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.
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
> And contamination would be a real problem at this scale
Copper is a monster. It's basically plague rats for a fab. You essentially have to split every line into two logical islands (Cu and non-Cu). This automatically complicates anything by an additional dimension.
> there are ordinary looking trucks puttering about on Taiwan's highways with likely tens of millions of dollars of chip inventory sitting inside. Like what happens if one of them gets into a crash? Will Nvidia miss earnings because of this?
Wow that's amazing. When he was talking about needing to move wafers around I was thinking surely they can't just like truck them around but nope, that's the whole thing
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)
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
> And contamination would be a real problem at this scale
Copper is a monster. It's basically plague rats for a fab. You essentially have to split every line into two logical islands (Cu and non-Cu). This automatically complicates anything by an additional dimension.
> there are ordinary looking trucks puttering about on Taiwan's highways with likely tens of millions of dollars of chip inventory sitting inside. Like what happens if one of them gets into a crash? Will Nvidia miss earnings because of this?
Wow that's amazing. When he was talking about needing to move wafers around I was thinking surely they can't just like truck them around but nope, that's the whole thing
I believe they're packed in such a way that those trucks can get into very serious wrecks without damage to the wafers.
https://patents.google.com/patent/US10217655B2/en
[dead]