Hi all,
I just read @quozl’s retrospective on OLPC (great project!) and his list of parts they had to change due to EOL or huge MOQ was quite interesting! I’d love to hear more stories about this and how people have solved them!
My first products are only 4 years old, but I’ve already been hit by parts going EOL. I used a Thermocouple IC from Microchip (MCP9600 I/MX) that suddenly changed version number (MCP9600 E/MX). According to Microchip, there is no change in silicon, but the new version will hang/crash if used with anything but Atmel and Microchip MCU’s. I’ve worked around it by just not doing specific calls to the device, but this cost me and the customer more than 40 hrs trying to solve and follow up with Microchip support. The customer knows it’s not my fault, but still kind of blames me. Reading James list in the post above, I feel that we’ve kind of been lucky not having had more parts cause problems than just this one?
Good questions, and I look forward to seeing what others say.
May I just observe that we could be operating in different market segments. If you’re not in the laptop business, the rate at which your parts EOL may be quite different.
OLPC worked with the CM Quanta Computer, which was at the time making almost all the laptops being made. Most of our part selection was done by them using their leverage. Our own part selection was mostly CPU or things that other laptops didn’t have. Of the other parts, some of them were parts that weren’t known to us until we had extracted the datasheets from the vendor via the CM. They hadn’t appeared on web sites, for example. The coupling between the CM and the vendors was tight and competitive, with multiple vendors being the preferred position at every stage.
Our sudden huge MOQ can be and was explained as the cost of replenishing a production line, when the tools have all worn out. That’s important for a physical part, like an LCD display, plastic casing, or LiFePO4 cells.
But most of the Microchip range is silicon chips, and those can be produced as wafer, tested, binned, and then stored in wafer for a while before they are broken out, packaged, tested and binned again. The tooling and production line is mostly common to all these products, so replenishing the line is covered by all the product revenue instead of just one part.
My guess is most of the future of Microchip’s silicon will depend on the forecasts they have made, and the orders they get. It should be something you can talk to your sales guy about, to pay more to extract some sort of quantity assurance.
I’ve worked on Medical devices that had 20+ year product cycles, the EOL that was most common was NOR Flash from Spansion. They were always doing process node shrinks to cut their costs but each shrink changed the timing (usually making it faster). We needed to re-qualify it every time.
Now I’m doing consumer and there is a part shortage so there is also the need to qualify alternative parts to keep the production line fed with millions of components. I’ve tested what seems like 20 different derivatives of a photodiode circuit for the same function.
TI is notoriously good about EOL, other vendors can have other reputations.
Microchip is one who revs their chips more often post release. Some of their parts are on like rev E. It seems like they are more fast and loose on their pre launch verification.
These days I guess there are not that many tier1 silicon vendors to choose from.
Like @quozl suggested, parts in the PC and mobile supply chain are far more susceptible to EOL and shortages, sometimes they are made for a single customer and produced to order one fab boat at a time instead of forecast+inventory. When the single customer goes away or next gen silicon is out, there is no reason to keep the old part around.
Had a BLE project a few years ago. I like Microchip generally and designed in one of the modules they’d bought from ISSC, what a disaster - the thing never worked reliably. Found the “Bobcat” from ACKme Networks, which worked perfectly from the get-go. ACKme was then bought by SiLabs who completely neglected the line. The part was EOL’d before the product could get to production.
Some other notable startup EOL acquisition victims:
TI - Luminary Micro "Stellaris"
This is how TI got into the ARM game, they bought luminary micro and then found out that there was an issue with the Flash in the chip causing them to EOL all of their parts.
Linear Tech - Silicon Dust RF Chips
I guess this tech never fully worked and it was a total dumpster fire.
“This is how TI got into the ARM game, they bought luminary micro and then found out that there was an issue with the Flash in the chip causing them to EOL all of their parts.”
Or alternatively, they just renamed it to Tiva and got rid of the Stellaris name, which wow the errata on those chips when TI first got them. They upgraded some, and others are pin for pin compatible. Plus TI had ARM support way before they bought Luminary, OMAP would be one example
Ah. I was doing some development on the Stellaris LM4F120, and it was weird. I went to look up some information a while later and it had just vanish. It seems TI tried really hard to scrub it from the Internet and pretend the Stellaris LM4 never happened. The Tiva-C TM4C123 is practically identical, except it has an option quadrature decoder input, and the timers got proper PWM support.
I still have an LM4F120 Launchpad, and I can see the MCU has a X part code.
Haha, I still have a Luminary branded kit from pre acquisition, such a collectors item.
Clarification is that Luminary got TI into the ARM Cortex-M game, yes OMAP & DaVinci precedes Stellaris.
I think Tiva was a port of the Stellaris RTL to a TI Fab and fixed the flash issues but I could be wrong. Would be interesting to know the full story and why they went to such lengths to erase the name.
Yes Tiva fixed a bunch of issues with the Luiminary.
My guess is they changed the name because the chips were horrific, TI seeded us a bunch of dev boards and I recall being super excited about using them since they had such a great feature set, but darn as i mentioned, the errata for those chips just made them pretty unusable, at least for us.
The Tempest /Inferno LM3S were at least some of the ones with flash corruption and were discontinued.
About a decade ago, I was working on the control panel for an “industrial” product and wanted to design in a small TFT color LCD module.
After evaluating various alternatives, we picked a really nice LCD. The LCD manufacturer had previously sold the product to Apple for use in their small iPods – but with the next generation of iPod switching to a different LCD (from another supplier), the previously exclusive LCD became available to the general market.
This was great for a short period of time, as the LCD module was well made, performed well, and had the right price point.
A few months later, though, we got an EOL notice – the LCD manufacturer wanted to keep offering the product, but as the LCD was no longer being sold by the millions to Apple, the forecast for the LCD was too small for the IC manufacturer for the LCD controller that powered the LCD module. There was a lifetime buy window and a promise that they were going to look into sourcing a replacement controller through other means (from the grapevine, I understood they ultimately managed to do so, but with a two year gap in availability!)
That LCD module manufacturer eventually got acquired by a much larger Japanese display manufacturer… In the meanwhile, we had to scrap our original design. The bean counters showed up on that redesign effort and the color LCD as well as a few other “nice features” got tossed out.
The replacement product functionally was as good as the original, but was widely panned for being unsexy by customers… Oh well.
Thanks @ToyBuilder. That matches scenario for OLPC’s LiFePO4 cells; which were semi-cylindrical; imagine taking two AA cells side by side and wrapping them in tape, and following the outline of the tape. The cells were available because of only one other customer of the battery factory (we didn’t know). Cells were capacity binned. We were taking the cream; the high bins. The other customer was taking the low bins. After a while, the other customer pulled out (we didn’t know), and our cells started to show 80% of our specified capacity. An unpleasant discovery. We guessed the cells were unsold stock. After a while, the EOL with big MOQ came in.
Well done, Trinamic!