TSMC 3nm Process Packs 250 Million Transistors Per Square Millimeter

[btarunr]

Imagine being able to shrink a Pentium 4 processor die to the size of a pin-head (if you can figure out how to place 478 bumps on it). TSMC revealed that its future 3 nanometer silicon fabrication node has a development target of 250 million transistors per mm². Called N3, the next-generation silicon fabrication node succeeds TSMC's N5 family of 5 nm-class nodes (that's N5 and any possible refinements).

TSMC CEO CC Wei confirmed that development of the 3 nm node is on-track, with risk production scheduled for 2021 and volume production commencing in the second half of 2022. Perhaps the most startling revelation is that TSMC has decided to stick with FinFETs for N3 owing to the maturity of the technology. Experts are of the opinion that sub-5 nm nodes will require major innovations with materials and structures. TSMC claims that N3 will provide a 10-15% speed improvement at iso-power or 25-30% power reduction at iso-speed, compared to N5.



View at TechPowerUp Main Site

 

[zlobby]

 

[Ferrum Master]

Hats off...

 

[Fluffmeister]

Certainly impressive, but I kinda miss the good old days when real men had fabs.

 

[pjl321]

That is pretty surprising, I was pretty sure TSMC would go the GAAFET route like Samsung.
It will be interesting see these two companies take different approaches with roughly the same timescales, although Samsung has yet to make me believe they can deliver.

 

[Ferrum Master]

Certainly impressive, but I kinda miss the good old days when real men had fabs.

I miss the old days when we had a competion of UMC and TSMC on the same product...

 

[birdie]

I miss the old days when we had a competion of UMC and TSMC on the same product...

I miss Intel leading the fabrication process. They are so behind now I'm actually frightened for them.

 

[Ferrum Master]

I miss Intel leading the fabrication process. They are so behind now I'm actually frightened for them.

Ke? Why it should bother you? I wish good luck to any competotor out there, thus having balance.

Intel leadership did not do any good. It brang stagnation to the company.

 

[birdie]

Ke? Why it should bother you? I wish good luck to any competotor out there, thus having balance.

Intel leadership did not do any good. It brang stagnation to the company.

That's exactly the issue Intel is having right now: they are not in contention and they haven't been for a year now while in the past they were usually two years ahead of the rest of the industry.

 

[pjl321]

I miss Intel leading the fabrication process. They are so behind now I'm actually frightened for them.

Why? They did nothing with it.
They just kept giving us quad core after quad core. If we were lucky we got a 5% boost from one generation to the next!
Watching Intel running around trying to caught up is great, I think Rocket Lake and Alder Lake won't be too bad, they will keep AMD within reaching distance, until Intel can get a proper new CPU out on their 7nm process. Then we will see some competition again.

 

[midnightoil]

If AMD are using N5P for Zen 4, it should be an absolute bloodbath vs what ever unholy mix of 14nm+++++++++++ and 10nm product stack Intel are going to have in 2021 and most of 2022.

Zen 2 @ 7nm --> [N5 +15% perf -> N5P +7% perf] = Zen 3

14nm++++++++++++++ is big, and insanely power hungry. It's also questionable if it'll keep a clock advantage over TSMC's N5P EUV.

10nm and 10nm+ are a bit less power hungry, but hit the same kind of clocks that GF 14nm did.

7nm ain't shipping in 2021. Wouldn't be surprised if it's late 2022 now (Covid-19 blamed). Whilst it's expected to clock better than Intel 10nm, I don't think it's expected to get anywhere near 14nm+++++++, so TSMC will probably have a significant clock lead on N5P (or 3nm depending on when Intel 7nm ships).

Can only see it getting worse for Intel, until Keller's new architecture debuts in 2023 or later.

 

[Ferrum Master]

Can only see it getting worse for Intel, until Keller's new architecture debuts in 2023 or later.

Absolutely... by no means faster. It just does not work other way. 5 years are the cycle of bringing new arch.

 

[pjl321]

If AMD are using N5P for Zen 4, it should be an absolute bloodbath vs what ever unholy mix of 14nm+++++++++++ and 10nm product stack Intel are going to have in 2021 and most of 2022.

Zen 2 @ 7nm --> [N5 +15% perf -> N5P +7% perf] = Zen 3

14nm++++++++++++++ is big, and insanely power hungry. It's also questionable if it'll keep a clock advantage over TSMC's N5P EUV.

10nm and 10nm+ are a bit less power hungry, but hit the same kind of clocks that GF 14nm did.

