During the NVIDIA GTC 2012 conference, we spoke with several sources from the world of semiconductor manufacturing and the figures given were not… encouraging.
During the NVIDIA GTC 2012 conference, we spoke with several sources from the world of semiconductor manufacturing and the figures given were not… encouraging. Furthermore, parallel to GPU Tech Conference, there was a Semico Impact Forum being held at the Double Tree Hotel in San Jose, CA.
While NVIDIA was publicly discussing the 28nm availability issues (yields are very high, wafer availability is low) and TSMC was the only semiconductor maker at the conference, Semico Impact Forum faired a grim message to the fabless chip designers.
According to sources with an extensive knowledge of the industry, the shift from 28nm to 20nm will take place over the course of next year, starting in the second half of 2013. It is expected that 20nm process node will only expand during 2014, with the shift to 14nm being “the big one”. But going from 28nm to 20nm is not going to be a walk in the park either. In order to amortize for the 20nm shrink, you will need to amortize costs over at least 100 million devices.
This is also the primary reason why the shift to 20nm will only be bulk (no SOI) with less options than 32/28nm. NVIDIA, Qualcomm, TI, AMD ship products in sufficient volumes to be able to afford the switch, but that will not be the case with smaller manufacturers.
On the other hand, the cost of 40/32/28nm production is going to drop significantly after the big players migrate to a new process node and leave ample capacity at older nodes.