Hitting 3 nanometers to cost chipmaker TSMC at least US$20 billion Taiwanese chip heavyweight TSMC has announced the location for its future 3 nm chip-making factory in late September and has now put a figure on how much that's going to cost: a cool $US20 billion. Outgoing chairman Morrish Chang (last week, he announced he will retire in June 2018) revealed the figure in a chat with Bloomberg …
10nm was/is about the limit for traditional DUV lithography. Everyone is switching to EUV at 7nm or soon after, and that will push things out down to 1nm or so. Beyond that it is probably a matter of escalating cost for the process rather than technological limits, since each generation seems to cost about 30% more than the last.
When chipmakers started getting down to the really small scales (sub 28nm), correspondence between scale and technology name desynchronized.
22, 20, 16, 14, 10, 7, 5, and 3nm are generation / technology names (keeping consistent with previous generation naming), but no longer connected to feature sizes like gate length, etc. In other words: they no longer mean what they used to.
When we went from 28nm to 22nm, transistors stayed pretty much the same size, what happened was they got rotated vertically 90 degrees - so instead of lying flat on the surface, they stood vertically on their edge.
That's been true for a long time, but node names still have meaning as far as density since each full node iteration is roughly double the density of the previous one. If you merely think of them as a placeholder for representing transistor density it is still useful even if the number no longer represents a physical gate dimension.
New generation logic processes from Intel, TSMC, etc. don't quite double density for real chips but that's not due to cheating on transistor size, rather the ever more restrictive design rules they're forced to use with high NA sources and SADP. For memory fabs utilizing simple patterns like NAND, each full node step still truly doubles the bits per mm^2.
Instead of TSMC trying to make concessions to the United States in order to be allowed to build a fab in the United States, I would think that a lot of countries would be willing to make concessions to them to get them to build a fab in their country. Including suitable countries with reliable electricity.
Or, at least, a lot of countries should be doing this if they had any sense. I can think of one that should be doing this right off the top of my head: Canada.
The presence of some of the world's most advanced silicon fabs in Taiwan is a matter of government policy, and forms a part of their don't-get-invaded-by-China policy. Apparently the Taiwanese government said (publicly, quietly but just loud enough) quite a while back (I think China was being somewhat beligerent at the time) that they'd blow up the semi-con fabs if one single Chinese solidier's boot touched the sand on a Taiwanese beach.
It's quite an effective policy. We all depend, China especially so, on the chips that flow out of the Taiwanese economy; China would be badly effected economically if the world's microelectronics industry took a big hit like that. This was certainly the case at the time when the Chinese didn't have so many of their own fabs, or the expertise. Even today there's not exactly that many advanced fabs around the world with capacity to spare to take up the work load of all those Taiwanese fabs, so it would be a long lasting blow. And in more recent times the relationship has become a little more normal (flights between the two countries, etc.).
Getting to be one of the world's few 3nm centres of excellence is simply a way of continuing that status quo.
So any discussions with Trump about letting TSMC move to the USA is going to involve the Taiwanese government asking questions like "so what are you (Trump) going to do to guarantee our security given that you won't let us arm ourselves?". I can't imagine that any positive answer to such a question would be convincing.
Trump in the Far East
Trump has single handedly destroyed the trust between the Japanese government and the US. For decades Japan has been paying a lot of money for the US to guarantee its security. Trump's election campaigning on the matter ("I'm bring the troops home") was welcomed in China and caused astonishment in Japan. However in the very same week of taking office the new US Secretary of Defence got packed off to Japan to repair the damage, reassure the Japanese that it was situation-normal. This marked an end to China calling Trump a "wise man", etc. (Of course the visit went largely unreported in the US, which was handy for Trump so that no one in the US would be reminded of the matter...)
However it seems that whilst the Japanese have gone along with that reassurance from the US for now, they're tooling up, looking to change their constitution, and forming new defence treaties with other countries (particularly the UK). This all indicate that Japan no longer fully trusts the US. I suspect that Taiwan has reached a similar conclusion, and letting TSMC leave Taiwan was a "do-anything-to-keep-them-here" thing.
You make some fair points about the geopolitics of this, and Trump is indeed a dangerous moron (I'd be moderately surprised if there is a single regular reader of El Reg more ignorant or foolish than the Mad Orange Infant), but "you won't let us arm ourselves" is an odd observation: last time I counted, the US had sold many tens of billions in military kit to Taiwan, including pretty much the full spectrum across air, sea and land.
For this reason, and geography, Taiwan would be an immensely costly invasion for the mainland and, indeed, might be a no-win: even a combined arms parachute- and amphibious-assault across the straits would be a slaughterhouse: for one thing, it's relatively easy for the defenders to concentrate massive artillery on a landing zone with short distances and supply lines, while the aggressor has air- and sea-borne lines to keep up. A Chinese invasion, even with the inevitable aid of fifth columnists and infiltrators, is not a guaranteed success, even *if* the US Pacific Fleet doesn't stick its oar in.
