The great semiconductor drought may be about to break
theregister.com
theregister.com
Not sure that means much. That 2% improvement is down near the noise threshold. There are a large number of new fabs under construction, but few come onstream before next year. We'll probably have a semiconductor glut by 2024, if China doesn't attack Taiwan.
The current exceptional shortages are mostly in legacy nodes, whereas newly constructed fabs will mainly be producing on cutting-edge nodes. Building new fabs for legacy nodes simply isn't economically viable.
In some cases one can do it[1], but I imagine it requires a fair bit of money to port the mask sets and qualify the new parts. A specific process at a foundry might be sufficiently unique that there's few if any alternative foundries, I believe this was the case for AKM's factory which burned down[2].
If an old part still sells it's often high-volume "jelly bean" stuff, in which case it makes economic sense to let the old machines run as long as they can as they're written off entirely, but not to invest a lot in upgrades. If it's low-volume niche stuff, the price per unit is high, but that means any upgrade cost would drive up prices a lot due to low volume. Thus any customers still hanging on to some old design requiring that part might consider such a price hike a good time to redesign their circuit instead.
[1]: https://www.latticesemi.com/-/media/LatticeSemi/Documents/Pr...
[2]: https://www.diyaudio.com/community/threads/akm-factory-fire-...
Heck, some chips even cost way less. For example, Diodes Incorporated's 74LVC1G06FZ4-7 currently sells for $0.016 in quantities of 10.000. It's an absolutely trivial component logic-wise, so there is nothing to gain by shrinking the die. In fact, due to the current capabilities and ESD tolerance shrinking it would make it worse. Chips like that are never going to move to a different node - they will be produced until the fab breaks down or becomes unprofitable, at which point the chip is discontinued.
The real problem though is that 300mm wafer size: A 40nm chip that was made for 200mm wafers still needs to be redesigned for 300mm wafers. It seems like that wouldn't be necessary--you'd think it would just mean you could cram more chips into a bigger space--but it's not so simple. It's not as difficult as designing for a completely new process (e.g. 5nm) but it's still a non-trivial change.
Meanwhile, TSMC has publicly stated that they are not going to build more 40nm fabs and is actively trying to move customers to 28nm or newer[0]. There isn't any profit in building 40nm fabs anymore, let alone even more legacy ones.
[0]: https://www.extremetech.com/computing/337567-tsmc-tells-its-...
CHIPs act and other government initiatives changes the profitability calculation since the government is paying a large part capital costs
That means things like having a de-facto embassy but not calling it an embassy. Having a one China policy, and doing a gigantic amount of trade with China, while not agreeing to Beijing's claims of sovereignty over Taiwan. Going 35 years without a phone call between leaders - but selling them tanks, drones and jet fighters. Having a "robust unofficial relationship" whatever that means.
People think the US might not defend Taiwan because the foreign policy of the US is that they might not defend Taiwan.
[1] https://en.wikipedia.org/wiki/Policy_of_deliberate_ambiguity...
So if you make ultimatums, be sure you can back them.
How many wars, how many deaths, how much suffering before we see them for crooks they are.
“PRACTICAL POLITICS”—AND THE DEATH OF A WORLD
“Let me give you a lesson in practical politics.” Senator Burt looked at his wristwatch, leaned back and smiled. “It is a mistake,” he said, “to suppose that the public wants the environment protected or their lives saved and that they will be grateful to any idealist who will fight for such ends. What the public wants is their own individual comfort.
“Now then, young man, don’t ask me to stop the Pumping. The economy and comfort of the entire planet depend on it. Tell me, instead, how to keep the Pumping from exploding the Sun.”
Lamont said, “There is no way, Senator. We are dealing with something here that is so basic, we can’t play with it. We must stop it.”
“Ah, and you can suggest only that we go back to matters as they were before Pumping.”
“We must.”
“In that case, you will need hard and fast proof that you are right.”
“The best proof,” Lamont said stiffly, “is to have the Sun explode.”
Building a fab in close proximity of russia or other unstable areas is probably not a priority right now.
Becoming a leader in high tech manufacturing means giving up other leading positions.
But it's also become painfully clear over the past few years that this tendency is deeply harmful not only to society as a whole, but also to that very industry when any kind of shock to the system occurs, because all redundancy and resiliency to shocks has been optimized away as "wasteful". This includes having more than 1 employee in any given role for far too many companies (see "lean staffing"), which means that if that 1 employee gets COVID, or gets hit by a bus, or quits, or (horror of horrors) even goes on vacation, suddenly there's no one who can do that work at all, and potentially no one who even knows how it's been done.
