Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel) - [2020 - 2025]

[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.

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

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.

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

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.

---------------------------------------------------------------------------------------------------------------------------------------------------


FEEL FREE TO CREATE A NEW THREAD FOR 2025+ OUTLOOK, I WILL LINK IT HERE

 

[511]

TSMC is doing relatively well, but I don't think they are the epitome of that. They have 80K employees with much bigger 3rd party customer count. Intel can probably do it for 30K on that metric alone. Extract efficiency, and maybe 20K. Then you have 10K for everything else. Bare minimum just to make things work. By having less people, you communicate faster, cut down the garbage as well.

Like read about how efficient companies can be. Cyrix had 1/3rd of the development time. Made x86 processors for $10 million. But they really ran on bare minimum. Assembled test chips by hand. Ran simulations on overclocked, off the shelf, desktop platforms, rather than overpriced servers. They had 70 or so employees total, and 50 core engineering group. They quoted one guy that "would do the work of ten". On Youtube you have some guys doing reballing DRAM chips by hand, one solder ball by one solder ball using tweezers. The amount of patience and persistence is insane.

Intel has lots of SW People and designers which TSMC doesn't have also TSMC and Intel has many businesses that's not funny they still have a internal EDA from what I recall for older nodes.

 

[johnsonwax]

He doesn't have technical knowledge about electric cars either. Yet here we are with Tesla, that single-handedly made electric cars viable to masses. Previous to that electric cars barely got 100km, and couldn't outperform a Prius, with a design that was solely optimized for weight and looked like a electric bicycle with a hood. Delays and problems they are facing doesn't change the fact his company achieves things that others couldn't.

But it's not like there was some massive technical achievement here. Tesla started out using standard 18650 cells, available over the counter. Most of the technical breakthroughs really came from the quadcopter research community - they needed improved power/weight ratio motors which required rare earth magnets, and improved power/weight ratio batteries, both of which were also necessary for EVs. That's where the technical breakthroughs came from.

The only reason why Tesla succeeded is that they didn't have $20B in sunk costs and debt attached to ICE factories and engineers like the other automakers did. Any of them could have done it, but they had the curse of incumbency - go ask Nokia how that goes. Blockbuster. The list is almost endless. The key innovation he brought was his own money. Nobody was going to capitalize a new automaker because US finance is insanely conservative when it comes to investing in manufacturing. None of them were going to help finance a new automaker, but Musk would. And he deserves credit for that, but that's really the thing he brought to Tesla (remember, he didn't found Tesla - he bought it).

Why hasn't ULA landed a rocket - same problem. They're wired head to toe to succeed in a disposable rocket market. Meanwhile, Blue Origin just proved that other players can do it too, and I'd argue they're somewhat ahead in the sense that I think the feds will be more comfortable with their landing approach. Again, the billionaires can do it because they don't have sunk costs in the old market and they can capitalize the effort.

 

[Doug S]

Intel themselves said that taking 24 months for a new process node was a choice motivated based on profit. Then gave comparisons to different timeframes.

12 months, and 36 months. 12 months is suboptimal because you aren't fully taking advantage of the revenue, and 36 months is well after the optimal profit point. Hence, they chose 24 months for new process technology. Intel took 3.5 years to go from Intel 7 to 18A. As chaotic it was, speeding up is to a point, possible.

He doesn't have technical knowledge about electric cars either. Yet here we are with Tesla, that single-handedly made electric cars viable to masses. Previous to that electric cars barely got 100km, and couldn't outperform a Prius, with a design that was solely optimized for weight and looked like a electric bicycle with a hood. Delays and problems they are facing doesn't change the fact his company achieves things that others couldn't. Yes, not all his claims and goals come to fruition. Yet maybe it's that ambition and dreams that allows it to get ahead of others, because others have much less of a target to attain.

The last time someone had made serious investment in electric cars prior to Tesla was GM in the early 90s, using far more limited NiMH battery technology. The increase in range was obviously going to happen when you make a generational change in batteries.

You can't seriously be comparing electric cars - something that dozens of companies are doing at scale now and probably hundreds have done at least at smaller scale in the past couple decades before they went under or got bought out - to leading edge chip fabrication which has only three companies producing at scale. That's not a matter of "just" needing enough money to do it, it is a matter of all the patents, trade secrets, supplier relationships, and specialized knowledge.

Compared to designing and building a leading edge fab building with all the specialized plumbing that needs to handle all sorts of different types of gases and liquids, many of them toxic and/or corrosive, air handlers to filter air to insane levels, etc., building an auto manufacturing plant is the equivalent of a kindergarten project with tinkertoys.

