The chip shortage could lead to an era of hardware innovation
staceyoniot.com
staceyoniot.com
> turning instead to software to handle functions that had historically been done in hardware
That generally means doing things less efficiently. Devices will get slower and battery lifetimes shorter.
> There’s room for new startups and new ways of doing things. More of the car’s features are accessed using a touchscreen as opposed to a knob or button, for example, which turns components that once were hardware into a software program.
This has got to be one of the worst times to do a hardware startup. Doing R&D at an established company is already painful, between skyrocketing costs and 6 month plus lead times on standard microcontrollers (not to mention more specialized ICs). If you're just starting out instead of "only" facing trouble keeping production going you're going have to hope that the chips you're prototyping with will be available when you go into production. If things keep going the way they have this year, that's a very risky bet no matter what components you select.
> And now, in light of the chip shortage, suppliers to the automotive industry are going to be looking for more ways to turn something that used to be an electronic part into software.
Not in the automotive field but from what I've heard it's the exact opposite issue, the [lack of] processors they'd like to use in those fancy touchscreen infotainment systems are a big part of their problem.
The central thesis is that the shortage will lead to innovation, but in my experience it has put a HUGE strain on everyone doing R&D. I suspect it'll instead lead to rising costs and stagnation as people wait for supplies. This is already basically what we have in the consumer hardware market, where last year's consoles still aren't consistently on store shelves.
I hope it does lead to debloating though, when mouse config software weights in over 200MB we're long overdue for some trimming of the proverbial fat.
For computers, desktops, laptops, servers and phone, we could start focusing on more efficient software. Replacing laptops every two or three year is no longer cost effective. We start to look at 5 to 8 year lifespan for phones and laptop, meaning that developers need to focus more on efficiency.
It's two different ways of dealing with rising chip prices: Use fewer chips, and use them longer. It's also a great way of shifting to greener computing.
> That generally means doing things less efficiently. Devices will get slower and battery lifetimes shorter.
Not necessarily. For example, the original idea behind RISC was too move lots of complicated functions (and instructions) from hardware into software, so that the hardware could concentrate on doing the most common and simple functionality quicker.
I can't imagine it's much better for other hardware-->software transitions, given the cost of specific hardware implementations they're generally only used when actually necessary.
I think there's some truth to creativity being enhanced by constraints. Certainly, if supplies are limited, and especially if the limits are uneven, there's going to be incentive to design around the chips that are limited, and some of that might be innovation that could be useful even after supply goes back to normal. Of course, CPU shortages are hard to design around, especially if all CPUs are in short supply; but some applications might be able to make use of different CPUs that might have more availability.
Microcontrollers aren't exactly interchangeable, even within the same product line. You could design for flexibility and use the arduino framework to run your code on most microchip/atmel, ST, and a million other chips but that comes at enormous cost -- to put it nicely that's an incredibly inefficient library if you're not doing anything demanding, and damn near worthless if you are. Any multiplatform framework that abstracts away the inner workings of microcontrollers is going to be too heavy to work for a huge percentage of people's power and complexity profiles.
It's not just MCUs and firmware either, any time you replace a component due to shortages you need to revalidate your design. Constantly redesigning and revalidating boards based on available stock is what people are doing right now to keep the lights on. It's hell.
If you don't need a microcontroller to do whatever you do, then sure. Pop it out and save a few bucks. But that's hardly innovation, it's more rectifying a mistake made when doing the initial design.
I'm thinking of stuff like (at least some) dishwashers with 'dirt' sensing don't actually sense dirt at all; instead the pump has a thermal overload, and they sense how many times the pump cycles to indicate dirtyness.
If you used to have a dirt sensor, but it was delayed 18 months, you might figure something like that out, and maybe that's handy. Or maybe there's some other thing you'd like to measure, but there's not a good way to measure it, but it causes something to misbehave and you can measure that; but you wouldn't have thought about it except that you ran out of dirt sensors.
1. Travel restrictions make it extremely challenging to evaluate suppliers, do joint development, and bring up manufacturing.
2. A bunch of recent high profile failures have caused VCs to cool on the space.
3. Google, Facebook, Amazon, and Microsoft are all hiring armies of hardware people, making the job market more competitive than it once was.
All that said:
1. There’s still no shortage of interesting and worthwhile problems to solve with hardware.
2. Like web companies that survived the dot com bubble bursting, the hardware companies that make it through 2020 and 2021 are going to have been strengthened by it.
