[simias]

I don't know but maybe one of the factors is that given how cheap microntrollers have become it's not uncommon to use an "overpowered" integrated chip just for ease of development. Suppose that you have to drive some LEDs on a washing machine, do you bother developing some optimized bespoke circuitry with discrete components or do you just slap a ~2$ 100+MHz 32bit Cortex controller that will let you implement all the logic in C and just reflash if you find an issue?

[AlexC04]

It makes me wonder if it would be possible to build a chip-manufacturing plant for any reasonable amount of money to produce these chips that don't need to be 7nm GPU powerhouses, but like the old clunker chips that can't get attention from the big guys.

Almost like starting a "generics" business in pharma medication but for older chipsets.

I'm sure there's a great trade to be had in producing the lower end stuff.

[baybal2]

> It makes me wonder if it would be possible to build a chip-manufacturing plant for any reasonable amount of money to produce these chips that don't need to be 7nm GPU powerhouses, but like the old clunker chips that can't get attention from the big guys.

> Almost like starting a "generics" business in pharma medication but for older chipsets.

There is actually a great interest in this business, but mainly from Chinese. World's biggest 200mm fab is in Shanghai. A decision to build a brand new 200mm fab would've never flew in the West.

Chinese 3rd-4th-n-th tier fabs been vacuuming the market for old equipment for last 5 years.

> I'm sure there's a great trade to be had in producing the lower end stuff.

At this very moment, production on 150mm-200mm wafers is actually few times more profitable than on the latest process because everybody is now ready to pay absolutely ridiculous premiums.

[reportingsjr]

Most of the chips in these shortages are being produced on either older process nodes, or on slightly specialized nodes. The typical micro that's been hit by this is using anywhere from a 28nm to 180nm node.

The trouble is, this is a temporary shortage, so it makes no sense to spend serious cash (you're talking hundreds of millions) to make a new fab when the demand won't be there in a year or two.

[baybal2]

> The trouble is, this is a temporary shortage

It isn't. Designs on 200mm were in dire shortage for half a decade, and Chinese foundries were making very decent money on decades old chips.

For the last 3-4 years, 200mm-180nm had a 12 month+ backlog across the whole market.

[pharke]

I wonder if there's a good business in the mix of these ideas. If a lot of manufacturers actually are using over powered chips because they are a) more available and b) easier to program with newer tooling then one might be able to find a niche making cheaper/simpler/older style chips if they also provided modern tooling making it easier to program them for simple tasks like weighing things, blinking lights, playing little tunes, reading a sensor, etc. I've heard good things about PlatformIO so leveraging that ecosystem could be a win as far as avoiding creating your own IDE. Producing great documentation for the products would also go a long way towards gaining adoption.

[baybal2]

No the challenge is exactly the opposite.

Tons of chips still made at >130nm, and 200mm equipment for simple reasons that companies don't make much money, or not having much volume in this stuff.

[anonymouse008]

> The trouble is, this is a temporary shortage, so it makes no sense to spend serious cash (you're talking hundreds of millions) to make a new fab when the demand won't be there in a year or two.

While true, one could say it’s a bet on inflation to borrow dollars now for productive assets.

[Tuna-Fish]

The problem with your idea is that you are competing against obsolete high-end fabs, which have already paid back all their capital costs long ago. In a normal market, it's pretty much impossible for you to match them in price if you still need to pay yours.

Still, GloFo basically made this their plan, when they pivoted from the very highest-end chipmaking into FD-SOI, which is less performant but cheaper to design for.

[lunixicityee]

You could dip your toe in by getting a design produced, if you're interested in the process.

Google and efabless accept submissions every few months for designs that use a free 130nm process development kit:

https://efabless.com/open_shuttle_program

130nm is plenty ancient; it's the same feature size as a >10-year-old STM32F1, I think. And I hear that those MPW runs are starting to accept ~$10K for a guaranteed spot with a closed-source design.

So you'd probably be looking at charging 6 figures per wafer. I don't have good insight into startup costs, but I would guess high 8-low 10 figures. Running costs would not be negligible either.

Is that possible? I haven't crunched the numbers and I don't have enough information or context to do so accurately. But my gut says that it might depend on how many billionaires you're on good terms with.

[baybal2]

> Is that possible? I haven't crunched the numbers and I don't have enough information or context to do so accurately. But my gut says that it might depend on how many billionaires you're on good terms with.

130nm is quite ancient, but there are digital parts from early nineties still on the market. They are way bigger than 130nm.

