Explaining the global chip shortage | Hacker News Reader

Explaining the global chip shortage

jabil.com

91 pointsnabakin54 comments

In some cases, the shortage has an obvious explanation. Many chips are made in exactly one fab, and may be made for a few days per month or a few months per year when a line produces multiple devices or device variants. Designs are not readily transportable to other fabs. So if there is an interruption from, say, a fire or a pandemic, the global supply of that chip is stopped cold until the interruption is resolved.

If a chip has been shipping for a while, the supply chain may be full of enough inventory that end users don't see a problem before production is restored. But imagine a case where the line is stopped for months and at the same time, the demand for that device accelerates. This is what COVID did to a lot of devices. Contagion mitigation strategies sent all the operators at the Infineon Kulim fab home for months, at the same time that auto sales surged. The power switch devices popular at US automakers, and purchased for 45 cents each, started seeing longer and longer lead times until Infineon's distributors started simply reporting 'availability unknown'.

This forced buyers into the perilous broker market. I say perilous because prices quickly went out of control for this chip. End customers were paying more than US$50 for many months and in some rare cases, over $100 for this 45 cent chip. These prices were paid by, among other companies, suppliers to semiconductor equipment makers. Without that chip, they'd not be able to sell their $10 million machines, which were also seeing demand skyrocket as chip makers scrambled to add capacity. Equipment makers would pay anything to keep their subsystem supply chain alive.

When the gray market got wind of the prices being paid, all kinds of scandalous activity happened. For example, unscrupulous resellers would gather many small lots of that chip, even units with different specification variants, paint over all the markings and re-label them all with the same, current date codes. This greatly increased their salability and value, but left the end customers with an unreliable product.

Eventually, the Kulim fab came back online and is back at full production. The supply chain is mostly recovered. And that 45 cent chip is now selling for 80 cents because Infineon saw first hand the high-value applications it was used in, and the price insensitivity of their end customers.

What are some of the intricacies in the 45-cent chip that makes it hard to produce by a large number of different fabs?

In the absence of more details, it seems to me that a cheap, 45-cent, chip should not be so complicated that only a small number of companies can produce it on-demand.

It's not a matter of intricacy. It's a 'low tech' power chip that is made on fully depreciated and outdated equipment. Most likely all Infineon lines were running at capacity at that time and they had no place to put it. Setting up a new line would not make business sense since a new equipment would cost in the realm of $250 Million for a line, and take a year or more to set up. Not to mention, lead times for new equipment was also extending. By that point, the crunch may as well have ended and they'd then have more capacity for that device than they needed, and no ROI on their $250M.

I am having trouble wrapping my head around a business that costs 250mm and a year+ to enter not increasing its prices earlier or even more than the current 80 cents. It must be that they are also limited in their ability to increase price by people just paying the cost to redesign to another chip in the medium to long term?

knowing it would cost that much ahead of time would let the purchaser and engineer work together to determine if it would be more cost-effective to swap that part out for a similarly specced but less expensive chip, or rework the circuit to use a slightly different chip, perhaps with two chips instead, or more jellybeans to make up the lost functionality

a ten million dollar machine missing a 45 cent part probably costs a few hundred thousand to redesign around that part. a mass market vehicle that needs multiple of those part per SKU would be cheaper to re-engineer

> knowing it would cost that much ahead of time would let the purchaser and engineer work together to determine if it would be more cost-effective to swap that part out for a similarly specced but less expensive chip, or rework the circuit to use a slightly different chip, perhaps with two chips instead, or more jellybeans to make up the lost functionality

And now you have not one but 2-3 chips that are in demand and price rises for those too.

Also in case of cars and similar equipment there would be a bunch of re-testing required so that drives cost of reengineering hard.

