TSMC Will Start Making 2nm Chips As Intel Tries To Catch Up (gizmodo.com) 83
Kekke writes: "Taiwan Semiconductor Manufacturing Co.'s new foundry will produce 2-nanometer chips," reports Gizmodo. "Construction on the plant in Hsinchu, southwest from Taiwan's capital of Taipei, is expected to start as soon as early 2022. TSMC's 3nm tech is reportedly expected to be put into production in late 2022 -- meanwhile, Intel will be rolling out 7nm chips toward the end of 2022 and into 2023." Will Intel have a genie in the bottle or a rabbit in a hat? Doesn't seem so to me. On Tuesday, Intel unveiled a comeback plan designed to help it reclaim processor manufacturing leadership within four years.
Jokes on you Taiwan (Score:5, Funny)
Intel is leap frogging straight to 0 nanometer.
Re:Jokes on you Taiwan (Score:5, Funny)
Nah, they'll have a few generations of 7nm+ and 7nm++ and then 7nm+++ and maybe a 7nm++++ before moving on, if their recent progress is any indicator.
Re: (Score:2)
The +, ++ and +++ don't mean nothing. There is a large performance delta between them.
Back in the day, going from 1000nm to 700nm was meaningful and proportionate. The 10, 7, 5 , 3nm numbers were just silly and led to very misleading comparisons, like the one in TFS. Those numbers didn't relate to anything in the silicon and were not comparable between manufacturers, which is why Intel dropped that naming recently. Manufacturers are pretty much constrained by the available equipment and so they all play in
Re: (Score:2)
and multipatterning. You forgot to say multipatterning, one of the most important variables. Otherwise great post.
Re: (Score:2)
and multipatterning. You forgot to say multipatterning, one of the most important variables. Otherwise great post.
Being in a company that manufactures silicon, I avoid getting into specifics of manufacturing technologies in public.
Re: (Score:2)
I did say 'cost' though. Which alluded to multipatterning.
Re: (Score:2)
Kinda. So does "pain".
Re: (Score:2)
Re: (Score:2)
these companies label the nm of the process by measuring a different structure
Correct. They actually measure the width of their anal sphincter in cm. Otherwise no, the nm of the process does not correspond to any physical dimension, it is pure marketing whimsie.
Re: (Score:2)
A lot of the leapfrogging in the CPU wars isn't actually due to the CPU. It's due to the lithographic process the CPU is manufactured on.
And a lot is due to the CPU. See here. [wikipedia.org] Brought the memory controller and IO controller on chip. IPC way up. Superior multithreading. Fine grained power management. Kicked out the video driver to make room for more cores. As you pointed out yourself, Intel had a modest process advantage at the time. AMD's initial win was due to better CPU design, plus hungrier pricing.
After that you're mostly on point, however AMD also keep up the pace of processor improvements. Also, don't forget about the impact of chiplet
Re: (Score:2)
Re: (Score:3)
Unfortunately the sources for this is not available in der8auer's bad English, only in German: https://www.youtube.com/watch?... [youtube.com]
Re: (Score:2)
I've seen - somewhere - a mention of dropping nm as a size measure and using Å(ngström) instead, though what the point is I have no idea - given 10Å is 1nm.
Re: (Score:3)
In one of the shown slides you can see Intel's supposed 20A, for 20 Ångström. I can understand why they want to get rid of 'size' measure, because in that industry everyone appears to measures their dick using a different and arbitrary scale, but yeah, replacing it with Å makes little sense.
Re: (Score:2)
Pat Gelsinger announced the new naming a few days ago, with informative descriptions. Check out TechTechPotato on Youtube - he has good coverage of the announcement.
20Å is 2nm and is one of the names intel is using on the roadmap. You can see what it means in terms of process.
Re: (Score:2)
Gelsinger will also have ticktock clocks installed in every office. All sales staff will be required to tattoo "TICK" on the left cheek and "TOCK" on the right cheek going forward. This is really going to work.
Re: (Score:2)
Those would be marketing angstroms. Or in technical terms, the dimensions of an angel's asshole.
i don't think (Score:3)
Re:i don't think (Score:5, Funny)
they are really 2nm. If I remember correctly it's the equivalent of 2nm right? They stack them vertically or something, the feature size is not 2nm
"Our new chip is a half-inch square. And fourteen feet tall."
Re: (Score:2)
B. But it should make cooling a chip a bit easier, or at least more interesting. That's a lot more surface area you can draw heat from.
Re: (Score:3)
I think the math works out that you have less surface area per volume. A cube is worse off on cooling than a flat plate. And when you start making stuff 14 feet tall you run into the limitations of the square-cube law [wikipedia.org].
Re: (Score:2)
There is such a thing as a vertical fet https://kelifisica.wordpress.c... [wordpress.com]
Re:i don't think (Score:5, Informative)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
but how do they increase the transistor / mm then?
They do decrease the transistor footprint as well. But as the footprint reduces, you need to go higher in order to compensate for the degradation in on/off performance with reducing footprint.
