STM32 10 year commitment and planned obsolescence

[diyaudio]

How does one go about designing product using microcontrollers like the STM32 having a production life cycle of 10 years ? compare that, to Microchip's mature product line that promises product availability exceeding 10 years. example PIC16F series and the like..

I ask this question because it looks like the rate of obsolescence is so deeply rooted into the silicon, it even forces the brain to think short term. This is detrimental as products sit in R&D for 5 years or even longer (with the MCU already in the design) and, by the time a product is ready to ship the MAIN mcu chip has reached half of its production life cycle.   

Cheers.

[krho]

Usually the chips are very similar, so you can just upgrade 4 years into the development.

[nctnico]

Usually the chips are very similar, so you can just upgrade 4 years into the development.

IMHO that is one of the things ST sucks at particulary bad. Their peripherals are allover the place. NXP is a much better choice in that respect.

[donotdespisethesnake]

How does one go about designing product using microcontrollers like the STM32 having a production life cycle of 10 years ?

There are no easy solutions, you need extra development work somewhere. (Assuming it is impractical to buy a lifetime supply of chips at any point.) On some projects I have done main development with similar but older chip, and port to pre-production/engineering samples as soon as they are available (help if you are a large purchaser), with the aim that volume production coincides with volume production of the target chip. That exposes you to bugs and limited quantities of the pro-prod chips, as well as delays in volume availability.

Another option is the mid-life re-spin, but obviously that relies on having an easy migration path to a similar chip, ideally there is a newer chip that is a pin-compatible replacement. Usually some software and hardware redesign is required. Probably the new design needs re-certification which is a significant exercise.

Cheap chips designed for consumer apps are attractive for the unit cost, but you really have to weigh that against development effort over the life cycle. Of course, promises of longevity are really just aspirations, no business can guarantee supply, so there is always need for contingency in the development budget to allow for component obsolescence. In some cases I have seen a product respin put out to an external design house, but that was really a last resort.

In both cases, having portable software and to whatever extent a hardware design that can migrate to different devices is essential. Avoid picking chips with special features that are only found in one manufacturer.

[Jeroen3]

You get automotive parts they're available longer. But developing with those is more expensive. More proprietary tools, features and compilers.
Anyway, ST now says: 10 year commitment, starting January 1st 2017.

When you are approaching end of life, you just do a large last buy. Note that a last buy isn't always from stock!
You should always have enough supply. And your new design should be almost done by then, if you plan to continue it.

Some vendors will respin a product if your Apple or Samsung.

[stmdude]

Some vendors will respin a product if your Apple or Samsung.

"respin"?  If you're talking about producing more of an EOLed product, for some vendors, you don't have to be that big. I've seen MOCs as low as 10K for the vendor to produce a fresh batch of chips. (AllWinner in this case)

[diyaudio]

You get automotive parts they're available longer. But developing with those is more expensive. More proprietary tools, features and compilers.
Anyway, ST now says: 10 year commitment, starting January 1st 2017.

When you are approaching end of life, you just do a large last buy. Note that a last buy isn't always from stock!
You should always have enough supply. And your new design should be almost done by then, if you plan to continue it.

Some vendors will respin a product if your Apple or Samsung.

Thinking aloud..

This is an undocumented feature explaining exactly why engineering (for certain long term products) stick with microchip or start using them,
its not just about a chip using the latest and greatest such as an ARM core ect.., for an initial product or R&D task, the design effort invested in the STM32 platform from a software development R&D cost is massive!, however a second stage product using the same platform will be based on previous experience, i.e using the same toochain, custom written libraries (hopefully written in portable C fashion)   

Planned obsolescence is really annoying and getting worse and worse by the day.

[stmdude]

Usually the chips are very similar, so you can just upgrade 4 years into the development.

IMHO that is one of the things ST sucks at particulary bad. Their peripherals are allover the place.

Yea, if you think that is bad, try using their drivers/libraries for more than a few years, spanning a few so-called similar chips.

[Jeroen3]

Some vendors will respin a product if your Apple or Samsung.

"respin"?  If you're talking about producing more of an EOLed product, for some vendors, you don't have to be that big. I've seen MOCs as low as 10K for the vendor to produce a fresh batch of chips. (AllWinner in this case)

It depends. A big reason why products are ended is because the fab contract ended, the machines are sold, or a more profitable product can be made.
When it's just a more profitable part that gets priority, you can just pay more. When the machines are sold, then you're looking into millions to even get 1.

And with the merges going on lately, expect the sale of equipment and loss of contracts be a leading factor.
I don't know if you've noticed, but a lot of Nexperia parts are sold out. They don't have unlimited access to NXP's fabs anymore.

