Likely production life of an ST ARM CPU?

[peter-h]

Looking at this

https://www.st.com/content/st_com/en/about/quality-and-reliability/product-longevity.html#10-year-longevity

it shows the 32F417 at 10 years, since January 2021.

How actually likely is it to be that short?

I know this is a different era but the Z180 ran for ~30 years, the H8/323 ran for ~25 years (and then had at least another 10 years of free availability from the US "cowboy" resellers, suffering only at the very end from chinese fakes which were an empty package) and the only chips I have used which were significantly shorter were from Atmel who would struggle to get a 10 year life out of anything.

I have just come across a chip called ESP32-WROOM-32E which claims a 12 year life but is so specialised that I would be amazed if any product based on it would be made for that long. But that CPU is unbelievable, especially for the incredibly low price of £2.16 one-off. Never heard of Espressif Systems...

[DC1MC]

If it's not automotive, you can forget about comitted production life of anything semiconductor of more than 10yrs or around, the industry is too dynamic and in the era of fabless and "we do everything at TSMC", the manufacturing technology itself may disappear and the chip must be redesigned.
 
Of course, for the especially successful chips, they may be be produced for an extended time or pin/feature compatible clones may appear, but there is absolutely no warranty and you can't base your business plan on it. The safe bet for many companies is to estimate the production and support necessary inventory and stock it near the end of life of the chip.

Cheers,
DC1MC

[peter-h]

I am sure the whole chip is designed in VHDL or whatever, so it is fab-portable. That is how the chinese can make fakes which actually work (not necessarily timing-exact).

Why would 'automotive' run for longer? ECUs and such can be changed. I have just got my hands on an ECU from a KIA car which uses a load of Siemens chips, including one 16 bit uC, and most of it is discontinued. But the ECU is current production, looking at date codes.

[DC1MC]

Automotive usually runs longer (IMHO) because, one, there are regulations in place for availability of parts for an extended period of time and two, the cars are usually a long term purchase, unlike the mobile phones and other consumer electronics and there is a significant demand for parts.

Speaking of production of chips, from the VHDL to actual usable silicon, there is so much pain and effort, usually non-transferable foundry and technology specific know-how, that saying "they have the VHDL, a chip can be produced at any time..." is a bit naive.

Did you look at the chip date codes on a current date ECU, you may be very well see dates around the EOL of those controllers that have been stocked (there is no problem for Siemens or Bosch to stock a couple of hundred thousands of chips for a specific in demand design).

 

[peter-h]

" saying "they have the VHDL, a chip can be produced at any time..." is a bit naive."

except... I didn't say that.

I wonder what controls how long past the last time buy a chip remains available in the "cowboy" market? I have seen about 10 years. I bought tens of thousands of the H8/323 at $6 when the official price was GBP 9, which was surprising since the cowboy outlets normally sell at a premium.

Another was the Atmel 90S1200. We last paid GBP 0.50 and I would buy a few k again if I saw them. Atmel replaced that one with another chip but by the time I got around to changing the design they discontinued that one too I will need to revisit that sometime... or find some more 90S1200.

[DC1MC]

" saying "they have the VHDL, a chip can be produced at any time..." is a bit naive."

except... I didn't say that.

I wonder what controls how long past the last time buy a chip remains available in the "cowboy" market? I have seen about 10 years. I bought tens of thousands of the H8/323 at $6 when the official price was GBP 9, which was surprising since the cowboy outlets normally sell at a premium.

Another was the Atmel 90S1200. We last paid GBP 0.50 and I would buy a few k again if I saw them. Atmel replaced that one with another chip but by the time I got around to changing the design they discontinued that one too I will need to revisit that sometime... or find some more 90S1200.

The last remains in UK:
https://www.silicon-ark.co.uk/at90s1200-12pc-cmos-8-bit-microcontroller-by-atmel

I think the "after-EOL" of a chip depends if it was used in some "millitary strategic project", where they've stoked large amounts and after some years, when the devices using them are fully retired they sell the stock to brokers.
Same with large corporations, that when finally the product they were compelled to support is gone, they're clearing the stocks and of course, the out-of-business situation. But this really does not have any kind of predictability and one can't base his business model on it. It's more of a desperate measure, like when NASA was asking for ceramic 486-25 for some radiation resistant computation module or such and some "cowboys" made bank.

In the end, either your product is important enough and have margins to assure you have enough parts for the whole life-cycle, or not, and then have to deal with the possibility of your original parts drying up and the need to redesign. For the consumer stuff that's no big deal, it may even come cheaper and more features can be added, but for projects where there is a significant certification and/or regulatory burden, one can be severely bitten by neglecting the supply chain, "cowboys" are not always available.

