TSMC 2 nm Wafer Output Projected to Reach 80,000 Units Per Month, by End of 2025

[T0@st]

Earlier in the year, we heard about TSMC being ahead of the game with its speculated trial production run of cutting-edge 2 nm (N2) silicon. Taiwan's premier foundry company is reportedly prepping its Baoshan and Kaohsiung plants for full-on manufacturing of next-gen chips. The latest insider whispers propose that TSMC is making "rapid" progress on the 2 nm (N2) front, as company engineers have moved onto an "intensive" trial production phase. Taiwan's Economic Daily News has picked up on compelling projections from industry moles; the Hsinchu Baoshan facility's current monthly production capacity is (allegedly) around 5000 to 10,000 2 nm wafers. The other 2 nm-specialist site—Kaohsiung—has reportedly moved into a small-scale appraisal phase.

TSMC declined to comment on recently leaked data points, but they released a general statement (to UDN), emphasizing that: "(our) 2 nm process technology is progressing well and will go into mass production as scheduled in the second half of this year." The Baoshan plant could ramp up to 25,000 2 nm wafers per month, once it moves into a mass production phase. Combined with the same estimated output from its sister site (Kaohsiung), insiders reckon that the combined total could reach 50,000 units per month. Following a predicted successful "second phase" transition, TSMC's most advanced facilities have a "chance" to pump out 80,000 2 nm parts (combined total). The latest murmurs suggest that this milestone could be achieved by the end of 2025. Industry watchdogs believe that Apple will have first access dibs on TSMC's upcoming cutting-edge offerings.




View at TechPowerUp Main Site | Source

 

[Space Lynx]

I hope they do a 2nm mac mini by the end of 2026, I will 100% grab that for a permanent desktop work/business rig.

 

[freeagent]

4nm is kind of a bear to cool when pushed close to its limits.. I can only imagine what 2nm will be like.. should clock well though

 

[kondamin]

I hope they do a 2nm mac mini by the end of 2026, I will 100% grab that for a permanent desktop work/business rig.

It’s not going to be magically better than the m4 just because it’s made on an other node.
and I kinda remember reading apple not moving to 2nm right away.

 

[Space Lynx]

It’s not going to be magically better than the m4 just because it’s made on an other node.
and I kinda remember reading apple not moving to 2nm right away.

I'm in no rush. My HP OLED laptop is good enough anyway. I do eventually want to sell it on though and get a Mac Mini for work, as Apple products are what the company I work for uses, so it would just make things 'easier' so to speak.

 

[N/A]

N3 plus 10-15% clock speed and density. This could mean 4GHz for GPU and 5.5-6.0 GHz CPU. N3 compared to N4 is plus 15% clock and 60% logic but 0% SRAM, 30% overall.

 

[john_]

34000 for Apple, 34000 for Nvidia, 2000 with lottery tickets, for anyone else rich enough to pay the prices.

 

[Denver]

In designs with limited SRAM, 2nm can deliver nearly a 2x density improvement over 4nm, Modern GPUs should see a +60-70% improvement with ease.

However, the high $30K wafer cost deters potential customers, including Apple and Nvidia. For some reason, I believe Zen 6 could use 2nm for the dense cores(Zen6C), given the incredibly small size of the CCDs and AMD's commitment to "always using the best node for the EPYC line moving forward." The Appetite For Datacenter Compute Is Ravenous

 

[Nhonho]

Do Americans understand what a nm is? They are adamant about moving to the International System of Units.

 

[TheGeekn°72]

Could N2 coming to market lower N4 prices ? Idk how it would do that exactly but maybe someone with better knowledge of the market would have an answer for me ?

 

[N/A]

Could N2 coming to market lower N4 prices ? Idk how it would do that exactly but maybe someone with better knowledge of the market would have an answer for me ?

Double the N4 prices.

 

[TechBuyingHavoc]

Do Americans understand what a nm is? They are adamant about moving to the International System of Units.

The 2nm node has no features that are actually at 2nm. The nomenclature stopped indicating the transistor gate length (the traditional measure of what the number of the node should be based on) back in 2011-2013. All the nodes after that have been diverging more and more from the actual transistor gate length.

Americans don't need to know what a nm is or what a angstrom is. It's all marketing fluff at this point anyway. The only thing that matters is benchmarks of the software you care about and what price you are getting it at.

 

[alwayssts]

N3 plus 10-15% clock speed and density. This could mean 4GHz for GPU and 5.5-6.0 GHz CPU. N3 compared to N4 is plus 15% clock and 60% logic but 0% SRAM, 30% overall.

We should get those clocks from 3nm (4/6), if just barely.

As for prices, people say '30k per wafer' right now, but I would imagine by the time most large chips are made on them (probably a good 3-4, if not 6 years from now) it will be closer to current pricing.

So yeah, double at launch (for mostly Apple), but just like 3nm has dropped (to ~20k) it'll probably be current 4/5nm prices before most use it for stuff consumers buy; same will probably be the case with 2nm.

 

[Clawedge]

Does anyone have a side by side comparison on the past few node shrinks? Would like to see the difference visually

 

[Vayra86]

Do Americans understand what a nm is? They are adamant about moving to the International System of Units.

OMG you might have just hacked the internets here.

That's why all these nanometer designations for nodes are off by so much. They're actually nano INCHES!

 

[Nhonho]

OMG you might have just hacked the internets here.

That's why all these nanometer designations for nodes are off by so much. They're actually nano INCHES!

Relax, you're taking the Internet too seriously...