[donotdespisethesnake]

Planned obsolescence is really annoying and getting worse and worse by the day.

They are following consumer demand, who want cheaper products with more features. That means going to smaller wafer technologies, and obsoleting old stuff. It's too expensive to maintain old fabs for a few parts.

If consumers bought products because they want the exact same thing they bought 20 years ago, and didn't care about new features or cost, then the strategy would change.
The law of supply and demand pretty much guarantees that cost must reduce and features must increase.

I tend to work in the industrial sector, and those customers *do* want they exact same thing they bought 20 years ago. However, it's a relatively small market by volume compared to consumers products. Instead of shipping millions of units, the volumes are thousands. And industrial users expect things to work for 20 years, and they don't buy replacements every 3 years like consumers.

So for industrial products, we look to vendors specializing in that market, e.g. Infineon.

[diyaudio]

Usually the chips are very similar, so you can just upgrade 4 years into the development.

IMHO that is one of the things ST sucks at particulary bad. Their peripherals are allover the place. NXP is a much better choice in that respect.

exactly, this pin to pin compatibility is a joke and hardly practical.     

[westfw]

it looks like the rate of obsolescence is so deeply rooted into the silicon

Can you explain that phrase in a bit more detail?  It's not making a lot of sense to me.

[danadak]

Take any manufacturers commit to EOL with a grain of salt.

Many mergers have trashed such commitments, and product lines.


Regards, Dana.

[MT]

EOL is not always entirely EOL!

Example, TI announced many many years ago PCM54, PCM56 "definitively" EOL and LTB, then TI stretched that with some months and then it vanished.  But suddenly slightly less then a year later they came back and not to far ago PCM54 went to the graveyard while PCM56 still with us with a slight change of product presentation, qoute: This product continues to be in production to support existing customers.

I wonder which "specific" industrial segment still motivates TI to run new wafers (assuming not NOS) of this old lady?
PCM54/PCM56 and other was once a workhorse DAC for the audio industry but i can imagine industrial control is another.

http://www.ti.com/product/PCM56

[Tomorokoshi]

Thinking aloud..

This is an undocumented feature explaining exactly why engineering (for certain long term products) stick with microchip or start using them,
its not just about a chip using the latest and greatest such as an ARM core ect.

This is a selling point the local Microchip reps never fail to bring up. At this point it seems to be true, as they continue to produce a processor with only one customer. We use that part. I think the other remaining customer finally moved on.

Anyway, there are reasons for wanting to stay with the same part well after it's two generations past its prime. This includes variations on:
1. Approval issues where the cost of re-approval may be prohibitive.

2. It's one board in a system of several, and changing the software would percolate to the other boards in the system.

3. Engineering resources are dedicated to replacement products, instead of minor upgrades to older ones.

[donotdespisethesnake]

I wonder which "specific" industrial segment still motivates TI to run new wafers (assuming not NOS) of this old lady?
PCM54/PCM56 and other was once a workhorse DAC for the audio industry but i can imagine industrial control is another.

Then on the other hand, TI unexpectedly dropped the whole family of LM3S MCUs, which pissed off a good number of designers.

[Jeroen3]

TI cancelled those because the flash was seriously flawed. Only 10 years data retention at room temperature is very low. Most other flash is 10 years at 125 C.

Flash memory endurance cycle specification is 100 cycles
Description:
The Flash memory endurance cycle specification (maximum program/erase cycles) is 100 cycles.
Failure to adhere to the maximum number of program/erase cycles could result in corruption of the
Flash memory contents and/or permanent damage to the device.
Workaround:
None. Because the failure mechanism is a function of the third-party Flash memory technology used
in this device, there is no workaround. This third-party Flash memory technology is used only in the
affected 130-nm Stellaris products and will not be used in any future devices. All other Stellaris
products use Flash memory technology that exceeds industry quality and endurance cycle standards.

[nctnico]

I wonder which "specific" industrial segment still motivates TI to run new wafers (assuming not NOS) of this old lady?
PCM54/PCM56 and other was once a workhorse DAC for the audio industry but i can imagine industrial control is another.

Then on the other hand, TI unexpectedly dropped the whole family of LM3S MCUs, which pissed off a good number of designers.

Perhaps but Luminary was about the first to use an ARM Cortex-M core and their products wheren't stellar to begin with (IIRC long errata sheets as well). I think in the end buying Limanary was a very bad move for TI because in the end they had to come up with their own products (MSP432) from scratch anyway.

[technix]

I ask this question because it looks like the rate of obsolescence is so deeply rooted into the silicon, it even forces the brain to think short term. This is detrimental as products sit in R&D for 5 years or even longer (with the MCU already in the design) and, by the time a product is ready to ship the MAIN mcu chip has reached half of its production life cycle.