The auto industry (and not only) tried to apply the "new supply chain paradigms", with "elastic stocks" and "just in time ordering" and they're kicking themselves over it, the loses incurred because they didn't want to stock semiconductors and other parts are now way bigger than just buying the stuff and keep it in the storage. So many other in the industry that were doing TED talks with a superior smile saying how cool they are that they've moved the burden of stock to the manufacturers/distributors and how they order only what and when is needed. This model only works with a perfect running supply chain and even a small disruption will crash whole industries.

[JPortici]

I am sure the whole chip is designed in VHDL or whatever, so it is fab-portable. That is how the chinese can make fakes which actually work (not necessarily timing-exact).

Why would 'automotive' run for longer? ECUs and such can be changed. I have just got my hands on an ECU from a KIA car which uses a load of Siemens chips, including one 16 bit uC, and most of it is discontinued. But the ECU is current production, looking at date codes.

- Don't forget about the analog side of things. ADC, comparators and such. VHDL -> Portable between fabs is about as false as ARM -> Portable between families
- Automotive last longer because of regulations. There is a long and costly validation cycle for automotive (and simillar industries), they can't change anything unless it's validated so they need the assurance the part will be available at least for N years

[Doctorandus_P]

Do you have any evidence that suggests that somebody in the '80-ies made a commitment to support the Z80 upto 2021?

That they made a commitment to support it for the next 10 years does not mean the product ceases to exist after that time.
It's more likely that that list is revised every year and the dates pushed a year forward for the forseeable future.


And with STM32, there are at least 6 different "second sources" from China 

[peter-h]

Great comments on JIT supply chain management. It is a huge fallacy and has bitten a lot of people. However, it is always the (small company) supplier who gets shafted into carrying the stock

No, the Z80 "just was" available. Multi sourced, which was a fashion back then, before the industry realised it makes everybody poorer I don't ever recall seeing any production life published. The 10 years in my OP is the first I have ever seen published openly. The Z180 was also sold as a Hitachi 64180 although not in the same packages.

Yes it is interesting where the big stocks of the H8/323 came from. Every cowboy in the US was selling them. Unusually they were going for around 50% of the official disti price. I was buying them until the fakes arrived and then suddenly the stocks dried up. Actually you could tell the fakes from the labels on the outside of the sealed bags. And the forger was so stupid that he even mis-marked it



The 90S1200 I am using is SMT. But I have a fair bit of stock.

[thm_w]

it shows the 32F417 at 10 years, since January 2021.

How actually likely is it to be that short?

Thats just the guaranteed time.
If its popular, like the stm32f103, it will be around a lot longer. That ones in production for what 14 years now? and its still "active" on ST site.

[westfw]

Never heard of Espressif Systems...

Really?  They're the company behind the ESP8266 chip whose various modules heavily 'disrupted' the IoT market, and subsequently is coming pretty close to revolutionary changes in the way that Chinese chip vendors interact with "small players" and hobbyists in the western world.

The ESP-01 apparently first appeared in 2014, so their actual behavior WRT 10-y product lifetimes should be apparent soon.

[peter-h]

Does IoT actually exist?

Regarding the "six" ST 32F chinese lookalikes mentioned above, how close are they and is there a 32F417 version?

I have a bit of a background in digital design including FPGA and looking at the functionality of this device it would take an extremely clever guy to write the code to emulate one of these and to get the timing anywhere near comparable e.g. float multiply in one cycle (obviously there is a 32 bit barrel shifter).

[Kjelt]

I am sure the whole chip is designed in VHDL or whatever, so it is fab-portable.

The fabs only get reticle-sets, so if you stay to the same process technology probably.
If not you have to make new reticles.
Perhaps this is not that big a deal for the digital domain, the core is from Arm so yes. Digital peripherals probably too.
But how about the analog part of the uC ? 
I do know that manufacturers have to redesign their entire layout whenever they change to a smaller process technology and also the used voltage levels change.
Redesign, oompf this is a huge risk for new errors to be introduced and verifying a microcontroller is so difficult , it explains the large errata sheets that grow in size after introduction of a part.
I am not an expert but this is one of the reasons the "old" fabs are still fully booked and also new fabs use older processes. And also a reason newer microcontrollers are using lower voltages to operate or have dc-dc converters on board.

[tszaboo]

How actually likely is it to be that short?

There are two reasons for this:
1) TSMC and others change their fab technology. These are on older processes, but they change nonetheless. New process is usually bettererer, so eveyone wants that.

2) The mans gotta eat. If we design a product that will be on the market for 100 years, where are we going to work in 20 years? So lets just add firmware features and change the hardware ever slightly every few years.

[Siwastaja]

Nobody designed in a Z80 because they though about it being available 30 years later. No one guaranteed such availability.