Could you expand on this?
When a gold rush is on nobody gets yelled at for buying shovels even if they lose money. But going against “conventional wisdom” requires either entirely independent VC or investors who really trust you.
5. China-West tensions, including HK/mainland border being locked down, are creating additional difficulties.
6. Current China anti-platform/monopoly drive has cooled domestic VC activity and appetite for vertical integration.
(Full disclosure: Cross-border hardware venture founder in China.)
Which ones?
If we think it's bad now, wait until the dozen or so AI chip startups have their moments of reckoning.
VCs have been "cool" on the hardware space since 2001.
Why fund hardware which can take 5-10 years to go somewhere when you can fund the latest social garbage fad and it will take off or fail completely within 36 months?
There is indeed a lot of demand for chips and electronic components of all kinds but I understand that manufacturers claim is not real demand, is just a spike caused by uncertainty in international supply chains. https://www.taipeitimes.com/News/biz/archives/2021/04/22/200...
Just like we all saw a lot of people buying toilet paper like crazy at the beginning of the pandemic, well, there is also a lot of hoarding by big tech companies that make this shortage likely to last well into 2022.
So, they're not exactly willing to expand capacity for a demand that is most likely to go away eventually.
So indeed, it is a challenging time to design and manufacture hardware; all the big players are buying it all.
The only advice I could give to anyone that is starting to design their electronics to make the design as flexible as possible, leave room on the PCB for 2 or more alternative components.
You could even try to make 2 different PCB layouts and choose one or the other based on price and availability of components, oh and try to delay components selection as much as possible, that's the most useful advice I've read https://titoma.com/blog/reduce-component-lead-time
>That generally means doing things less efficiently. Devices will get slower and battery lifetimes shorter.
I'm personally not at all convinced hardware accelerated GUI toolkits are more power efficient than software rendered ones. Weather it's due to the abstraction or some weird idea that drawing is now "free" you end up with way more drawing being done once something is hardware accelerated and it tends to more than offset the efficiency gains. The only place it really works are video games because they're actually paying attention to where the limits are. (and they don't really care about power efficiency.)
It really depends on what is being drawn and how. If a GUI toolkit is all flat colors, basic shapes, and doesn't use antialiasing then it can be very power efficient on modern CPUs, even low power embedded ones. If it instead uses gradients, alpha blending, antialiasing, and drawing complex shapes it's going to far less efficient all in software. The difficulty increases with increased demands for fill rate (display size x frame rate).
On modern CPUs (even embedded ones) a software rendering to a QVGA display would be no probably and likely no less efficient than hardware acceleration. However as the fill rate demand and drawing complexity increases software rendering will quickly hit a wall.
> That generally means doing things less efficiently. Devices will get slower and battery lifetimes shorter.
Possibly, but the point is to reduce the number of expensive integrated circuits. A classic example:
From the Wikipedia article: "The floppy drive controller was built with 8 ICs, one of which is the PROM, containing tables for the encoder and decoder, the state machine, and some code."
(My emphasis.)
What made it at all possible was moving most of the logic to the CPU:
https://retrocomputing.stackexchange.com/a/7188
The IWM chip, used in the prototype Mac with double sided "Twiggy" drives, and later in the original production Mac 128k with Sony 3.5" diskette drives, is basically Woz's original disk controller design (converted by Wendell Sander and Bob Bailey to NMOS), with the addition of buffer registers to allow the processor to run a bit more asynchronously. But, except for not having to run (* almost) synchronous with the disk data rate, the 68k CPU did all the raw data decoding (GCR nibbles to bytes, etc.) and head stepping in software, very similar to the 6502 code in the Apple II (DOS) RWTS.
The opportunity from either exploiting the shortage, or ameliorating it is now long gone.
> That’s because electric cars require a lot more computer and chip-based components than those powered by gas
Electric cars require a lot less electronic components!!!
> “I saw a design the other day of a lightbulb with 12 components,” said Carrieres. “[My team] asked me if this was an idea, and I had to tell them that it was a real product.”
Recently we had a rush project for redesigning a control panel for a certain machinery. The task was to replace, or remove MCUs currently experiencing shortage.
After looking at the project for a few minutes, we realised MCU was used basically just for blinkey LEDs, and an interlock.
Our engineer slapped few 555s, flip flops, and registers in under an hour. Client engineers' jaws fell off.
