How to ensure long-duration manufacturing of central components?

Question

This question is related to project planning and minimizing risks in the future. Say that Company X builds this clever device (I do not own a company - I am just curious). The clever device takes advantage of a micro controller unit (MCU) as the central most important component. This MCU is manufactured by Company Y. Today, in 2013, how does one ensure that Brand Y and model is still produced X number of years in the future, either by Company Y or someone else?

Are there currently any specific brands+model families (or just general architectures), that one can rely on being available into the (un)foreseeable future? Any brands/model-families/architectures known to be uncertain? I guess Intel and Atmel must be producing certain model families, that are quite certain to remain in production for a number of years/decades. But which model or architectures exactly?

This question appears to be off-topic because it is not about electronic design. — Leon Heller, Aug 07 '13 at 14:07
Then I am not sure where to post such a question. Any other SE site more suitable? I will say that it is somewhat related to electronic design. — Ole Thomsen Buus, Aug 07 '13 at 14:12
I think it is, at least in the sense of "design for manufacture". — pjc50, Aug 07 '13 at 14:14
Exactly - This question is more related to actual serious manufacturing of many copies of the same design. — Ole Thomsen Buus, Aug 07 '13 at 14:17
@Leon - untrue, designing for supply lifetime is a **key** part of electronic design. You cannot just make your circuits in the lab using whatever you like and throw them over the wall assuming production will magically take care of all the real world issues! — Chris Stratton, Aug 07 '13 at 14:25
Components with multiple sources are obviously more secure. With processors, you could argue 8051 derivates are secure as they have been around forever, but a lot of the work is in handling the peripherals - for an embedded project written in C, porting the whole project to a different CPU may not actually be harder than porting it from one 8051 derivate to another. For something in assembly, that may not be true. Though there's always the possibility of simulating an older part on an orders-of-magnitude more capable newer one. — Chris Stratton, Aug 07 '13 at 14:31
Reason for asking this question: A company (a real one; not mine) is designing and selling two model families A and B. Family A is working fine, and the only reason family B exists, is that the MCU used in A, reached EOL some time ago. B uses a completely different brand and architecture (AVR 8-bit). As a completely revamped design in B, they also jumped (the shark) from a clean software implementation in C to a complex C++ design. The change of MCU thus inspired new unnecessary software complexity. Did I mention that products in the B family have weird critical bugs? :) — Ole Thomsen Buus, Aug 07 '13 at 15:07
Software rewrites (rather than ports) are _nearly always_ a disaster! — pjc50, Aug 07 '13 at 15:39
This previous question covered many of the same issues: http://electronics.stackexchange.com/questions/50910/question-about-long-term-support-for-legacy-board-designs — The Photon, Aug 07 '13 at 16:18

score 22 · Accepted Answer · answered Aug 07 '13 at 14:21

Although this is not stricty an electronic design question, it is important to most design engineers. Component sourcing is one of our biggest headaches, and most companies are smart in letting a separate person deal with it instead of causing severe depression and anxiety in the engineers.

There are three ways to combat this, aimed at three different tiers of products:

Not-terribly-hard-to-make products should just be adaptable. Say you are making a custom board with expected 100-1000 units production per year. Just design in whatever you want, and when you get a product change notification from the manufacturer that one of the parts is going out of production: use another component and just eat the engineering hours. Terrible as this may sound, this is often economically the best idea in this respect. Even large production runs work well with this model; just produce a new variant of your product that is functionally the same. This is being done in the consumer and professional space all the time.
Small-run, specialist products that took a lot of man hours to make. For instance specialist scientific tools. The best course of action is to do a good estimate of your required components during the service life of the product and buy twice as many components as you will ever need. Cost is rarely a factor, so even though this will cost you quite some money in advance, as well as space to safely store it all, this will be alright. Don't underestimate storing cost: they need to have very specific, tightly controlled atmospheric conditions, especially to ensure solderability.
Medium to large run long term support products. Here, you will want to get a direct line to the manufacturer of your chosen product and ask them to either (a) produce a special version for you with a specified service time or (b) when the PCN goes out, ask them to make those chips specially for you. All MCU companies do this last bit. If you want at least 10.000 chips, even ones that have gone out of production for 20 years, they will happily make them for you - at a nominal fee. However, this is only possible if you need at least in the order of 10 000 units, often even at least 100 000.

Very few companies guarantee any kind of long term support on their components. Even so-called 'design for long term use' automotive parts from Microchip are only guaranteed production parts for 10 years, which is nothing compared to the lifetime of some specialist gear. You will always need to check in directly with manufacturers to ensure availability in the long term.

