Global Chip Shortage Hits Home
Discouraging news from the BMOW manufacturing front: the global chip shortage seems to be growing worse, and is now causing major problems for some of my vintage computer products. This is a change from a year or two ago, when the shortage mostly affected new high-end semiconductors that BMOW doesn’t use anyway. Since then I’d noticed that prices were rising on lower-end semiconductors, with some parts shortages, but I was mostly able to absorb the costs or find alternative parts. In the past three to six months, the landscape has changed. Many of my essential parts have become difficult or impossible to find. This jeopardizes the continued viability of several BMOW products, including Yellowstone and the Floppy Emu.
When searching for semiconductor parts using a specialized search engine like Octopart, at first it may look like the major suppliers are sold out but secondary suppliers still have stock at reasonable prices. Unfortunately I’ve learned not to trust this, and some of the secondary suppliers seem to be intentionally deceptive. They list parts they don’t have, at prices they’re not actually willing to accept.
Twice in the past week, I’ve placed an order with a secondary supplier for thousands of dollars in hard-to-find chips. The order is made and confirmed and paid. My credit card is charged $4000. Feeling good that I’ve sourced the hard-to-find part, I go ahead and buy all the other parts needed to begin manufacturing. And then three days later I get an email from the supplier saying “those parts are already sold, sorry”. Or worse, “the price on those parts you already bought for $5.17 is retroactively increasing to $20.97.” I try not to take it personally, but I can tell you I really blew up over that second one.
Yellowstone’s First Manufacturing Run May Be Its Last
I’ve been working on the design of the Yellowstone universal disk controller for Apple II for more than four years. I’m finally almost to the point of starting manufacturing. At the time I started the design, the Lattice MachXO2-1200 FPGA was a widely-available part for about $5. Now it’s basically impossible to find. A few of the questionable secondary suppliers may have some, for a much higher price, or their listings may be fiction. I could substitute the next higher member of the same FPGA family, the MachXO2-2000, but that’s also unavailable.
Will MachXO2-1200 availability improve any time soon? I’m not counting on it. I have about 250 of them that I bought last summer, so in theory I can manufacture at least that many Yellowstone cards, if I can also find the other necessary components. Even very common parts like a 74LVC244 bus driver are becoming difficult to find, so that’s not guaranteed. Unfortunately this means Yellowstone’s first manufacturing run may also be its last manufacturing run, at least for a while.
The Future of the Floppy Emu
The inability to build Yellowstone cards would be very unfortunate, but the inability to build Floppy Emu boards would be disastrous from a business standpoint. But I fear this may be where things are headed. The two primary chips used in the Floppy Emu are the ATMEGA1284 microcontroller and the XC9572XL CPLD, and both of them now have limited or no supply. I’m currently in the midst of starting a new production run of Floppy Emu boards, and it’s been very tough to source all the parts. I was eventually successful (assuming suppliers don’t retroactively cancel or modify my paid orders for a third time), but in six months when I go to do this again, there’s a good chance the parts simply won’t be available anywhere at any price.
Forcing a Redesign
Faced with zero availability of a key part, the only real solution is to redesign the product to use a different part. That costs lots of time and money, could introduce new bugs, and is definitely not something I’m eager to do. But if the alternative is retiring the product, do I really have a choice? There could be some upside too, as different parts or newer parts might eventually help support new product features.
If I’m going to redesign an existing product around new parts, it’s absolutely critical that the new parts are free from supply risk themselves. I can’t spend months redesigning a product, only to have the new parts become unavailable just like the old ones did. So if I’m choosing a new part now, I want to see that it’s available from several different authorized suppliers, and that the suppliers have thousands of them currently in stock. That will tell me it’s a popular high-volume part, not some niche part that may have its own supply problems in six months.
