This is horrible. What keeps me up at night is thinking about TSMC. One company. Just needs an earth quake, fire, explosion, something and the whole world is crippled. I know we have samsung, intel, etc other fabs. But if TSMC ever went down it would make the thailand floods for hard drives look like a joke.
Taiwan is currently having a drought. Water supply to TSMC has been limited. TSMC scrambling to secure enough water. Fabs are chemical plants that require lots of water.
It might even be a good investment to build a fab just to clean up in the event of a shortage. Kind of like how if anyone in Texas had the good foresight to winterize they could have made a year of profit in a few days...
Having worked in the semiconductor industry, it would be good investment for Apple, Google, or Microsoft to get into fabbing stuff for themselves and others just on the basis of being able to outcompete everyone else in operational efficiency. There's a killing to be made from low-hanging fruit all over the place, at least for anyone who's able to front costs of opening or acquiring a billion dollar fab.
> it would be good investment for Apple, Google, or Microsoft to get into fabbing stuff for themselves and others just on the basis of being able to outcompete everyone else in operational efficiency.
They can try...
Surely they have cash, and so did the Chinese who were signing 10 digit cheques for anybody claiming to be able to setup a modern fab.
Running a fab is not like running a cat video website, or making a cookie cutter electronics from reference designs.
On other hand, if there is a company which can turn off the lights on a dotcom companies like them, it will be a big fab business.
TSMC's Morris Chang's integrity, and softness, completely uncharacteristic to the big ambition, ego, and testosterone-driver semi industry, were what made TSMC stood out. Dirty play in the semi industry was everywhere in nineties.
People who are taking over TSMC from Chang may not share his adherence to morals bordering on altruism.
P.S. Saying what that industry was is by no means a meme. Early silicon valley businesses would've given a heart attack to any modern politically correct business exec. Stuff like this was rather common:
Just read some memories of Fairchild Semi "alums." Stories of men from semiconductor companies stealing each other's wives in the most literal sense, gallons of expensive liquors, contraband Cuban cigars, break-ins into labs, and fisticuffs at business meetings were the world of the valley back then.
> Running a fab is not like running a cat video website, or making a cookie cutter electronics from reference designs.
You're squaring up against a stickman. The companies cited are not fly-by-night startups "running a cat video website". Dave Cutler got his start on the factory floor, working out better processes for a toilet paper manufacturer. Apple has plenty of hardware chops and the vertical integration bug. Google shook up telecoms by scaring the shit out of the industry incumbents (although they could have stood to get their hands dirtier, Cutler-style[1], and stay in the game).
Although none of this is necessary. The fruit hangs so low here that all it would take is someone half competent who can see the enormous gains to be had by just not doing so many things in the dumbest way possible. This isn't bleeding edge stuff. We're talking catching up on things that that wouldn't even be considered state-of-the-art 25 years ago. Hell, they could start just by drastically cutting back on how many hires they source from the defense industry and have had their minds warped by (and numb to) the bureaucracy and bottomless budgets from that world.
> if there is a company which can turn off the lights on a dotcom companies like them, it will be a big fab business
Not a chance. The semiconductor industry, perversely, has no idea how to put computers to use, and the leadership and organizational structures are aligned against that kind of achievement ever coming from within (or being airdropped in by a leadership hire who cuts across the lanes from another industry) and making the top-down sweeping changes that would be required.
> Not a chance. The semiconductor industry, perversely, has no idea how to put computers to use, and the leadership and organizational structures are aligned against that kind of achievement ever coming from within (or being airdropped in by a leadership hire who cuts across the lanes from another industry) and making the necessary top-down sweeping changes).
I've seen attitudes like this in the industry many times, and every time such "bright idea man" comes, a previously well faring company dies.
Take a look at Sun Micro, people now will never believe that they once had ambitions to "corner" the server/internet market the very same way those bright "financial strategist" boys suggesting chip companies to "go undercut this, and that, and we will own the Internet!"
You're ignoring everything I'm saying. (Probably because you don't want to consider the idea that emperor really isn't wearing any spiffy new clothes.) We're not talking about bright ideas. We're talking about dumb, dead simple ideas. Dumber than UPS's fabled "let's try to minimize how much time we waste on left turns".
