Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel)

[DisEnchantment]

TSMC's N7 EUV is now in its second year of production and N5 is contributing to revenue for TSMC this quarter. N3 is scheduled for 2022 and I believe they have a good chance to reach that target.


N7 performance is more or less understood.


This year and next year TSMC is mainly increasing capacity to meet demands.

TSMC Radically Boosts CapEx to Expand Production Capacities, To Reach $14B For 2019

www.anandtech.com

For Samsung the nodes are basically the same from 7LPP to 4 LPE, they just add incremental scaling boosters while the bulk of the tech is the same.

Samsung is already shipping 7LPP and will ship 6LPP in H2. Hopefully they fix any issues if at all.
They have two more intermediate nodes in between before going to 3GAE, most likely 5LPE will ship next year but for 4LPE it will probably be back to back with 3GAA since 3GAA is a parallel development with 7LPP enhancements.

Samsung’s Aggressive EUV Plans: 6nm Production in H2, 5nm & 4nm On Track

www.anandtech.com

Samsung's 3GAA will go for HVM in 2022 most likely, similar timeframe to TSMC's N3.
There are major differences in how the transistor will be fabricated due to the GAA but density for sure Samsung will be behind N3.
But there might be advantages for Samsung with regards to power and performance, so it may be better suited for some applications.
But for now we don't know how much of this is true and we can only rely on the marketing material.

This year there should be a lot more available wafers due to lack of demand from Smartphone vendors and increased capacity from TSMC and Samsung.
Lots of SoCs which dont need to be top end will be fabbed with N7 or 7LPP/6LPP instead of N5, so there will be lots of wafers around.

Most of the current 7nm designs are far from the advertized density from TSMC and Samsung. There is still potential for density increase compared to currently shipping products.
N5 is going to be the leading foundry node for the next couple of years.

For a lot of fabless companies out there, the processes and capacity available are quite good.

 

[naukkis]

Matters at 6 GHz but not 4 GHz? Is it a square factor, i.e. F^2 so 6 GHz has double the loss of 4 GHz?

Or are they mainly counting on the fact that HPC will run flat out 24x7 while phones/laptops/PCs will run at max frequency a tiny fraction of their operating life, and has to do with the temperature of the silicon and wiring?

If current feed is limiting factor for clock frequency improving it will increase fmax. Design targeting lower fmax probably focus on reducing system overall capacitance which doesn't benefit much from improved current feed.

 

[sdifox]

I'm not sure you're looking at it from all angles with this take.

Both TSMC and Samsung are essentially treated as part of the government by their respective nations (Samsung especially so). They're viewed as their national champion and their representative to the tech world. Samsung Foundries in particular has almost no obligation to turn a profit, their financials are rolled into other segments of Samsung and shielded from public scrutiny. It's ran almost entirely for developing the requisite manufacturing knowledge for the benefit of South Korea. For Taiwan, TSMC is the lifeblood of the entire country. It's viewed as more important and prestigious than their own military. Any young Taiwanese that is particularly gifted intellectually is pushed by society to join TSMC and contribute to Taiwan's success. Asianometry has a lot of excellent videos on the TSMC / Taiwan topic.

Intel is not treated by the United States the same way. Yes, we favor it of course but it's not our national identity. People joining Intel aren't looking to satisfy their patriotic duty to their country, it's just one of many large multi-national companies within our borders. We welcomed Samsung and TSMC to build fabs in our country and even subsidized parts of it. Their increased costs are due to regulations that are actually enforced (instead of handwaved away their government) and the differences in working culture from the West and Asia.

Intel got a lot of hand out from US governments no?

Subsidy Tracker

Subsidy Tracker, produced by the Corporate Research Project of Good Jobs First, is a wide-ranging database on misconduct by large and small corporations throughout the United States.

subsidytracker.goodjobsfirst.org

 

[H433x0n]

Intel got a lot of hand out from US governments no?

Subsidy Tracker

Subsidy Tracker, produced by the Corporate Research Project of Good Jobs First, is a wide-ranging database on misconduct by large and small corporations throughout the United States.

subsidytracker.goodjobsfirst.org

I’ve got news for you - So is Samsung, TSMC, ASML, etc.

Samsung is practically a government entity, they account for ~20% of South Korea’s GDP. The Republic of Samsung is a meme for a reason.

Governments treat this industry as a top strategic priority.