Looking for alternatives to the XC9572XL, I searched for other programmable logic parts with a similar amount of logic space (72 macrocells here). I eliminated all the parts in BGA packages, because I’m just not prepared to tackle BGA prototyping or assembly – I need chips with actual exposed pins. Then I eliminated all the parts that don’t have widespread availability and thousands of stock. That didn’t leave many options remaining. The leading choices are probably the XC9536XL, which is the same part I use now except with half the logic space, or else some low-end members of the same MachXO2 FPGA family that I’m using for Yellowstone.
Microcontrollers – The New Unobtanium
Next I turned my attention to finding an alternate microcontroller, and this is where my heart really sank. I searched for microcontroller options with:
- at least 40 I/O pins
- at least 16K RAM
- exposed pins
- a mainstream architecture like ARM or AVR
- stock level in the thousands
The result of my search was basically nothing. I was floored.
What happened to all the Atmel microcontrollers? The STM32 stuff? Texas Instruments? Microchip? Anybody? Hello?
I tried going directly to the ST Microelectronics store, and viewed their STM32 Mainstream MCU section. There are 1152 different microcontrollers in this section, of which only 16 are in stock! How is that possible? Has their entire factory shut down?
OK, you’re right, it’s a lie to say there are no microcontrollers that satisfy my search. DigiKey shows 932 results when I ignore the stock level. But if I’m going to bet the farm on a new MCU, I want to see a stock level in the thousands, and that limits the choices to just a few dozen, virtually all of which aren’t what I’d consider “mainstream”. It’s 2021, am I really going to redesign my whole product around a dsPIC or some weird thing from Renesas? Does anybody actually use the AVR32 stuff for real products? The ARM Cortex parts dominate the microcontroller space these days, and they’re available from several different manufacturers. If I’m starting a new design, that’s probably the most sensible choice to ensure long-term availability. So let’s make that a requirement.
Despite their huge popularity, ARM Cortex MCUs matching my requirements just aren’t available in large quantities from anyone right now. What are other businesses using for their new microcontroller-based product designs? I don’t know.
If I eliminate microcontrollers from obscure vendors I’ve never heard of, the only parts I can find that might possibly work and are widely available are:
- Atmel SAM3 and SAM4 families
- Texas Instruments Tiva C family
That’s the whole list. I don’t really know anything about the Tiva C or the Texas Instruments dev tools, so that leaves the Atmel SAM stuff as the only choice. The available options aren’t the most desirable ones, though. If I were choosing an ATSAM part, I’d probably choose a Cortex-M0, or something like the popular ATSAMD21 or ATSAMD51. But the available options are in the less popular members of the SAM family like the SAM4L and SAM3S. So… yeah. I’m not sure what to do. I’m afraid we’re headed for a difficult time ahead. Here’s hoping the global chip shortage starts to ease soon.
Read 24 comments and join the conversation24 Comments so far
Leave a reply. For customer support issues, please use the Customer Support link instead of writing comments.
Any mileage in a £10 raspberry pi zero? That isn’t going anywhere any time soon and could add cool stuff like WiFi and a larger GUI .not sure if it’s fast enough to bit bang what you need, but it has been used as a 2nd processor on a BBC micro so it’s certainly pretty capable. Good luck. I
I don’t think so, that would be a very different type of product. The Pi Zero also has limited availability and single unit quantity ordering limits at most stores. I’m looking to keep the same general product design, just replace the components with similar ones with better availability.
I’ve had to switch some of my low volume projects to the RP2040, which is the chip at the heart of the Raspberry Pi Pico. It has a lot of IOs as well as some bit-banging peripherals. It still seems to be available, maybe because it’s new enough that it’s not being used in large quantities.
The big problem? I can’t get level shifters!
Redesigning the product for a temporary chip shortage seems extreme. I guess it must be considered if your livelihood depends on it.
ATSAMD21 is readily available from Microchip. Yes, they’re “out of stock”, but they’re regularly restocking and will tell you what date your order will ship. One thing nice about Microchip is that you can order direct in any quantity and they don’t write you off for being small.
https://www.microchipdirect.com/product/ATSAMD21J18A-AUT
I guess I could have worded it better than “readily”. It’s relative. I have had to wait up to 3 months for Microchip orders during the shortage.