The EUV stuff is all about feature size reduction and reducing the number of masks (I.E. cost).
The funky shaped transistors is all about making really small transistors actually work.
Re:i don't think (Score:5, Informative)
2nm is a marketing term.
Basically TSMC uses smaller numbers than Intel for about similar size processes. Intel is behind, but not as much as the numbers would suggest.
Beaten at their own game (Score:2)
2nm is a marketing term. {...} TSMC uses smaller numbers than Intel for about similar size processes.
Yes, but it's funny seeing intel beating at the "numbers" game that they started initially with their MHz wars.
Re: (Score:2)
Actually I think "Pentium Rating" system by AMD(and maybe Cyrix?) that could be more properly credited with it.
Re: (Score:3)
Jokes on you! Wait until you see the peak wafer output Intel achieves with their 7nm er 4nm process.
https://twitter.com/chiakokhua... [twitter.com]
20wkpm in 2023? hahahah good job Intel.
Re: (Score:2)
Ouch.
Re: (Score:2)
This is what happens when you don't order enough EUV equipment from ASML. Allegedly Gelsinger is correcting the issue by getting first-in-line for ASML's next-gen machines, but those won't be delivered and operational until . . .2024 or 2025. Intel won't be able to muster any significant volume of their 7nm or 5nm processes in a fashion that is timely to the market.
Re: (Score:2)
Re: (Score:2)
It's not even that, they have basically said that the numbers are pure marketing and have no relation to anything.
Re: (Score:1)
Intel reaping what they have sowed. (Score:5, Interesting)
Spending money on dividends and stock buybacks when they should have been investing in their own manufacturing have bitten them hard in the posterior.
Re:Intel reaping what they have sowed. (Score:4, Funny)
All they have to do is rename 10ESF from 7 to 1 and they'll be ahead again!
Re: (Score:3)
Honestly, given their long history of anti-competitive behavior, it couldn't have happened to a nicer company.
Re:Intel reaping what they have sowed. (Score:4, Interesting)
have bitten them hard in the posterior.
Lightly nibbled at best. Intel recorded record net cash flow last year and just shy of $6bn in profit. Revenue figures show the slight drop in manufacturing investment hasn't affected them in the slightest.
As for "when they should have been investing in their own manufacturing" I can only conclude you have no idea what you're talking about. Intel's share buybacks and dividends were made with free cash *AFTER* investment in their capabilities. And if you graph investment you'll see it has been on a steady and consistent upwards trend. At no point during their shareholder splurge did they ever invest less than $13billion into R&D, so the notion that they somehow cut corners in that department is absurd.
Now as to how they actually ran their R&D division there's plenty of argument to say the investments haven't really borne fruit, but to claim they haven't been investing is just admission that you don't read their quarterly reports.
Re: (Score:3)
the slight drop in manufacturing investment hasn't affected them in the slightest.
When you ignore what other people are talking about because you care about something different, it doesn't mean they're wrong, it means you can't hear them.
Your comprehension sucks.
Re: (Score:2)
No one ignored anything. Intel's manufacturing issues are lack of R&D successfully making the 7nm node viable.
If you're going to act like an arse it helps to have a clue what you're talking about.
Re: (Score:2)
Isn't that ironic? You still can't hear anything, how would you think you have a clue? And how would you know if I have a clue, when you can't hear me?
Re: (Score:2)
Well when you post something relevant we'll discuss further if you have a clue. But don't bother, it's clear you have nothing at all of value to add online, I won't make the mistake of reading your drivel again.
Re: (Score:2)
Re: (Score:2)
Capex matters only if you have a viable technology. Intel isn't suffering simply because it can't crank out chips fast enough (at least not more than anyone else). Their 7nm process just didn't work. You need a working design before you start talking capex. TSMC has been spending shitton of Capex precisely because they had finished R&D on the next node.
They don't care (Score:5, Insightful)
Execs get rich and that's all that matters. If the company folds they move onto the next victim. Boeing used to be run by former engineers and now look at how the MBA crowd does things.
Re: (Score:2)
Intel will run crying to the government for help. It's the American way, to give the government the power to grant wishes.
Re: (Score:2)
Major corporations are in the business of making investors happy. TSM pays a few cents more in dividends than INTC.
Re: (Score:1)
The final nail? (Score:5, Funny)
Re: (Score:3)
Re: (Score:1)
Needing better exception handling?
It's Atari Jaguar all over (Score:3)
Remember the time when computer consoles were all about bits? And how the Jaguar taught us that it means jack shit?
I have a hunch this isn't much different.
Re: (Score:2)
Compared to other manufacturers processes, sure, but between generations of TSMC processes it allows differentiation.
Re: (Score:3)
If I may rebound on your mention of the Jaguar console, as it reminds me of what I was saying:
"The Jaguar showed us it isn't enough to have the best console, you need the best games.
The DreamCast showed us it isn't enough to have the best console and the best games, you need the best marketing(*)."
(*)marketing in Sony language translates as lies and damn lies in english.