The expected R&D time for STM32 products is less than 6 months. This is the ugly norm in China, as neither the boss nor the investors of most of the users of STM32 chips are patient enough to wait for a product to be well polished, and those rushed product usually have a half life of 6 month to 1 year.

[Jeroen3]

Tell me then. What should one choose for long term?
Apparently is not STM32 (10 years), NXP LPC* (10 years), Freescale (10 years) or Silabs 32 bit (10 years).

Are you really limited to 8 bitters and chips behind NDA?

^{*some old ones were 15}

[mikeselectricstuff]

If you're designing for a long lifetime, then go with an MCU manufacturer with a track record of long-term availability, and backward compatibility of newer devices.
It may be that a newer, compatible device will appear that can be used with minimal effort & save some cost.
But also bear in mind that other parts of the design may well go obsolete before the MCU so work on the assumption that some redesign may be necessary at some point.
The most important thing is to document everything as well as possible, both hardware and software. Having to reverse-engineer an old, poorly documented design will make any redesign all the more painful.

 

[mikeselectricstuff]

Tell me then. What should one choose for long term?
Apparently is not STM32 (10 years), NXP LPC* (10 years), Freescale (10 years) or Silabs 32 bit (10 years).

Are you really limited to 8 bitters and chips behind NDA?

^{*some old ones were 15}

8 bit would seem a bit risky.

Microchip ( 8,16 and 32 bit) is probably a good bet, though maybe avoid the ex-Atmel parts for the moment in case they get purged.

It appears Microchip have always been set up to do smaller runs, hence the huge variety of parts. This means they can keep making older parts as long as people want them, unlike others who need a high volume commitment to justify larger batches.

You can still buy PIC16C54's, introduced at least 20 years ago, as well as numerous later pin compatible, and nearly sofwtare-compatble parts. 

[Jeroen3]

Well, poor me just finished a design on an STM32F103. Yes yes, I know it's an mature product. But it's an available product. When you're not shoving many parts a month, you kind of rely on the distributors keeping stock.

Should I instead have chosen Renesas, who claim PLP until 2036 on some parts? I don't know.
The 103 is very typical and does not have fancies that you can't replace by literally any other cortex m3 or beyond on the market.

If instead I would have chosen the F373 (16 bit delta sigma) or an L4/F7 with DFSDM, then I would have a major problem on EOL of the chip.

[donotdespisethesnake]

NXP Longevity program https://www.nxp.com/products/product-information/product-longevity:PRDCT_LONGEVITY_HM

Infineon were promising 15 year lifetime for their XMC products, they appear to have backed off that and any mention of lifetime has been expunged from their website.

STM32F103 has been cloned by GigaDevices. Possibly it might become like 8051, someone, somewhere will still be making them many years from now.

The fact is though, 10 years is a long time in the semicon industry, making business plans that far ahead is just guesswork. We've tried to pry info out of vendors, but obviously information about their demand and profitability is very commercially sensitive. And whatever roadmap info they will share with us under NDA goes out of the window when they are acquired by another company

[nctnico]

I usually look at what is being stocked in reasonable quantities at companies like Farnell, Mouser, Digikey, etc. These devices are appearantly popular (or will be due to availability). BTW the first ARM devices NXP came up with (LPC2103/LPC2104) are still available from Farnell after 10 years or so.

[mikeselectricstuff]

I usually look at what is being stocked in reasonable quantities at companies like Farnell, Mouser, Digikey, etc. These devices are appearantly popular (or will be due to availability). BTW the first ARM devices NXP came up with (LPC2103/LPC2104) are still available from Farnell after 10 years or so.

Though NXP did some nasties along the way with slighly "improved" /01 (I think that's what it was called) versions of the earlier ARM7 parts which were slightly incompatible.
In one case, they fixed a bug that was 100% work-roundable with a trivial code change, and so never needed "fixing" but of course the fixed version of the chip broke code that had the workaround.

[chickenHeadKnob]

Infineon were promising 15 year lifetime for their XMC products, they appear to have backed off that and any mention of lifetime has been expunged from their website.

This is a timely thread, does anyone know whats going to happen to the XMC4200? I was looking at it today comparing it to the STM32F334 as both have high resolution PWM, but mouser warns:
End of Life: Scheduled for obsolescence and will be discontinued by the manufacture
for most parts in the 4200/4100 series. Just checked again and at least one QFN48 packaged part doesn't have the warning but the QFP64's do. Huh?