The ATMEGA1284 and XC9572XL are already both “legacy” status, so it’s not obvious whether their supplies will ever bounce back. But in general the chip shortage seems to have gotten much worse in recent months. The ATSAMD21J18A that you linked says “Order now, can ship on 05-Dec-2022”. Virtually the whole ATSAMxxxx product line is the same way, and I didn’t find anything with an estimated date earlier than July 2022. There are still a few specific packages and variants that have some remaining stock, but they’re like the three cans of beef stew left on the supermarket shelf after a blizzard warning induces lots of panic buying. I’m expecting they’ll be gone too soon. So like it or not, I think we’re going to be dealing with this chip shortage for the next year or more.
I’m a fan of the STC line of chips (8051 architecture). Sadly they now cost a few dollars each (!) and whilst they can go to 64K of ram you need to use a separate RAM chip (they top out at 8K internally). I’ve never tried that and I suspect it eats all your pins.
Misc notes of rambling anyway, just in case you’re curious:
* Toolchain is really simple and easy to assemble: sdcc compiler (FOSS), stcgal programming software (FOSS) and you use a cheap USB UART as the programmer. I usually put these programs into my source code folders, so anyone else on the project can just run compile.sh (or compile.bat) and immediately program the chips on their device.
* Have been around for many years, lots of cheap chinese gadgets use them. Multiple generations (STC15, STC8, etc) all compatible but with better clocks-per-instruction figures (many instructions are now 1-clockcycle).
* Docs are mostly OK, they provide English manuals with example code for each feature in both C and asm. IIRC some of the newer chips sometimes only have Chinese docs, but there are only small changes between generations so you can read the wrong manual and everything still works xD imagine doing this with an ARM part…
* 8051 means there are lots of compilers that support it.
* 8051 means the IO is mostly standardised (timer addresses, feature bits, etc); so porting from one 8051 brand to another is typically not too difficult.
* 8051 means it has 4 different types of RAM. AVR really wins against this PITA. Most of the time you don’t have to notice, but it’s worth specifying C pointers to point to one type of RAM only (eg __xdata) so that the processor doesn’t have to fork into different versions of code at runtime depending on the destination pointer’s type.
https://lcsc.com/products/Microcontroller-Units-MCUs-MPUs-SOCs_11329.html?brand=151
I wonder how responsive micro companies are to directly ordering? Eg asking for X in package Y (perhaps even with a higher-than-normal spec of memory)? Might be worth trying to contact AVR/microchip directly to order a reel?
N.B. this post is a mixture of frustrating and hilarious to read Steve, thankyou for writing and sharing it. I used to read stories of people’s orders getting second-sold even after they pay for them, that’s absolutely crazy and ridiculous, but I guess in this climate there is no need to build up customer loyalty/respect.
I’m currently working a project that (used to) use MK20DX’s. They’re complete unobtanium. Their sister model processors (eg in bigger packages) are in stock, but I tried those and I can’t get them to boot up properly. Implementation manuals are both 1500 pages long, with slight differences sprayed everywhere.
https://www.eevblog.com/forum/microcontrollers/clock-not-runningtrying-to-move-teensy3-2-project-bare-mk20dx/
I gave up when we realised I wasn’t clever enough to get this working within our schedule. I’m now looking at maybe using Espressif ESP32’s instead, they seem to be able to hold better stock levels (?) than other people. Or just manufacturing in a much smaller scale, minimising the goals of our research project and using off-the-shelf Teensies as plugin modules instead.
You’re right, I’m an idiot, and misread the year. Six months ago, Digikey stock was a year out but Microchip would ship much quicker. Now we’re to the point of Microchip being a year out, and Digikey being “unknown”.