While the real reason client was at the mercy of their supply chain was not that individual MCU, it still shows just how much blind following of "one chip to rule them all" can eat away at supply chain resilience.
If you have a part uniquely fitting your unique needs, the likeliness it being some rare widget, and having poorer availability increases.
Sounds like it's more an issue of this work being outside the client engineer's competence than any kind of engineering breakthrough.
I suspect replacing the MCU with discrete components would have resulted in a more expensive product if there was no shortage.
It was likelly cheaper, but now they paid likelly hundredfold of what could've ever been saved.
And what? Complexity, and BOM size are not a problem by themselves alone. You should not view it in isolation from the context.
This is exactly the thinking which led the client running into this — "Trying to save money at any cost"
The point I make is that they easily spent a multiple times the price of this panel entire production run as a punishment for that.
Blinkey leds, and an interlock in between few buttons is really an overkill for MCU, and should've been treated as such no matter how many cents can be saved.
You can save a few dollars on a screwdriver if you can put in screws with a hammer, but you are certainly better off not doing that.
Sure. So what's important about context? Is it impossible to have a 555 shortage?
> Blinkey leds, and an interlock in between few buttons is really an overkill for MCU, and should've been treated as such no matter how many cents can be saved.
Would you refuse to use 1% resistors if they were cheaper than 10% resistors? Overkill isn't bad by itself.
> You can save a few dollars on a screwdriver if you can put in screws with a hammer, but you are certainly better off not doing that.
It's more like a set of 100 drill bits being cheaper than the specific 20 you need all wrapped in individual boxes.
this is a fun one, because there is no shortage of knobs and buttons, just chips, and knobs don't need chips to work. Even if you do need a chip for it to work (eg. rotary encoder to canbus), most of them are still available, because the "old tech" is good enough for them (compared to newer, smaller transistor sizes, needed for modern cpus to process a lot of data - eg graphics cards, modern cpus and even touchscreen radios for cars).
The problem is that we did make rich people the global arbiter of what people value, by giving them more units which can be used to value things.
This is a fundamentally undemocratic system. Ten billionaires have way more ability to decide on something's value than ten people making minimum wage.
(This is also the fallacy in "If you don't like it, why not short it and make money instead of complaining?" I don't like it, and I know my short positions aren't going to influence the market as much as some rich guy's much larger long positions.)
All modern democracies are representative. People themselves don't decide about things, their representatives do. This is the price we are willing to pay to get effective governance.
One of the usual goals of a democracy is equal representation ("one person, one vote"). You can transfer your representation to someone else - either directly, in the sense of proxy voting, or indirectly, in the sense of voting for a representative - but you don't give it up permanently. When elections next happen (and if elections don't regularly happen, it's not a democracy), you get your vote back and can freely choose what to do with it, again. If your representative isn't doing what you want, you can choose another one.
That's not true of the system of money. Even if it were theoretically been true at some point, once you transfer money to someone else, the associated power of that money stays with them or whoever they choose to transfer it to. If you bought Windows 95 for $209.95 a quarter century ago, that fraction of your ability to "vote" on how society values things left you a quarter century ago, after spending it exactly once on valuing Windows 95, and it probably never returned to you (unless you became a Microsoft employee or supplier).
So, someone who makes something that is (rightly or wrongly) highly valued thereby accumulates the ability to value things in the future, and have disproportionate influence on what is then going to be highly valued. This process repeats continuously and has repeated for centuries.
More simply, the US Constitution was not approved by "one person, one vote", not even with a dozen caveats and asterisks.
One person, one vote is certainly gameable. One dollar, one vote is even more so.
The overwhelming majority of valuation / price discovery does not take place in a coordinated or intentional manner. People buy what they want or need and in aggregate this determines price. Billionaires in aggregate are a minor portion of transactions that take place -- they can participate in larger transactions than most people but can't just will any market configuration into existence. They can create localized distortions, but if they're doing something unsustainable then they won't stay billionaires long and it will end up a blip.
However, I suspect your position may be a cryptic way of saying "There are some human activities / endeavors that I don't like and it's easy to blame the rich people invested in that / making their money off that." This would not support your point. It takes more than just some rich people to create and sustain an industry.
My point is precisely that this assumption is not true.
A more precise way to say that you're "doing something unsustainable" is that you're persistently overvaluing something. But how is it determined that you're overvaluing something? The rest of the market assigns it a lower value than what you assign.