Right now I have an ongoing project with a company making aircons. Their kit supplier uses a really, really ancient, and rare Hitachi MCU made on 600nm, and they are paying few dollars for it — more than some modern ARM SoCs.

They really want to change their kit supplier, or compel the chip supplier to cut cost, but the kit supplier itself can't migrate from Hitachi MCU because they don't have firmware sources as they themselves only copypasted the firmware as a binary for decades..

[pbourke]

> but the kit supplier itself can't migrate from Hitachi MCU because they don't have firmware sources as they themselves only copypasted the firmware as a binary for decades..

That’s seems like a rather existential problem. If I’m understanding correctly, the kit supplier makes the control board and the manufacturer does final assembly?

[baybal2]

Yes, and the Chinese kit supplier seemingly got the tech from a Japanese aircon maker somewhere in nineties, and then copied the board verbatim ever since.

[pbourke]

I wonder if you could run the firmware in emulation on a more recent CPU.

[KirillPanov]

> And I hear that those MPW runs are starting to accept ~$10K for a guaranteed spot with a closed-source design.

That is absolutely not true.

[CoolGuySteve]

I suspect the ultimate outcome of RISC-V is that it will be the commodity CPU the same way any fab can make DRAM.

[maltalex]

I think that's exactly what some of the old fabs are doing.

When a new process node comes out not all fabs are immediately upgraded. Fabs with older tech simply start producing simpler chips while the new ones pump out cutting edge ones.

[makapuf]

You can also put a 16MHz 8bit which can cost you a few dozens of cents max (Outside of shortage)

[ed25519FUUU]

You gotta take a second and respect how powerful these chips are that usually costs a few pennies each.

[AnimalMuppet]

And not need a 32-bit bus, so it saves on board cost, too.

[judge2020]

With the amount of horrible infotainment systems in the wild i honestly doubt they’re using overpowered chips. I’m sure any consumer grade APU (ie. CPU with an iGPU) from the past 5 years would do better than the chips currently in cars.

[simias]

I've worked in that industry. The problem with infotainment systems (be it in planes or in cars) is that they're usually designed years before the planes/cars enter production, they have very strong constraints in terms of price and component choice (you need automotive-certified parts, not smartphone parts, and they need to last a long time even if they have to go through Arizona summers) so they're already outdated by the time the car comes out.

These systems are also usually integrated with other systems to provide additional functionality using largely custom code that somewhat prevents quick iteration and code reuse, especially since the people writing the code are largely not in-house but various contractors (that's where a company like Tesla has the upper hand since I suppose that they control the software stack a lot more than the average).

Beyond that these systems suffer heavily from design-by-committee and worse yet, committees whose core competence really isn't computer UI.

[delecti]

Don't underestimate the ability of lazy, incompetent, or (most likely) rushed developers to fill the headspace given to them by overpowered hardware.

[danielmeskin]

How much of that is the APU? I’d imagine the bottleneck would lie with manufacturers using the cheapest panels and digitizers they can.

[kube-system]

I think the panels and digitizer used in automotive applications are pretty specialized and relatively expensive. They have environmental requirements that far surpass that of typical consumer products.

[krapht]

Eh, no, that's not how it works in high-volume manufacturing. There are 70 million washing machines sold per year. Suppose your large conglomerate employer sells 0.7% of that total, or 700,000 units. It doesn't take much of a per-unit savings to pay for the salary of a FTE to optimize the design.

[simias]

Maybe, you have to see if the cost of having independent components (dev time, prototyping etc...) is worth the few cents saved on the BoM.

Then you have to consider that IC designs are usually easier to reuse since they're more flexible, if you can have a single design with different firmwares for your entire line of products vs custom hardware for every design. Even if you sell 700k units/year you probably have a few models in your inventory, each selling for a fraction of that.

Beyond that it's pretty common for modern appliances to come with so-called "smart" features that require more processing and more IO capabilities. It's not rare for modern coffee makers to come with a color screen instead of the good old 7 segment displays.

So really the equation is not that simple, especially for higher end models that will have a more expensive BoM overall and a lower number of units sold.

[dehrmann]

It doesn't matter for this, but it's definitely the case in the hobbyist segment. Look at how many people use Raspberry Pis for things better suited to a microcontroller.

[speed_spread]

It's true that RPi are often overpowered but I'd contend that Linux is the platform being targeted more than the RPi itself. Development is much easier if you can assume a full fledged OS is running.