How crazy can we get in computing?
The U.S. Department of Energy (DoE) does a lot of crazy things in its labs, including basic research. Some of which may hold the key to faster and better computing that shows up in data centers in the future.
How crazy? Current thinking is Moore's Law is dead. Silicon-based technologies -- just the stuff that runs everything from your mobile phone to the massive data centers operated by Amazon, Google, Microsoft, Facebook and others -- are expected to be good for another five years. State of the art 10 nanometer (nm) chips are expected to start shipping in 2017, with new tricks using different materials and processes to shrink chips down 7nm in 2018 or later, with the end of the road at 5nm processes in 2020.
Image from: breakingenergy.com
For your average cell phone or server, getting only another bump or two of speed and/or better performance per watt may not seem to be a big deal, until you start looking at the trifecta explosion of Big Data, Analytics and Artificial Intelligence (AI). Speed is life for practical applications of the BD/A/AI trinity, so hitting the wall with only two chip generations left would effectively spell an end to the rapid explosion of IT advances we've seen over the past four to five decades.
Back in October, DoE's Lawrence Berkeley National Labs built a transistor with a working 1nm gate, using a combination of high-tech carbon nanotubes and relatively low-tech molybdenum disulfide, an engine lubricant sold in auto part shops. The combination of nanotubes and engine lube demonstrated you can build the same type of computing gate used by current silicon, showing that there's more "room at the bottom" the IT industry could effectively use.
The proof of concept demonstration is a long way from figuring out how to pack billions of transistors onto a chip and crank out millions of those chips for commercial use, but Berkeley Labs demonstrated there's a way forward beyond 5nm.
Longer-term research is trying to figure out the mysteries of superconductors. Today, for select applications, you can buy "high-temperature" superconductor (HTS) wire to replace traditional aluminum or copper cabling, with HTS wire transmitting 10 times more power than conventional cables. A lightweight compact DC power superconductor cable can be used to separate power room and server bays for an almost unlimited distance without a voltage drop -- just keep the liquid nitrogen cold.
DoE has poked away at the mysteries of superconducting for decades and among its efforts is funding advanced manufacturing of superconductor wires at the University of Houston and Stony Brook University. Producing better superconductor wire means lower cost cabling and more opportunities to apply the technology -- and its power savings across the board -- into both data centers and high-speed computers.
The holy grail is being able to find materials that are superconductors at room temperature without the use of liquid nitrogen or more extreme cooling solutions, enabling faster and more powerful computers, and dramatically more energy efficient data centers. Researchers around the world have been trying to find a working physics model for how superconductors work, but every time there's a partial answer, someone stumbles across another curve ball. DoE supercomputing power is being used to puzzle out the complexities of superconductors, but the phenomena remains one of the craziest aspects of physics yet to be solved.