Go watch that Dave Cutler clip and pay attention to what he says near the end of the video about being a leader and about there being no task too small when it comes to getting the job done.
"let's try to minimize how much time we waste on left turns"
A tangent, but I would really like Google Maps, or some viable substitute to permit "only right turns" as an option.
There are routes where every time it wants me to make a left onto a busy highway from a side street with no traffic light. Knowing where the lights are would be one solution, but "only right turns" wouldn't depend on data.
> Just read some memories of Fairchild Semi "alums." Stories of men from semiconductor companies stealing each other's wives in the most literal sense, gallons of expensive liquors, contraband Cuban cigars, break-ins into labs, and fisticuffs at business meetings were the world of the valley back then
Err..
I don't think you've ever been out with sales execs from SV companies. "gallons of expensive liquors" sounds like an average Friday night.
> contraband Cuban cigars
Of course no one in SV these days does any contraband drugs./s
None of those companies have any expertise in running a semiconductor fab, though. There probably is low-hanging fruit, but you have to be really familiar with the process to know where it might be.
When I worked in the semiconductor industry, it turned out that the same set of people often worked for multiple different vendors as contractors and consultants, setting up lines and troubleshooting process problems, etc. So presumably you need to tap into those people.
But TSMC's Arizona plant will make 5nm chips in 2023, by which time 5nm wont be the leading-edge node - 3nm will be. It's nice to see some geodiversity however.
In any case, TSMC is dependent on ASML, which has a monopoly on leading-edge EUV tech.
There is a decent amount of competition at older nodes, like > 40nm. So we'll still be able to make some chips if TSMC completely goes out of business, but obviously it wont scale to the demand and will seriously hamper business, science, and government.
I'm comparing against GP's suggestion "to build a fab". Point is, SMC and TSMC fabs will be online faster than a direct fab investment.
> science
I believe science has access to more compute than it needs. I could be wrong now, but last I checked, a few years ago, the huge computational clusters (both public and private, e.g. @ pharma companies) were under-utilized, especially on legacy-hardware-that-is-still-pretty-damn-good, and scrambling to find ways to engage with outside clients to fill out capacity.
There will still be plenty of demand for 5nm chips in 2023. There will still be plenty of demand for 5nm chips in 2030 for that matter. Not everything uses the latest process node.
You make it sound like anyone with the will can just take ASML product and create a leading fab. If that was the case Intel wouldn't be in the mess it is in.
Some military planners thinks the Chinese will not invade but will enforce a maritime blockade around the island to force the issue.
Ships would be inspected and food and medicine will be allowed through but all trade goods will be blocked.
But that is all depended on the US and it's allies having a stomach for a naval confrontation or that the Chinese mainland umbrella of long range missiles and air defense can be an effective deterrent against that.
The first US-flagged vessel stopped and searched would trigger WW3. I think there's little question about popular sentiment in the US. Backing down would be political suicide for any president.
Does the chinese navy have experience with maritime blockades? Besides, for chips, worst case scenario is load those chips onto ruslans and offboard them in guam or okinawa. A minor marginal expense. Will china shoot down taiwanese planes?
No idea but it would be preferable optics wise to stage as an internal Chinese policing action against a rebellious province rather than a full on military invasion of another country and all that would entail.
Makes you wonder if those buildings would be rigged already. Conspiracy level stuff, for sure but it is such a strategic asset. In WWII bridges that were still in daily use in Europe were already mined well before the territory changed control with the leavers trying to deny the new arrivals the use of the bridges.
In the modern era, it probably is just gathering precise positioning for a missile launch. I expect some computer wiping will be in the cards too. That might be the modern version of rigging the buildings already.
Given what a strategic asset they are, I would expect its not really "conspiracy level stuff" on various plans. After all, the US had attack plans for the UK between the world wars.
I wouldn’t be surprised if TSM has self-destruct features built into their operations as a deterrent, or at least a plan to let the USAF bomb the hell out of the foundries if China invaded. The former might be a bit too risky :)
*and South Korea. Maybe not an explicit attack, but if there is supply chain disruption due to tensions between china and japan not having contested sealanes (getting out of SK could be at risk) in your chip pipeline is probably a good thing.