Especially if said part was used as "jellybean" across a lot of modules just because it's cheaper to buy same part in bulk

if a part started off costing 45 cents, it can't be too difficult to reproduce using transistors or similar components from competitors. re-certification for simple subsystems like that (or just using a competitor's chip) would still be less expensive than paying $100 per part on a SKU/car that you are producing 10,000+ units of. TI's 74AHCT595 is probably functionally similar to Nexperia's/NXP/Diodes Incorporated (digikey shows those 4 vendors for that chip)

there is also the option of the buyer just saying 'nope, not worth it', and cancelling the product line the chip was bound for, which would be worse for the chip manufacturer who just sunk millions into producing the part

it's an intricate balance :) regulatory bodies aren't fans of price gouging either, and might step in if the delta is actually that drastic (tangential but relevant https://en.wikipedia.org/wiki/Martin_Shkreli#Daraprim_price-...)

I guess if those chips were much more expensive for a long time, car manufacturers would find other solutions than using them. A shortage makes them wait rather than redevelop the whole product. So for the chip manufacturer it's better to keep prices low to avoid losing customers altogether.

Because prices have zero, or close to that, to do with costs. If your costs are too high for the market to accept, you simply go out business.

Oh, I mean, if it costs enough to enter the market that other people don’t step in during times of high demand and scalping, why don’t the original manufacturers raise their prices to closer to what the scalpers/resellers are getting, at least until some competition arrives. there must be some mechanism making that unattractive / not work is all I mean - maybe the fact that people can redesign to a different chip.

Production is cheap, engineering is expensive.

If you raise your prices too much, people will switch to competitors. This means that nobody is buying your product, so you'll never get to recoup the engineering expenses. Even worse, you'll become known as a manufacturer with unreliable pricing, making your chips very unattractive for new designs.

Having a steady and well-known demand might be a lot more valuable than the short-term profit you can squeeze out of the current shortage.

> Setting up a new line would not make business sense since a new equipment would cost in the realm of $250 Million for a line, and take a year or more to set up. Not to mention, lead times for new equipment was also extending.

Here's where this could get ironic: what if they wouldn't be able to set up that line anyway, since the equipment they'd need itself requires the very chip they just stopped producing?

I wonder if such circular dependencies are already a consideration in the semiconductor business.

I feel like the number of such chips required for “production scaling” would be so low that they could find a cache of said chips for said purpose if they wanted/needed to.

The problem is that the different fabs are, well, different. While on a high level they are basically doing the same thing, the actual process itself can vary quite a bit.

While you start with a fairly generic design, the final production artifacts are highly specialized to the fab. The fab essentially provides a "library" of components which are used to translate the generic design into something ready for production.

Switching fabs means re-engineering your design, re-running prototypes, re-validating basically every single part of your chip. This is a process which can easily take many months, perhaps even a year. Not to mention significant engineering costs.

Doing all this once is acceptable because you can just keep using that single design for decades. That's why each individual chip can be so cheap: production cost is near-zero, and engineering cost can be amortized over a giant volume.

It's an interesting article overall - but it seems to be half written now, and half written earlier in 2022. There's no publish date on the blog article that I can see.

Google seems to think it was posted 3 days ago - but in the article you have sentences like "However, some of the most in-demand phones are expected to become more available by February 2022." Which adds to the confusion of the currency of some of the statements / observations.

I'd go further and argue that this contains no new information and it's basically a sales pitch for outsourcing your electronics manufacturing to Jabil.

You're correct. You can wayback machine it and see they first archived a version in 2021 and it's been updated multiple times since then.

https://web.archive.org/web/*/https://www.jabil.com/blog/glo...

This video seems to be doing much better job at explaining the problem [1].

Long story short there's high demand for trailing edge (50nm+) tech, but it is not worth investing in building more of these for number of reasons mentioned in the video (huge investment, long time, and despite high demand it's still only low percentage of profit compared to cutting edge)

I suppose prices must increase but likely there are some long term contracts in place.

1. https://www.youtube.com/watch?v=YJrOuBkYCMQ

One other thing that's not mentioned is qualifying the new fabs and porting old designs. Each design (even the old ones) would need to be requalified. That could easily be $Millions for a $0.1 die. Until you've made >>10 million parts (1000s of wafers) you won't be cost competitive even in a converted old depreciated fab... Now add the costs of building a new fab and bringing it on line?!

Realistically, I can see building new 28nm 12" fabs and designing new die (for the same purpose) for the smaller geometry, but anything older/larger will just make due with what is available. The ecosystem existed because newer fabs were always taking up the demand slack, while the older lower cost fabs were fully depreciated and didn't take much engineering effort to keep running the same old process.