 

[511]

Agree with Dough Chip Manufacturing is not that easy especially without a base IP

 

[maddie]

The last time someone had made serious investment in electric cars prior to Tesla was GM in the early 90s, using far more limited NiMH battery technology. The increase in range was obviously going to happen when you make a generational change in batteries.

You can't seriously be comparing electric cars - something that dozens of companies are doing at scale now and probably hundreds have done at least at smaller scale in the past couple decades before they went under or got bought out - to leading edge chip fabrication which has only three companies producing at scale. That's not a matter of "just" needing enough money to do it, it is a matter of all the patents, trade secrets, supplier relationships, and specialized knowledge.

Compared to designing and building a leading edge fab building with all the specialized plumbing that needs to handle all sorts of different types of gases and liquids, many of them toxic and/or corrosive, air handlers to filter air to insane levels, etc., building an auto manufacturing plant is the equivalent of a kindergarten project with tinkertoys.

It's sort of implied, in "specialized knowledge", but just as electrical output cannot be wished into being, the quantity of specialized employees is vital. Time, and the items associated with it, is the only resource money can't control. Back to STEM schools folks, in mass this time.

We'll see soon with the rare earths issue. Narratives are easy, the real world isn't.

 

[digitaldreamer]

Cybertruck shipped 2 years late. The easiest thing in the world to do is the thing you know the least about.

Yep. Nothing is impossible for the man who has never done it himself. You can't beat the physics in chip making and you'll go bankrupt trying.

 

[MoistOintment]

Intel themselves said that taking 24 months for a new process node was a choice motivated based on profit. Then gave comparisons to different timeframes.

12 months, and 36 months. 12 months is suboptimal because you aren't fully taking advantage of the revenue, and 36 months is well after the optimal profit point. Hence, they chose 24 months for new process technology. Intel took 3.5 years to go from Intel 7 to 18A. As chaotic it was, speeding up is to a point, possible.

He doesn't have technical knowledge about electric cars either. Yet here we are with Tesla, that single-handedly made electric cars viable to masses. Previous to that electric cars barely got 100km, and couldn't outperform a Prius, with a design that was solely optimized for weight and looked like a electric bicycle with a hood. Delays and problems they are facing doesn't change the fact his company achieves things that others couldn't. Yes, not all his claims and goals come to fruition. Yet maybe it's that ambition and dreams that allows it to get ahead of others, because others have much less of a target to attain.

Why are we acting like Musk's leadership revolutionized the EV industry? Using massive amounts of small, pre-existing batteries to make an electric car, first starting with a converted Lotus, then moving on to a low volume, high performance, luxury sedan, then using that platform for a luxury crossover, then following up with a low cost platform to build a high volume smaller sedan and smaller crossover with a lot of parts commonality was 100% Martin Eberhard and Marc Tarpenning's plan and design.

Musk provided the funding and then eventually ousted the two, just executing on the designs and decisions they pioneered. He occasionally interfered, like insisting on Gullwing doors that led to delays and cost overruns. Once the original founders' plan was executed and the decision making was 100% on Musk, we got such brilliant strategic decisions like an 18-Wheeler, removing LIDAR, and the Cyber Truck.

Also can't forget his brilliant plan to fix traffic: Divert public funding from rail projects to build single lane, one-way 25mph roads underground with a maximum throughput of 4,500 passengers an hour...populated only by his company's cars

Musk succeeds most when he provides funding, hires brilliant staff, let's them do most of the decision making, and then use his mass cult of personality to provide his ventures a ton of publicity. His companies seemed to all operate just fine without him while he worked full time in DC for half a year.

If Musk is able to actually start a new fab company from the ground up and directly compete successfully with TSMC in density and performance, I'll post myself eating my own socks

 

[jdubs03]

Keep an eye out on this. Could be quite the snafu if it has legs.

https://twitter.com/x/status/1990961575398945227

 

[511]

Keep an eye out on this. Could be quite the snafu if it has legs.

https://twitter.com/x/status/1990961575398945227

> Stole TSMC Tech
What if the tech was something he developed

 

[Josh128]

Why are we acting like Musk's leadership revolutionized the EV industry? Using massive amounts of small, pre-existing batteries to make an electric car, first starting with a converted Lotus, then moving on to a low volume, high performance, luxury sedan, then using that platform for a luxury crossover, then following up with a low cost platform to build a high volume smaller sedan and smaller crossover with a lot of parts commonality was 100% Martin Eberhard and Marc Tarpenning's plan and design.