[MT]

In one case, they fixed a bug that was 100% work-roundable with a trivial code change, and so never needed "fixing" but of course the fixed version of the chip broke code that had the workaround.

Seams to apply to ST as well, 10 year old erratas (e.g STM32F429 now into Rev15) never get fixed ,will break code but then ST surprisingly just transfers F429 bad erratas to "so called new designs" such as F446 and the F405, F407,F446 etc terrible ADC gets fixed with F415, F417 etc!?! Huh, i dont like ST's half arsed fixing behavior, why just bother to fix these two and not 446!?

[technix]

In one case, they fixed a bug that was 100% work-roundable with a trivial code change, and so never needed "fixing" but of course the fixed version of the chip broke code that had the workaround.

Seams to apply to ST as well, 10 year old erratas (e.g STM32F429 now into Rev15) never get fixed ,will break code but then ST surprisingly just transfers F429 bad erratas to "so called new designs" such as F446 and the F405, F407,F446 etc terrible ADC gets fixed with F415, F417 etc!?! Huh, i dont like ST's half arsed fixing behavior, why just bother to fix these two and not 446!?

If a chip is a strict subset of another, it is entirely possible that the subset chip is packaged from the full-featured die, with some featured disabled. This way a chip with partial failure can be salvaged and sold at a reduced price instead of going straight to the garbage bin.

[coppice]

In one case, they fixed a bug that was 100% work-roundable with a trivial code change, and so never needed "fixing" but of course the fixed version of the chip broke code that had the workaround.

Seams to apply to ST as well, 10 year old erratas (e.g STM32F429 now into Rev15) never get fixed ,will break code but then ST surprisingly just transfers F429 bad erratas to "so called new designs" such as F446 and the F405, F407,F446 etc terrible ADC gets fixed with F415, F417 etc!?! Huh, i dont like ST's half arsed fixing behavior, why just bother to fix these two and not 446!?

If a chip is a strict subset of another, it is entirely possible that the subset chip is packaged from the full-featured die, with some featured disabled. This way a chip with partial failure can be salvaged and sold at a reduced price instead of going straight to the garbage bin.

These days a single MCU die is generally used in quite a large range of distinct part types. It is seldom economic to bin them out, according to where defects are found, though. They generally just test a wafer against the spec of the final part it is going to become, and package everything that passes.

[westfw]

If you want to be really paranoid, you could worry that the sale of ARM and MIPS may cause the elimination of (relatively un-profitable) 32bit microcontrollers entirely.

[technix]

In one case, they fixed a bug that was 100% work-roundable with a trivial code change, and so never needed "fixing" but of course the fixed version of the chip broke code that had the workaround.

Seams to apply to ST as well, 10 year old erratas (e.g STM32F429 now into Rev15) never get fixed ,will break code but then ST surprisingly just transfers F429 bad erratas to "so called new designs" such as F446 and the F405, F407,F446 etc terrible ADC gets fixed with F415, F417 etc!?! Huh, i dont like ST's half arsed fixing behavior, why just bother to fix these two and not 446!?

If a chip is a strict subset of another, it is entirely possible that the subset chip is packaged from the full-featured die, with some featured disabled. This way a chip with partial failure can be salvaged and sold at a reduced price instead of going straight to the garbage bin.

These days a single MCU die is generally used in quite a large range of distinct part types. It is seldom economic to bin them out, according to where defects are found, though. They generally just test a wafer against the spec of the final part it is going to become, and package everything that passes.

What I mean here is that the ones not quite passing isn’t tossed either.

[donotdespisethesnake]

This is a timely thread, does anyone know whats going to happen to the XMC4200? I was looking at it today comparing it to the STM32F334 as both have high resolution PWM, but mouser warns:
End of Life: Scheduled for obsolescence and will be discontinued by the manufacture
for most parts in the 4200/4100 series. Just checked again and at least one QFN48 packaged part doesn't have the warning but the QFP64's do. Huh?

I think that specifically refers to AB revisions, which are replaced by rev BA due to manufacturing site change. See https://media.digikey.com/pdf/PCNs/Infineon/2015-064-A.pdf

As far as I can tell, all the XMC parts are still active.

[coppice]

These days a single MCU die is generally used in quite a large range of distinct part types. It is seldom economic to bin them out, according to where defects are found, though. They generally just test a wafer against the spec of the final part it is going to become, and package everything that passes.

What I mean here is that the ones not quite passing isn’t tossed either.

Actually, they are usually tossed. With raw die yield percentages in the 80s and 90s, which is what you get for a typical MCU sized die in a reasonably mature process, there aren't enough "this fails if you test it as a die for X, but we could reuse it as a die for Y" situations to justify the cost of handling them. With really big dies, like Pentiums, respinning a partially defective die becomes more interesting. Even there it seems to be mostly done just for speed grading these days. We don't see 3 core parts from AMD or Intel any more.