We desperately need to realign our industries around a) what supports human beings and their needs, b) what ensures continuity of business, and c) what ensures redundancy in the entire system, including the supply chain, so that we can avoid these kinds of utterly foreseeable difficulties. But proposing anything other than what makes the already-wealthy shareholders' currency-denominated high scores go up as much as inhumanly possible is, if it is considered at all, seen as effectively political suicide for anyone not already as openly left-wing as, say, Bernie Sanders.
If we, the people, want this kind of change, we need to signal it by saying so openly, and by voting out the politicians at all levels and in all countries who prioritize business profits over humanity and common sense.
But aside from the very small number actively benefiting from this shift, too few people are willing to admit—or even understand—that it has taken place. They're just too invested in the Red vs Blue football game to be willing to do anything but blindly support "Their Side".
https://www.prnewswire.com/news-releases/globalfoundries-pla...
https://blocksandfiles.com/2022/10/05/fabulous-new-york-stat...
TI just bought $1.5 billion worth of old fab capacity from Micron, too. In other TI news, they're ramping up production in a new 300mm wafer fab now, with the first wafers coming just last month. They're also building four new fabs at Richardson, Texas. https://news.ti.com/blog/2022/09/29/tis-new-300-millimeter-w...
Samsung is looking to have a new fab near Austin by late 2023, at about $17 billion.
https://www.anandtech.com/show/16483/samsung-in-the-usa-a-17...
The US government passed the CHIPS Act that invests government money heavily in securing new domestic semiconductor investment. The state of New York and Onondaga County are investing heavily, too. The county is putting $10 million into Syracuse University and $5 million into its community college for research and education around semiconductors.
I don't think anyone would put a new industrial facility in Kings County.
This was literally the theory that guided European (and particularly German) relations with Russia for the past couple decades. It was disastrously wrong.
Russia knew that the consequence of launching a war would be the evaporation of trade with Europe, and they still did it. They won't attack NATO because it's NATO, with overwhelming nuclear and conventional military forces, it has nothing to do with trade.
It's a fallacy to focus on the one failure of the policy and ignore the other cases when it was working.
As for NATO its expansion all the way to the borders of Russia didn't prevent the war either, and it remains to be seen whether we won't end up with a war between Russia and NATO.
In times of war it's tempting but dangerous to feel like militarisation of the world is the solution. History doesn't show that to be the case in general.
The mistake is on believing that the effect was symmetric. Russia benefited very little from the trade (yes, for internal reasons), so they made no effort to preserve it.
Yes, one of the main goals of the world bank and western global economic policy after world war 2 is to create an interconnected world where countries like Germany are dependant on Russia in the hopes that this forces peace talks by making war more costly. The idea is that they are also dependent on us and we can sanction countries like Russia or Iran or North Korea into working with us instead of fighting us.
The issue with this policy is that we need to make the first move in good faith but if countries don't return honestly then our sanctions remove luxury goods like McDonalds while their counter sanctions remove necessities like heating. The second part of the issue is that once sanctions have been applied there no second step and the sanctioned country will just ally itself with other sanctioned countries as a matter of necessity thus pushing them further away from us and closer to all out war.
To protect our selves we need to be self sufficient in these matters which makes the whole thing not work as we will have no leverage other than war.
This is, of course, a gross generalization with lots of exceptions. Especially in the realm of "defense" technologies. But given this philosophical trust in markets to determine the whos, whats, whens, wheres and hows for industries, they tend to intervene late or weakly even when the public clamors for action.
Also... this philosophy has created the wealthiest societies in the history of humanity. And while past performance is not necessarily indicative of future performance... it's certainly something.
Yeah, and Intel announced similar for Ohio...and how is that going. A lot different to announce an eventual fab than to actually invest that money next year (which is the TSMC portion of the story).
Investing now makes sense - not for the next quarter or year - but for the next decade. Is anyone really going to bet the chip market won't be larger in 10 years?
Yes, it's interesting how we're so short sighted in the west compared to some parts of Asia and Europe.
For example: my friend works for a Japanese company and their roadmap and outlook for success of a new product is 10 years. I was listening to a podcast and they were interviewing a guy from luxury conglomerate LVMH and they think about how today's products will affect their brand in 100 years. That's an extreme example of course, but it's still an interesting perspective to consider.
