1. Yes, everyone cancelled orders and idled capacity. Everyone was anticipating a deep and long recession until the fiscal/monetary bomb of March/April 2020. See #7 above.
2. The 120nm legacy nodes and similar are right at the center of the whole shortage. This has been underinvested for years because it has been low demand growth and low margin. Automakers are particularly vulnerable, because for years, every new feature has gotten its own cheap microcontroller to run it. There are now dozens in new cars.
3. Graphics cards have been affected by crypto miners sucking up supply at any cost, and everyone not named “Apple” competing for TSM’s top end processes.
4. iPhones don't just come with Apple A-Series chips made on TSM’s top-end node, but a host of others, some made on legacy processes where the shortage is. Likely we are talking about some cheap analog power chips or something like that. Apple has decided to push as much supply chain problems to iPad as they can.
5. For the vast majority of what automakers want, 7nm is overkill. Once there are more high end SoCs running multiple systems including self/assisted-driving, that will come into play. But that is not where the shortage is concentrated.
6. Apple was Intel’s 4th biggest customer before they left, so it has had a small impact, but only at TSM at their top node. Samsung has picked up Qualcomm and AMD as a result.
Crazy times in the semi world.
On #1 also bear in mind that the supply chain is deep and complicated. OEM (GM, Toyota, Tesla, Ford etc) orders a number of COMPONENT X that contains a chip from a Tier 1 supplier. Tier 1 supplier order subcomponents from other suppliers. (Continue this until you have built a very confusing interdependent supply web).
At any level of this, anyone could have (and did) cancel their orders in 2020. As in OEMs cancelled some orders, Tier 1 suppliers cancelled some orders, Tier X suppliers cancelled some orders, etc.
I like to say that reality cannot be reduced to a single variable. It is a complex multivariate problem.
Most reporting out there on any subject reduces that being reported to a single variable. Example: Container shortage > bad > supply chain crunch. Etc.
The overwhelming majority of people have no practical understanding of business and, in particular, manufacturing. The complexities involved --the variables-- do not exist in their realities. What you explain about the supply chain is a revelation to almost everyone reading it. Even with this, reality is far more complex than what any of us can explain in a couple of paragraphs.
I have lived through a couple serious supply chain issues in the last several decades. This one is a bit different because almost everything in technology products is affected. It's one thing to re-spin a design because one chip becomes 60-week unobtainum, quite another when you can't get 75% of the BOM for 40 weeks.
One thing I learned working in manufacturing is to think about movement and management of empty carriers. If you fill something, it will be empty at some point and it doesn't just disappear. You have to get it back to where you fill it again.
> 5. For the vast majority of what automakers want, 7nm is overkill. Once there are more high end SoCs running multiple systems including self/assisted-driving, that will come into play. But that is not where the shortage is concentrated.
There was a discussion about this a few months ago. It's not just that newer nodes are overkill, it's that auto components apparently have very long and expensive validation requirements, which would be another explanation for why they are still depending on these older manufacturing processes.
Tesla purchasers, are, to some degree, buying into the idea of the brand. This works if your brand is nice and new and flashy, but I think that traditional automakers would probably suffer reputational damage if they tried this approach, and it is very hard for them to shake old reputations if they manage to get one of unreliability.
I'd say Tesla ignored that future debt, but I think it's more nuanced. I'd hazard they made a wilfull decision that the relative simplicity of electric cars will allow them to avoid a maintenance nightmare down the road by simply replacing whole components (a la Apple).
We'll see how it turns out.
Also, if you're on a 7nm process, odds are there's not a decent way to e.g. drive the transistors for your power window actuator. Samsung's 7nm process has an 3 sigma intrinsic breakdown voltage around 4.8V, which precludes any kind of 5V tolerance and driving 2.5Vt power FETs.
Little 3V3, 5V tolerant microcontrollers are good for a lot of things, and they belong on bigger process. While 10nm might be able to do it, there's not really transistors for it in most libraries.
