AToday two things determine the quality of a chip:
-- its design, in summary the performance of its architecture.
-- its etching, in a nutshell the design implemented in silicon. The latter is now largely determined by the minitization of the etching, but also by other processes providing products that consume more or less and/or will be able or not to escalate the frequency.
The price of the fabrication facilities has become so important that many smelters are called Fabless, they do not have a plant and deal with specialists.
TSMC and UMC are two giants in this field, and are content to make the factories and produce processors ordered by their customers.
TSMC announced that it would spend $5 billion for the establishment of a new manufacturing plant as well as the R&D for etching in 32, then 22, then 15 nm.
The goal is to satisfy its large customers like ATI/AMD, NVidia or Sun who are very demanding, who constantly need to put out ever more powerful products.
Currently, apart from TSMC and UMC, only Intel has the financial means to build such plants. AMD is in a delicate situation it has become increasingly difficult for it to do the same, which partly explains their difficulty and delay in their move to 45nm. It must be said that in the field of manufacture of processors, AMD has lost its expertise, for the most part delegated to IBM, which also marking time.
[translation by jeremy]