The 90 nm node question is still very much in the air. Intel went to great lengths (and many millions of dollars) to get their 90 nm process up and running, AMD is only now starting serious production of 90 nm parts which will show up in late Fall of this year. If you have watched closely in the past, TSMC has followed up full scale production of each node change about 6 to 8 months after AMD.
Now, TSMC does not have the budgets for process tech advancement that Intel and AMD does, even though being a foundry IS their business. Overall product margins for AMD and Intel chips are much higher than the margins that TSMC has by producing a 3rd party chip. Spring of 2005 is when we can expect to see the NV50/R520, but many in the industry are skeptical that the commercial foundries will have a solid 90 nm process up and running for full scale production. I have talked to Intel Fab engineers, and they have stated that the jump from 130 nm to 90 nm was far, far more costly in both time and money than the jump from 180 nm to 130 nm. We all know the problems everyone had getting 130 nm up and running, now multiply those problems by a factor of 3 and you can understand the jump to 90 nm.
Oddly enough, the commercial foundry that will probably have a full scale 90 nm line up and running is IBM. They already have 90 nm parts in full scale production. While yields and speed bins are not good for the process right now, in 9 months the IBM 90 nm should be solid enough to produce high end GPU's at a good price with decent yields and speed bins. NVIDIA has the advantage here by working with IBM. I have heard very little about TSMC's 90 nm node, but what I have heard is that it is nowhere near production quality (and very far away from what IBM is able to do now).
If TSMC has considerable problems with its 90 nm node, most likely they will continue to work on improving the 110 nm node. Currently the transistor performance of the 110 nm node is nearly identical to their 130 nm FSG line, and due to the 110 nm node just shrinking some of the features of that line, yields and speed bins should be nearly identical to the 130 nm FSG. Now, if TSMC wants a stopgap measure to please certain of its clients, it may integrate Low-K into the 110 nm line, which would decrease power draw and heat production, and would allow a chip to run faster (just as it does with the 130 nm Low-K line vs the 130 nm FSG).
IBM is the wild card here. If IBM can get its 90 nm commercial line up to spec and running as well as Intel's or AMD's lines, then NVIDIA could have a leg up on ATI, unless ATI also decides to look into IBM. Still, a large sticking point to working with IBM is that it is a large company, and it has its own products that it wants to produce on its own lines. So if IBM wanted to dedicate most of its 90 nm line to its PowerPC product, then all other 3rd party products will have to fight for space on that line. IBM's products get priority here.
If you want to get an exact answer, I would suggest asking Sireric (though he will not tell you a thing). Other than that, just take a look at the foundry scene and do your best to imagine what is going through the engineers' and bean counters' heads. Going 90 nm is a sticky proposition at this point in time.
Now, TSMC does not have the budgets for process tech advancement that Intel and AMD does, even though being a foundry IS their business. Overall product margins for AMD and Intel chips are much higher than the margins that TSMC has by producing a 3rd party chip. Spring of 2005 is when we can expect to see the NV50/R520, but many in the industry are skeptical that the commercial foundries will have a solid 90 nm process up and running for full scale production. I have talked to Intel Fab engineers, and they have stated that the jump from 130 nm to 90 nm was far, far more costly in both time and money than the jump from 180 nm to 130 nm. We all know the problems everyone had getting 130 nm up and running, now multiply those problems by a factor of 3 and you can understand the jump to 90 nm.
Oddly enough, the commercial foundry that will probably have a full scale 90 nm line up and running is IBM. They already have 90 nm parts in full scale production. While yields and speed bins are not good for the process right now, in 9 months the IBM 90 nm should be solid enough to produce high end GPU's at a good price with decent yields and speed bins. NVIDIA has the advantage here by working with IBM. I have heard very little about TSMC's 90 nm node, but what I have heard is that it is nowhere near production quality (and very far away from what IBM is able to do now).
If TSMC has considerable problems with its 90 nm node, most likely they will continue to work on improving the 110 nm node. Currently the transistor performance of the 110 nm node is nearly identical to their 130 nm FSG line, and due to the 110 nm node just shrinking some of the features of that line, yields and speed bins should be nearly identical to the 130 nm FSG. Now, if TSMC wants a stopgap measure to please certain of its clients, it may integrate Low-K into the 110 nm line, which would decrease power draw and heat production, and would allow a chip to run faster (just as it does with the 130 nm Low-K line vs the 130 nm FSG).
IBM is the wild card here. If IBM can get its 90 nm commercial line up to spec and running as well as Intel's or AMD's lines, then NVIDIA could have a leg up on ATI, unless ATI also decides to look into IBM. Still, a large sticking point to working with IBM is that it is a large company, and it has its own products that it wants to produce on its own lines. So if IBM wanted to dedicate most of its 90 nm line to its PowerPC product, then all other 3rd party products will have to fight for space on that line. IBM's products get priority here.
If you want to get an exact answer, I would suggest asking Sireric (though he will not tell you a thing). Other than that, just take a look at the foundry scene and do your best to imagine what is going through the engineers' and bean counters' heads. Going 90 nm is a sticky proposition at this point in time.
