The race between Intel, Samsung, and TSMC to ship the first 2 nm chip | Hacker News Reader

The race between Intel, Samsung, and TSMC to ship the first 2 nm chip

arstechnica.com

20 pointshimaraya23 comments

What's the avg spacing on atoms in a silicon matrix anyway? Something on the order of 0.2nm (2A)? So are we now nearing a point where 1 extra atom is 10% error in a single dimension?

Would be grateful if someone who actually understands this stuff would care to illuminate my speculation further?

2nm is a commercial name that doesn't have much to do with actual nanometers. The actual distance between transistors (gate pitch) in the 2nm process is expected to be 45nm.

I’m all team America now. Having a lot of our chips made in a factory on the other side of the world is appalling and hugely distressing.

I'm currently investing in Intel as a way of hedging the risk of Chinese invasion of Taiwan. Once the TSMC factories are bombed the competition for Intel will basically disappear overnight.

I'm investing in INTC as well, but I really hope this bombing and war never happen and would rather my INTC plummets...

Samsung exists. China won't invade any time soon though.

When they get below "1nm" what measure will the switch to; and will it have even less relationship with any reality?

The nm nomenclature is nominal and not an actual measurement. i It's a concept of the level of improvement. Also not what TSMC uses

https://www.pcgamer.com/intel-renames-process-nodes-intros-a...

Sounds like bullshit has replaced a once useful physical measurement. They should've found a new, real and objective measurement or just given up and used a date or arbitrary version scheme.

The shape and construction of the gates and transistors changed. Minimum feature size ceased to correlate with density/performance in some senses. This scheme has been somewhat arbitrary since about 20nm or so afaik, but expectations are set around what the number communicates, so here we are.

Intel is using angstroms, so their "2nm" node is "20A" and the "1.8nm" node is "18A". These of course still don't have any more relationship with the physical size of the transistors than previous Intel, TSMC, or Samsung nodes but it keeps it relatively clean without needing a decimal point in the name.

Why is Intel on this list?

Why not? As they all intend to ship their version of 2nm in ~2025.

I thought Intel was getting out of the fab game and outsource to tsmc

The design (term?) half of Intel is allowed to outsource to other fabs, the fab half is allowed to seek other design customers.

Is intel anywhere near 2? I thought they’re stuck somewhere in 7/10 land

3 and 2 from TSMC are divorced from the actual component spacing - they’re all pretty close in terms of actual technological advancements, they just don’t all use the same marketing for process names.

MHz and Megapixel race all over again!

Intel 4nm ( Meteor Lake ) is coming out soon ( in a few days time ), Intel 3nm expected next year. And Intel 2nm in (late) 2025. So in a similar time frame as TSMC.

I hope Intel and Samsung will have much success, because TSMC betrayed customers by exclusively providing all top-level manufacturing capacity only to Apple and handling anyone else as a second class citizen.

Everyone else is a second class citizen. Apple brings in more than TSMC's next 3 largest customers combined.

Uhhh they provide plenty to anyone who bids. Nvidia uses them.

TSMC's success is partially that they're the only company who promises not to compete with their customers, because they don't make their own chips like Intel and Samsung do.

Reply on news.ycombinator.com