[MT]

In one case, they fixed a bug that was 100% work-roundable with a trivial code change, and so never needed "fixing" but of course the fixed version of the chip broke code that had the workaround.

Seams to apply to ST as well, 10 year old erratas (e.g STM32F429 now into Rev15) never get fixed ,will break code but then ST surprisingly just transfers F429 bad erratas to "so called new designs" such as F446 and the F405, F407,F446 etc terrible ADC gets fixed with F415, F417 etc!?! Huh, i dont like ST's half arsed fixing behavior, why just bother to fix these two and not 446!?

If a chip is a strict subset of another, it is entirely possible that the subset chip is packaged from the full-featured die, with some featured disabled. This way a chip with partial failure can be salvaged and sold at a reduced price instead of going straight to the garbage bin.

Thats not what i'm talking about, rather point is 446 is not entirely a "dont-toss-re-labeled part" since it has power regulator,quadspi,SPDIF-RX ,etc the other F4's/F429 dont have so i assumed to get these features ST had to do new mask sets/new wafers. There is a migration paper for 429 to 446, so in this case of 429 vs 446 situation is reverse else why would ST fuse out/omit features in older F429 and then enable them in later products such as F446 if they both came from same wafer as you assume? New type of (reverse) marketing strategy?  I suspect they do not come from same wafer! For 415-417, well since ST mentioned in a post on their forum i think it was, they had fixed the ADC in them while 405,407,429,446 had not...but:

Disclaimer: I do not know to 100% if ST telling the truth or not about ADC fix i just saw a post about it.

[technix]

In one case, they fixed a bug that was 100% work-roundable with a trivial code change, and so never needed "fixing" but of course the fixed version of the chip broke code that had the workaround.

Seams to apply to ST as well, 10 year old erratas (e.g STM32F429 now into Rev15) never get fixed ,will break code but then ST surprisingly just transfers F429 bad erratas to "so called new designs" such as F446 and the F405, F407,F446 etc terrible ADC gets fixed with F415, F417 etc!?! Huh, i dont like ST's half arsed fixing behavior, why just bother to fix these two and not 446!?

If a chip is a strict subset of another, it is entirely possible that the subset chip is packaged from the full-featured die, with some featured disabled. This way a chip with partial failure can be salvaged and sold at a reduced price instead of going straight to the garbage bin.

Thats not what i'm talking about, rather point is 446 is not entirely a "dont-toss-re-labeled part" since it has power regulator,quadspi,SPDIF-RX ,etc the other F4's/F429 dont have so i assumed to get these features ST had to do new mask sets/new wafers. There is a migration paper for 429 to 446, so in this case of 429 vs 446 situation is reverse else why would ST fuse out/omit features in older F429 and then enable them in later products such as F446 if they both came from same wafer as you assume? New type of (reverse) marketing strategy?  I suspect they do not come from same wafer! For 415-417, well since ST mentioned in a post on their forum i think it was, they had fixed the ADC in them while 405,407,429,446 had not...but:

Disclaimer: I do not know to 100% if ST telling the truth or not about ADC fix i just saw a post about it.

Is F439/417 newer than F446? If this is the case the ADC fix did not make it towards the deadline of F446 tape-out date so got delayed to the F439 instead. F415/417 are simple cut down versions of F439.

[pigrew]

What is the F446 ADC problem? I'm thinking about using the nucleo board in a project for a client. (I'm building five prototypes for them). The errata sheet mentioned the control registers shouldn't be changed during conversions, is that it? 

[MT]

Is F439/417 newer than F446?

Who knows what came when, some dates are mentioned in https://en.wikipedia.org/wiki/STM32

If this is the case the ADC fix did not make it towards the deadline of F446 tape-out date so got delayed to the F439 instead.

In reality nobody outside ST have the slightest clue how ST handel this, we can only guess whats going on.

F415/417 are simple cut down versions of F439.

Previously i had assume 446 was cut down/fuse off peripheral/same wafer of 429 but it seams not to.

What is the F446 ADC problem?

All F4 series have a noisy ADC, there are 2 papers from ST, survey the web its lot about the issue.

I'm thinking about using the nucleo board in a project for a client.

Dont, your just asking for major problems, to get ADC reasonable you need to do a well thought out PCB layout.
Im still battling my design, noise is partly software dependent, great isn't it?.

(I'm building five prototypes for them). The errata sheet mentioned the control registers shouldn't be changed during conversions, is that it?

Certainly not.