Inflection Points And Changes Ahead

It’s hard to justify throwing away a well-oiled machine and replacing it with a new one. It works, it’s predictable and it’s low risk. And nowhere is this more evident than in the semiconductor industry. The doubling of transistors every two years for nearly five decades has created a $300 billion chip industry, reduced the price of processing by orders of magnitude, and made possible electronics that were considered science fiction when the industry first began.

There’s no end in sight, either. Moore’s Law will continue unabated for at least the next several process nodes and demand for new chips is growing in brand new markets such as medical, industrial and automotive. But how we develop these chips will have to change, and that shift will prompt other changes—all of which will have some rather interesting repercussions from the front end of design to verification to back end manufacturing and test.

There are several possible inflection points ahead, each with its ramifications:

1. Lithography. It’s no secret that the current lithography tools aren’t sufficient. Immersion and 193nm wavelength lasers were supposed to have been replaced by extreme ultraviolet lithography at 20nm. The industry will be lucky to see it at 14nm, and it’s certain there will be at least double patterning, and possibly even triple patterning, required at 14nm. Beyond that—the next node appears to be 10nm—either EUV reaches commercial viability or self-assembly templates will replace masks.

IBM says it is close to commercialization of self-assembly. It’s not the first choice. Why disrupt an approach that has served the industry almost since its inception? But without EUV it may be the only choice because double patterning won’t work at 10nm. And that’s where changes begin, from front-end architectural planning of layout and materials and physical effects to design tools to design for manufacturing. In theory, at least, self-assembly may be able to add significant density to designs even at existing nodes.

2. More IP. The amount of commercial IP in designs is growing. Given that these are largely black-box technologies, that will fuel a push toward pre-verified subsystems complete with complex verification IP and hooks into just about every possible configuration and interface imaginable—at least at first.

This all makes sense from a business perspective. Time to market is more critical than building something from scratch that may or may not be as good as what’s commercially available—or at least that’s the theory. But the reality is that to remain competitive chipmakers will have to focus on where they can really differentiate, and that most likely will be in much narrower areas of development—software, analog IP and maybe some digital IP.

3. Stacked die. Some industry experts dismiss stacked die as simply a packaging solution similar to the old system-in-package or multi-chip module. While that’s probably true for the big IDMs, it’s a completely different story for fabless companies, IP developers and interconnect vendors—and for consumers.
Shorter wire distances, the potential to mix and match chips from different process nodes, bigger pipes for signals and lower power to drive those signals all add up to a potentially significant inflection point—and one that could radically change the competitive landscape for companies that can quickly assemble chips for specific markets with different power and performance characteristics and targeted IP.

There are kinks to work out, of course. Who’s responsible if something goes wrong? How do you test die effectively? What’s the best method for cooling these devices? But all of these are being worked on, and most companies involved in the R&D work don’t see these problems as insurmountable.

What’s interesting about all three of these inflection points is that any one of them has the potential to dislocate the industry from business as usual to looking at entirely new approaches. Taken together, they could force even more significant changes because remaining competitive will require innovation that is way beyond the normal way of doing things. And when there is no clear road map, no lane markers, and no standard vehicles to drive, the changes could be well beyond what we expect today and possibly from market sectors and geographies that have never played a significant role in the past.

It may be time to put away the oil can.

–Ed Sperling