7nm ain't shipping in 2021. Wouldn't be surprised if it's late 2022 now (Covid-19 blamed). Whilst it's expected to clock better than Intel 10nm, I don't think it's expected to get anywhere near 14nm+++++++, so TSMC will probably have a significant clock lead on N5P (or 3nm depending on when Intel 7nm ships).

Can only see it getting worse for Intel, until Keller's new architecture debuts in 2023 or later.

Jim had quite a hand in Golden Cove and a full Keller architecture is Ocean Cove. I think you will be surprised how fast Intel comes out with 7nm. They are already on the second iteration of the process and things are looking massively better than 10nm as all the masking issues are gone.
I would love Zen 4 to land in 2021 but I can't see that happening with the delays Zen 3 is having plus the power issues 5nm appears to have currently.

 

[evernessince]

Jim had quite a hand in Golden Cove and a full Keller architecture is Ocean Cove. I think you will be surprised how fast Intel comes out with 7nm. They are already on the second iteration of the process and things are looking massively better than 10nm as all the masking issues are gone.
I would love Zen 4 to land in 2021 but I can't see that happening with the delays Zen 3 is having plus the power issues 5nm appears to have currently.

Jim Keller is a great CPU architect but as the man himself has said, he's just one guy on a team. Ultimately, Keller alone won't save Intel, it's a team effort. He gave major props to AMD engineers when he left which is why AMD is still projecting nice improvements on architectures Jim had no hand in.

In any case it doesn't seem to matter how fast Intel comes back. They have a tight grip on the OEM market apparently regardless of competition.

 

[zlobby]

I miss Intel leading the fabrication process. They are so behind now I'm actually frightened for them.

Why would you be actually frightened for them?

Jim Keller is a great CPU architect but as the man himself has said, he's just one guy on a team. Ultimately, Keller alone won't save Intel, it's a team effort. He gave major props to AMD engineers when he left which is why AMD is still projecting nice improvements on architectures Jim had no hand in.

In any case it doesn't seem to matter how fast Intel comes back. They have a tight grip on the OEM market apparently regardless of competition.

Well, Keller sort of shot himself in the foot, IMO. Even back then it was clear intel don't care for superior products. All they care abouy is undermining the competition with shadowy tactics. They didn't care about having a trully superior product, only how it will look in benchmarks.

 

[pjl321]

Jim Keller is a great CPU architect but as the man himself has said, he's just one guy on a team. Ultimately, Keller alone won't save Intel, it's a team effort. He gave major props to AMD engineers when he left which is why AMD is still projecting nice improvements on architectures Jim had no hand in.

In any case it doesn't seem to matter how fast Intel comes back. They have a tight grip on the OEM market apparently regardless of competition.

Every company he has been at he has created amazing chips that changed the course of the company and computing, original Athlon, MIPS, Tesla chip, Apple mobile chips...

 

[birdie]

Why would you be actually frightened for them?

FML, seriously, people? You're the second person in this thread who's asking this question. Don't y'all like viable competition? Or you prefer AMD to become Intel like they did, albeit for a short while back in the days, when they released the AMD64 architecture which trumped Intel Prescott? Have you all forgotten the Athlon 64 FX-62 CPU which was released for staggering $1032 in 2006? A mere dual core CPU which was only slightly faster than other AMD64 CPUs due to its frequency? It would be great if the obsession with hating Intel and licking the boots of AMD stopped but as far as I can see we are very far from it. Don't y'all think for a second that AMD is in here for giving you cheap fast CPUs because they love you. No, competition made them do so. If Intel stops competing AMD will suddenly become slow to innovate and will increase their prices - you can trust me on this.

 

[$ReaPeR$]

FML, seriously, people? You're the second person in this thread who's asking this question. Don't y'all like viable competition? Or you prefer AMD to become Intel like they did, albeit for a short while back in the days, when they released the AMD64 architecture which trumped Intel Prescott? Have you all forgotten the Athlon 64 FX-62 CPU which was released for staggering $1032 in 2006? A mere dual core CPU which was only slightly faster than other AMD64 CPUs due to its frequency? It would be great if the obsession with hating Intel and licking the boots of AMD stopped but as far as I can see we are very far from it. Don't y'all think for a second that AMD is in here for giving you cheap fast CPUs because they love you. No, competition made them do so. If Intel stops competing AMD will suddenly become slow to innovate and will increase their prices - you can trust me on this.

Are you joking? Intel didn't lose their grip on the market during the Pentium era and sure as duck they won't lose it now. They have 80% of the market, so calm down and don't worry for a billion dollar company. Even if they lose that's the name of the game, capitalism, unless you would like to try something else.

