All Indicators Point North

Designing and producing chips has always been difficult, but the number of things that conspire to make it harder at 20nm is the longest in the history of the semiconductor industry. The list will grow longer still at 14nm and beyond, not to mention so expensive that one mistake will kill a company.

While system engineers and architects look at the challenges on the front end, the problems further down the line are compounding at a much faster rate. Design teams view this from the standpoint of what they encounter at each new node—things like routing congestion, physical effects such as electromigration, ESD, and heat, more rules that restrict what they can or cannot do, not to mention the challenges of writing efficient software and threading it across multiple cores.

Further down the line, things have gone from difficult to ugly. On the lithography front, EUV may never be ready for prime time, and the cost of double, triple or quadruple patterning will render Moore’s Law uneconomical for the vast majority of planar SoCs. In fact, the only thing that will continue to scale well will be geometrically redundant shapes, such as memory or processors—and better separately than together.

To make up for some of the cost of multipatterning, foundries and equipment makers are considering a shift to 450mm wafers from 300mm. This isn’t so simple, and the transition to 300mm at 130nm raised a slew of complaints about the cost of insufficient yield. Add to that new structures such as finFETs and tunnelFETs, new process technologies, more data to process, and some new materials for insulation the price and complexity increase even further.

And just to put this in perspective, the introduction of high-k/metal gate technology back several nodes ago is just now beginning to take hold. Chipmakers have been delaying the adoption of any new process technology as long as possible. Now they’re facing a wall of must-do’s before they can move on.

The obvious solution is to limit exposure to these new rules and technologies by implementing them only where absolutely needed, and in limited doses. That likely will drive 2.5D in the short term, and ultimately 3D stacking in the long term. Economics alone is a powerful driver, but economics plus technology are overwhelming. The push to go vertical is now coming from all sides, which will transform it from in the very near future from a possibility into a necessity.

—Ed Sperling