Musk provided the funding and then eventually ousted the two, just executing on the designs and decisions they pioneered. He occasionally interfered, like insisting on Gullwing doors that led to delays and cost overruns. Once the original founders' plan was executed and the decision making was 100% on Musk, we got such brilliant strategic decisions like an 18-Wheeler, removing LIDAR, and the Cyber Truck.

Also can't forget his brilliant plan to fix traffic: Divert public funding from rail projects to build single lane, one-way 25mph roads underground with a maximum throughput of 4,500 passengers an hour...populated only by his company's cars

Musk succeeds most when he provides funding, hires brilliant staff, let's them do most of the decision making, and then use his mass cult of personality to provide his ventures a ton of publicity. His companies seemed to all operate just fine without him while he worked full time in DC for half a year.

If Musk is able to actually start a new fab company from the ground up and directly compete successfully with TSMC in density and performance, I'll post myself eating my own socks

You missed the most important part scaling up manufacturing and making the business profitable. Being a brutal boss about worker productivity, working 7 days a week on the production lines, etc. Creating publicity and increasing the market cap 100x from 2014 till now. If you think anyone could do that, we have nothing further to discuss. Not that I believe he can compete with TSMC anytime soon, or ever, but what Musk did with Tesla, SpaceX is undeniable.

 

[Nothingness]

> Stole TSMC Tech
What if the tech was something he developed

You're not the owner of the technology you develop in a company.

 

[511]

You're not the owner of the technology you develop in a company.

But I can develop similar tech at other company if not bound by a clause.

 

[dacostafilipe]

But I can develop similar tech at other company if not bound by a clause.

Not too "similar" as that's protected, even without non-complete.

 

[511]

Not too "similar" as that's protected, even without non-complete.

why do you think company poaches employee from one other for their talent ofc he is an ex Intel Employee as well this logic can be applied their as well

 

[Nothingness]

why do you think company poaches employee from one other for their talent ofc he is an ex Intel Employee as well this logic can be applied their as well

Talents are poached, not what they did. I wrote a patent, my name is on it, but it belongs to my company. If I move to another company and implement it in another CPU, my new company could be prosecuted. I'd have to come up with new dissimilar enough ideas. So basically I couldn't reuse the tech in my patent.

 

[511]

Talents are poached, not what they did. I wrote a patent, my name is on it, but it belongs to my company. If I move to another company and implement it in another CPU, my new company could be prosecuted. I'd have to come up with new dissimilar enough ideas. So basically I couldn't reuse the tech in my patent.

But I can create another one dissimilar and they have not even given the proof that he has actually stolen anything from TSMC

 

[oak8292]

But I can create another one dissimilar and they have not even given the proof that he has actually stolen anything from TSMC

TSMC has already been down this path with Liang Mong Song a prior R&D leader that went to Samsung. Liang is currently the CEO of SMIC.

“Under the conditions that Samsung Electronics offered Liang three years' salary equivalent to his ten years of service at TSMC,[21] Liang agreed to join Samsung[22][23] and at the same time took away more than 20 employees, including those from his old TSMC engineering department.[20][18]

To comply with the period stipulated in the non-compete clause, Liang began to work as a visiting professor at Sungkyunkwan University under Samsung in October 2010, and actually taught at Samsung's internal corporate training university - Samsung Semiconductor Institute of Technology.[24] On 13 July 2011, Liang officially joined Samsung as the chief technology officer of Samsung's LSI department and was also the executive vice president of Samsung's wafer foundry.[25]

At that time, Samsung was at the R&D bottleneck of switching from the 28 nm process to the 20-nanometer process. Liang advocated that Samsung abandon the 20 nm process and directly upgrade from the 28 nm process to the 14 nm process.[21] In the end, Samsung's 14 nm process was mass-produced about half a year earlier than TSMC.[26]”

Liang Mong Song - Wikipedia

en.wikipedia.org

From the perspective of ten years in hindsight it does look like Liang took a lot to Samsung about FinFETs. It does not look like his research methods stuck and it was just the FinFET tech.

 

[Nothingness]

But I can create another one dissimilar and they have not even given the proof that he has actually stolen anything from TSMC

Yes, definitely. But you have to be very careful about the ideas you bring to your new company and be sure you don't infringe on the patents your previous company had. OTOH beyond patents some things you hear or read in a company are under restrictions too and you should not use these information. Ha well, I hope I won't change company before retirement, that'd be hard to keep everything secret

And in the end, as you said, it doesn't seem there's a proof any such thing happened in that case. Moving from a company to a competing company doesn't imply some secret was stolen.