The thing about designing around a parts shortage is that we have to change the design for each batch, based on availability. And when the shortage is over, we’ll want to redesign again, because who wants to be stuck paying premium TI prices when a Microchip part costs 4x less.
There have been conversations at eevblog and such places where small electronics creators are essentially deciding to change industries until the shortage is over.
> The thing about designing around a parts shortage is that we have to change the design for each batch, based on availability.
Right, exactly! I’d be willing to do one round of redesign, if it put me on a much stronger footing for the future. But when virtually all of the common microcontroller families are facing shortages, there’s no obvious choice for a “stronger footing”, and it would be crazy to attempt a redesign every six months to chase whatever is available at that moment.
The whole situation is just hard to believe. Look at STM32, which is an extremely popular microcontroller family. Digikey lists 2809 different STM32 parts in their catalog, of which only 23 are currently in stock. That’s over 99% of the whole product line unavailable. Wow.
I don’t really want to change industries, but maybe I need to get a second job. Or maybe I should have accepted that supplier’s $20.97 revised price for the part I need.
I ran into the supply chain blues buying passive components for repair jobs. Things like capacitors, nothing fancy, have lead times of 6-12 months if they don’t happen to be in stock at the usual vendors.
The vintage electronics communities I frequent strongly suspect a ton of companies are going to trim their offerings, which usually means slow sellers like transistors commonly used in vintage stuff, are going to get the axe. I would not be surprised if anything marked “legacy” or “not for new designs” lands up getting killed off for good.
Thing is, where is all this inventory being used? Clearly there is demand for these micro controllers, so discontinuing them would inconvenience a lot of customers. Chip shortages of cutting edge 7nm CPUs and video cards gets all the press, but these humble micro controllers run far FAR more of society.
Have a look at EFM32. There is a good stock of some of them at Mouser. When it comes to the 5V CPLD I think 5V parts are becoming obsolete anyway. You could consider a Coolrunner CPLD (thousands in stock at Mouser) with external level shifting.
Stay away from NXP/Freescale parts like Kinetis (they go into cars). And stay away from parts that have automotive qualifications like AEC-Q100 (there is a parameter at Mouser for that).
As far as I know, the supply of semiconductors and other parts hasn’t really decreased. The problem is that demand has gone crazy, and the global shipping network is struggling, so the supply can’t keep up. Then just like toilet paper under COVID lockdowns, companies are pre-buying large volumes of parts and stockpiling them for later, which makes the shortages even more severe. Eventually this should all unwind and return to normal, but my prediction is it’ll get worse before it gets better. I agree a general shortage of microcontrollers will have far-reaching effects for a huge range of products.
I would expect legacy parts to continue being produced for as long as it’s profitable for the manufacturers. I’m no expert, but I don’t think old assembly lines can easily be converted to new products and processes. So they probably wouldn’t discontinue a legacy part in order to make more of a newer part – I don’t think it works that way.
M0AGX have you worked with the EFM32? That’s one of the “obscure vendors I’ve never heard of” that I eliminated from my search. Maybe I can’t afford to be so choosy under these conditions, but I’d prefer to stay with a mainstream vendor with wide community support. Cool-runner II could definitely be an option for a CPLD replacement, good suggestion. The other options that I mentioned currently offer better value in logic resources per dollar, but that could change.
I use EFM32 a lot, basically every time I can. I like these parts because Silabs’ software libraries (emlib) have the right abstraction level and just works. Not a single big ball of mud like STM32Cube and not loose bag of random parts like Freescale (at least in the old days). Just simple things like uart_init(UART0, 9600) and off you go without studying all possible clock options. Getting a UART or I2C fully working is trivial. Demos are also pretty useful. The libraries are also easy enough to take out from Simplicity Studio and move into a full custom Makefile project that I tend to use. Jumping between MCU families/sizes is also very easy, peripherals (and emlib APIs) are practically the same. Silicon Labs is not that obscure but certainly less known than ST or Atmel. I find their support forums quite okay.
