versal AI Edge. Each checked against the vendor's own page or datasheet on the day, not written from memory; urlChecked records that. All four were genuine holes rather than refinements. MCX is where NXP's general-purpose roadmap went after LPC and Kinetis; MSPM0 is TI's low-cost 32-bit line; STM32N6 is an MCU-class part with a 600 GOPS NPU and no internal flash, a category the list jumped over entirely; and the fpga_soc class held only Zynq-7000 and PolarFire, with no entry for the current AMD adaptive SoC. |
Each of these runs entirely in your browser. Nothing is uploaded, stored or sent anywhere, and none of them needs an email address.
See all 51 engineering tools →
We add tools here fairly often and write up the things worth writing up — a stackup that behaved oddly, a standard that turned out to be obsolete, a calculator that was quietly wrong. Join and you get told when something new lands.
One email a month at the very most, and usually less. No drip sequence, no sales cadence, no sharing your address with anyone. Unsubscribe whenever you like.