Five years until transistors stop shrinking, predicts last Moore’s Law roadmap

By 2021, the historical tradition of Moore's Law, of transistor size being cut in half every two years, may come to an end

Five years until transistors stop shrinking
Rudolf Getel

Moore’s Law, which says the number of transistors within an integrated circuit will double every two years, had a good long run, but its end may be near – very near – potentially a mere five years away. By 2021, even if chip makers could further shrink and add more transistors, the high cost of manufacturing would make it financially impractical.

That’s not the only problem, according to the last installment of the International Technology Roadmap for Semiconductors 2.0 (pdf). By 2020 to 2025, it will be “practically impossible” to reduce device dimensions. While one solution is to stack the transistors, it can’t get so hot that it burns up either.

That doesn’t mean Moore’s Law has one foot in the grave, just that the historical tradition of transistor size being cut in half every two years will come to an end.

Memory device manufacturers were called the “real drivers of Moore’s Law,” and they will continue being the real drivers in the future. 3D power scaling, which allows the density to continue to increase without the transistor size decreasing, has already become a solution for NAND Flash.

The report states:

The combination of 3D device architecture and low power device will usher the (Third) Era of Scaling, identified in short as “3D Power Scaling.” Increase in the number of transistors per unit area will eventually be accomplished by stacking multiple layers of transistors.

In explanation, the report compared the vertical dimensions of Flash memory transistor density to the approach of densely packed cities using skyscrapers to deal with space limitations. The vertical trend of cramming memory transistors on top of each other in a single circuit was predicted to be an industrywide trend in the next decade. That method could “accelerate the level of transistor density beyond Moore’s Law traditional trends.”

In addition, the authors listed examples that indicate by the year 2020, several new devices will be available “to work in conjunction or better than CMOS on some specific applications.”

A goodly portion of the report hones in on heterogeneous integration and heterogeneous components, which use multiple varying applications, but each application has different device performance requirements. The mobile device market has pushed component manufacturers into producing smaller components that offer more performance and functionality, so too has the “Internet of Everything.”

Back in the day, numerous companies developed logic chips, but now there are only four: Intel, TSMC, Samsung and GlobalFoundries, which gobbled up IBM chip fabrication plants. Each of those companies has its own roadmap and are “fiercely competitive.” As 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.”

Even if Moore’s Law does end, that doesn’t mean chips won’t keep getting better. This last report is long and gives plenty of examples of what was, what is and what shall be. Keep in mind, however, that it’s only a roadmap. IEEE is taking over, somewhat, and will hold its first international Rebooting Computing conference in October.

Tom Conte, co-leader of the IEEE Rebooting Computing Initiative, said the new IEEE roadmap will continue to track “Moore’s Law to the bitter end.” He added, “This isn’t saying this is the end of Moore’s Law. It’s stepping back and saying what really matters here—and what really matters here is computing.”

New! Download the State of Cybercrime 2017 report