Future Technology Trends

Superconducting supercomputers have been sought after for years, but their viability still remains unproven. Some of these technologies are kind of like nuclear fusion, always existing at a juncture 15 or 20 years into the future. Lots of money has been poured into research yet thus far it has mostly been relegated to vaporware status. Single flux quantum (SFQ) and reciprocal quantum logic are two possible superconducting methods to create ultra-fast CPU’s. SFQ currently has useful niche applications but  doesn’t come close to competing with standard complementary metal oxide semiconductor in most markets. Reciprocal quantum logic is an even newer approach to the problem. There is hope that some offshoot of these technologies could be used to power future systems at an exaflop/s or beyond. It may be necessary as the energy consumption of the fastest machines is growing ever higher (superconducting for high end computing).  

The United State’s Intelligence Advanced Research Projects Activity agency has a new goal of developing a ten to one hundred gigahertz 32 bit SFQ CPU that has a million gates and an integrated megabyte cache. Currently systems based of SFQ logic have been constrained by a lack of compatible random access memory. The ram just hasn’t scaled to the same degree as the SFQ gates. To work properly these processors would need to be cooled to a temperature between four and ten kelvin. That is just a few degrees above absolute zero. The memory has to be able to function in these extreme conditions. The transistor-less microchip utilizes the element niobium as a main ingredient. SFQ logic can be fabricated using a 90 to 250 nm node. This number compares to 22 nm process that is being used for CMOS in 2011. Intel’s densest devices have over a billion transistors, so a SFQ processor will be much less complicated. IARPA is searching for RAM elements that use about 10 to 17 joules to either read or write information. The specifications require a relatively high bit density per square centimeter.

Silicon has made it possible to create extremely inexpensive and quick desktops, laptops or tablets. They can be owned by nearly anyone who wants them. Unfortunately, you will never be able to buy these types of exotic circuits at an ordinary electronic store. A personal superconducting computer is out of the question. The cryogenic aspect means it will only be used for government or perhaps company mainframes. It is somewhat of an uphill battle to try to get the novel science to the HPC market.  The dearth of mass produced items ups the cost to make them. The speed though still makes this an attractive option to pursue further. In 2007 it was called a crackpot tech idea that could transform enterprises. Over ten years ago it was hoped that superconductors could be used for petascale systems. This early promise just never panned out and CMOS managed to fulfill the role well enough. A disruptive development may lead to zettaflops but this is an unlikely outcome at present. Future devices may contain combined neuromorphic, digital and quantum fundamental parts to far surpass anything that exists today.

Europe also has its own superconducting research called “S-PULSE“. The FLUXONICS society is a European foundry for superconductive electronics. They mention that RFSQ can be implemented using a rather simplified process. New complexities had arisen that made it more difficult to design the architecture. It doesn’t appear as if the project pages have been updated recently. There may be far too many engineering obstacles that need to be overcome for this to become a reality. However, the payoff is very high and worth studying.

See this older paper about cryogenic memories for an RSFQ ultra-high-speed processor.