You can't count on manufacturer lifetime guarantees either. 5 or 10 years ago Intel prematurely retired a number of CPUs that they'd originally promised would have had a number of additional years of availability for use by embedded system manufacturers. — Dan Is Fiddling By Firelight, Aug 07 '13 at 18:26
And in the case of (3) you will also be notified when the products are going end-of-life and given an opportunity to make a "last time buy" where you estimate all of your future needs and stock up, or at least buy enough to cover the time required to redesign. — Ben Jackson, Aug 07 '13 at 21:14
@BenJackson You will be notified, maybe, possibly, if you asked nicely enough. I have numerous examples of when I asked nicely, but wasn't informed of the LTB. — , Aug 08 '13 at 00:34
@DanNeely and more interestingly, Intel kept on producing 486 processors until somewhere in the 00s, but they were restricted to just a few companies and they would never be allowed to enter the free market. So something was still very much in production and available, something that was in the free market just a decade earlier, but absolutely impossible to get a hold of regardless :P — user36129, Aug 08 '13 at 07:51

score 9 · Answer 2 · edited Aug 07 '13 at 15:58

9

The bottom line is this: If you want to guarantee that you can make your product far into the future then you (or your company) need to do things to make sure that happens. While your various suppliers might assist, in the end it is all up to you. It shouldn't be that way, but it is. Here are some steps you can take to help:

Maintain a good relationship with your distributors and manufacturers reps. These are the people who can best watch your back, and notify you about potential supply issues.
Check, frequently, for End-Of-Life (EOL) notices from all of your suppliers. While your rep or distributor should tell you about these things, you cannot rely on them. Check on a monthly basis on the various manufacturers web sites.
Expect to keep extra stock on hand of the critical parts. Not only does this mean having capital tied up in stock, but having the infrastructure to store these parts (humidity controlled chambers, etc.)
Expect to make a large last-time-buy of critical parts. This means having the money or credit available to do that and, again, having a place to store the parts when you do get it. Depending on the part cost, and your volume, this might require purchasing anywhere from US$1K to US$1million worth of parts in a single purchase!
Expect to redesign the product sometime in the future. This means not only budgeting for this, but also requires that you plan ahead and archive key documents, files, and equipment. I would go so far as to setup a development PC with all of the tools and then STORE THE ENTIRE PC. Nothing is worse than having to redesign a product only to find out that you need a copy of Windows that is unavailable and won't run on a modern PC, or you need an I/O port that is obsolete and unavailable. Make backups of the files, and revisit the backups once a year to prevent bit-rot (making sure that the thumb drives, CD's, and hard drives have not deteriorated or are too old to use in a modern PC).
Do the obvious things like choosing components that have a high chance of being around in the future, and made from a mfg that will be around. But this does not override any of the other things on this list. Just because a mfg says that a product will be manufactured for 10 years doesn't mean that they won't go out of business tomorrow.
Be paranoid and proactive. Ultimately you are the only person/entity that cares enough to guarantee things into the future.

edited Aug 07 '13 at 15:58

JYelton

32,302
33
134
249

answered Aug 07 '13 at 14:42

2

I like the idea of storing the ENTIRE PC :) I kind of expected that this was a critical and important detail in electronic design (for manufacture). Thanks for these hints. – Ole Thomsen Buus Aug 07 '13 at 14:56
+1 ... Item 6 might include getting to know a manufacturer's reputation for obsoleting parts. Also, although no guarantee, it often helps to choose parts with multiple sources (2 or more manufacturers) – Tut Aug 07 '13 at 14:57
1

@OleThomsenBuus Yesterday a coworker dug up an old PC that he had to use. It requires the old style PC-AT keyboard with the large DIN type connector. He doesn't have one. Even ubiquitous things like keyboards go away. Store EVERYTHING! In another anecdote: I know of a purchasing guy who knew of a last-time-buy for almost a year. He failed to act on it and was fired. His mistake could cost his employer $5-10 million in revenue. There is serious money to be lost due to lack of being proactive. – Aug 07 '13 at 15:04
@DavidKessner - And what would be the reason to stall on the last-time-buy option? Was he acting too carefully with purchasing resources? – Ole Thomsen Buus Aug 07 '13 at 15:21
@OleThomsenBuus Stalling is a financial issue. In business, having lots of money tied up in stocked parts is money that can't be used to make more money. It's like putting money in a savings account that has 0% interest, or worse because you are paying for the floor space. You also might want to wait until the last minute in order to get the most up to date sales forecasts (so you don't buy too many parts, and never use them). This particular guy just screwed up and didn't buy the stuff, and it wasn't his first big mistake. – Aug 07 '13 at 15:31
A slightly nicer option to storing the entire PC is to make SURE you have a VM that can run all the relevant tools (**absolutely test this**), and storing the VM in lots of places, so it won't get deleted accidentally. Save the VM software too. Then, test it against any new VM software regularly to make sure it's still workable while you have access to an older computer on which it works as well. – Connor Wolf Aug 07 '13 at 23:42
1