 

[TheinsanegamerN]

Every company he has been at he has created amazing chips that changed the course of the company and computing, original Athlon, MIPS, Tesla chip, Apple mobile chips...

Every company he has been at has amazing engineering teams and poach great employees or outright purchase companies that are doing great things. Keller is talented, but he cant do anything with a team that has no interest in competing

 

[pjl321]

Every company he has been at has amazing engineering teams and poach great employees or outright purchase companies that are doing great things. Keller is talented, but he cant do anything with a team that has no interest in competing

Motivating or creating a talented team might be one of his strengths and he might be doing that at Intel right now. Let's not forget the sh!t-show he inherited at AMD when he first arrived, they were in much worse shape than Intel are even now, and AMD had much worse products than Sky Lake but look at them now.

I fully believe Intel will be back, I am really glad AMD have and are having such a great run. I think it's in AMD's hands as to who wins for the next 2 years but starting in 2022 Intel will have some good CPUs again and it will be down to how well AMD have released Zen 3 and Zen 4 and what kind of performance lead they have managed to create during this Intel restructuring phase as to whether Intel can walk straight back into the lead again or whether they need to battle for it.

I would be willing to bet that by the end of 2024 the status quo has returned and Intel has the best best chips in most segments for most workloads though.

The only thing that I can see upsetting this is if AMD manages to nail monster APUs, like I mean proper beasts though. Basically a chip double what the Xbox Series X is but using Zen 4 and RDNA 3.0 on 5nm and with 32GB of HBM3 as the unified system and graphics memory.

 

[Imouto]

Motivating or creating a talented team might be one of his strengths and he might be doing that at Intel right now. Let's not forget the sh!t-show he inherited at AMD when he first arrived, they were in much worse shape than Intel are even now, and AMD had much worse products than Sky Lake but look at them now.

To be honest Intel looks like an elephant graveyard. They went on a hiring/buying spree a while back but I don't think it is going to have much of an impact. I mean, they hired Raja Koduri and Ryan Shrout so I wouldn't be so sure that they know what they are doing.

 

[londiste]

I miss Intel leading the fabrication process. They are so behind now I'm actually frightened for them.

Why would you be actually frightened for them?

Not for them. Us. We definitely do not want a monopoly.
If Intel is behind or out of the race, we only have two competitors left and one of them has not had a good track record lately either.

FML, seriously, people? You're the second person in this thread who's asking this question. Don't y'all like viable competition? Or you prefer AMD to become Intel like they did, albeit for a short while back in the days, when they released the AMD64 architecture which trumped Intel Prescott? Have you all forgotten the Athlon 64 FX-62 CPU which was released for staggering $1032 in 2006? A mere dual core CPU which was only slightly faster than other AMD64 CPUs due to its frequency? It would be great if the obsession with hating Intel and licking the boots of AMD stopped but as far as I can see we are very far from it. Don't y'all think for a second that AMD is in here for giving you cheap fast CPUs because they love you. No, competition made them do so. If Intel stops competing AMD will suddenly become slow to innovate and will increase their prices - you can trust me on this.

While I agree with the point, you are looking at the wrong competition. When we talk about semiconductor manufacturing (in a thread about 3 nm process on TSMC's roadmap) its TSMC, Samsung and Intel. All of them are putting $10+ billion yearly into R&D and really more in tune of $15 billion. We are not going to get new competitors into that easily.

We all know the state of TSMC and Intel. Samsung has been kind of there with SoCs and seems to get its ducks in a line for 7nm process, albeit quite late. Their presence in high-performance manufacturing has been weak though.

 

[Emu]

Are you joking? Intel didn't lose their grip on the market during the Pentium era and sure as duck they won't lose it now. They have 80% of the market, so calm down and don't worry for a billion dollar company. Even if they lose that's the name of the game, capitalism, unless you would like to try something else.

Intel held on to their market lead during the Pentium 4 era using shady techniques like giving steep discounts to OEMs who only shipped products with Intel CPUs (if a OEM shipped products with other CPUs then they would be forced to pay retail prices for Intel CPUs). Intel is still filing appeals on the court case that they lost about this.

 

[john_]

I miss Intel leading the fabrication process. They are so behind now I'm actually frightened for them.

Intel falling behind in manufacturing technology opened up the doors to AMD and ARM. It was a good thing.

 

[medi01]

TSMC: Oh, hi, Intel, look what I've got!
Intel: <runs away shouting random nonsense>