> Realistically, I can see building new 28nm 12" fabs and designing new die (for the same purpose) for the smaller geometry, but anything older/larger will just make due with what is available. The ecosystem existed because newer fabs were always taking up the demand slack, while the older lower cost fabs were fully depreciated and didn't take much engineering effort to keep running the same old process.

Also, power matters. Smaller process means jack shit if the part needs the same amount of current running thru it. Anything analog is harder to shrink (as you care about the detailed parameters of transistors, not just the on/off state, so most likely full redesign), and anything that needs to pass enough current and enough voltage just can't be shrunk in the first place.

See your humble 5A linear regulator die shot

https://zeptobars.com/en/read/LM338K-5A-LDO-TO-3-TO-220

only the upper part could be shunk, the bottom one, the power transistor, can't be

Luckily, a lot of the really high power (moderate voltage) stuff is moving to SiC and GaN which is smaller and much higher power density. So there's some reason to hope for new fabs to ease capacity constraints.

The global chip shortage reminds me of the "just in time" training that corporate mandated for all engineers at my large cap aerospace company about ten years ago.

"Is it always better to avoid having spare parts? Maybe in some cases..."

"No," the older engineer a few seats away from me interrupted. "Keeping extra inventory is a waste of money."

He actually thought that redundancy is a waste and selecting parts doesn't involve any consideration of uncertainty...

That's one of the main presuppositions in Lean Manufacturing that is so entrenched and so insidious. What's even more annoying is when managers who have worked in a Lean environment come over into our industry of steel fabrication (high-mix low-volume, where lean is low-mix high-volume), and they assume that 1) we're screwing up because nothing is consistent in our charts and graphs and 2) that if they just make us do Lean stuff, it will magically improve.

It leads to disaster every time.

Lean manufacturing is a lot about avoiding waste, but it doesn't necessarily mean removing all buffers and stocks, which is a type of waste, if this is causing a much worse type of waste, which is making the entire production line useless because it is missing input material. I really think "lean manufacturing" = 100% JIT is a misconception and an over-simplification.

And wasn't it notably Toyota, inventors of lean which had a decent stock of components? (Not huge perhaps, but better than a lot of competitors.)

The argument becomes how much buffer and stock do you need? How long can you fade a falloff in supply? One hour? One day? One week? What is "wasteful" vs. what is wise?

Exactly, which is a much more sane and useful discussions to have, and something any serious process engineer will do using for example a simulation.

> "Keeping extra inventory is a waste of money."

But he is right! Extra inventory has a cost. Each cost increases the final price.

And the customers, they don't buy from you because you have extra inventory in case of a global pandemic (as well as they don't buy from you because you employ local staff or because you pay them well or because you make an actual effort to reduce your carbon footprint): no, customers buy cheap.

Whenever you are tempted to criticize a corporation about some decision they took, do consider those pesky, cheap customers first.

So perhaps a good counter would be:

"No, keeping spares is insurance."

This way, the costs of keeping extra inventory is acknowledged, while underscoring it also buys you something. Whether or not you need it, or how much, is for discussion, but treating spares as pure waste by definition doesn't feel responsible, and is how you end up in trouble if there are supply chain disruptions - whether global or local.

And in more general sense: money is indeed cheaper to store than physical items, but can only substitute for them if, at the time you need those items, you can actually buy and get them.

I've dealt with US companies as European for a couple of things, and I think that I've been avoiding companies from there for the past couple of years due to speculation about this chip shortage, I've 2 bikes from US one super 73 and one bird bike, for the bird bike in particular, in order to have the engine replaced I've had to wait for 2 months, due to chip shortage, while the shop I've bought it from has had the whole time new bird bikes delivered to sell, so yeah, I think a good part of the global chip shortage when it comes to post-sales components is speculation, I've got a vanmoof since then and it takes a week get damaged pieces (not needed replacement parts so far, but I've friends with vanmoof who had)

But yeah as a rule of thumb stay away from super73 and birds bikes

Reply on news.ycombinator.com