Daniel Nenni has started a conversation on his creation Semiwiki with the provocative title: "Are 28 nm Transistors the Cheapest...Forever?" (http://bit.ly/16HR8D7)
It appears that after (thanks EDAC) having addressed only technology issues for fifty years, the industry is finally realizing that we are all in this in order to make a profit. Daniel deals almost exclusively with fabrication costs, but I think that there is a significant, and growing, body of evidence to show that development costs also justify his conclusion.
The work involved in preparing a circuit for fabrication, given the highly restrictive, and foundry dependent, design rules, force designers to spend significant more time creating a floorplan that can be diffused and exposed with a high probability of acceptable yields. The 28 nm process is the last node that has some relationship with the shrinking methods that have followed Moore's "law" governing semiconductors manufacturing.
Daniel chooses its words very carefully by stating that the 28 nm process "may be the cheapest per million gates" which is the most factual meter. It is no longer news that all processes after this one require new "inventions" and cannot be achieved with a "simple" shrinking because the manufacturing technology is drastically different from previous ones.
This, of course, does not mean that it will be financial folly to use a process below 28 nm. It just means that one must be able to justify the additional development and production costs with increased revenue. And the revenue must come from actual end users benefits.
The implication is that electronics companies, EDA vendors, and foundries all will have to consider a new business model that will be in addition to, and not a replacement of, the existing model. If one considers the 90 nm node that is used to produce the largest volume of digital chips today and the 28 nm node, the pricing model (not the actual prices) is practically the same. But after this, the pricing model must change or profits will turn to loss for some of the segments in the industry. Of course I am particularly worried about the EDA segment, and even more when considering small vendors offering only one tool.