As far as I know Samsung in Texas is not yet a done deal. They're requesting complete tax abatement for twenty years in addition to the inducements offered by local governments and I believe they're waiting to see if they can pull it off. Their current facilities in Austin have remained closed since the big snow storm.
> Kind of like how if anyone in Texas had the good foresight to winterize they could have made a year of profit in a few days.
This fundamentally mis-states the situation in Texas during the recent cold catastrophe. Gas suppliers that had winterized had their electricity cut off and could not pressurize their distribution. Generators who had winterized had their gas supply curtailed. And so on.
It was a systemic failure by the PUCT and the RRC, who are the regulators. Ultimately, the Governor who appoints those boards and the Legislature who grants their authorities failed.
It wasn't some yahoo generator too dumb to winterize, although there were plenty who didn't find it financially convenient to do so.
> Kind of like how if anyone in Texas had the good foresight to winterize they could have made a year of profit in a few days...
Hmm. When would this foresighted person have made their preparations? What would installation have cost? What would maintenance have cost? More or less than a year's worth of profits?
> What would installation have cost? What would maintenance have cost? More or less than a year's worth of profits?
Power that's selling right now for $25/MWh was selling at the $9,000/MWh cap. A 36,000% increase in price can surely pay for enough to, say, keep your pressure sensing equipment from freezing over. (Which is why the 1,280MW STNP Unit 1 went offline at a $11.5MM/hour+expenses cost.)
Not to mention the lives and property damage that might have been spared.
> When would this foresighted person have made their preparations?
As far as _when_, maybe after the 2011 TX cold snap that took generation offline for similar reasons. Or the one in 1989.
This kind of analysis assumes the price would go that high if people had winterized their equipment. The price only went that high because the grid had basically already failed. It's extremely unlikely it would go that high if the grid was functioning. So the incentive to get that payout only exists if the grid falls over anyway. Basically this either allows smaller actors (who can't prevent the crises) to take a bet with long odds and a large payout (already not a winner as far as infrastructure investments go), or it basically removes its own incentive if enough of the larger actors take steps to avoid it.
Basically I think this idea of ultra-high pricing during a crisis providing an incentive to prepare for one (as opposed to paying for the capacity to that the crisis doesn't occur in the first place), is not necessarily well-formed: it's feels like it's assuming the market remains efficient in one half while it also requires that the market has effectively failed due to the crisis we're trying to structure the market to prevent.
A few days in advance, and no installation or maintenance costs. It costs about $4000 to rent a 1000kW diesel generator for a week. It consumes about 70gal of diesel per hour, and it costs $2.90 ish for diesel in Texas. That comes to $4000 + $203/hr running costs.
At one point, Texas electricity prices hit above $9000/MWh. Running the generator for 1 hour would net you $9000. Not bad for peak profit. The average profit will be a bit lower; here's some very rough back of the envelope math:
> On a typical February day, she pays Griddy less than $2.50 for power. But the one-day cost spiked to hundreds of dollars after the storm. In all, she was automatically charged $1,346.17 for the first two weeks of February
So $2.50 a day over 2 weeks is $35. The spike to $1346 is 38x her normal cost, and Griddy uses wholesale prices. I see wholesale costs around 6.7cents/kWh, so at 38x normal pricing, that means electricity would cost around $2546/mWh. Not as high as $9000, but still a nice profit averaging for 2 weeks. Profit per hour would be $2343/hour; multiply that by 24/day and 14/two weeks, gets you $787,248 total. Subtract out $4000 for the rental, and you get $783,248.
Close to a million dollars over 2 weeks is not a bad haul. Even if you divide the price by 2, it's still pretty good.
Nothing would really happen. We’d keep using 2 year old processors from Intel, and pay more for them. The GPUs, whose price is high because of cryptocurrency, will not be missed. These auto manufacturers could pay a little more for a general purpose CPU instead of ASICs.
