TSMC Uses Old Fabs to Make New Chips [video]

Discussion (3 Comments)Read Original on HackerNews

ggm•about 4 hours ago

I always wondered why there wasn't more visible "second string" production. In the current market, I get that for a given purity of silicon you make more money sending it out the door as AI chips, but the demand side for memory and other things surely would justify taking a fab line on granularity generation -2 or -3 and churning out older, but high yield things.

It has been pointed out to me when I say this that you cannot just switch a fab designed for VLSI gates in CPU into giant RAM making machines: The fabs all have their own peculiarities. They need to be tuned. Nothing is free, nothing is quick. The number of layers, the reagents, the specific dopands, they all vary. And contamination would be a real problem at this scale so purging a line made for process-A chemicals to become process-B must be quite an exercise too.

I believe current RAM and SSD stuff is very multi layered too. The older lines might just not "do" that amount of stacking.

(Not a VLSI engineer or a brain surgeon)

smb06•about 3 hours ago

Former VLSI engineer here. Fab process engineering is a highly tuned artifact. You are talking about getting down to a few nanometers. You are talking less than 100 atoms at that scale. Changing anything requires reworking the whole fab to be highly tuned again. Back in my day (over a decade ago) you would run any process for months before being confident that it wasn't going to vary too much

josh-wrale•about 1 hour ago

I wonder how TSMC in USA manages without the historical pool of prior node fabs...

TSMC Uses Old Fabs to Make New Chips [video]

⚡ Community Insights

Discussion (3 Comments)Read Original on HackerNews