One Step At A Time
The next rev of Moore’s Law—14nm—will be driven less by the performance/power/area gains of a classic process shrink than by three-dimensional transistors.
FinFETs are a game changer for Moore’s Law. In fact, the promise of these transistors is so significant that it raises questions about the value of moving from 28nm to 20nm. This is why foundries are now focused on creating hybrid processes that match finFETs with 20nm investments they already have made.
The value of a 3D gate structure is that it allows chipmakers to dramatically reduce leakage by increasing the surface area, which means chipmakers can more effectively shut off transistors when they’re not in use. That accommodates designs with higher frequencies, because transistors can be quickly turned on and off when they’re not in use, while also improving energy efficiency. Less leakage equals longer battery life.
None of this is free, however. In fact, the total cost of a design per transistor will likely rise. Manufacturing these chips will require double and triple patterning, more exotic materials for insulation, more design tradeoffs, and tougher engineering decisions because of the impact of extra margin on power and performance at advanced process nodes. That explains why companies such as Intel and Samsung are pushing for 450mm wafers. More die per wafer can help offset that cost.
But none of that should diminish the advantages of finFETs, which will help with the power/performance equatiuon for at least a couple process nodes. The three-fin design will almost certainly evolve into more fins, before moving in more exotic directions such as carbon nanotube FETs and tunnel FETs. The big challenge on all of these devices is being able to manufacture in volume and for a reasonable cost.
What’s a “reasonable” cost is a relative term, of course. In conjunction with all of this, there will be concerted push to re-use these advances as much as possible. Platforms, integrated subsystems and new approaches to networking between chips are the likely choices for the future, and work is underway in all of these areas. But the path forward looks certain, with finFETs serving as an essential bridge. Now the big question is how the various pieces will be assembled, and by whom.