Kliment

Not really. You can't reliably make large features with a small process node, and because the smaller processes are dramatically slower and more expensive and have lower yield, unless you can counter all that with more dies per wafer you end up with a whole lot of defective parts at enormous expense. The bigger the die, the higher the chance of a defect landing on it, and making it useless. Also small feature processes have much lower capacity than larger feature processes, so even if you were able to switch there wouldn't be nearly enough wafer capacity. The fact it's 4-5 orders of magnitude more expensive doesn't help either.

This article is sadly incorrect (not poor journalism necessarily, but based on statements from vendors who are defending their own interests).

Big name manufacturers are refusing to invest in fab expansion because they believe (correctly) that total demand for semiconductors has barely moved. What has changed is the confidence in the market (and this was destroyed by a major fuckup by car manufacturers deciding to screw over their suppliers at the expense of literally the entire industry). Nobody is actually producing more electronics (many manufacturers are producing less because they need to ration supply and can't allocate as much to specific products and risk not being able to produce others). Everyone is buying more, because they can't be confident they'll be able to get stock when they need it.

Chip manufacturers know that anything they put on the market now will be gone immediately, at any price. They also know that one day everyone will have enough stock stockpiled and will stop buying for a while. If they increase production now, they have added expenses and a limited time to recoup them before there's oversupply on the market. If they keep production at current levels, they get more revenue as the prices are higher, and lower costs so they can weather the coming overproduction phase better. It makes zero economic sense to expand production.

Some chip manufacturers are asking customers up front for demand visibility several years out to be able to plan better. Others are being absolute dicks and doing the same thing in an openly-hostile way by holding customers hostage (yes we know you ordered two reels of this part and we promised delivery by date X. We won't deliver on date X unless you also order and pay for another ten reels which we won't commit to a delivery date on, and also your order for the two reels is non-cancellable and non-refundable). I can't name this company because it would put the customer at risk, but they're a household name in the industry.

The core problem is everyone is operating in the dark and there's no visibility into future demand, supply, pricing, or availability. In this situation, large capital expenses make sense on the buyer side (to ensure future supply and widen their visibility horizon) but not on the supplier side (who then shoulder the risk of future demand cratering and being left sitting on their capex).

Not quite. You need bigger feature sizes to get good yields and good voltage resistance and performance over wider temp ranges, so a modern design would still need to be built on large-feature nodes. The car industry did cause this, but they caused it with their whims, purchasing practices, nasty contract terms, and general fucking over of suppliers, not with their allegedly outdated designs.

Do it! I switched from Eagle years back because I needed a guarantee that I'd be able to open a decade-old design and the subscription model destroyed that confidence. Kicad was annoying but manageable then, it's an absolute joy to use now.

See my other comment in this thread - it makes no sense to ramp up production because the current demand increase is almost entirely due to lack of market confidence and there's no guarantee it will last long enough for new capacity to go online. That's why ~nobody is building new capacity for the affected nodes.

Graphics cards and gaming consoles are heavily affected by availability of many other things than their headline processors. There's a ton of other parts in there, power management ICs, level shifters, muxes, switches, all sorts of tiny cheap parts that are out of stock. Suppose you're making a game console. You're missing one power management chip, and suddenly you can't produce any units unless you redesign your board for another part, assuming you can get it. It's enough for a single part in your bill of materials to be out of stock and you can't manufacture anything. This dependency means you will tend to buy as much as you can of all the parts you might need, draining the market of any excess capacity. Apple needs lots of chips other than their SoC - they have chips for power management, radios, digital microphones, sensors, antenna amplifiers and switches. Even if supply for their SoC is guaranteed for years, they can't build a single device unless they have every single part that device needs.