I don’t know if this is any *new* thing with places like digikey, mouser, etc. I’ve had the exact same experience you did many times over the last decade. They just don’t care about you if you’re not incorporated. It might not be related to the global chip shortage at all. It sounds like their standard practice to me.
Had you look at the AVR128DA and AVR128DB family? They are still 8-bit AVR but are a huge improvement over old ATMEGA1284 and, even if they are not available in the older TQFP 44, you can find them in the TQFP 48 package. And they are even cheaper.
Heeeeey that AVR128DA/DB might be a winner! Thanks, I don’t know how I missed it in all of my searches. It has the best current availability of any microcontroller family I’ve looked at recently, although that’s not saying much. I will say that redesigning around another 8-bit MCU seems slightly backward-looking, but it has the advantage of being similar to what I’m already using (I think?).
Do you have a link to anything that gives an overview of the AVR128 and how it compares to the ATMEGA family? I’m not finding anything but Microchip’s own pages that are full of marketing fluff and short on useful objective info. I would have thought the introduction of a new AVR part would be significant news and there would be some articles about it on Hackaday or AVR Freaks or something.
I haven’t found any comparison document too… so from what I remember comparing the datasheets, both family has 128K flash and 16K ram but the newer ones (DA and DB) come at 24MHz instead of 20Mhz, have more timer/counters and more USART. The distinctive trait of DA family is that it features a 10 bit DAC where instead the DB family has up to 8 pin on a different voltage from the main one. I used one of those to be able to interface an old AY-3-8912(5V) and an SD card (3.3V) from the same MCU.
Good overview of the AVR DA and DB series in comparison to the older ATMEGA stuff is here: https://github.com/SpenceKonde/DxCore This is written by the author of the DA/DB series Arduino core.
I know someone that works for one of the large chip houses that makes ARM compatible chips. I asked him about the extreme shortage of chips you are seeing. He says they have far more orders than they can fulfill right now and in many cases they are quoting 52 weeks, only because that is as far out as they can quote things. They are heavily managing who gets what right now and prioritizing things like supplying customers such that they won’t get designed out of products. But that isn’t the only thing affecting the supply. Their ability to provide chips is hampered by their own supply chain issues and their access to fab time. And when designing test fixtures for chips, they are having problems finding the components THEY need to build those, just like you are, which further slows everything down.
He says he doesn’t see things improving much until Q3 or Q4 of next year.
Yes yes, hard to find parts. This is not news and has been the case for over one year for all of us indie sellers. Not to mention the international shipping issues caused by the pandemic. Here’s the thing though, you’re not alone in being unable to source (and ship) parts.. so who cares if it takes another year or two to release your product? Just stick to your good and working design, and wait it out. It’s not like Toyota or some who does have your parts will suddenly copy your idea and start selling it by the boatload. Sit back and chill, up your skills, design something new, and have a warm beverage. It’ll be fine.
NotJohn, you’re missing the point. Steve can’t just wait for things to eventually show up, as he needs product to sell in order to make a living. It’s the same issue that is affecting car dealers here, on the east coast of the US. There are extremely few cars being delivered, so that dealers have nothing to sell, and that’s pushing them out of business.
Redesigning a product for a temporary chip shortage seems extreme. I guess if your livelihood depends on it, it has to be considered.
ATSAMD21 is available from Chimicron. Yes, they are “out of stock” but they restock regularly and will tell you the ship date of your order. One of the great things about Chimicron is that you can order any quantity directly and they won’t turn you away because of your small size.
https://www.chimicron.com/product/06463538-ATSAMD21E17D-AUT.html
I suppose “temporary shortage” is relative, but we’re talking about key components being potentially unavailable for a period of multiple years. For example the Lattice FPGA used in Yellowstone has been out of stock since 2021 and is current estimating availability of new parts in mid-to-late 2023. At that level of shortage, you either redesign or exit the market.