One of the topics we’ve covered multiple times at ExtremeTech is the difficulty of continuing to scale semiconductor technology, and the related problem of improving chip performance without increasing clock speed. While Intel and other manufacturers continue to search for long-term solutions to this problem, no known next-generation technology is expected to restart silicon scaling and allow for a return to traditional clock speed gains.
Researchers at the California Institute of Technology think they may have a solution to this problem — one that involves returning to a very old technology to solve the problems of existing methods. Vacuum tubes, according to Dr. Axel Scherer, could be key to improving transistor performance and lowering power consumption.
Chances are, when you think vacuum tubes, you think of old radios or possibly Aopen’s AX4B-533 “TubeAmp” motherboard. The systems that Dr. Scherer and his research team are working on are nothing like classic vacuum tubes — according to the team, the structures are roughly 1,000x smaller than a human blood cell, which would make them 6-8nm. One problem with modern CPUs is that they suffer from significant amounts of electricity leakage — Scherer’s designs would use leakage current to flip states on purpose, thereby improving efficiency and overall performance.
One reason for this research is that Scherer thinks the microprocessor teams scaling below 10nm will encounter problems. The properties of silicon apparently change at that point, becoming both elastic and emitting light. “It’s a different material, and it gives you this different behavior,” Scherer told the New York Times.
Dr. Scherer isn’t trying to reinvent the transistor or replace the silicon economy. Boeing is funding his research due to its potential applications in space and aviation technologies, and silicon will obviously be the gold standard for everyone for years to come. It’s still interesting to consider the question: Could such a fundamentally different technology, shrunk to a microscopic scale, solve the problems of transistor scaling and performance?
Maybe — but there’s a lot of problems to be solved between here and there. First, there’s the question of manufacturing — can we crank out tens of thousands of vacuum-based processors in a month? What does it cost to build these solutions, switch out manufacturing hardware, and build an ecosystem around them? Can they built quickly enough to maintain current production rates, and how will they integrate into existing product lines?
These might seem like boring questions compared to a technology’s fundamental promise, but the boring questions are what ultimately determine whether or not tech comes to market. When we talk about Intel not being able to build faster CPUs, it doesn’t mean silicon is the fastest semiconductor ever. It means that Intel can’t find a method of building faster chips that’s cost-effective, scaleable, and likely to last multiple product generations.
Miniature vacuum tubes could evolve into a major driver of PC performance, particularly if they can be manufactured at scale, but the cost and manufacturing challenges are a huge roadblock to any different technology establishing itself as a silicon competitor. Neither carbon nanotubes nor graphene have done so, despite huge initial hype. There’s something satisfying in the idea that a century-old technology could be adapted and improved to the point that it boosts modern computing, but it’s going to take an awful lot of expensive work to prove it can do so.