TSMC to Begin Producing Chips on 5nm Process Late Next Year

DooKey

[H]F Junkie
Joined
Apr 25, 2001
Messages
10,566
If the news out of Taiwan is true it seems TSMC is going to begin mass production of 5nm chips late next year or early 2020. Annual production is estimated to be over 1M 12-inch wafers. That's a lot of 5nm chips for sure. Of course promises are made to be broken and anything can happen between now and next year, but if they stick to their schedule we'll be getting some nice gains in GPU's and SoC's. I love the progress.

Speaking at a TSMC technology forum in Hsinchu, Wei said the company will pour in NT$750 billion (about US$25 billion) to build a plant in the Southern Taiwan Science Park in Tainan, construction of which already started earlier this year, Wei said.
 

Zarathustra[H]

Extremely [H]
Joined
Oct 29, 2000
Messages
33,137
Wow, 5nm is pretty tiny.

Whats the smallest we can theoretically go on silicon again?
 

mesyn191

2[H]4U
Joined
Jun 28, 2004
Messages
2,983
Whats the smallest we can theoretically go on silicon again?
Potentially under 3nm but the costs are getting so high that in practical terms 5nm might be the smallest we go for a very long time. Some products will probably never even use less than 5nm it seems.

Great article on the issues here: https://semiengineering.com/big-trouble-at-3nm/

tl&dr: big fundamental problems for circuit design at 3nm (industry seems to be going nanosheets instead of nanowires but nothing set in stone yet due to uncertainty), chip design costs alone at 3nm are projected to be ~$1-1.5 billion (for a large high performance GPU like NV would design) and actual production costs will be staggeringly higher with estimated costs for 40K wafer starts per month being ~$15-20 billion.

Expect lots of delays in process development past 5nm and expect much much higher costs no matter what. Personally I'm expecting 7nm and 5nm to each both stick around for longer than 28nm did.
 

mesyn191

2[H]4U
Joined
Jun 28, 2004
Messages
2,983
I think everyone elses 7nm will be equal to intels 10nm, 5nm will be ahead.
For 7nm yes, for 7nm+ TSMC, GF, and Samsung might actually have a nice but not definitive edge overall, but Intel's 10nm probably won't be fixed and in high volume production until late 2019 at best, more realistically mid 2020 it seems.

If TSMC sticks to their schedule than Intel's 10nm will be competing with TSMC's 5nm. And Intel's 5nm is rumored to be delayed as well to 2021-2022 for HVM.
 

umeng2002

Gawd
Joined
May 23, 2008
Messages
923
No.... we're running out of "nm"s

Quantum computers won't save us. A quantum computer is only good at specialized tasks.
 

mesyn191

2[H]4U
Joined
Jun 28, 2004
Messages
2,983
Rumor mill is now saying Intel will now pretty much abandon 10nm and go straight to 7nm. They might release some products on 10nm to recoup some development costs but expect them to be very late and low clocked low end stuff. They're stuck on 14nm+/++ for performance parts until after 2020.

That is why suddenly they're going MCM for server parts and stuff in late 2019 and 2020.
 

Uvaman2

2[H]4U
Joined
Jan 4, 2016
Messages
3,143
The mythical Intel 10nm.
Yeah, they can accuse others of playing with numbers all day long, but it doesn't change the fact that they are falling behind. If anybody's "7nm" parts start shipping in volume any time soon, they will be clearly behind, which sucks.
 
Top