Is 7nm the Last Major Node? | Hacker News Reader

Is 7nm the Last Major Node?

semiengineering.com

30 pointsSemiTom6 comments

I wondered how 7nm compares to the size of single silicon atoms. Google says the van der Waals radius of a silicon atom is 210pm. This translates to around 30 atoms for 7nm, considering about 20% spacing between atoms. However I am not a chemist, so perhaps I am wildly off. Anyone here can give some elucidation?

If I am not mistaken, a 14nm feature is about 20 or so Si atoms wide, so a 7nm feature would be about 10 or so Si atoms wide. If this is correct, clearly the next process node after 7nm (3.5nm, presuming halving each time) would be the last process node that allows an integer breadth (5 Si atoms). Beyond that you'd be down to non-integer feature breadths (or alternatively, non-integer average breadths with wide variances).

In terms of TSMC Roadmap, We have

10nm > 2017 7nm > 2018 7nm+ > 2019 7nm EUV / 5nm > 2020

Is the same thing they said 16nm was the last major node, but TSMC, after 4 - 5 iteration of 16nm, with one of them now called 12nm, has gotten cheaper. And it seems they have plans to further reduce the cost of 16/12nm.

I expect the same for 7nm. Both 12nm and 7nm will be much more affordable in 2020. Right now the uncertainty is 5nm and 3nm. Both are technically achievable, the problem is cost. At the moment 5nm+ cost structure looks like exclusive to certain industry and clients. Unless they bring in 450mm wafer but i have no idea how that is working out.

And 28nm remains the cheap and cherrful solution with no replacement in sight.

Is 28nm still the cheapest when lifetime power consumption is factored into the cost?

Cheapest in terms of Design and Manufacture at Scale. As a matter of fact 28nm prices is going up due to supply constraint, and 22nm ULP is not out and wont be as cost effective yet, and i doubt it will be in the future, it will be a tool for TSMC to keep higher margin.

Probably.

Reply on news.ycombinator.com