If you and your billionaire buddies just decide to value something highly, then simply because you have money already, you can do that, and you can influence enough of the market to make sure that you're not overvaluing it relative to the market, and thus, it is sustainable.
The more money you have, the more leverage you have to pull off this trick. Sometimes it works, sometimes it doesn't, but the average person on a minimum-wage strategy cannot even attempt it.
People don't solely buy what they want or need. They also buy, in very large quantities, things that they expect to be able to resell later. Sometimes this happens at small scale (think toilet paper or gasoline, or even think houses); sometimes this happens at large scale (think commodities markets or mergers and acquisitions). These sorts of secondary markets have a huge influence on price discovery.
"Despite ordinances against organ sales, this practice persists, with studies estimating that anywhere from 5% to 42% of transplanted organs are illicitly purchased."
The window is closing rapidly on this option though. The more it becomes just another financial asset class with trillions in market cap, the harder that option becomes. However, if a country like the US ever decides it's a real threat to it's sovereign monetary policy, it could take that option anyway and simply bail out those impacted financially.
I think the most likely scenario is simply that it becomes regulated like any other part of the financial system. Heck thinks like CTR's SAR requirements. It will be difficult to ever get Bitcoin or others to their dream of decentralized currency unbound by governments when governments can make regulations controlling it's usage. The average business owner isn't going to risk their business and jail time trying to work outside those regulations.
1) ban computer games to free up people’s time, and grow the economy,
2) ban ML research to improve overall privacy.
All these will also affect GPU prices favorably.
No computers No chip shortage
Fixed :D
That will spur innovation on approaches that don't have that requirement.
There’s a long history of countries imposing capital controls. And it’s rarely favorable to the home country’s exchange rate. What you are proposing is basically just capital controls towards the crypto economy.
Yes it will make it harder for new buyers to bid up the price of Bitcoin. But it almost certainly will induce current holders to hoard any Bitcoin they own, thus reducing supply and putting upward pressure on the price. Without a liquid market, miners are likely to hoard their rewards, creating even more Bitcoin scarcity.
It also may create panic buying before the law comes in effect, or by people in foreign jurisdiction fearing the export of the policy.
How do you define one? What attributes does a pyramid scheme have?
One could easily ask what's a currency that's backed by nothing, is inflated to oblivion continuously robbing marginalized peoples' life savings, is often used for illegal drug/weapons/sex-trafficking/and worse crimes, is involved in a good percentage of the world's pollution, and that may soon be worth very little?
Like, it’s not a bunch of gold bricks or a briefcase full of paper money...It’s not controlled by a bank who can deny your transfer. Anyone can take all of their crypto anywhere on Earth, and can do so instantaneously and virtually anywhere with an Internet connection.
Its a squeeze, no different than what happens to rental prices when there is 1% vacancy rates versus .5% vacancy rates
Some of these technologies managed to produced something useful. On the other hand, I cannot use crypto to anything but speculation.
This video covers it quite well: https://www.youtube.com/watch?v=Z7QkIECEkVc
Automakers cancelled their orders during the early days of COVID-19 and lost their place in the queue (and now are crying foul).
Also many companies have double-booked orders from multiple suppliers making the shortage look worse.
There's a huge amount of semiconductor capacity coming online in the next few years, much of will be barely profitable once this current shortage dissipates.
What's really happening is that I (and likely many others) realised that Chinese-brand chips (which are not affected by the shortage) works just as fine as the western ones.
The other thing I realised was that non-authorized sellers works just as fine as well. Yep, I had to implement more thorough testing/QC, but I would not call that extra work innovation.
I think most assembly-houses self-source some parts, and they do it in large quantities (like 1M+ parts/year). As such, they buy at extremely good prices. They overbuy themselves, and in order to get the same good pricing next year, they keep overbuying. Some of their inventory is then sold locally, and ends up in the Shenzhen market or at unauthorized resellers.
For the quantities I buy (sub-1000 pcs typ.) the pricing is usually 0.2-2.0 times the known good parts from known (authorized) sellers.
The problem is that I can not be sure what I get, and I have to implement really extensive testing for each batch.
I have ended up raiding the warranty return bin for fixable boards in my small startup. Its easier to repair the returned boards right now than order new ones.
Or perhaps there is already a lot of recycling, but it is not apparent to me and probably not apparent to most people.
Is pulling and sorting chips from e-waste inefficient? Damages too many chips?
