So why are graphics cards, gaming consoles etc out of stock? Why is Apple attributing fewer iPhones sold to semiconductor shortage?
Can automakers not use a 7nm fab?
How much of the chip shortage can be attributed to AMD and Apple taking the laptop market away from Intel? And if this is true, does this mean that Intel now has unused capacity?
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
1. There was real and relatively sudden increase in demand:
a. COVID-driven in part, all students and many additional employees had to buy computers to study/work from home,
b. a., driven by video, lead to increased demand on Internet & telecom infrastructure.
c. Electric vehicles consume an order of magnitude more chips than ICE cars. EV demand is also skyrocketing.
2. There were fabs (semiconductor factories) shut down due to COVID. This one in Malaysia [0] makes power chips and was particularly problematic for ICE auto makers.
3. The first indication of problems manifested in chip lead times pushing out. As buyers started to see more and more of this, they started increasing orders and soon adopted a hoarding mentality, similar to why many of us couldn't find toilet paper in stores early in the pandemic.
Result: As a personal anecdote, a chip that sells for $0.47 from that fab in Malaysia recently sold for $75 on the broker market.
Note that no, the state of the art fabs cannot be re-outfitted to make legacy chips. The old equipment they were developed on is generally no longer available and the profit from making those inexpensive chips would not justify the capital spend on new equipment.
[0] https://www.reuters.com/article/infineon-results-malaysia/in...
https://twitter.com/wired/status/1335445557788233728
https://twitter.com/BitcoinMagazine/status/14149943888208977...
https://twitter.com/dmgblockchain/status/1438547780646019086
AFAIK, the cost to add significant additional manufacturing capability for most products means that increased demand has to be forecast for a really long time to be worthwhile. Hence why it was still hard to find quality nitrile gloves, isopropyl alcohol, and Clorox wipes during 2020 even though everyone knew there was a bigger market for them for potentially quite awhile. I'm sure the chip and hardware manufacturers don't want to invest millions or billions in assembly line machines and then have buying levels drop back to pre-cryptocurrency-mania levels.
So whether automakers could use a 7nm fab is a theoretical question -- there is but one 7nm fab, the one at TSMC and they already sold every wafer they can start. They would be able to sell more if they could start more but there are only so many EUV machines ASML can make so there are only so many wafers TSMC can make. All the gaming consoles? yeah, that's 7nm chips. With AMD and Apple moving to the next line, there'll be some capacity freeing up but the demand is still very, very strong.
Also, another effect I do not see here is the container disruption. Read https://www.vox.com/recode/22832884/shipping-containers-amaz...
https://www.prosoundnetwork.com/business/akm-factory-fire-sh...
There are other companies that make similar chips, but in much smaller quantities, and they are really in no better position than AKM to fill the void.
A recent "update" : https://evertiq.com/design/51031
This caused a bunch of chips that are old and cheap to be complete unattainable, due to a lack of safety stock across the industry. Cars, power supplies, and other "large" things are highly dependent on these older chips, and have very long redesign timelines.
Newer generation is actually simpler, its just demand growth outstripping supply. I'm sure 2020/2021 also impacted the ability to bring more capacity online but probably only to a small extent, these are long term investments that aren't responsive to instantaneous demand.
Intel offered car manufacturers 16nm chips "as many as they wanted", but they'd have to be re-engineered from, from what I read, 150nm and certified for harsher conditions (temperature, humidity, acceleration, probably noxious fumes if installed under the hood) than consumer electronics, which adds to the cost.
In addition to that, with the upcoming switch to electric cars they probably wouldn't produce enough of them to recoup the investment.
This massive jump in demand has caused backlogs. Covid supply chain issues didn't help.
Meanwhile the cryptocurrency boom provided a means to turn graphics cards into virtual money, so those all got bought out.
All while staffing, mobility (can't fly troubleshooters out so easily!) and other aspects of working have got harder, and a related physical shipping crisis is causing an increase in inventory both at factories and in transit.
Its just a bottlenecked queue, that then became closed off, which amplified the size of the queue so much faster.
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.
People who are more likely to be concerned with covid (more educated) are also more likely to consume electronics, so demand on electronics is particularly increased.
Supply side issues are secondary and by now they aren't a big deal at all.
You cannot take a, say, 150nm or 45nm design and just rescale it in Photoshop to make a 7nm design.
Not only does electrical characteristics change which might require changes, such as leakage currents, but in order to reach the densities of "7nm" things such as transistor construction has changed. For example older, larger nodes used planar transistors while a modern node likely uses non-planar transistors like FinFET[1]. A nice write-up can be found here[2].