I've thought for a while that if things get too nasty, it's a lot more likely that the mainland would try an air and naval blockade of the island instead. Taiwan's biggest port is on the western, mainland side of the island; the ranges work even for air interdiction; China can claim to be "peacekeeping" until it opens fire; and it has some political cover given that even the US has always (gutlessly, IMHO) sustained the one-China policy. If such a thing occurred, you'd have to rely on Trump to conduct a clever, mature, nuanced approach to the crisis to avoid triggering a major war ... in other words, there would be a major war, hindered by President Tiny Hands sticking his even smaller fingers into strategic matters he understands about as well as garden snail.
So, yeah, let's hope that Taiwanese fabs can maintain a magic "hands-off" aura ...
Japanese have gone along with that reassurance from the US for now, they're tooling up, looking to change their constitution, and forming new defence treaties with other countries (particularly the UK).
Why the UK? We don't even have a military capable of defending the UK, and Sir Michael Fallon and Philip Hammond are working tirelessly to further reduce the capability in the forthcoming "defence review". So we have no functioning aircraft carriers, the Navy's single helo carrier is about to be flogged to anybody who'll buy it, along with their two assault ships (which are only about twelve years old). The Royal Marines have been flagged for significant cuts, but with only 1,000 men anyway, that won't leave much. It has been hinted that the entire Leonardo Wildcat fleet (only just bought) will be mothballed or retired, and the Army may have to forgo all of its spotter helicopters.
The only help we can offer Japan on defence is to say "Look at what we do. Now do the complete opposite".
Why the UK? We don't even have a military capable of defending the UK,
UK homeland defence has been based on the "Moscow Criterion" for half a century - and you can substitute any other capital city for Moscow. Carriers and so on are for force projection and interference in the affairs of other parts of the world. Actual defence depends on submarines and the same logic applies to Taiwan, so that makes the UK and France the only viable allies if the USA keeps going nativist.
Mo' Chang, mo' problems.
But seriously. Keep in mind if the line width is 14 atoms then the oxide thickness is usually 1/10 that so 1-2 atoms thick.
Someone must be thinking "Y'know perhaps we should try making these by building things up, rather than chopping stuff away."
Of course, that's when it gets really expensive.
The issue with sub-10nm tech is not the possibility to etch it - that can be done. The issue is with long term stability of the product. When you have a feature which is less than let's say 50 atoms wide, a couple of atoms popping out of their places in the crystal lattice or swapping places with their neighbours is enough to turn the cart.
Atoms are not electrons, they do not move around easily. However, they do too. It is simply a matter of probability and some heat.
So while adding more transistors, etc 3nm smells like adding some planned obsolescence too. 28nm chips can run for decade(s). It is the capacitors which give up the ghost. At 3nm... Dunno... If it is regularly running at 70C+ (fairly common in a CPU under load) I have some doubts it will survive for more than 5-6 years.
At 3nm... Dunno... If it is regularly running at 70C+ (fairly common in a CPU under load) I have some doubts it will survive for more than 5-6 years.
I would guess (from a position of ignorance) that once you get down to 3 nm, there would be a lack of resilience to imperfections that currently do not cause too many problems, and even before they end up in devices, production yields at the fab would fall dramatically.
Any thoughts from those with knowledge of these matters?
"Smaller geometries mean more transistors in a given space, and only a few of the industry's giants have pockets deep enough to work at the smallest sizes (Intel, STMicroelectronics, IBM, Samsung, Global Foundries, and TSMC among them)."
Probably right for all others, but STM, no way they have the finances to invest 20 Bn USD !
Their revenue and cash position is not enough for this, at all. Many years back, when they had a lot more revenue and cash, they never were able to even cash out 2-3 Bn for new fabs equipment, so probably even less now.
<q>even a combined arms parachute- and amphibious-assault across the straits would be a slaughterhouse:</q>
There is an old and somewhat racist joke about a Chinese general hearing the news about a battle against <enemy>: "Terrible news, we lost 10,000 troops, and they only lost 1000". The general replies "Good, good". This goes on for several days, and eventually he asks why he keeps saying "Good, good". His reply is "pretty soon, no more <enemy>".
The PLA could almost certainly take Taiwan, even if it was at a terrible cost.
His reply is "pretty soon, no more <enemy>".
They have a similar battle plan now, just with equipment. If you count the amount of anti-ship missiles concentrated in the area - you will gulp. It is enough to wipe through sheer numbers several USA carrier groups. Same for other ordnance and equipment. Sure, it may not be superior to either the Russian originals (where they copied it from) or some of the USA stuff, but there is a ridiculous amount of it in place.
Another Register spin trying to make it sound bad that TSMC must keep ahead in fabrication. Actually, since Scamscum will do this, Register could put it as the price to compete with that Korean company. So scam scum are the bad guys in this.
But actually to suggest that it is either Apple or scamscum is rubbish - the ever shrinking fabrication processes due to Moore's law is what all fabricators are chasing. It's just a matter of their customers want it and the competition is also trying to do it.
So Register's attempt at a negative portrayal of Apple here is just more childish garbage.
But I'll balance that with some kudos because most Register articles are good and informative - just drop the negative Apple thinking that brings out the Apple bashers.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