Auto manufacturers can't just put in a general purpose CPU. Embedded CPUs have functionality general purpose CPUs don't have. Not to mention it would take months to design new hardware and write new software.
The chips being put into cars today were selected at least 5 years ago.
As someone having worked in the automative industry, I can confirm the latter part. It takes about 3-5 years from the start of a new design over going through all qualifications up to shipping a car with it. Even switching to a mostly compatible chip (same ISA, same pinout, etc) will take more than 1 year.
The stock market would crash guaranteed if Apple had no inventory of phones. AMD no gpus or cpus, etc. It wouldn't just be like oh sorry out of stock. It would touch every aspect of your life.
You can sleep a _little_ better: the narrative around this ended up being focused around TSMC, but it's simply because that's the most visible part of the chain, they're first to a new node and all the newest gadgets need the newest node, so all the newest gadgets are on TSMC. They're not solely responsible for like 90% of chips or anything.
TSMC might not be responsible for 90% of the chips, but estimates are they account for 28% of the mature market [1]. They don't only do new nodes, but older nodes as well. Between them and UMC (also Taiwan based), those 2 companies account for 41% of all modern IC production. Micron and Powerchip are also big fabs out in Taiwan and if someone were to say that 50-60% of all 45nm+ ICs were made in Taiwan I would not be surprised.
If there was a geopolitical struggle and the fabs in Taiwan were impacted for months or years, we would see the effects of that for years or even decades. Its why China, US and EU are all trying to build some more resiliency in the supply chain. Its also why Taiwan has resisted allowing tech/expertise to leave, since the amount of leverage they have lets them punch way above their weight and makes them such a strategic partner to the rest of the world.
TSMC is also a bit more geographically diverse than a lot of people assume. Their largest fabs are split between three different cities in Taiwan, so a fire isn't going to knock them all offline. And their smaller or less-advanced fabs include several more locations. So it's not quite like the semiconductor industry has put all their eggs in one basket, even for leading-edge logic nodes.
As someone with a new phone, managed to find an nvidia card at msrp, and who likes computers made by team red, no sorry, I'm still not going to sleep good at night. You tell the population they cant get new iPhones, you watch the anarchy. Not just plebs, but the stock market react to Apple saying no new iPhones, actually no iPhones in stock at all. It would be devastating not from a "I can't have the shiny new gadget waaa".
Apple can quell the stock market crash, depression, and ensuing national mental health crisis by going back to what they were doing, industry standard practice, dual sourcing from Samsung fabs.
AMD/Team Red...well, same thing. They're not due to be on the TSMC node you're worried about until 2022 at the _very_ earliest
Not to mention the PLA and the turning of mainland back to Maoist policies.
Taiwan's critical role is the tech sector is IMO a deterrent bigger than Nuclear weapons at this point, and its invasion may well be a casus belli in the coming years.
We really need to localize manufacturing of fabs. Even if the tech is making something closer to more primitive forms of computing from 30+ years ago. A lot can be done on those chips with well written software.
Unfortunately just trying to manufacture a transistor is extremely difficult and dangerous.
I wonder if there's something more accessible that could help localize circuit logic as a primitive building block of computation - one that isn't worried about being the fastest thing but the most accessible. If not targeted per individual then per a big neighborhood/small town.
As someone who has had to endure over a year of very close to the metal (in this case, imagine centuries old, gnarly, rusty random pot metal, maybe) development on a Renesas platform, I'm tempted to think "fewer Renesas products? Awesome!!" but obviously that would be childish.
I'm sorry for all the people hurt by this, fire is no joke.
I used Renesas with GHS debugger. It required _constant_ access to license server. If I used VPN to access another virtual machine the IDE would throw dialog on top of other windows with a countdown. Using GDB CLI after that felt like living in the future!
We as a society are going to have to reckon with how fragile our silicon supply chain is eventually... It seems like every other year we hear about a fab fire here, or a fab flooding there, and then seeing prices rise by 200% for months.
As I read elsewhere (don't have the link) this is supposed to be the #2 chip supplier for the automotive industry, and the #1 and #3 closed down because of the Texas Power outage and are still not operational.
