Actualy density is all TSMC can offer , their electrical parameters being vastly inferior to either intels or glofos
most refined processes.
You are comparing two nodes which were unavailable to any external customer, regardless the electrical parameters, for the vast majority of the time in which TSMC's 28nm has been commercially available and shipping in volumes.
Comparing Intel or GloFo to what TSMC offers is purely an academic argument if you were/are a fabless company for the past two years. Regardless what technical deficiencies you may assign to TSMC's 28nm, the bottom line is it was the only game in town for fabless companies for nearly 2 solid years.
Availability trumps inaccessible, unreliable, and inconsistent any day of the week when you are spending 2-3 yrs designing a product.
Currently prices being the main factor they have plenty
of cost restricted customers but if ever the competition
increase , and it is with intel targeting the lowly priced chips ,
any foundry that has superior process will highjack their
customer base , as it happened with qualcomm switching
to glofo to benefit from a more advanced process.
We have yet to determine what, if anything, is really afoot with Qualcomm.
But what we do know is you cannot transcend the time requirements involved in redesigning and laying out your chip (yet again) for a second foundry. There is no such thing as "hijacking" a customer at this late stage of the game.
If Qualcomm is producing chips at GloFo then Qualcomm made the decision to do so some 2 years ago (minimum) as that is the time required to port an existing vetted IC design from one foundry to another.
In the foundry business, the only time you have to intercept customers and steal them away from the competition is during the early stages of the node's development.
If GloFo wanted a customer for 2013 (like Qualcomm) then GloFo would have had to convince that customer to start designing their chips for GloFo's process flow no later than 2011.
If that is "hijacking" then it has to be the slowest darn robbery there ever has been.
You can forget the two years dev cycle for both 10nm
and 7nm , theses are no more trivial nodes , current tools
are to be replaced completely as they wont be extendable
past 14nm.
The development cycle for nodes has NOT been 2 years since roughly 1992.
Node development has historically been a 4 year effort, at least for the past 20 years.
Nodes are developed in parallel. A new node is formally* started in development every 2 yrs, with two nodes perpetually in the pipeline, staggered by two years so they come to production every two years.
At Intel, for example, 14nm has been in "development**" for ~3 yrs, 10nm has been in development for ~1.5yrs, and 7nm is just entering development at this time.
This is an industry standard approach to R&D, every IDM and foundry does it this way.
* "formally" is an official designation when a node's project team is picked and the financial commitment is outlaid and supported by the BoD. Of course there are years and years of pathfinding research that precedes an official node from being commissioned and started. Finfets for example were under "Pathfinding" at Intel for a good 6yrs before being officially transferred to the 22nm development team...so one could legitimately argue that 22nm was in R&D for 10+ yrs, but technically/officially it was only under development for 4 yrs, the finfets themselves were under research for 6yrs.
** while everyone casually refers to it as "R&D", the "R" and the "D" happen sequentially and are handled by two separate groups (and this is industry wide practice as well). At Intel the "R" happens in the so-called pathfinding phase, the "D" happens within the process node development team itself (where they make the "R" manufacturable and yieldable). At IBM the "R" is handled by various research labs such as TJ Watson whereas the "D" is handled by the process node development team at Fishkill. The "development" phase of a node is typically 4 years, the research phase can be anywhere from 2-8 yrs depending on the specific material, component, or electrical parameter under discussion.