I suspect many companies have bought insurance and it is about to bite them in the ass with excess inventory.
STM32 also seems to be counterfeited more than other lines. Kind of glad I don't have to sift through that right now.
Some sectors,like HDD and memory, rely on the annual floods (hehe) to keep stuff from getting cheaper, but not all of them.
I have the money but also tons of patience. No, 2000+ euro for a video card is not a good deal. Even half that is too much. Sadly, I know a good amount of people who fall for the hype and actually do buy overpriced stuff because it's cheaper than a few months ago
I worry the trend mentioned in the article doesn't reflect better supply, but lower demand due to incoming recession.
I am glad you are happy with your purchase and yet I wouldn't pay that for a new 3060, even if I wanted one
I'd say 250 euro would be fair for a brand new 3060 but it costs almost twice as much, according to my brief search on amazon.de
Again, glad you like your card but prices suck across the board and it's not just inflation. Again, I am glad you like your purchase but I say more patience is likely to pay off. What are they going to do, halt all production?
It's going to be interesting to see the report for October to see if the trend is continued.
"You get a Raspberry Pi, and you get a Raspberry Pi, and you get a Raspberry Pi, ..."
I am tracking the strike price on eBay, with greater supply the price should edge down towards the market price, looking to see at what point the price will drop down.
Although price is one value, it would be handy to have an idea of the depth of the market at that price, it isn't as though there are thousands listed.
The real shortage is energy, and that is why the US HAS the strangle the dollar liquidity.
FIAT money is only worth something if it's backed by scarcity and violence.
It's downhill from here for eternity, I suspect you wont be able to buy anything off the shelf in a couple of months/years.
Which is why you should buy last generations quality hardware now at peak architecture/lithography (14nm Atom, 14nm Xeon 4-core (larger has memory contention and is too power hungry), 28nm Raspberry 4).
This generation and the next is/will be less open/repairable and have higher max TDP which is a problem with unreliable power.
Sucks locking up that much money in inventory AND contributing to a toilet-paper run, but for him it beats unemployment, and for me, I can buy and build his kits which is better than not being able to...
In a greater economic policy way, its an inflation-reducer. He has a huge pile of money locked up in his closet in the form of unprogrammed Arduino chips, instead of hyperinflating stuff he could have bought like a better spectrum analyzer or real estate or whatever else.
It’s come in handier than I expected for the MCU shortages, but it’s also been really nice to not have to even look at stock/lead times when considering/prototyping new products.
I wouldn't be surprised if keeping 180 days of high-running, common parts on hand becomes the norm for high-margin products.
Anyway for cars part of the problem originated when with covid they cancelled all orders and the assembly lines for their stuff got converted.
Then their managers layoff workers and periodically go on the news, taking great care of never mentioning that it was partly their own fault.
That meant that all buffers we're removed and any interruption caused cascading amplifying consequences.
You get something like a pandemic and the consequences mean that interdependent feedback systems get broken.
You read all these 1990s TPS books that are like "we switched to this system and there's no costs! Everything was better"
This is the fundamental risk tradeoff coming back. It's always there, as reliable as a natural force.
The classic works but like Womack always had that red flag. It's like 300 pages of skimpy on exacting procedural details and zero talk of downside. Everything works as long as 100% of your arbitrary selection of infinite supplies can magically appear on the loading dock perfectly on time (well usually it actually can, except when it can't)
Fun fact: Toyota was among the least affected manufacturers, so much for TPS doesn't work.
Second fun fact: When it comes to supply chains everybody not working in SCM seems to know better when somethung goes wrong ehile ignoring all the times stuff just "magically" shows up. This is true for all backbone domains so, from utilities over supply chains all the way to waste removal and the medical profession. The latter less so due to all those drama series.
Well, just like agile, most companies aren't actually implementing TPS but something that looks similar on the surface.
"A Toyota spokesman said one of the goals of its lean inventories strategy was to become sensitive to inefficiencies and risks in supply chains, identify the most potentially damaging bottlenecks and figure out how to avoid them.
“The BCP for us was a classic lean solution,” he said."
https://www.reuters.com/article/us-japan-fukushima-anniversa...
It's much more about seeing problems faster than just cutting down on inventory for the sake of cutting down on inventory
On a national level, I think we are seeing policy makers (re)reaching your conclusion, however.