They designed it in because of what it was, and hoped it would be still available in 5-10 years. Continued availability was thanks to continued demand.

It's exactly the same now. You can't count on most components being available in 10 years, but they may be, just like what happened with Z80.

[peter-h]

The Z80 came out when there were very few other micros, and it got a lot of design-ins. Also a lot of designers were proficient with it and everybody (except fresh from college chaps ) prefers to re-use existing expertise.

The analog parts of a 32F4, yes, but they aren't exactly state of the art. They are "12 bit" which means 10-11 on a good day. This tech is about 20 years old.

What continues to amaze me is that there are people with the brains and the time to write VHDL (or whatever is the current HDL) for the fake chips. For a Z80 it is probably a few weeks' work. For a 32F4 - ? Must be years. Unless it has leaked from ST.

[wraper]

The analog parts of a 32F4, yes, but they aren't exactly state of the art. They are "12 bit" which means 10-11 on a good day. This tech is about 20 years old.

But will need to be redesigned nonetheless. The effort is quite close to spinning a new chip.

What continues to amaze me is that there are people with the brains and the time to write VHDL (or whatever is the current HDL) for the fake chips. For a Z80 it is probably a few weeks' work. For a 32F4 - ? Must be years. Unless it has leaked from ST.

They are not fake. They are compatible chips which are legally produced. They buy CPU IP from ARM and add their own peripherals. Sometimes original marking is removed by counterfeiters and then marked as if they were ST.

[nctnico]

it shows the 32F417 at 10 years, since January 2021.

How actually likely is it to be that short?

Thats just the guaranteed time.
If its popular, like the stm32f103, it will be around a lot longer. That ones in production for what 14 years now? and its still "active" on ST site.

Yep. The trick is to pick a chip which is popular (or is likely to become popular). Using one which is on a manufacturer's development board or another popular dev kit is a good choice because it will get a lot of design ins. The processor chip from TI on the original Beagle board is still in production a decade later.

[peter-h]

In that case the 407/417 should go for a long time It is on the ST STM32F4DISCOVERY dev board.

UM1472
User manual
Discovery kit with STM32F407VG MCU

And "everybody" copies the dev kit schematics

In fact I now recall the decision to go ST ARM against Atmel/etc ARM, made about 4 years ago, was the likely production life. Atmel chopped their chips fast in recent years. The 90S1200 replacement lasted barely a few years. And now one can get an ARM for the same money, in some tiny package with 4 pins, which is hilarious

[nctnico]

Don't rule out NXP either. You can still buy a lot of their old ARM based microcontrollers. You can still buy the LPC2106 which AFAIK is their first ARM MCU from 2003 (judging by the revision history in the datasheet). Even better: a lot of peripherals in the much newer Cortex-Mx parts are the same (or backwards compatible) so software porting effort is minimal.

[peter-h]

I also wonder how much people get committed to a chip according to how much library code there is for it, and how much work they had to do to get the bugs out.

I reckon anybody who has designed in the 32F4xx and had to spend months or longer fixing the vast number of bugs in the ST libraries for USB and ethernet will never change, out of fear of having to do it all again for some other chip

One spends months trawling forums for fixes for these bugs (which have been separately discovered by lots of people) and most of the time the question appears but there is no fix. Partly this is because anybody developing code commercially isn't going to post the fixes they developed.

In the days of a Z80 or H8/300 there were no libraries because, ahem, there wasn't anything much on the chip You could change a uC in perhaps a week.

[Siwastaja]

It's a very funny remark that people who used hardware abstraction libraries fear to design in new parts because they know hardware abstraction libraries didn't abstract hardware.

Meanwhile, bare metal do-it-all-yourself guys are just fine redesigning. We have control and we love it. Just finished an emergency redesign to a completely new chip. Manufacturing stopped for two weeks.

I know in certain areas long certification cycles are required and part substitution is not allowed. If you do that and do not do it in close co-operation with the part manufacturers who promise the life cycles, you are doing it wrong.

For rest of us, emergency redesigns are not nice and without risk, but doable. But having to redesign every ten years after receiving part life status notification and last buy notification a year in advance? Come on, just go on with the technology.

[nctnico]

It's a very funny remark that people who used hardware abstraction libraries fear to design in new parts because they know hardware abstraction libraries didn't abstract hardware.

The main problem is that the API for the hardware abstraction library becomes obsolete quicker than the chips it supports.

[westfw]

The 90S1200 replacement lasted barely a few years.

I thought the 90S1200 replacement was the ATtiny2313, which is still "active" (or ATtiny2313a)
(the tiny is a substantial superset of the 1200, most of which can be ignored.)

[peter-h]

The 2313 went obsolete a long time ago, but maybe just the one type of package?