TSMC set to begin mass production of 3nm chips

[Daniel Sims]

In context: While everyone in the tech industry is already looking forward to CES at the beginning of January, TSMC has one more big announcement before the end of 2022. The start of 3nm semiconductor production comes as no big surprise, but the company's announcement might be a bid to calm geopolitical tensions.

Taiwanese news outlets are reporting that TSMC will hold a ceremony on December 29 to officially announce the launch of 3nm mass production. Apple will be the primary customer for the first two series of 3nm semiconductors from the Tainan Fab 18, which are expected to go into upcoming products like the iPhone 15.

TSMC will start the New Year with a limited-capacity N3 node process before moving to the more stable and efficient full-production N3E later in 2023, followed by N3P in 2024. That year, TSMC will bring its 2nm GAA process into trial production at a Hsinchu plant for mass production in 2025.

The announcement will likely dispel talk of poor yield rates for 3nm. It could also be an attempt to quell Chinese concerns about TSMC's perceived drift toward the west.

Must read: How did TSMC get so good?

The chip maker is set to invest $40 billion into two Arizona fabs which will start manufacturing 4nm semiconductors in 2024 and 3nm nodes later on. Chinese media reacted angrily to the development, accusing the US of tricking TSMC into the deal and stealing technology from "our Taiwan region." China claims authority over the self-governing island.

Celebrating the opening of the Tainan fab communicates that the company's bleeding-edge node development remains in Taiwan. In addition to the Hsinchu 2nm node, another TSMC facility in Taiwan is charting a course to 1nm, which could arrive before the decade closes.

Samsung began manufacturing 3nm semiconductors in June and is preparing a more energy-efficient 3nm node for 2024. The company's chips will be used in Nvidia graphics cards, Qualcomm mobile processors, IBM CPUs, and Baidu cloud server chips. Samsung is currently a distant number two, behind TSMC, in terms of global semiconductor market share.

Meanwhile, Intel is trying to catch up with its 20A "Ångström" node, planned for 2024. The company broke ground on a new Ohio facility in September which is expected to start up in 2025. Intel hopes its current roadmap will make it more competitive with Samsung and TSMC by then.

Permalink to story.

https://www.techspot.com/news/97103-tsmc-set-begin-mass-production-3nm-chips.html

 

[koblongata]

And then nVidia will still manage to introduce a 500W graphics card based on 3nm

 

[toooooot]

If it means GPUs that are x2 fast and the same power usage that would be great.

 

[Srksi]

Oven is ready, lez get cookin!

 

[Srksi]

And then nVidia will still manage to introduce a 500W graphics card based on 3nm

Make it 1000! Nice round number for future!

 

[Mr Majestyk]

Apple is brain dead. Bleeding edge node costs a bomb per wafer. Prices will have to rise (again). Nvidia made the same stupid mistake with Lovelace paying premium for 4nm node rather than use the refined 5nm node and why we end up with $1200 4080.

 

[Burty117]

Apple is brain dead. Bleeding edge node costs a bomb per wafer. Prices will have to rise (again). Nvidia made the same stupid mistake with Lovelace paying premium for 4nm node rather than use the refined 5nm node and why we end up with $1200 4080.

Didn't AMD use 5nm and 6nm but still cost the same?

Here in the UK at least, 7900XTX is £1150 and 4080 is £1200.

 

[neeyik]

Here in the UK at least, 7900XTX is £1150 and 4080 is £1200.

Good luck even finding a 7900 XTX for sale in the UK; the few places that do have them, list them well over MSRP. Disappointingly, it's actually cheaper to get a 4080 or 3090 Ti.

 

[Burty117]

Good luck even finding a 7900 XTX for sale in the UK; the few places that do have them, list them well over MSRP. Disappointingly, it's actually cheaper to get a 4080 or 3090 Ti.

Good to see using that "cheaper and more mature" 5nm and 6nm process has really paid off!

It's a shame though, Very hard to go the AMD route here when the better product is basically the same price.

 

[neeyik]

Good to see using that "cheaper and more mature" 5nm and 6nm process has really paid off!

Probably has for AMD; not so much for the end user, though.

 

[bexwhitt]

Didn't AMD use 5nm and 6nm but still cost the same?

Here in the UK at least, 7900XTX is £1150 and 4080 is £1200.

nobody has the 7900XTX in the UK so prices are meaningless.

 

[Burty117]

nobody has the 7900XTX in the UK so prices are meaningless.

My point was, everyone keeps banging on about 5nm vs 4nm savings, but it hasn't done anything to save the customer money.

Everyone also keeps saying that because the 5nm process is more mature they should have less failed chips etc... Yet stock is lower than Nvidia's on 4nm?

Just pointing out what people say vs what actually happened in real life.

 

[Ludak021]

Are they ready to produce 3nm SRAM lol NOPE. Not even close.
"let’s not confuse ourselves with the name of the node with what the technology actually offers.” - Philip Wong

 

[waclark]

Good luck even finding a 7900 XTX for sale in the UK; the few places that do have them, list them well over MSRP. Disappointingly, it's actually cheaper to get a 4080 or 3090 Ti.

From the recent news about the XTX it looks like no one will be getting them or, perhaps, no one should be getting them until they resolve the temperature issue. That's sure to help the 4070Ti sell well.

 

[HardReset]

From the recent news about the XTX it looks like no one will be getting them or, perhaps, no one should be getting them until they resolve the temperature issue. That's sure to help the 4070Ti sell well.

If that was only issue on reference cards, it does not affect cards with custom cooling. And near future almost all cards will eventually be custom ones.

 

[waclark]

If that was only issue on reference cards, it does not affect cards with custom cooling. And near future almost all cards will eventually be custom ones.

That appears to be the case, though I'm not sure all the results are in yet. I think some AIBs use the reference design as well, but many do not.