Moore's Law Is About To Be Reborn

2021. This is the year that the 2015 International Technology Roadmap for Semiconductors, released earlier this month, predicts that it will no longer be economically viable for companies to continue shrinking the size of transistors, the fundamental units of computer processors that have historically dictated the maximum component density that can be squished onto a wafer of silicon.

Instead, the report forecasts, manufacturers will find more benefit in such strategies as arranging/orienting transistors vertically rather than horizontally. This may seem obvious enough, but it comes with its own challenges, particularly the ever-increasing difficulty of heat dissipation. As processors become more 3D, a trend inspired in part by the development of stacked memory architectures, thermal density increases while the amount available surface area to shed that heat falls. 3D scaling will then require lower power devices that are better at shedding heat.

Moore's Law has been more or less successfully predicting the advance of computing since it was first put forth in a 1965 paper by Intel co-founder Gordon Moore. Its definition 50 years later has come to be whatever the press decides it is at any given time, but the gist is that we can expect the maximum transistor density possible within an integrated circuit to double every two years. What this has meant historically is that the size of the transistors themselves is being halved every two years, but that's not necessarily the case. 3D power scaling, as it's properly known, achieves the density increase with no decrease in transistor size required.

The 2015 ITRS is the final report of its kind. A consortium of international semiconductor organizations has been producing the ITRS annually for 20 years—largely thanks to a shared interest in pushing semiconductor technology forward in a unified fashion—but as industry participation has waned, it's become less and less relevant. For one thing, the recent trend has been toward semiconductor requirements being dictated by chip consumers themselves, tech giants like Apple and Qualcomm, rather than semiconductor manufacturers like Intel and Samsung.

"They don't want to sit in a room and talk about what their needs are," VLSI Research analyst Dan Hutcheson told IEEE Spectrum. "It's sort of like everything's fun and games when you start off at the beginning of the football season, but by the time you get down to the playoffs it's pretty rough."

To some degree, the ITRS will be replaced by a new roadmapping effort by the IEEE called Rebooting Computing. Indeed, a good deal of the new, final ITRS report echoes a preliminary Rebooting Computing paper from last fall. That paper contains the following problem statement, which is worth repeating for its concision:

So, while it's possible to achieve and even outpace Moore's Law by stacking transistors, the fundamental challenge remains of how to do that without incinerating your integrated circuits in the process. Photonic computing remains one possible solution, but that's hardly a foretold technology.