Other filters will be same part but different temperature range; same MCU but different footprint (which would require a board re-spin but not normally the end of the world).
You'll also get SKUs where different peripherals and flash is available which is generally the same die, but not everything worked when tested at the fab so it will get binned appropriately.
There's other filters but I guess I am saying the total number of SKUs of die that are getting made is not anywhere near 3000.
You are right with the 730 day lead time. Actually anything beyond a year you can interpret that way.
Even when there is stock, the parts will be on allocation anyway. Normally the biggest existing customers get first dibs and everyone outside of that circle is given an absurd lead time. Sometimes you can push your way in if your order is big enough.
I feel for the PlayDate team. It's not a fun time when this happens. Even my usual rule of "make sure there are at least two alternatives if possible" is failing in this shortage.
We've even resorted to ordering dev kits and scavenging parts from there which is good fun :)
Salvage is ok for R&D or high cost assemblies with big FPGA's.
[1]: https://mecrisp-stellaris-folkdoc.sourceforge.io/stm32f1xx-d...
On other hand, I've seen very expensive STM32s being used in near semi-disposable goods.
Car companies are doomed because they required special "automotive grade" MCUs, which are usually just very long running die series with a lot of testing.
Being on old die users, they can't share the benefit of much larger mainstream MCU die batches. They are now in the end of the queue waiting for big batches of dies for mainstream MCUs to finish manufacturing.
IMHO, lots of auto parts uses automotive grade MCUs without any particular necessity. It's just because they were well moneyed, and they can. Some Mercedes car literally have one MCU per button, which just sits on CAN bus, and blinks an LED.
Like a single-use disposable electronic COVID test: https://www.youtube.com/watch?v=UvArprBmdFA (though this one uses an nRF52)
So it seems possible that "two years" is just an overflow in inventory management restocking projections.
I.e. "more than we can measure", "unknown but not soon", etc.
Digi-Key is great for prototyping and small-scale production runs, but once you get into mass manufacturing your CM (contract manufacturer) is buying these parts for you through more direct channels.
STM can produce small numbers of the various different SKUs and send them off to stocking distributors like Digi-Key to support small-scale prototyping and smaller production runs, but they don't aim to stock everything in high quantities everywhere.
Unfortunately, when supply chains get tight the smallest players suffer the most. The big manufacturers went out and locked up their supply contracts when this all started. Companies like STM don't mind because it's much easier to do big manufacturing runs for a few customers than to try cater to small orders all over the place.
Of course, they'd prefer to have the capacity to serve everyone and sell as many chips as possible, but obviously we're bottlenecked on the production side relative to all of the demand.
[1]: https://www.raspberrypi.com/for-industry/
[2]: https://www.industrialshields.com/industrial-plc-raspberry-p...
One of a million examples: https://pipci.jeffgeerling.com/boards_cm/edatec-cm4-industri...
Having your power connector come loose on a vibrating machine or having your SD card die is a serious problem when you're trying to use a machine to make money reliably and consistently.
Anyway that's unrelated to GP really, raspberry pi sucks too much power for playdate, I think.
Hah, I’d never thought of the power connector coming loose but I can see how it will be a real issue in an industrial environment, or on vehicles. I’ve also heard a lot of stories of the Pi’s SD card corrupting or dying, though I agree that actual products might use eMMC instead.
I think the Pi is better suited for less portable products like NAS, or TV boxes, although we already have popular inexpensive SoCs for those.
> unsuitable for consumer devices that need computing power a SBC offers
PLC is not "consumer". "for industry" devices often have service contracts and dedicated staff whose job it is to ensure reliability and spares, and vastly different requirements than a plug and play "consumer" device.
You've compared two very different use cases as if they were alike.
Consumer devices actually have much stricter reliability requirements than many industrial applications - if you look at margins, you'll see that any return is a huge dent. Versus industrial, where swapping out a $50~ sbc part once a year is still vastly cheaper than "high reliability" solutions which cost into the thousands. Maybe you lost half a day of work, but that's probably acceptable for a small shop running a lathe that saved $$$$ on upfront capex.
And, there's no way anyone's using raspberry pis for high availability, such as in manufacturing.
But back to those consumer requirements - there's a lot of software and hardware requirements that you get "for free" with more expensive, traditional solutions. For example,
- verified boot (pre-boot checksum/signing verification, and chain of trust)
- multiple boot partitions with fallback to prevent against bad updates
- image based updates and recovery
- out of band network recovery modes
- guaranteed flash failure rate and yield
- minimal form factor with reduced components for unneeded features
and a bunch of other small stuff that all shaves off failure rates by .5% here and there, and yet is completely worth it to anyone deploying consumer devices. Some of that you could probably implement yourself, but make a mistake versus using a tried and tested solution... well, that's more margins and profit lost.
By the way, in another comment you said "Pi is better suited for less portable products like NAS, or TV boxes". Well, yeah, 'cause it's basically a set-top box SOC that was put into a hobbyist board. ;) The Pi has a TON of "undocumented" features and modes around being a set top box. Broadcom will sell you basically the same GPU/CPU package with a lot of those features listed above.
I may be way off here but my observation is that chips that are very configurable are the hardest to purchase. For example, the STM32 series have something like ~3000 SKUs ( https://www.digikey.com/en/products/filter/embedded-microcon... ).
Normally, this allows every designer to pick the most suitable part and have the lowest total BoM cost. However, in today's challenging supply/demand environment, the total number of unique parts that must be manufactured must be causing a huge problem.