 

[sdifox]

I’ve got news for you - So is Samsung, TSMC, ASML, etc.

Samsung is practically a government entity, they account for ~20% of South Korea’s GDP. The Republic of Samsung is a meme for a reason.

Governments treat this industry as a top strategic priority.

Oh I know about Samsung, I am saying Intel is not that different.

 

[trivik12]

Considering amount of capex intel has invested in its fabs that 8B from 1993 is peanuts compared to what TSMC or Samsung have gotten. That is why Pat lobbied to pass Chips act. not sure how much they will get for Ohio or Arizona fabs.

 

[H433x0n]

Interesting article from WikiChips about TSMC N3.

TSMC N3, And Challenges Ahead

A Look At TSMC N3 Process

fuse.wikichip.org

Specifically TSMC N3B HP Cells compared against Intel 4 HP cells.

 

[Dayman1225]

Jensen saying that Intel test chip data is looking good for next generation process node, probably 18A

https://twitter.com/x/status/1663501320907620353

 

[Henry swagger]

Jensen saying that Intel test chip data is looking good for next generation process node, probably 18A

https://twitter.com/x/status/1663501320907620353

Probably 5000 series or 6000 series will use 20a node... 😃

 

[FangBLade]

Probably 5000 series or 6000 series will use 20a node... 😃

But only the cheapest, least significant models. Not even Intel uses its own manufacturing process for graphics, so certainly Nvidia won't either. They will utilize TSMC's capacities for the AI market and high-end desktop graphics, while Intel's will be focused on the lower to mid-range segment. Let's not forget that there will be delays, that's Intel's trademark.

 

[gdansk]

I don't want a TSMC leading edge monopoly for a decade. No one benefits from that. Promising to hear 18å is going well. But it's Jensen so who knows.

 

[Hitman928]

Intel nabbing Nvidia as a customer would be huge news but I have my doubts that it will actually happen, at least not at 18a with any significant product line. Who knows though, we'll see what happens.

 

[clemsyn]

Jensen saying that Intel test chip data is looking good for next generation process node, probably 18A

https://twitter.com/x/status/1663501320907620353

I think he is using Intel to get better prices from TSMC

 

[DrMrLordX]

I think he is using Intel to get better prices from TSMC

That's what they tried with Samsung. Didn't quite work though.

 

[KompuKare]

I think he is using Intel to get better prices from TSMC

The problem with being that transparent is that eventually someone might call his bluff!

 

[trivik12]

its possible NVDA might make specific chips for US Government contracts at Intel. That might mandate it be manufactured state side.

 

[sdifox]

It would be hilarious if NVidia loses out on TSMC allotment because Jensen decided to mess around.

 

[gdansk]

It would be hilarious if NVidia loses out on TSMC allotment because Jensen decided to mess around.

TSMC is selling N3(e?) wafers to Intel of all people. I don't think they're the type to care if Nvidia is looking around other than working on making better processes to keep Nvidia hooked.

 

[Doug S]

Nvidia had a lot of different products, they might choose one or two to go with Intel to see how things go before they'd risk going all in with everything. If Intel could even handle Nvidia's needs, I think Nvidia would have no choice other than to pick and choose a few smaller volume designs.

Look at how fabbing modems for Apple drove them into shortage on 14nm in consecutive years 5-6 years ago or whatever it was.

 

[Geddagod]

Nvidia had a lot of different products, they might choose one or two to go with Intel to see how things go before they'd risk going all in with everything. If Intel could even handle Nvidia's needs, I think Nvidia would have no choice other than to pick and choose a few smaller volume designs.

Look at how fabbing modems for Apple drove them into shortage on 14nm in consecutive years 5-6 years ago or whatever it was

Isn't Nvidia planning to put some of their hardware in a random handheld? I don't remember lol

 

[H433x0n]

Wouldn’t something like 18A be significantly more expensive than TSMC N3E?

Nvidia doesn’t really have to be picky with node performance since even with a worse performing node they can still outperform competitors with their ecosystem and software stack. I would think Nvidia be among the last to switch to a more expensive and more performant node.

 

[Doug S]

Wouldn’t something like 18A be significantly more expensive than TSMC N3E?

Nvidia doesn’t really have to be picky with node performance since even with a worse performing node they can still outperform competitors with their ecosystem and software stack. I would think Nvidia be among the last to switch to a more expensive and more performant node.