Automakers caused this. The issue is not just that they fucked up and ruined the market that one time in mid/late-2020. The issue is that that one fuckup destroyed the confidence in the market. So you can no longer rely on stock being there when you need it, meaning you need to buy as much as you will ever be likely to need, because you don't know when you'll be able to buy again. So you buy as much stock as you can afford to, and if you can't afford to you borrow money and buy it anyway. This is a big problem, because the signaling mechanism of the market is broken. Any stock that shows up now will be gone instantly, but the aggregate demand for parts hasn't really moved - people don't need significantly more parts, but they do need to be certain they'll have them when they need them, which is why they buy more than they need now. At some point you have enough stock and don't need to buy again for years. This is why it doesn't make sense for semiconductor manufacturers to build out additional capacity - by the time it's up and running 14-18 months from now, there's no guarantee anyone will be buying parts in anywhere near the volumes that are needed right now. So this demand shock at the moment is temporary and nobody can predict how long it will last, but you need to know demand way out if you're going to set up a fab expansion. This is why the bullshit move that automakers pulled is having so much outsized impact on the entire industry.

To answer your other questions:

All complex devices with large number of parts in them are super vulnerable to one tiny part halting the production of the entire device. This is why complex devices are poorly available. For GPUs and game consoles in particular there's an extra market-distorting factor of bitcoin fuckers and scalpers, respectively. Apple builds complex devices with large numbers of parts in each.

You can't use a 7nm fab to build things you would normally build on 120nm. Not only is there way too little 7nm capacity, and it's much more expensive per chip, but the smaller the node size the higher the defect rate is going to be, and the lower voltages your parts are going to be able to handle. Automotive and power electronics need to handle high voltages. You will never see 7nm process parts being used for switching mains voltages, for example, or driving 600V motors. In addition, designing parts for newer processes is a lot more work and has very long setup times.

None of the chip shortage can be attributed to AMD or Apple or Intel. They are a totally different category of part/process than the ones that are currently most affected. Intel is still operating near their capacity and selling countless server and laptop CPUs. I don't think this is a factor at all here.

That's not been true for the CDU for at least three decades.

I also scored far on the left and I'm doing okay in life. You don't need to be evil to get shit done. Just get shit done.

There's exactly one thing I still prefer to do in Gimp - splitting a single image into multiple images, one per RGBA/HSL channel. There's also some image formats that Gimp supports that Krita does not. But that's it. There's nothing else I miss, and Krita is just incredibly far ahead on usability.

Here's a good English-language explanation of the Mankala model https://www.lexology.com/commentary/energy-natural-resources...

If you're into foodstuff-related destruction, the other things to search for are "flour explosion" and "molasses disaster"

The energy densities of modern batteries are a couple factors lower than that of gasoline, and much lower than TNT. Gasoline is incredibly energy-dense, but its energy density is comparable to butter. Do you worry about sticks of butter exploding on you? No, you don't. Gasoline is dangerous because it really likes to evaporate and the mixture of gasoline fumes and air ignites at the slightest provocation. Batteries are dangerous because they can go into thermal runaway from an internal short if the barrier between the electrodes fails. The energy density is not what matters here.

Devices of that size almost universally use prismatic lithium-polymer cells.

There's very little fluorine-containing anything in the size of cell used in wireless headphones. Note that the article you linked used many amp-hours worth of batteries for each test, and they took multiple minutes to get to over 50ppm of HF. Scaling that down to headphone cell size (~600mAh) it would take tens of minutes to produce any significant amount. My theory is outer material containing vinyl decomposing to chlorine which caused the direct reaction. Would also make sense as thermal decomposition of PVC is only dependent on temperature and would be in direct contact with the person holding it.

There's already head-only MRI machines. Like all MRI machines, they are phenomenally expensive to operate, eat shittons of energy, and only make sense if they're in constant use. But yes, reducing bore size improves resolution.

Yes and no. You can order Comirnaty in Germany if you're a doctor's practice, and the deliveries will come from the central distribution facilities in each German state. Those facilities are receiving shipments from the Marburg production facility that BioNTech owns and operates. Similarly, if you were to do the same as a Belgian physician, you would receive shipments that originate in Pfizer's production facility in Belgium. It's exactly the same product though. I don't know if there's a difference in the labeling beyond specifying production site. The production lot labels you get for each dose to be added to patient vaccination records look exactly the same (other than the lot serial numbers of course).