The point is, moving from one node to another, especially if they're far between, might require a huge redesign, along with all the validation and testing that entails. A new set of masks for the latest nodes is also very expensive[3].
[1]: https://en.wikipedia.org/wiki/FinFET
[2]: https://semiengineering.com/knowledge_centers/manufacturing/...
[3]: https://anysilicon.com/semiconductor-wafer-mask-costs/ (not current but illustrative)
Consumer electronics including GPU and consoles demand have increased dramatically due to pandemics. And they use components that uses bigger nodes which are in demand.
>Can automakers not use a 7nm fab?
They could but even 7nm are in short supply.
>How much of the chip shortage can be attributed to AMD and Apple taking the laptop market away from Intel? And if this is true, does this mean that Intel now has unused capacity
Both AMD and Intel are at full capacity. Intel still have server capacity to fill.
Read the Bullwhip effect. [2].
All of that is excluding crypto, and investment / trading ( buying and hoarding ) in semi- components.
This caused a scramble where customers found alternate paths to satisfy requirements. Buying up stock of other, similar chips (which now creates a shortage of chip "b"). Reserving time with a new fab. Or deciding not to have navigation in certain car models. Or halting sales for a while. All of which drives another cycle of churn into the schedule.
This repeats itself over and over, like an earthquake with aftershocks, until things smooth out.
I think there are other reasons for the shortage, but they've been exacerbated by the inflation issue.
I guess I'm asking if this theory has any credibility outside of the various conspiracy theory blogs/sub-reddits?
Automobile manufacturers are used to ordering suppliers around wrt price, lead times, and so on. Chip makers used to put up with that, but when things got tight and other folks started waving dollar bills around, the chip makers said "hold it - we can get paid more AND not have to deal with those jerks?"
Price cant adjust: Long term contracts, regulation etc.
Supply cant adjust: Move to newer chips (so essentially misplaning), shortage of people, congested supply chains etc.
1) The pandemic resulted in a significant increase in demand by businesses buying laptops/PCs for employees so they could work from home. Lots of GPUs there and they are still buying.
2) Lots of people as individuals buying gaming PCs and consoles since they were spending much more time at home and wanted something to do. They too are still buying in record numbers.
3) With AI being a hot field right now, lots of GPUs being purchased for deep learning by researchers and businesses. Maybe this is a permanent increase in demand, maybe not... time will tell.
4) With cryptocurrencies being white hot right now, lots of GPUs being purchased for mining.
So you might say 'well why not just build more fabs/factories/whatever and they can all make lots of money?' Well sure, for now. Fabs cost billions of dollars and take a couple of years to come online. The companies that have to invest in them likely spent much of 2020 and 2021 doing everything they could to increase production in the short term while asking themselves 'how much of this demand is long term?' as all of the items listed above are relatively recent changes, rather than things proven to be structural changes. It sounds like by mid- to late-2021 companies like TSMC decided to take the plunge after securing some up front payments and volume commitments from major customers.
In addition to concerns by companies like AMD and nVidia about how long term items 1 and 2 are, they know for a fact that 4 isn't likely to be sustainable given that there was a major crypto crash in the last 5 years and likely to be another one soon. When this happens, this will not only decrease demand as crypto miners stop ordering, but it will likely also increase supply as miners dump their GPUs on the used market (Ebay etc.) to cut their losses. If you follow the GPU news, you'll notice that AMD and nVidia are being very strategic in releasing new mining-specific GPUs. This is partially to offer products that specifically target the needs of miners, but I suspect more to minimize how much of this product can be dumped on the used market hurting GPU demand from consumers down the road.
In addition to all this going on specific to their industry, they're dealing with all of the other manufacturing and logistical issues everyone else is. So they're also having to figure out how to get that PITA power/memory/whatever controller that they need manufactured in quantity and in time to build their own products and so on.
So these companies are responding: they probably did as much as it financially made sense for them to in the short term while also adjusting their longer term plans. I'm expecting them to come online with significant new supply (for the new products released in 2022) just in time to see it dry up soon after (at least in the short term) since that's typically how these things go.
Processors are a commodity and everything but there just aren't enough suppliers with good enough product at a low enough price. The process is toxic and complex and requires heavy investment.
It's almost a miracle we get to see the type of efficiency in development that we do today (although others may call this miracle a spiraling debt crisis..)
