Article TSMC surpasses 1 billion 7nm chips

[Kenmitch]

That's a whole lot of chips! I wonder if Intels broken 100,000 functional ones yet?

TSMC Celebrates Making Over a Billion 7nm Chips

That’s enough silicon to cover 13 Manhattan city blocks.

www.tomshardware.com

Each of TSMC’s 7nm chips also has at least one billion transistors inside it, meaning that the company has made more than one quintillion 7nm transistors overall.

 

[A///]

And 3nm is a go for 2H22. Exciting times. Or maybe exciting energetic times (EE Times)... I'll see myself out.

 

[NTMBK]

And 3nm is a go for 2H22. Exciting times. Or maybe exciting energetic times (EE Times)... I'll see myself out.

I'll believe it when I see it! TSMC have been on a roll lately, but shrinks are getting incredibly hard. It wouldn't surprise me if any of the remaining big players stumbles.

 

[JoeRambo]

Hopefully it will be better than 20nm tech that was end of road for planar. Not being able to properly scale down SRAM? Interesting times ahead, certainly a chance for both Samsung / Intel to make those innovative transistors work ( or fail completely ).

 

[A///]

I'll believe it when I see it! TSMC have been on a roll lately, but shrinks are getting incredibly hard. It wouldn't surprise me if any of the remaining big players stumbles.

If shrinks were relative to the display name, you'd have a point. TSMC could increase density by 1.3x and still call it 3nm when their previous shrinks and density increases hovered around 1.5-1.8 or so. A more conservative approach than Intel's 2x shrink per node with a large overbite for 10nm. As you would recall, the 14nm to 10nm plan was a 2.7x increase in density while going to a shrunk node. Intel felt overly confident. Intel went beyond their normal shrink and increase for 7nm and failed that, too, unless a simple bug is what set them a year behind. Or if that's placating investors.

TSMC has been patenting aggressively the ever since their bungle. TSMC's future is their major premiere partners. Apple, AMD, Qualcomm, et al. It's healthy to be skeptical, just as I'm skeptical Intel will get their chiplets out for "client 2.o" or whatever they claimed on their Architecture Day. I'll believe it when I see it, so far Intel's been lying for nearly 6 years.

I'm not financially or ethically beholden to either company. I haven't owned an AMD product in 20 years. I prefer seeing physical products. So far Intel's got nothing except a few lame 10nm laptops.

 

[Doug S]

I'll believe it when I see it! TSMC have been on a roll lately, but shrinks are getting incredibly hard. It wouldn't surprise me if any of the remaining big players stumbles.

Given how well TSMC has executed the last few nodes I don't see why you doubt them. Their announcement is saying they've figured out how they're going to mass produce at the given specs - risk production will begin next year. Figuring how to do it and actually doing it are different things, but they appear to be doing better (i.e. lower initial defect density for N5 vs N7, and N7 vs N10) with each step.

Some might complain well "it isn't a full shrink" since they only went 1.8x with N5 and are going 1.7x with N7. Do the math, a 1.7x shrink every two years is better than a 2x shrink every three so there is little to be gained by trying to be more aggressive but taking longer to roll it out as a result. They seem to be pretty well aware of how much they can bite off, and aren't shooting for the moon in their shrink and by making FinFET work for another generation they are giving themselves more time to get GAA or nanosheets figured out for 2nm in 2024.

 

[CluelessOne]

1 billion chips. That's a lot of phones. How many for GPUs and CPUs?

 

[Vattila]

Some might complain well "it isn't a full shrink" since they only went 1.8x with N5 and are going 1.7x with N7. Do the math, a 1.7x shrink every two years is better than a 2x shrink every three so there is little to be gained by trying to be more aggressive but taking longer to roll it out as a result.

True. Unfortunately for Intel, being aggressive is inherent in their vertical business model. A big point of owning your own fabs is to drive down cost and achieve higher margins. They need to make each process node sufficiently significant to make sense (cost-effective). Since they do little manufacturing for others, it also needs to be perfectly scaled to their future product needs. As a consequence, any development (or contingency planning) that involves spending and lower margins is very painful for Intel. As competition starts to bite, they are more inclined to do cost reductions and return excess cash to shareholders in the form of dividends and share buyback.

 

[moinmoin]

True. Unfortunately for Intel, being aggressive is inherent in their vertical business model. A big point of owning your own fabs is to drive down cost and achieve higher margins. They need to make each process node sufficiently significant to make sense (cost-effective). Since they do little manufacturing for others, it also needs to be perfectly scaled to their future product needs. As a consequence, any development (or contingency planning) that involves spending and lower margins is very painful for Intel. As competition starts to bite, they are more inclined to do cost reductions and return excess cash to shareholders in the form of dividends and share buyback.

Indeed, very well summarized. On the one hand just as you write it's pretty logical and completely predictable how Intel is behaving up to now, which started half a decade ago imo. On the other hand the whole prolonged process is taking way too long but is still way too ingrained, making it more and more possible that Intel's business crosses the tipping point after which building fabs for new nodes just isn't financially feasible anymore. I really hope Intel finds a way out of this conundrum, but so far Intel appears to fail to actually recognize the full extend of the situation it's in.