4 ms·

The lead times for the STM microcontrollers have gone way beyond unreasonable. "730 days" I think is a polite way of saying "we may never supply you with this c

by laydn 5y ago

The lead times for the STM microcontrollers have gone way beyond unreasonable. "730 days" I think is a polite way of saying "we may never supply you with this chip". 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-microcontrollers/685?s=N4IgTCBcDaIM4BcC2BmCBdAvkA 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.

twarge 5y ago
STM32H7 is the upgrade path, but it doesn't seem to be any more available.
- Inviz 5y ago
  I bought one from their site.
thrtythreeforty 5y ago
The part number explosion may not be as bad as you think: ST very likely does some relabeling/fuse programming of the same die to make multiple part numbers. As the most classic example, the STM32F103C8 promises 64KB of flash, and the F103CB (note CB vs C8) has 128KB. But it turns out that some of the 64KB actually have 128KB [1], and they've just been fused to report the smaller part number. [1]: https://mecrisp-stellaris-folkdoc.sourceforge.io/stm32f1xx-diagnostics.html https://mecrisp-stellaris-folkdoc.sourceforge.io/stm32f1xx-d...
- janekm 5y ago
  True, and then every variant in every package version. But it does cause problems in the current climate as each of those variants do have to make it through the production pipeline (after die fab) which is also constrained at the moment. And of course ST is prioritising "high value customers" (like car manufacturers).
  - thrtythreeforty 5y ago
    Oh for sure, it's not like fab capacity is the only bottleneck at this point. Seems like everything is the bottleneck.
  - baybal2 5y ago
    Car companies are just few percents of MCU market. A car may have a lot of MCUs, but that's probably less than 30 per economy car. The entire car industry is a drop in the ocean for The Big Semi. 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.
    - mastax 5y ago
      > On other hand, I've seen very expensive STM32s being used in near semi-disposable goods. Like a single-use disposable electronic COVID test: https://www.youtube.com/watch?v=UvArprBmdFA https://www.youtube.com/watch?v=UvArprBmdFA (though this one uses an nRF52)
  - varjag 5y ago
    Contrary to the popular opinion (which can be traced back to one or two newspaper articles), it's not the car companies causing this. All sorts of material has evaporated from markets, in grades beyond automotive.
PragmaticPulp 5y ago
> 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 https://www.digikey.com/en/products/filter/embedded-microcon... ). 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.
sizzle 5y ago
Can they not use an off the shelf board like Raspberry Pi?
- renton_wal 5y ago
  Raspberry Pis, as are most consumer sbcs, are extremely unsuitable for an embedded, high reliability consumer device. For a whole bunch of reasons, from reliability, to form factor, to pricing, to technical concerns (such as IO).
  - xyx0826 5y ago
    Raspberry Pis have commercial and industrial use cases so they can have pretty good reliability. There’s an official “for industry” page [1], and there are PLC products made with the Pi [2]. I’m curious on why do you think the Pi (including the Pi Compute Module)’s form factor make it unsuitable for consumer devices that need computing power a SBC offers? [1]: https://www.raspberrypi.com/for-industry/ https://www.raspberrypi.com/for-industry/ [2]: https://www.industrialshields.com/industrial-plc-raspberry-pi-202007 https://www.industrialshields.com/industrial-plc-raspberry-p...
    - PostOnce 5y ago
      Generally, actual industry would use the pi compute module instead of the pi itself, and you plug the compute module into a carrier board with better power, storage, and IO options (i.e., rs485, 24V DC, eMMC instead of SD, etc) One of a million examples: https://pipci.jeffgeerling.com/boards_cm/edatec-cm4-industrial-computer.html 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.
      - xyx0826 5y ago
        > 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. 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.
quesera 5y ago
"730 days" is, of course, exactly two non-leap years. 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.
negative_zero 5y ago
A little bit off. If you are interested: The link you have there includes several different ways of ordering the same part but packaged differently: Cut Tape, Reels, Tubes, Trays etc. Some of those have will different sizes of packaging (eg tray of 50 vs tray of 200). I'd divide that 3000 by 3-5. 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 :)
- klintcho 5y ago
  There has been a number of submission on HN discussing the chip shortage - and I’ve yet to see anyone discussing using second hand chips, scavenging from used electronics. Obviously as described in the post: redesign is always going to be expensive and cause delays if even possible. But I’m still curious as to why this would not be possible or at least create a much bigger immediate market for old chips?
  - RF_Savage 5y ago
    Reliability and in general it not beimg able to scale salvaging 5 chips to salvaging 50 000 chips. Your margins well be gone if there are any significant failures in the field. Salvage is ok for R&D or high cost assemblies with big FPGA's.
  - rcxdude 5y ago

by laydn 5y ago

twarge 5y ago

STM32H7 is the upgrade path, but it doesn't seem to be any more available.

