Intel to Outsource a Part of 2023 Processor Production to TSMC

[Valantar]

It's hardly different than what they've done in the past or continue to do in Tibet ~ or is that old news now? There's also Vietnam just in case you forgot, the West turned a blind eye towards the invasion of Tibet nearly half a century back so why does it matter now? If the powers that be were actually serious about democracy or rule of law, human rights & all that BS we wouldn't have the Taliban, Iraq-Iran war, overthrowing democracy in Iran so on & so forth!

... so you missed the part where I said

This is of course not to say that the US is any better, but it does (currently) pose less specific risk in these areas.

?

Which is to say that it's no big deal for China ~ just that there's an effin big sea in between & an actual invasion will unlikely be like a bloodless coup. Their only hope to integrate Taiwan is if the rulers actually hand over the key to the kingdom themselves.

I honestly don't think they'd care, if it came to that. Authoritarian regimes generally don't mind the need to exert violence on the population in order to assert their control. The response to the HK protests showed us that on a small scale within a nominally democratic and autonomous system, so if global conditions changed to such a degree that the Chinese ruling class saw grasping control of large parts of the global chip supply, those things would already be accepted as necessities of the process of taking over. The need for violence is in no way what would hold them back; the likelihood of the benefits being drowned out by the negative consequences is what does.

Which, of course, is why most superpowers throughout recent history has preferred covert action, like the US propensity to stage coups against democratically elected governments and putting totalitarian dictatorships in their place, or the Soviet propensity for industrial espionage, kidnappings and assassinations. It's just much easier to get away with that kind of stuff. But when conditions are right, authoritarian regimes (and the US does deserve some inclusion in that term) do not shy away from violent intervention - Iraq, Afghanistan, Crimea, Ukraine, and a whole host of other situations in even very recent history demonstrate that clearly.

Oh, and as for that "effin big sea" - it's, what, 100km across? That's really not a lot. Commercial ferries cover distances like that in a few hours; warships could likely do it quite a lot quicker. Heck, that distance could probably be covered with landing craft alone if one wanted a covert approach, even if it would be riskier and slower.

 

[medi01]

Anyone other than me notice that 2023 is still two years out? Even if Intel 7nm were to maintain their typical naming sicrepancy to competing nodes, Apple has been on 5nm for a year, and everyone else is bound to be following them within the next year. So if Intel 7nm is equivalent to TSMC 5nm, they'll still be well behind. Makes complete sense for them to outsource, though I can't imagine that being good for the overall chip supply situation.

Intel 14nm transistor => 24x24 nm.
TSMC 7nm transistor => 22x22 nm

At its face value, Intel 7nm could easily beat TSMC 7nm, 5nm, 3nm and even, wait for it, 0nm.

Or it might not.

It depends.

 

[Valantar]

Intel 14nm transistor => 24x24 nm.
TSMC 7nm transistor => 22x22 nm

At its face value, Intel 7nm could easily beat TSMC 7nm, 5nm, 3nm and even, wait for it, 0nm.

Or it might not.

It depends.

Please define "transistor" in that size comparison - not all transistors look the same or are even remotely comparable after all.

For reference, here are some numbers from WikiChip (1, 2):

	TSMC 7nm	Intel 14nm
Fin pitch	30 nm	42 nm
Fin width	6 nm	8 nm
Fin height	52 nm	42 nm
Gate length	?	20 nm
Contacted gate pitch	64 nm HP / 57 nm HD	70 nm
Minimum metal pitch	40 nm	52 nm
HP SRAM bit cell	?	0.0706 µm²
HD SRAM bit cell	0.027 µm²	0.0499 µm²
LV SRAM bit cell	?	0.0588 µm²

So, your numbers, whereever they come from, are at best misleading. SRAM density is 2x on TSMC 7nm, and fin area is much smaller (height doesn't affect area after all). It's closer than one might expect - hence the adage that Intel's processes are typically named one tier higher than their competitors (i.e. Intel 14nm being roughly equivalent to TSMC 10nm, Intel 10nm supposedly (but not in real life, given their troubles) being roughly equivalent to TSMC 7nm, etc.), but there's still a significant difference.

 

[medi01]

Please define "transistor" in that size comparison - not all transistors look the same or are even remotely comparable after all.

Transistors used for CPU L1 cache.
Measured by some German dude using some insane equipment.

 

[Why_Me]

The country has always been Taiwan. The old ROC government/KMT was technically an invading force that's occupying the island. Most KMT supporters want to either go back to or reunite with China. Most Taiwanese have zero interest in China. This is largely the fault of the US, who "liberated" Taiwan from Japan and then quickly forgot about the island. Taiwan had it's own government after the Japanese invasion ended and before the KMT invaded. But what do you do when you're a tiny nation with no army when you get invaded by a foreign force?