When he placed that trade embargo on China, many companies were forced to change their orders to factories in other locations, such as Taiwan. These factories were already running at capacity.
Edited for source: https://www.theverge.com/2020/9/26/21457350/us-tightens-trad...
- Long time push for "just in time" supply chains. A lot of manufacturers stopped keeping a local stock of chips they needed for the next 6 months of manufacturing to save on storage costs. This is fine as long as there are no hiccups in any earlier part of the supply chain (this transition has been happening over the last 15-20 years).
- Crypto-coin mining and the sudden extreme demand of graphics cards. This has been going on for a long time. A lot of low-nanometer demand goes to a large number of graphic cards and ASICs for mining crypto-currency. This led to the graphics card bottleneck that existed for the last 3 or so generations of graphics cards.
- pre-COVID there were several (I want to say Taiwanese) fab plants that were destroyed due to Typhoons. This led to reduced production capacity.
- There was contamination of shipments of silicon wafers which disrupted production in several fabs. This was a pretty short delay more than anything but it was a contributing factor.
- At least one US based Fab was closed. I believe some other international fabs also closed down either temporarily due to COVID or permanently for other reasons. Some fabs were acquired by major companies and likely switched their production pushing more pressure on other fabs.
- Global politics. There was a lot of aggressive changes to global policy during this time frame that forced a redistribution of where certain chips were being manufactured adding more work for all fabs (retooling, reworking designs, getting a design working in a new fab).
- The canceling of car chips actually relieved the pressure on the fabs as they were able to take other fab orders. I don't think this slowed down fabrication in general, but when the car manufacturers realized they actually did need that capacity and more, the capacity had already been sold to other customers. This is entirely on the car manufacturers and is only kind of related to the overall chip shortage.
- Global distribution issues. This isn't just in the final stage of consumer products that has been affected but also raw supplies affecting the rate some chips can be manufactured.
- Increased demand overall. Consumer consumption of chips has increased pretty dramatically over the last decade pretty consistently as everything wants to be connected to the internet now from your toaster, to your butt plug.
- Complexity and difficulty in building up a new fab. These are incredibly specialized facilities costing in the 9 figures to setup the facility. The people that work in these facilities are highly specialized not just in the profession but to the intricacies of that specific facility and their pipeline. Training up these extremely skilled labor tasks can't be done quickly.
It's important to understand that many manufacturing industries operate on a "JIT (just-in-time) manufacturing" model[0], where goods used in a manufacturing process are only ordered and received as needed, minimizing inventory and storage costs. (e.g., instead of having a warehouse full of screws sitting around waiting to go into products, you order X screws from another manufacturer every month to fulfill Y number of effective orders) This means that many industries (especially the automotive industry) operate on a constant stream of supplied products, in many interdependent chains.
There are two downsides to this approach:
1. The JIT model has to be at least partially predictive. You can't necessarily order components right as you receive orders, because it takes time to produce those components (remember, no one has warehouses full of those components ready to ship, because it's often JIT all the way down) and you don't want to keep customers waiting as components get shipped to you. Instead, you need to at least somewhat predict how many orders you expect to fulfill in order to balance between keeping customers waiting ("shortage") and having excess stock laying around which might never sell (waste)
2. The other main downside is that JIT production is highly interdependent on other industries and can be fragile. If you expect to be able to order X screws this month from your supplier, but some external event causes them to only be able to supply you with Y screws (Y << X), you're shit out of luck. You literally cannot produce your product because the underlying components simply do not exist, which leads to a shortage of your product
Both of these downsides come to a head in the presence of some global event that causes uncertainty and inefficiency. If a low-level JIT process depended upon by many industries suffers from an inefficiency (e.g. workers can't come in because they're sick), many higher-level processes start to slow down (e.g., no one has screws to build their products with). This effect is multiplicative across industries, since many manufacturers specialize in specific products and downtime can affect many "downstream" clients.
Along with this, such global events can dramatically change customer demands, and when you've only predicted needing to produce X products but you now have Y demand (Y >> X), customers are out of luck because the products literally do not exist to be sold (shortage).
Considering that computer chips are now present in almost all major consumer products these days, it doesn't take much uncertainty and inefficiency at the "base" for a domino effect to significantly slow down a lot of industries.
Most shortage todays is on 200mm.
Why? Trump's sanctions on SMIC brought down like half of world's 200mm capacity.
> Can automakers not use a 7nm fab?
No, process, and cell libraries are incompatible. Chips will have to be redesigned.