Inviz 5y ago
I bought one from their site.

thrtythreeforty 5y ago

The part number explosion may not be as bad as you think: ST very likely does some relabeling/fuse programming of the same die to make multiple part numbers. As the most classic example, the STM32F103C8 promises 64KB of flash, and the F103CB (note CB vs C8) has 128KB. But it turns out that some of the 64KB actually have 128KB [1], and they've just been fused to report the smaller part number. [1]: https://mecrisp-stellaris-folkdoc.sourceforge.io/stm32f1xx-diagnostics.html https://mecrisp-stellaris-folkdoc.sourceforge.io/stm32f1xx-d...

janekm 5y ago
True, and then every variant in every package version. But it does cause problems in the current climate as each of those variants do have to make it through the production pipeline (after die fab) which is also constrained at the moment. And of course ST is prioritising "high value customers" (like car manufacturers).
- thrtythreeforty 5y ago
  Oh for sure, it's not like fab capacity is the only bottleneck at this point. Seems like everything is the bottleneck.
- baybal2 5y ago
  Car companies are just few percents of MCU market. A car may have a lot of MCUs, but that's probably less than 30 per economy car. The entire car industry is a drop in the ocean for The Big Semi. 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.
  - mastax 5y ago
    > On other hand, I've seen very expensive STM32s being used in near semi-disposable goods. Like a single-use disposable electronic COVID test: https://www.youtube.com/watch?v=UvArprBmdFA https://www.youtube.com/watch?v=UvArprBmdFA (though this one uses an nRF52)
- varjag 5y ago
  Contrary to the popular opinion (which can be traced back to one or two newspaper articles), it's not the car companies causing this. All sorts of material has evaporated from markets, in grades beyond automotive.

PragmaticPulp 5y ago

> 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 https://www.digikey.com/en/products/filter/embedded-microcon... ). 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.

sizzle 5y ago

Can they not use an off the shelf board like Raspberry Pi?

renton_wal 5y ago
Raspberry Pis, as are most consumer sbcs, are extremely unsuitable for an embedded, high reliability consumer device. For a whole bunch of reasons, from reliability, to form factor, to pricing, to technical concerns (such as IO).
- xyx0826 5y ago
  Raspberry Pis have commercial and industrial use cases so they can have pretty good reliability. There’s an official “for industry” page [1], and there are PLC products made with the Pi [2]. I’m curious on why do you think the Pi (including the Pi Compute Module)’s form factor make it unsuitable for consumer devices that need computing power a SBC offers? [1]: https://www.raspberrypi.com/for-industry/ https://www.raspberrypi.com/for-industry/ [2]: https://www.industrialshields.com/industrial-plc-raspberry-pi-202007 https://www.industrialshields.com/industrial-plc-raspberry-p...
  - PostOnce 5y ago
    Generally, actual industry would use the pi compute module instead of the pi itself, and you plug the compute module into a carrier board with better power, storage, and IO options (i.e., rs485, 24V DC, eMMC instead of SD, etc) One of a million examples: https://pipci.jeffgeerling.com/boards_cm/edatec-cm4-industrial-computer.html 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.
    - xyx0826 5y ago
      > 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. 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.

quesera 5y ago

"730 days" is, of course, exactly two non-leap years. 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.

negative_zero 5y ago

A little bit off. If you are interested: The link you have there includes several different ways of ordering the same part but packaged differently: Cut Tape, Reels, Tubes, Trays etc. Some of those have will different sizes of packaging (eg tray of 50 vs tray of 200). I'd divide that 3000 by 3-5. 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 :)

klintcho 5y ago
There has been a number of submission on HN discussing the chip shortage - and I’ve yet to see anyone discussing using second hand chips, scavenging from used electronics. Obviously as described in the post: redesign is always going to be expensive and cause delays if even possible. But I’m still curious as to why this would not be possible or at least create a much bigger immediate market for old chips?
- RF_Savage 5y ago
  Reliability and in general it not beimg able to scale salvaging 5 chips to salvaging 50 000 chips. Your margins well be gone if there are any significant failures in the field. Salvage is ok for R&D or high cost assemblies with big FPGA's.
- rcxdude 5y ago