Yes the US did liberate Taiwan with thousands of American lives .. at about the same time Sweden was supplying Nazi Germany with raw materials and yes the US has helped Taiwan in regards to that country's defense.

 

[fgk]

The country has always been Taiwan. The old ROC government/KMT was technically an invading force that's occupying the island. Most KMT supporters want to either go back to or reunite with China. Most Taiwanese have zero interest in China. This is largely the fault of the US, who "liberated" Taiwan from Japan and then quickly forgot about the island. Taiwan had it's own government after the Japanese invasion ended and before the KMT invaded. But what do you do when you're a tiny nation with no army when you get invaded by a foreign force?

What you wrote is simply false. The ROC wasn't an invading force. Japan ceded the island to the ROC in 1945 as part of their war reparation to China. The US even played a vital role in the ROC getting the island.

The Cairo declaration was signed by the US, UK, and the ROC in 1943, and it declared that "all the territories Japan has stolen from the Chinese, including Manchuria, Formosa, and the Pescadores, shall be restored to the Republic of China".

 

[TheLostSwede]

Yes the US did liberate Taiwan with thousands of American lives .. at about the same time Sweden was supplying Nazi Germany with raw materials and yes the US has helped Taiwan in regards to that country's defense.

I think you missed my point here.
The US forgot about Taiwan, as the nation that was founded before Japan invaded.
The US was helping the KMT, i.e. the invading exile government of the ROC.
That's two quite different things.
Maybe read some history before you make stupid comparisons?

What you wrote is simply false. The ROC wasn't an invading force. Japan ceded the island to the ROC in 1945 as part of their war reparation to China. The US even played a vital role in the ROC getting the island.

The Cairo declaration was signed by the US, UK, and the ROC in 1943, and it declared that "all the territories Japan has stolen from the Chinese, including Manchuria, Formosa, and the Pescadores, shall be restored to the Republic of China".

Of course the KMT was an invading force, it's the ROC government. Taiwan was never a part of China, regardless what the PRC claims and before Japan invaded it was it's own nation called the Republic of Formosa.

Republic of Formosa - Wikipedia

en.m.wikipedia.org

They should've been given the nation back, but instead it was given to a dictator, in 1952, despite the KMT invading Taiwan since 1945 and killing tens of thousands of Taiwanese during the white terror era.
You're making some really flawed claims here, as Taiwan was never ruled by the ROC until that point, nor by any previous Chinese dynasty.
Japan didn't cede anything, they simply packed up and left. They did later sign a paper that said they laid no claim to Taiwan, but that's hardly the same thing.

In accordance with the provisions of Article 2 of San Francisco Peace Treaty, the Japanese formally renounced the territorial sovereignty of Taiwan and Penghu islands, and the treaty was signed in 1951 and came into force in 1952. At the date when the San Francisco Peace Treaty came into force, the political status of Taiwan and Penghu Islands was still uncertain.

None of the post–World War II peace treaties explicitly ceded sovereignty over the covered territories to any specific state or government

The Taiwan government had no say in any of this, so in other words, they nation was stolen from them again.

 

[pjl321]

Do we still think there is room for TSMC manufactured 5nm monolithic CPUs this year and 3nm next year for Intel like so many prominent industry insiders are suggesting before we move onto tile based CPUs where Pat said Redwood cores and beyond would be node agnostic to allow for the likes of TSMC to help with production of different tiles?

 

[Valantar]

Do we still think there is room for TSMC manufactured 5nm monolithic CPUs this year and 3nm next year for Intel like so many prominent industry insiders are suggesting before we move onto tile based CPUs where Pat said Redwood cores and beyond would be node agnostic to allow for the likes of TSMC to help with production of different tiles?

Pretty much all near future 5nm capacity at TSMC is already booked for Apple, and I would be very surprised if AMD didn't sign up to be next in line for that several years ago. It sounds extremely doubtful that Intel will be able to get their hands on any TSMC 5nm capacity for the next year, and maybe even longer.

 

[pjl321]

Pretty much all near future 5nm capacity at TSMC is already booked for Apple, and I would be very surprised if AMD didn't sign up to be next in line for that several years ago. It sounds extremely doubtful that Intel will be able to get their hands on any TSMC 5nm capacity for the next year, and maybe even longer.

I actually think the most likely way forward for this partnership would be for Intel to licence TSMC's IP so that Intel can use their own EUV machines (which are the same as TSMC's) and so create capacity that way.