@ConnorWolf That works _IF_ your software doesn't have to talk with old hardware (like, an old MCU programmer that connects to a parallel port). But like you said, absolutely test it. A VM is fine, so long as it all works. – Aug 08 '13 at 00:32
@DavidKessner - I hadn't thought of things like parallel port interfaced hardware. Good point, you are completely correct. – Connor Wolf Aug 08 '13 at 03:52

score 5 · Answer 3 · answered Aug 07 '13 at 15:25

There are no guarantees.

It does help to look at the past history of a company to see how it deals with obsoleting old products. Some companies go to great lengths to make sure the legacy products are taken care of. Others dump them as soon as newer versions are available.

Microchip is a good example of the former. You can still buy a PIC 16C54, although you'll pay a few dollars more for one than for newer parts that do more. Companies like Maxim are at the other end. Different companies do have different philosophies and cultures.

It also helps to look at the market a component is being sold into. If the primary customers are cell phone companies, for example, then the component may dissappear quickly. New models of cell phones come out so fast that cell phone companies don't much care about component availability 5 years after first release.

Avionics and medical are examples of the other end. In both cases, the cost to getting a product certified and accepted is long and expensive, and product lifetimes are long and the products are expected to continue operating in the field for a long time (10s of years). A company that makes parts for these markets will likely go to extra lengths to make sure old products are available in the future, even if the price will be uncompetitive at the time. It's a lot better to keep buying a PIC 16C54 for $5 for use in a $10k medical device than to redesign it to use a newer PIC.

Thanks for your answer. In my particular concern and reason for inquiry, the use of the MCU is not for short life time products such as smartphones, etc. These are devices built into infrastructure (houses). They might fail in the future, and, due to the devices being interconnected, the new replacement needs to be compatible with the remaining versions. This is why a replacement should be exactly as the old versions - at least in its interconnect interface and protocol. That's one scenario at least. — Ole Thomsen Buus, Aug 07 '13 at 15:42

score 3 · Answer 4 · answered Aug 07 '13 at 14:30

You could just ask them: most companies will have a product roadmap that tells you when products will be retired. They are publicised to a certain extent, e.g. http://www.eteknix.com/intel-discontinue-over-24-processors-by-q1-2012/

The US military has a special "second source" requirement, that every component be available from at least two manufacturers (and for a particular period of time). This increases their costs but guarantees supply. Obviously you can only get this if you're a big customer.

But generally technology moves on, and you have to update designs when parts go out of production. Sometimes there is a regulatory change that forces this, such as the RoHS elimination of lead in parts.

Otherwise you end up relying on "new old stock", which is no longer manufactured but has sat in a warehouse for years. Nixie tubes are a popular hobbyist example of this.

Microcontroller architectures are often remarkably persistent; parts using 6502 and 68000 instruction sets are still available.

Note that if you're planning for the long run you need to keep your design data, software source and toolchain "live". Otherwise you find that something vital is stored on a disk you can't read for a software package that is unobtainable.

IMHO, the whole "second source" thing is overrated/unrealistic, and the aerospace/military industry is slowly moving away from that. The bottom line is that there are a lot of critical products with no second source available. Other than RAM, decrete logic, and some Flash, modern chips mostly don't have second sources. Limiting yourself to CPU's that have second sources could also limit you to 1990's technology. — , Aug 07 '13 at 14:56
@David Kessner - But what if 1990s tech might actually do the trick? If all you need is a simple state machine to act on input inside a wall for 1 billion years - then an 8-bit 8051 might be just fine? I like simplicity and certainty if it realistically achievable. — Ole Thomsen Buus, Aug 07 '13 at 15:15
@OleThomsenBuus _IF_ the 90's tech works, then great. But that stuff is 20 years old already and you might want it to be around another 10-20 years. That might be a lot to ask. Also, if you use 20 year old tech, then your competition has had 20 years to do the same, or better. There are certainly cases where old tech is fine. I personally like to develop products that are innovative and haven't been done before. You can't do that very well with 20 year old tech. — , Aug 07 '13 at 15:26

score 1 · Answer 5 · answered Aug 07 '13 at 14:36

If you have a good relation with your supplier (high volume purchases help!) then he will let you know when a product reaches end-of-life, and offer you a last-buy option, so that you can buy enough to either

run your complete production in the remaining years of your own product's life cycle
run enough production to allow you time for a redesign

How to ensure long-duration manufacturing of central components?

5 Answers5