In China a lot of reverse engineered info on proprietary chips floats around among engineers. Are hardware engineers in the west too busy to adapt to recycled chips? Is the supply too precarious?
The only one I know of working in this area is https://en.wikipedia.org/wiki/Andrew_Huang_(hacker) I wonder if he gets lonely.
How about, when a couple of dead simple switches will do it, DO NOT stuff an embedded Linux SoC in. A toaster, for example, should not require one single bit of software.
Cars. Omg. Do we have to turn them into rolling smart phones? DO we? I dread the day when my car has to OTA update every 3 weeks. There should be no need for that. Yeah, I can see you might want something to handle engine timing or active suspension. But I do not need or want all these 'apps'.
We seriously need some sanity in product design these days. Don't blame engineers. Blame the product designers.
I agree that having to update too frequently would be a hassle, but we could make far better use of existing vehicles if they were more easily upgradeable and updateable.
Better yet - have the actual "car" layer (steering, acceleration, brakes, safety critical parts like airbags and ABS) entirely self-contained on a separate system that's written in a safety-critical language. Force all the "extras" (AC, seats, radio, windshield wipers, maybe blinkers etc.) to be controllable via a user-supplied module that might as well run on a consumer-grade electronics motherboard.
That way the part of the car that gets you from A to B gets to benefit from being self-contained and as optimised for emissions/performance as possible for the time the car was manufactured. The part of the car that merely provides comfort and convenience is something the users are free to monkey about with because it doesn't compromise any regulatory or safety standards.
About the industrial production of such analog objects, I have the intuition (maybe I’m wrong) that we lost the knowledge of how to build complex mechanical objects at scale. Those objects used to require a huge design work where now, all you have to do is a plastic case enclosing a chip and some software developers to bring it to life.
Maybe sewing machines? Typewriters? Linotype machines?
I admit that it's all a mixed bag. Common use of microcontrollers allows more cleverly designed electric hand drills at lower cost, but they are mechanically more poorly made than in days of yore.
Cars strike me as having the same problem, partly in the ability to design with computers, not just having a computer in the loop. Major components (transmissions, rear ends, etc.) used to be built in a much more beefy way partly because they were just guessing.
Luckily we are all free agents. We can buy old or we can buy new.
And this time round, the record can be made in a digital format, so we won't lose it for future generations.
from top of my head some superior analog devices:
* kitchen/body scale * oven * radio * light switch * thermostat
granted, it's mostly not per se because they are digital but that their UI is now a touchscreen with input lag and without any tactile response.
And I really can't remember any even though I grew up with them, weird huh.
Digital scales are smaller, lighter, more reliable. Radios and thermostats, too (those bimetal plates analog ones use are pretty bad).
Touch screens... I hate them but they're reliable, moreso than hardware buttons if not stressed beyond specs.
I've never used digital light switches, analog are great.
Ovens, yeah, I have a really old one that just won't die, I've worked on hundreds of them and it's always some electronics fault (high heat and poor cooling will kill a lot of components).
Good examples, thanks.
I will never miss carburetors and non-hydraulic clutches and brakes (shudder) though :D
Mechanical buttons: sure, if you're talking about those junky, flat momentary push-buttons then yeah - they will go out of service after a number of presses. But, a proper mechanical push-button or radio button, such as on old radios and on industrial machinery? Basically indestructible. They could likely last a century, and if they go bad, you can replace them because they are generally possible to service.
Somehow I have to upgrade all my hardware every few years, which of course increases the demand for chips, even though the software I use doesn't have much extra functionality over the software I used ten years ago. Nicer UI, yes, and I would argue that modern software is often more reliable than software ten years ago, but these things shouldn't require faster hardware.
This might be against the ethos here on HN, but I'm still of the opinion that 95% of the webpages would be much better if they were written in plain HTML.
Or, knowing that users won’t be purchasing new computers at as fast of a clip will encourage software made for a wider audience.
Sure, some of those energy savings are a result of better design and materials rather than clever electronics and software, but often the two must be used together - for example the washing machine that can be lightweight because it uses software to rearrange the washing in the drum for the perfect balance... And a lightweight washer reduces energy consumption to heat it up...
Imagine getting software updates that make your old phone faster rather than slower?
Crypto is here to stay, stop fighting it. Instead try to fix its issues.
Drugs are a consumable, crypto is an item of exchange. If you cannot exchange them, then they have no value.