See "The PS2 will connect to high-speed networks" or "It will do Toy Story graphics in real-time"
To get back to the topic, Transistors/mm^2
Re: (Score:2)
The Jaguar wasn't even the best console. It was all marketing. Basically it was 2 32-bit chips working in tandem which made them claim that it's 64-bit.
Also, it was a bitch to develop for. So even if it had been the best hardware on the market, if the programmers cannot utilize its power, it's meaningless.
Re: (Score:3)
But yes, back on topic.
In the end, the only meaningful metric will be how the chip performs when used to calculate something. I don't care how many nms your chip is done in, I care about the instructions it can do per second on an operating system that I can use to run relevant software on. If your chip is the fastest in the world but only runs with some proprietary OS that cannot run any relevant programs, it means jack shit.
Re: (Score:2)
Re: (Score:2)
PS2 is just as easy to pirate games for. Can find them everywhere in stands in Asia. That's why the PS2 is still one of the most popular consoles outside the west.
Re: (Score:2)
The Dreamcast shows us that the lack of a DVD player in your games console when stand alone DVD players where expensive kills your console stone dead. Basically you could buy a PS2 and get both a games console and a DVD player. Most people have a finite budget and that was a deciding factor for enough people to doom the Dreamcast.
Intel Needs New Fabs (Score:5, Insightful)
Let's be honest; Intel had to make this announcement. Anything other than "here's our awesome plan to get back on top" would have had shareholders foaming at the mouth after their 7nm process problems [tomshardware.com].
Until Intel actually manages to build their new foundries, they're going to be lagging behind as they depend on already overstretched external foundries [pcgamer.com]... and it looks like Intel are going to use the cash generated by leasing capacity at their existing fabs to pay for spinning up the new ones. Intel (well, everyone really) is absolutely gagging for foundry capacity at this point; they're even considering buying GlobalFoundries outright [techhq.com].
Clearly there's demand for another major contract foundry, and Intel are one of the few companies who have a chance of successfully operating in this space. They've indicated their intentions to build new foundries in the US and the EU, which, apart from creating jobs in those regions, would mean that less of the total global chip production capacity is located in Asia. Not having all the world's eggs in that particular basket would also be a good thing. As it stands now, we're only one natural disaster/international political fiasco/pandemic away from chip manufacturing oblivion.
I think their 2025 goal for regaining supremacy in performance and efficiency is somewhat optimistic, but I genuinely hope they make it, even if just to push AMD to keep upping their game. More foundries are exactly what we need to resolve the global chip shortage, assuming related supply chains also recover in the near term.
Re: (Score:3)
More like a 'cunning plan'!
Baldrick would be proud about the 'another fine mess that Intel has gotten itself into'.
I hope that Intel has got its place in the ASML product line confirmed. ASML will be very busy getting all the GF and TSMC orders satisfied.
Whatever Intel does I'm done buying anything they make.
Re: (Score:2)
With all that being said, Intel’s discussion point today is that it will be the lead customer for ASML’s next generation EUV technology known as High-NA EUV. ...
ASML’s latest update suggests that it expects customers to be using High-NA for production in 2025/2026, which means that Intel is likely going to be getting the first machine (ASML NXE:5000 we think) in mid-2024.
I think the relationship between Intel and ASML is still strong. Obviously, there's massive competition for orders, but I'm sure Intel will get their "fair" share.
I'm sure they paid through the nose for it, too.
Re: (Score:2)
Last round, Intel bought a bunch of cutting edge ASML machines then let them rust as they doubled down on their 10nm circus.
Re: (Score:2)
Re: (Score:2)
poor comparison. (Score:2)
Re:poor comparison. (Score:4, Insightful)
Intel vs TSMC nm are not like for like comparisons of the end product, TSMC are not as far ahead as the numbers would indicate as they are the WRONG numbers to compare. e.g. Intel 7nm has far higher density than TSMC 5nm.
The important bit, is that TSMC can actually make their 5nm parts.
Re: (Score:2)
Re: (Score:2)
nm are not chip density. (Score:2)
These numbers don't have a real meaning (Score:2)
These numbers have been meaningless for quite a while. They're just marketing spout at this point. It's like trying to focus on core speed without checking how many cores there are.
TSMC is about one year ahead of Intel. AMD has been there four times. Intel has crossed this bridge before, and will again.
Anandtech ... (Score:3)
This is sounding like the old timey MHz wars (Score:2)
If a 14 nanometer chip is faster, cheaper and more reliable I prefer that. We assume die shrinks have commensurate power reductions, but that may not be a valid assumption anymore.
Forget feature size and give me a terahertz clock speed please
Re: (Score:2)
What does 2nm actually mean? (Score:2)
What I have been surprised with lately is how quickly the size seems to be going down. Every month it seems I hear that that next year an even smaller processing will be available. 2nm is like 5-7 times smaller than Intel current CPUs.
That is insane. But what does that mean? Will be be getting CPUS with 5 times the current core count? 5 times the speed? 1/5th the price? in 2-3 years?