Also, Apple is moving off of 5nm as we speak, moving to N4 and N3. TSMC have already announced that their 3nm is way ahead of schedule with production of around 30,000 wafers per month starting in ~Aug. Apple has every single one of these but it does free up some 5nm for later this year which I believe AMD bought most of but at the same time I only hear bits and pieces from TSMC these days so other deals could be getting made too.

 

[Valantar]

I actually think the most likely way forward for this partnership would be for Intel to licence TSMC's IP so that Intel can use their own EUV machines (which are the same as TSMC's) and so create capacity that way.

Also, Apple is moving off of 5nm as we speak, moving to N4 and N3. TSMC have already announced that their 3nm is way ahead of schedule with production of around 30,000 wafers per month starting in ~Aug. Apple has every single one of these but it does free up some 5nm for later this year which I believe AMD bought most of but at the same time I only hear bits and pieces from TSMC these days so other deals could be getting made too.

Do we have any indication whatsoever that TSMC is willing to licence out their node tech? I haven't heard even a whisper of this at least. And the Intel news is quite explicit in stating that silicon will be made in TSMC's fabs. Intel also seemed pretty positive about having fixed their own 7nm node and being on track for bringing it to market in their most recent press event about this stuff.

As for Apple moving on from 5nm, that isn't very likely even if they are jumping into 3nm as fast as possible. Remember, they have iMac/Mac Pro hardware in the pipeline, at least one large-size SoC (though I would guess two: one for high end laptops and "entry" iMac Pro/Mac Pro SKUs, and one for high-end Mac Pro SKUs; and likely some form of dGPU or accelerator card to go with those in the Mac Pro). There's the M1, which will stay in production for a while yet - it's rumored to be the basis for the next-gen iPad SoC as well, so it'll be moving some volume. But most importantly, making those large monolithic PC SoCs on 3nm simply isn't feasible for such a young process node. Even with the astronomic prices demanded by Apple's pro gear, it's highly unlikely that 3nm will be for anything but iPhones in the first year or so - it's just that much more economical to make smaller chips on a newer node.

I'm sure there's some spare 5nm capacity that Intel might have snapped up, but I'd be shocked if it amounts to any type of notable volume any time soon. Much more likely they're getting in on 7nm as other high volume customers are moving to 5nm.

 

[pjl321]

Do we have any indication whatsoever that TSMC is willing to licence out their node tech? I haven't heard even a whisper of this at least. And the Intel news is quite explicit in stating that silicon will be made in TSMC's fabs. Intel also seemed pretty positive about having fixed their own 7nm node and being on track for bringing it to market in their most recent press event about this stuff.

As for Apple moving on from 5nm, that isn't very likely even if they are jumping into 3nm as fast as possible. Remember, they have iMac/Mac Pro hardware in the pipeline, at least one large-size SoC (though I would guess two: one for high end laptops and "entry" iMac Pro/Mac Pro SKUs, and one for high-end Mac Pro SKUs; and likely some form of dGPU or accelerator card to go with those in the Mac Pro). There's the M1, which will stay in production for a while yet - it's rumored to be the basis for the next-gen iPad SoC as well, so it'll be moving some volume. But most importantly, making those large monolithic PC SoCs on 3nm simply isn't feasible for such a young process node. Even with the astronomic prices demanded by Apple's pro gear, it's highly unlikely that 3nm will be for anything but iPhones in the first year or so - it's just that much more economical to make smaller chips on a newer node.

I'm sure there's some spare 5nm capacity that Intel might have snapped up, but I'd be shocked if it amounts to any type of notable volume any time soon. Much more likely they're getting in on 7nm as other high volume customers are moving to 5nm.

Yeah they won't be moving everything all at once for the reasons you stated, I believe the iPhone and probably M2 will be on N4 this year which is much closer to N3 than N5 but I think Apple wants to be able to say their phone next year is on a different node to this this year and a different one to last year so that is why it's N4 and not N3. But even this move frees up huge amounts of 5nm capacity.

I don't think anyone really knows what's going to happen, I guess it's just a case of watch this space. But Intel won't have 7nm CPUs until 2023 (and could also easily slip couldn't it), TSMC will be well into offering 3nm to high performance chips by then and we might even be looking at 2nm using GAA. I think it just makes sense for Intel to use TSMC until they definitely get back on track and caught up, so until 2025-2026 at the earliest.

Let's face it, I can't see Alder Lake's 10nm Enhanced SuperFin being anywhere near good enough to compete with Zen 4's 5nm and AMD could be on 4nm or 3nm before Intel get's to 7nm (ok not all nm are the same but 10nm vs 3nm is a huge difference). I think Intel will buy the TSMC wafers and then decide what to make with them, mobile chips, desktop, server, super computer or of course GPUs.