Most of the 8inch wafer supply (mostly >90nm) was no where near capacity before the pandemic and supply restrictions. China did buy up a lot of it (actually moving equipment) and perhaps those were closed down?
Much of the legacy 12inch supply was fairly constrained even before the pandemic (especially specialty 55 and 40nm) due to shifts in display driver needs for OLED. The relatively new 10 & 7nm were pretty well dedicated to known customers.
Much of Automotive is in the larger 180nm to 90nm range with some 28nm for SOC due to cost, reliability, and legacy test qualifications. Except for entertainment only Tesla really uses much smaller than 22nm. It's interesting that they seem to have been willing to use commodity processors much more, and that seems to have made their supply chain more resilient (at the prices they were willing to pay).
It's interesting because a lot of mixed-signal analog really doesn't shrink well 28nm, and requires a lot of redesign to be cheaper than 180nm. There are ways to get down to 14nm and below, but the design costs require huge economies of scale to pay off.
The last couple of years in Semi really has been a game of how much are you willing to pay for the old backlog of parts that were hard to sell earlier. Only now are some new designs really making it out a year later. It's supply chain, but it's also price gouging and just taking profits while you can (why build more when margins will only plummet?). I do worry about the risks of concentration of this production in China.
To give you an idea of what we are talking about, here’s Infineon’s product page. They are a big automaker supplier, and right in the middle of all this. https://www.infineon.com/cms/en/product/
Source: Spent many weeks fighting supply chain issues and leading redesigns in 2018-2019 due to the beginning of the chip shortage.
I know the shortage is associated with the pandemic because that's when most people started hearing about it, but it was a growing problem even before COVID came along.
Volumes were 1-10 million annually.
Demand in first half 2018 was super hot, and depleted inventories. Wholesalers and retailers built up inventories of finished goods in anticipation of The Greatest Christmas Ever. Manufacturers did not, and kept inventories thin.
But demand fell off a cliff in the back half of 2018, and the The Greatest Christmas Ever was a dud. 2019 saw canceled orders and capacity shutting down quickly in the first half of the year. By the holiday season 2019, I was driving around the ports of LA and Long Beach seeing containers stacked high in warehouse lots, and overflowing car lots, some with 2018 model year cars.
Anyway, 1-10m puts you in a weird pocket when things were moving so fast like that. In 2018, you are too big to fill the order, because bigger players are sucking all the air out of the room. In 2019, you are too small for them to turn capacity back on for.
I’ve been following this stuff since the 1980s, and 2018-now is the weirdest semi market ever, with two sharp upcycles and a sharp downcycle sandwiched in between, all in the space of 4 years.
I think it's worth emphasizing this - all of the other issues you describe are correct, but what's allowing individual disruptions to turn into a global crisis is a lack of buffering between components of the individual systems. This is a known issue with JIT, and in some ways it's factored into the models - assuming one bad year in ten, 9 very good years can still be more profitable than 10 pretty good years - but it widens the possible set of outcomes and allows for these kinds of cascading failures, which makes it harder for one institution to model or plan around.
https://techwireasia.com/2021/12/malaysian-floods-devastate-...
I think there are also new sources of chip demands in other industries, such as perhaps companies utilizing ML having their own chip designs. But I'm not sure how much that plays into the issue.
2. If you look at that chart, you will see that the decade previous to the pandemic was the worst ever for durable goods demand. Supply chains had adjusted to that.
3. Supply chains had also all gone to a JIT model to keep inventories lower. This was a huge source of efficiencies, but made them vulnerable to a demand shock.
4. There are new sources of chip demand in EVs.
5. There are new sources of chip demand for very cheap ARM and RISC-V chips. High volume, low margin stuff that has been underinvested for manufacturing, like the entire auto chip chain.
6. 2019 was a down cycle, and companies were idling capacity.
7. When the pandemic hit, companies projected lower demand, and idled more capacity. This put them in a huge hole from which they still have not extricated themselves.
8. COVID outbreaks in Asian factories complicate things
9. Transportation bottlenecks complicate things