Nvidia wants a divorce from TSMC?
Posted 18th July 2012 at 03:08 AM by abraxalito
This article is a fascinating peep behind the curtain of the current economics of semiconductor production through the eyes of a major IP player - Nvidia. More evidence that Moore's Law is grinding to a screeching halt I suggest.
Nvidia deeply unhappy with TSMC, claims 20nm essentially worthless | ExtremeTech
Nvidia deeply unhappy with TSMC, claims 20nm essentially worthless | ExtremeTech
Total Comments 4
Comments
-
I'm always up for a little armchair analysis of the semiconductor industry. My day job is such that I have to pay a certain amount of attention to microelectronics and fabrication technology. So, as a rebuttal:
1. Even if the transistor cost is the same on the next process node, there is still advantages to going there: smaller chips with lower power consumption.
2. Intel is going there, at least to 14 nm. So as a competing company, either you are too, or you bow out of Intel's market.
3. The "cost per transistor over time" can only be considered a very rough prediction into the future. It could well end up that technology advances reduce the costs at 22, 20 and 14 nm well below what is shown.
4. Nvidia ... is nvidia. They have a well known history of blaming others for their own problems when it suits them.Posted 21st July 2012 at 10:57 AM by rjm
-
Thanks for the stimulating comments. Let me see -
1) Its not necessarily so that its lower power - that was one of the points raised. But yes, smaller chips certainly. But why is smaller an advantage when the cost is the same? Because it allows smaller packages?
2) Intel is going there - but Intel has taken its eye off the ball and its going to pot (see earlier posts in my blog). Intel has become accustomed to large margins and is finding it hard weaning itself off them - classic case like Kodak.
Given that we've seen no advantage in going to 14nm then why bother to compete with Intel at 14nm? Customers don't demand a particular node, they just want functionality at a low enough price with a particular power consumption. They really couldn't give a hoot about what geometry's inside.Posted 21st July 2012 at 03:10 PM by abraxalito
Updated 24th July 2012 at 12:47 AM by abraxalito -
The discussion is predicated on the assumption that we've reached a tipping point where it really is going to be the case that the smaller nodes will cease to be cost competitive.
My point is only that I wouldn't bet on that. Similar comments were heard when 90 nm came out, with all the leakage problems. Then metal gate, high-k dielectrics, what not came on line and things have moved smoothly enough since. Now we are at another sticking point, with a lot of new, untested methods. Things look bad, but its far from impossible that once again predictions based on prior trends will proved to be incorrect.Posted 23rd July 2012 at 11:50 PM by rjm
-
I have made no such assumption, its rather looking like the conclusion though. Not definitive of course, but strongly suggestive.
This does not rely on prior trends to predict, rather its based on customer requirements. I note for example that its the equipment manufacturers pushing 450mm, not the fabs. That's a classic sign of a bubble - the end point demand simply isn't there.Posted 24th July 2012 at 12:45 AM by abraxalito
