Transistors will stop shrinking after 2021, but Moore’s law will probably continue, according to the final International Technology Roadmap for Semiconductors (ITRS).
The ITRS—which has been produced almost annually by a collaboration of most of the world’s major semiconductor companies since 1993—is about as authoritative as it gets when it comes to predicting the future of computing. The 2015 roadmap will however be its last.
The most interesting aspect of the ITRS is that it tries to predict what materials and processes we might be using in the next 15 years. The idea is that, by collaborating on such a roadmap, the companies involved can sink their R&D money into the “right” technologies.
For example, despite all the fuss surrounding graphene and carbon nanotubes a few years back, the 2011 ITRS predicted that it would still be at least 10 to 15 years before they were actually used in memory or logic devices. Germanium and III-V semiconductors, though, were predicted to be only five to 10 years away.
The latest and last ITRS focuses on two key areas: that it will no longer be economically viable to shrink transistors after 2021—and, pray tell, what might be done to keep Moore’s law going despite transistors reaching their minimal limit. (Remember, Moore’s law simply predicts a doubling of transistor density within a given integrated circuit, not the size or performance of those transistors.)
The first problem has been known about for a long while. Basically, starting at around the 65nm node in 2006, the economic gains from moving to smaller transistors have been slowly dribbling away. Previously, moving to a smaller node meant you could cram tons more chips onto a single silicon wafer, at a reasonably small price increase. With recent nodes like 22 or 14nm, though, there are so many additional steps required that it costs a lot more to manufacture a completed wafer—not to mention additional costs for things like package-on-package (PoP) and through-silicon vias (TSV) packaging.
This is the primary reason that the semiconductor industry has been whittled from around 20 leading-edge logic-manufacturing companies in 2000, down to just four today: Intel, TSMC, GlobalFoundries, and Samsung. (IBM recently left the business by selling its fabs to GloFo.)
The second problem—how to keep increasing transistor density—has a couple of likely solutions. First, ITRS expects that chip makers and designers will begin to move away from FinFET in 2019, towards gate-all-around transistor designs. Then, a few years later, these transistors will become vertical, with the channel fashioned out of some kind of nanowire. This will allow for a massive increase in transistor density, similar to recent advances in 3D V-NAND memory.
The gains won’t last for long though, according to ITRS: by 2024 (so, just eight years from now), we will once again run up against a thermal ceiling. Basically, there is a hard limit on how much heat can be dissipated from a given surface area. So, as chips get smaller and/or denser, it eventually becomes impossible to keep the chip cool. The only real solution is to completely rethink chip packaging and cooling. To begin with, we’ll probably see microfluidic channels that increase the effective surface area for heat transfer. But after that, as we stack circuits on top of each other, we’ll need something even fancier. Electronic blood, perhaps?
The final ITRS is one of the most beastly reports I’ve ever seen, spanning seven different sections and hundreds of pages and diagrams. Suffice it to say I’ve only touched on a tiny portion of the roadmap here. There are large sections on heterogeneous integration, and also some important bits on connectivity (semiconductors play a key role in modulating optical and radio signals).
I’ll leave you with one more important short-term nugget, though. We are fast approaching the cut-off date for choosing which lithography and patterning techs will be used for commercial 7nm and 5nm logic chips.
As you may know, extreme ultraviolet (EUV) has been waiting in the wings for years now, never quite reaching full readiness due to its extremely high power usage and some resolution concerns. In the mean time, chip makers have fallen back on increasing levels of multiple patterning—multiple lithographic exposures, which increase manufacturing time (and costs).
Now, however, directed self-assembly (DSA)—where the patterns assemble themselves—is also getting very close to readiness. If either technology wants to be used over multiple patterning for 7nm logic, the ITRS says they will need to prove their readiness in the next few months.
This post originated on Ars Technica UK