Tapeouts per year vs Process Node

[Marshall]

I have a thought-provoking question for the SemiWiki audience, particularly interesting in light of TSMC OIP next week:

Where can I find data that summarizes where tapeouts occur (or are anticipated to occur) sorted by

1) foundry,
2) node,
3) geographical location,
4) Company (top 10 -15 largest companies responsible for the majority of the tapeouts)

Regards,
Marshall


(source Semiconductor Technology Nodes ? History, Trends and Forecast)

 

[Paul McLellan]

That data is probably completely unavailable. Here is data Synopsys uses on tapeout by node, which is probably as detailed as you can get. They only track the first 500 (?) designs in any node, so there isn't really an abrupt falloff on the old nodes, people are still designing in 28nm (and 90nm too).

View attachment 15307

 

[IanD]

There is an abrupt falloff in 20/22nm -- lots of tapeouts (and still going up) in 28nm (lowest cost), but rapidly accelerating 16/14nm (lowest power) has killed 20/22nm stone dead.

But since 16/14nm is really just 20/22nm with FinFETs, if you add the two together it looks much more like all the other ramp-ups. What a pity the industry tried to pull the wool over everyone's eyes by calling the FinFET processes 16/14nm...