Or John's work is in frontline production use and Sally's and Mike's is not, so there's different exposure.

Anyone have a non-paywalled link to the article referenced at the end of the post?

Not at all, though it's great for that.

Location: Germany

Remote: Remote only

Willing to relocate: No

Technologies: electronics, part sourcing, PCB layout, hardware prototyping

Email: kliment@0xfb.com

Project/contract work only.

I do electronics, end to end (design from idea, part selection, layout, prototyping, testing, enclosure design). Lately I've been working mainly on adapting existing designs to the new reality of part shortages. If you have a device that you can't produce at the moment due to part shortages I might be able to help you redesign it for the current availability situation. Of course I'm also more than happy to help with new devices. Contact at kliment@0xfb.com or Kliment on liberachat.

Recent-ish example of this kind of adaptation: https://www.kickstarter.com/projects/arthurwolf/smoothieboar...

I won't work on weapons, advertising, cryptocurrencies or automotive (for motivation on the latter, see my analysis on the reason for the component crisis here https://news.ycombinator.com/item?id=26931498 ). This is non-negotiable, don't bother asking for exceptions.

SEEKING WORK | Remote only, based in Germany

I do electronics, end to end (design from idea, part selection, layout, prototyping, testing, enclosure design). Lately I've been working mainly on adapting existing designs to the new reality of part shortages. If you have a device that you can't produce at the moment due to part shortages I might be able to help you redesign it for the current availability situation. Of course I'm also more than happy to help with new devices. Contact at kliment@0xfb.com or Kliment on liberachat.

Recent example of this kind of adaptation: https://www.kickstarter.com/projects/arthurwolf/smoothieboar...

I won't work on weapons, advertising, cryptocurrencies or automotive (for motivation on the latter, see my analysis on the reason for the component crisis here https://news.ycombinator.com/item?id=26931498 ). This is non-negotiable, don't bother asking for exceptions.

One thing to be aware of with RNA viruses - they mutate all the time and almost all mutations are nonviable (unable to infect cells or unable to replicate) and die off immediately. The fact there are so few variants (and there hasn't been a new dominant variant since October 2020 ) illustrates how fragile its mechanisms are.

Sci-hub does in fact cache and store everything. They have articles available that are no longer available from the original source.

Of course we are, and have been doing for ages. The difference is that the trash cars end up being replaced by giant comic book spaceship cars that use the gains in fuel efficiency to add more bulk to the vehicle and still emit as much nasty. Switching to replacing the existing dirty new vehicles with electric transport won't entirely remove the problem, but it will at least stop making it worse.

I suspect it's either a typo or an extremely US-centric view.

There are <$5k EVs in China (when bought new). They're not super luxurious but they exist. The west just hasn't caught up yet, but it will if there's political will.

I use obscure parts. For example, just this month I've used a part that the manufacturer doesn't admit exists on their website, and a part where the manufacturer doesn't even have a website, and neither is anything widely used. But there's always a small community of people who use the part, and we cling to each other and compare notes vigorously.

Maybe it's just me. Vendor documentation in hardware tends to be kinda crap but there's usually plenty of other sources for information from other people who implemented something using the part, and that information doesn't get stale, ever, and documentation state improves over time. I've had problems finding reliable info on what's going on in webdev, most posts are "do this" solutions which don't really help you understand what is happening under the hood, and there are countless crappy tutorials. To be fair, I'm an extreme case, I touch web stuff once every few years, and only if there's really no way around it, and a lot happens in browser capabilities and javascript in a few years.

Strong disagree. I do embedded development of varying complexity. I often have to learn and figure out new things, but once I have something figured out, I have it figured out, and I can drill down to hardware register level to figure things out when something is misbehaving.

Every single time I touch anything web-related I discover that the entire landscape has shifted, all the tools I used last time are now abandonware, and I have to figure out everything anew. It may be more accessible if you're starting from scratch, but it's a constant treadmill that you instantly fall off of as soon as you do anything else. And when something doesn't work? Good luck finding out what's happening in the countless layers on top of layers on top of layers.

Apparently both kvm and serial