> because the facilities were completely shut down, clean rooms were flooded with particles and every piece of equipment was contaminated
Ouch. That suggests that they didn't have backup power for the positive pressure ventilation system. I hope adding that is on someone's todo list for next time.
Even if they had backup power for those kinds of systems, the cold in Texas took down utility-grade generation capacity. It wouldn't be a surprise if they had their backups die, too.
I'm not super familiar with this space and am a bit confused - why is this big on automakers and not other chip consumers? Also 300mm seems kind of big for chips? What am I missing?
300 mm is the wafer size. The wafers then get cut into chips. The whole production line has to be made so that it can process the wafer size. For a long time 200 mm wafers was the industry standard. That this is a 300 mm wafer fab implies that it is a high volume, fairly new fab that burned down.
>Automakers were already struggling with chip supply
TBH, that is their own fault.
When the pandemic started a year ago, they chickened out thinking people won't have money to buy anymore cars and halted production, cancelling orders for semiconductors. Then, when the demand for cars exploded, since nobody wanted to be on a subway/bus full of people carrying a contagious virus, they wanted their chips to resume production, but the semiconductor fabs have already allocated the wafer capacity the automakers have just freed up to other industries that were booming.
The automakers basically sold both their kidneys to the pawnshop for a quick buck and are now trying to buy them back during a kidney shortage.
I'm not sure we should be so quick to fault them. The unemployment rate when the pandemic really started building up steam was absolutely staggering so a lot of people quite reasonably expected a severe recession along with severe contraction of consumer demand to follow. That we didn't experience one is a minor miracle (and printing of a ridiculous amount of money).
The high unemployment rate during the pandemic was(and still is) mostly formed of people working in the service industry, travel, minimum wage earners and people with little to no skills and those people were having a rough time even before the pandemic so they were never on the automakers' radar of customers would buy a new car, not in a million years. Those people can only afford to buy used.
On the other hand, the typical new car buyer, the middle classer and above, saw little unemployment, on the contrary, many have seen their best years yet, which is why the demand for new cars (among other things) exploded.
You absolutely should be quick to fault them... They cut off nearly all orders from suppliers anticipating an economic downturn and had no risk mitigation strategy if they were to resume production. In essence they left the line at a concert after hearing rumors the band wasn't going to show.
Now they know the concert is definitely on, they want their place in the line back and the manufacturers told them to pound sand, since there are tons more people paying more money than they are.
Automotive manufacturers think they have much more weight than they actually do when it comes to purchasing electronic components. They definitely came to a rude awakening when the chip fabs and manufacturers told them to go to the back of the line.
For reference, the entire automotive market in 2020 spent 38 billion on semiconductors. Apple alone spent 53 billion in 2020.
If you are going to prioritize something, who do you think is going to get special treatment?
Already has... I know some people who work IT for a large automotive facility in Ontario that have been laid off for 3-4 weeks already.
It really feels like nobody appreciated just how critical to all aspects of our lives semiconductors have become - tiny little chips have caused heavy industry to grind to a halt in some places.
Not Toyota, but I don't know how large their mitigation buffer was set to. Seems like they took the logistical risk analysis more seriously than other makers.
Car safety regulations in quite a few places will require you to fully recertify a car even if you changed a stereo in it.
Some place limitations on safety critical electronics like ABS specifically requiring any modification to it being treated as a new car model for legal purposes.
Such laws are predatory, but laws they are
But even if they did find cheaper substitutes, it's not the more advanced chips which are in the most acute shortage, but really old ones, which were kept in production for decades specifically because of car industry's requirements.
It was really difficult for me to buy a new car last year. Customized orders are totally impossible right now. Whatever they might have in stock (or have arriving soon), that's what you can choose from.
No, 300mm is correct and refers to physical size of whole silicon wafer[1]. 300mm (commonly known as 12") is currently the biggest class of silicon wafers.
Naka factory was producing 300mm wafers using 28nm process [2].
They use 200nm later in the article, but wikipedia[0] says that 300mm is a standard wafer size, and fabs are classified by the diameter of wafer, so you might be right. I assumed it was the feature size before, too.