About 60%+ of market activity is in the USA: https://www.statista.com/statistics/1195753/bitcoin-trading-...
The USA could ban online exchanges in the same way it has banned online gambling.
Sure, trading could still occur via foreign exchanges, but the pool of people willing to put money into the network would shrink dramatically and the price - since the whole thing is a ponzi scheme - would also fall dramatically.
If the price of Bitcoin or Ethereum was 5% of its current value, the levels of electricity and advanced manufactured goods dedicated to mining them would also fall to the same extent.
Turkey (an economic minnow) banning exchanges, and a few tweets from Elon Musk have been enough to send prices tumbling 10-20%. Can you imagine if the USA banned the exchanges?
Its also likely that allied nations eg. European countries, Australia, Korea, Japan, India would follow a USA ban with their own.
I don't know where that report got its numbers from but they're bullshit; the majority of trading takes place in unregulated exchanges (that explicitly ban US customers) like Binance and Huobi: https://www.statista.com/statistics/864738/leading-cryptocur.... In large part because the US exchanges charge outrageous fees. The US exchanges are popular with large corporates because there's less counterparty risk, but the majority of retail volume is traded outside the US, on exchanges not hosted by US allies.
Plus, China is actively subsidising miners, and is deliberately supporting the crypto ecosystem due to its potential to further destabilise the US dollar's role as global reserve currency.
What is banned is 1) paying for stuff with cryptocurrency and 2) sending money to exchanges via "digital wallets", you have to use bank transactions.
Looks like they are paving the ground to trace and tax crypto investments.
So yeah, this supports your point anyway.
Thus there are some value to having a warehouse. Ie something between just in time delivery and keeping lots of stock.
I have been asked to optimize to return vms to the hardware pool ..
I’m fairly sure there’s enough old unused raspberry pies in peoples drawers that could be reutilised.
Chip X exists in Product A, which costs $10. Without Chip X, Product B cannot be sold. Chip X is suffering from a shortage. The gross margin on Product B is $1000. As a consequence, the manufacturer of Product B buys a ton of Product A on Alibaba in order to cannibalize it and harvest Chip X. Now the gross margin on Product B is $990 (not so bad), but Product A is nowhere to be found.
In other words, the shortage moves to the cheapest available finished product that has Chip X in it, until marginal profit reaches 0.
The consequence is obvious. Product C, which cannot be manufactured without Product A, eventually becomes unavailable. And so forth.
This is particularly gruesome in the semiconductor sector because supply chains for finished goods are so long.
Obviously this would only be viable in low end devices (tooth brushes rather than graphics cards). But there is already an industry based around desoldering old parts and selling them as 'fake' new ones - if this was done in scale with the parts tested and a company could order 20K there would be a market.
Every time you need a $3 charger chip (real price of isl9241, one at the end instead of zero) to fix a $3000 computer you have to buy a $130 apple gadget, desolder it, and throw that $130 apple garget including battery into garbage bin.
$3000 laptop turned to paperweight due to ISL9240 unavailability. https://www.youtube.com/watch?v=HJ2jyo7pAmE
There must be a huge amount of pis that ended up in some drawer after a few weeks, and because it's cheap it was never worth it to sell it. And if you wanted one it was so cheap that it was more convenient to order a new one than to deal with private sellers on Craigslist just to save five bucks.
It would be great if we could repurpose all of them in some halfway meaningful way, but at lest for commercial products it's just not feasible.
EVs have less fundamental systems than an ICE. I can see that EV designs being the most modern and entailing a lot of clean sheet thinking would result in more computers, but fundamentally I would think it takes less computers to run an EV.
A battery pack may be made up of 20 or more modules that each have their own charge management logic, and there are battery heating and cooling systems that need controlling.
Most systems are the same in ICEVs and EVs - cabin control, entertainment, steering, transmission, throttle and brakes, suspension, lights, safety systems, and so on. To me, the difference doesn't seem very large.
https://www.manualslib.com/manual/628934/Sapphire-Audio-Pure... (soldered CPU in the low res picture)
But it's not specific to AMD mbs, many Intel Atom have a soldered CPU.
https://www.mini-itx.com/2017/10/09/intel-atom-c3958-gets-be... (Atom C3958 [..] is a fully fledged server-grade (!!)SoC)
RAM is usually upgradable. For processors it depends on what you buy.
modular desktop computers were and are a niche.