 

[Doug S]

Yes, definitely. But you have to be very careful about the ideas you bring to your new company and be sure you don't infringe on the patents your previous company had. OTOH beyond patents some things you hear or read in a company are under restrictions too and you should not use these information. Ha well, I hope I won't change company before retirement, that'd be hard to keep everything secret

And in the end, as you said, it doesn't seem there's a proof any such thing happened in that case. Moving from a company to a competing company doesn't imply some secret was stolen.

I think it is the fact he had a habit of taking handwritten notes during meetings and when he left the company he took boxes full of those handwritten notes with him. That made his intent a bit too obvious IMHO. That is exactly the "some things you hear or read in a company" you mention. Really not all that different than if you left your current employer and took with you a copy of all the emails you'd sent/received and the contents of your Documents folder. It would be impossible to do so without keeping tons of privileged information that you aren't allowed to take with you.

Some companies will track (or even block) stuff like plugging in a storage drive or bulk emailing to external accounts - especially for employees who are leaving - for just this reason. TSMC is pretty paranoid about their secrets, I think it would be safe to guess they have stuff like this in place, making it nearly impossible for someone to take troves of data out the door. Making notes using paper and pencil gets around this - you don't get the same level of detail as if you were do a mass electronic copy but you'd evade any technical measures, so the company would have to rely on "analog" methods like cameras or security guards seeing what you take home with you. But you'd know what information to write down, versus what you can either assume you'll remember, be able to independently recreate, or anyone "skilled in the art" would be able to recreate, to fill in the gaps.

Just knowing "we tried X, Y, and Z before we settled on Y" for such-and-such is valuable info that avoids wasting a lot of time down blind alleys.

 

[Doug S]

On a different subject I thought this was interesting. Not because I care about cryptomining chips, but rather the deep subthreshold operation at 0.26v. Seems like there are a couple gotchas, though they promise the biggest one (SRAM) is fixed in their next version. I wonder if a bigger player might become interested in acquiring them, or licensing/buying the IP for that technology? Rather than targeting the most AI per rack, target the most AI per kilowatt?

https://www.eetimes.com/crypto-mining-asic-goes-deep-sub-threshold-on-3-nm/

 

[adamge]

I think it is the fact he had a habit of taking handwritten notes during meetings and when he left the company he took boxes full of those handwritten notes with him. That made his intent a bit too obvious IMHO. That is exactly the "some things you hear or read in a company" you mention. Really not all that different than if you left your current employer and took with you a copy of all the emails you'd sent/received and the contents of your Documents folder. It would be impossible to do so without keeping tons of privileged information that you aren't allowed to take with you.

Some companies will track (or even block) stuff like plugging in a storage drive or bulk emailing to external accounts - especially for employees who are leaving - for just this reason. TSMC is pretty paranoid about their secrets, I think it would be safe to guess they have stuff like this in place, making it nearly impossible for someone to take troves of data out the door. Making notes using paper and pencil gets around this - you don't get the same level of detail as if you were do a mass electronic copy but you'd evade any technical measures, so the company would have to rely on "analog" methods like cameras or security guards seeing what you take home with you. But you'd know what information to write down, versus what you can either assume you'll remember, be able to independently recreate, or anyone "skilled in the art" would be able to recreate, to fill in the gaps.

Just knowing "we tried X, Y, and Z before we settled on Y" for such-and-such is valuable info that avoids wasting a lot of time down blind alleys.

This brings back memories of Snowden.

 

[maddie]

This brings back memories of Snowden.

Nah, Bolton.

 

[DrMrLordX]

I'm hearing noise that Japan effectively cut off China from high-end photoresist compounds which will cripple semiconductor production there. Is there any truth to this?

 

[maddie]

I'm hearing noise that Japan effectively cut off China from high-end photoresist compounds which will cripple semiconductor production there. Is there any truth to this?

Cut off nose to spite face? Why did Dutch return Nexperia to Chinese control? Now Japan doing this? Hard to believe.

 

[DrMrLordX]

Cut off nose to spite face? Why did Dutch return Nexperia to Chinese control? Now Japan doing this? Hard to believe.

There are other factors in play. But. Politics aside, removing Japan's photoresist supply would shut down most of the PRC's fabs. It's not a material that can be stockpiled.