We don't know how many EUV layers 18A has or if it will be using the more expensive high NA scanners for critical layers, so it is hard to determine costs. More expensive, probably. "Significantly more", perhaps but perhaps not. Plus Intel probably has an incentive to discount early customers if they really want IFS to succeed.

They need a design with with someone big, and Nvidia is really hot right now - I don't count their previous announcement with Qualcomm since there was really nothing committed to, no product no node no timeline no quantity so I count that as meaningless market drivel. Being able to announce something with Nvidia that has specific dates and ideally some sort of dollar amount (probably in terms of what Nvidia will sell the resulting stuff for since it will sound more impressive) would go a long way, especially since the press will hear "Nvidia" and assume Intel is making their AI chips even though that's the least likely thing they would have Intel make with an unproven process due to their size.

 

[Doug S]

Timely article from the WSJ about Intel's problems trying to get IFS off the ground. No matter how well they execute on their process roadmap and even if they take back all the x86 share they lost to AMD they cannot survive long operating their own fabs if they are their only major customer. It is simply too costly to develop new processes and build new fabs if they have only their own chips to amortize those massive fixed costs against.

Once Mighty Intel Struggles to Escape ‘Mud Hole’

Rivals such as Nvidia have left the chip company far behind. CEO Pat Gelsinger aims to reverse firm’s fortunes by vastly expanding its factories.

www.wsj.com

 

[H433x0n]

Timely article from the WSJ about Intel's problems trying to get IFS off the ground. No matter how well they execute on their process roadmap and even if they take back all the x86 share they lost to AMD they cannot survive long operating their own fabs if they are their only major customer. It is simply too costly to develop new processes and build new fabs if they have only their own chips to amortize those massive fixed costs against.

Once Mighty Intel Struggles to Escape ‘Mud Hole’

Rivals such as Nvidia have left the chip company far behind. CEO Pat Gelsinger aims to reverse firm’s fortunes by vastly expanding its factories.

www.wsj.com

Their 2 big examples are from Tesla in 2021 and Qualcomm in 2022. I'm basically disregarding the comments from Tesla entirely. The Qualcomm comments were interesting since it sounds like there is a fundamental design difference that Intel's fabs need to sort out with low power mobile designs.

All of the information in this article is more than 6 months old, I'm not sure what the goal was writing it now instead of 6 months ago when the Qualcomm deal seemingly fell through.

Edit: A part of me is suspicious this article was pushed today in response to the stock bounce from Jensen’s comments. This article seemingly was designed to throw some cold water on Intel and say “Bad things happened in 2021 & 2022! The fabs are still bad!”

 

[controlflow]

Timely article from the WSJ about Intel's problems trying to get IFS off the ground. No matter how well they execute on their process roadmap and even if they take back all the x86 share they lost to AMD they cannot survive long operating their own fabs if they are their only major customer. It is simply too costly to develop new processes and build new fabs if they have only their own chips to amortize those massive fixed costs against.

Once Mighty Intel Struggles to Escape ‘Mud Hole’

Rivals such as Nvidia have left the chip company far behind. CEO Pat Gelsinger aims to reverse firm’s fortunes by vastly expanding its factories.

www.wsj.com

Nothing timely about it. Seems like old regurgitated information mostly. Pretty odd thing to publish now.

 

[SpudLobby]

Timely article from the WSJ about Intel's problems trying to get IFS off the ground. No matter how well they execute on their process roadmap and even if they take back all the x86 share they lost to AMD they cannot survive long operating their own fabs if they are their only major customer. It is simply too costly to develop new processes and build new fabs if they have only their own chips to amortize those massive fixed costs against.

Once Mighty Intel Struggles to Escape ‘Mud Hole’

Rivals such as Nvidia have left the chip company far behind. CEO Pat Gelsinger aims to reverse firm’s fortunes by vastly expanding its factories.

www.wsj.com

This is completely correct. it's overlooked how much of the shift to an open FS and outsourcing (hybrid scheme) is a delayed reaction to reality that should have taken place years ago. They simply do not have the scale to amortize this R&D and the fixed cost structures, the writing was on the wall by the 10's IMO. It might be more sustainable if Intel hadn't lost mobile and GPU's but ultimately same gig, they'd run into the same issues with fixed costs of development optimizing the process for different markets since the overlap is not perfect.