Intel has a two year lead in process technology over the rest of the semiconductor industry and could get further ahead if yields are satisfactory on its finfet-based 22nm process due for introduction later this year.
So said Mike Bryant, CTO of Future Horizons, at the recent IFS 2011 conference in London.
Intel has been shipping 32nm in volume since 2009, said Bryant, adding: “While the others are only just getting into volume production.”
TSMC has issued its 28nm design kits but, said Bryant: “This tends to happen a year ahead of full production.”
The worrying thing for the rest of the industry is, said Bryant: “If Intel’s 22nm process yields well, then they will move further ahead.”
Therein lie the challenges for Intel:
Thousands of test wafers will be needed to generate the data to simulate the size and aspect ratio of the fin and check accuracy.
Variance (the change in behaviour due to the placement of dopant atoms and line roughness) becomes more critical with the raised channel.
Spice models will have to be optimised using thousands of test wafers.
3D worsens layout dependent effects/ restricted design rules.
Parasitic capacitance extraction “just got a whole lot worse,” said Bryant.
Removing heat from the fin is another problem.
“Quantised sizing – instead of widening the channel you need to parallel them up so you can’t optimise transistors exactly to their operation,” said Bryant, “this also restricts Vt optimisation and other leakage reduction techniques.”
Finally Bryant pointed out that “an analogue mixed-signal methodology for finfets hasn’t even begun.”
The prize for cracking all these problems is that processors might get to 4GHz, and that Apple may use Intel for foundry to get an advantage over Android competitors.