Thursday, December 4, 2014

US intel agency is developing a superconducting exascale computer and cryogenic memory http://goo.gl/vOAtZh

Meissner Effect (superconducting magnetic levitation)

The secretive R&D wing of the US government’s intelligence efforts, IARPA, has announced a program to build a superconducting computer. IARPA will be working with IBM, Raytheon, and Northrop Grumman to develop this superconducting computer, but the exact financial details of the deal are not available. Ultimately, the purpose of the program is to build an exascale supercomputer — a computer that is capable of at least 1,000 petaflops (1 exaflop), or about 40 times faster than the world’s existing supercomputers. And yes, in case you were wondering, such a computer would almost certainly be used by agencies like the CIA and NSA for cracking encrypted messages.

As you’ve probably surmised, IARPA — or Intelligence Advanced Research Projects Activity to give its full name — is the intelligence version of the US Department of Defense’s DARPA. A quick look at IARPA’s research programs shows some similarities to DARPA, but with a bias towards social engineering, neuromorphic (brain-like) computing, and parsing big data (to dig some nuggets of intelligence out of the haystack of data). In this case, the superconducting computer is part of the Cryogenic Computing Complexity (C3) program.

The main task of C3 is to find a path to exascale computing that doesn’t necessitate extremely expensive power and cooling requirements. The world’s current top supercomputers consume about 10 megawatts of power to provide 20 petaflops of computation. While the underlying hardware is becoming more power efficient — thanks to advanced nodes like Intel 14nm and the monstrous parallelism of IBM’s Power8 CPU — these advances on their own aren’t really enough to make exascale computing feasible. What we really need is a new type of computing — an alternative to the humble CMOS transistor logic that doesn’t consume as much power or produce as much heat.

A superconducting chip from MIT, which uses Josephson junctions

A superconducting chip from MIT, which uses Josephson junctions

Enter superconducting logic — which is essentially a catch-all term for any kind of computer that uses superconductivity to reduce circuit/transistor resistance to zero, and thus massively reduce power consumption and heat generation. In this case, it looks like IARPA — and IBM, Raytheon, and Northrop — will be specifically investigating the use of Josephson junctions. Basically, when a copper wire is cooled to near absolute zero (-273 Celsius) — i.e. when it becomes superconductive — an electrical current will flow along the wire indefinitely, without any voltage applied. If you put a semiconductor in the middle of the wire, you can turn the Josephson effect into a very low-power switch. This approach is called single-flux quantum (SFQ) logic — and rather importantly, we can use it to drive a digital (binary) computer. (Superconducting computing can, and often does, refer to quantum computing — but not always.)

Read: World’s first superconducting power line paves the way for billions of dollars in savings, more nuclear power stations

IARPA’s C3 program also calls for the development of cryogenic memory, which would operate “in close proximity” to the superconductive CPU — but there are currently no details on what form this cryogenic memory might take. In either case, as we’ve discussed before, processing power is meaningless without the caches and main memory to back it up.

D-Wave's 128-qubit quantum processor

There are already some superconducting quantum processors, like this 128-qubit D-Wave chip — but they’re not capable of supercomputing-type applications

Moving forward, the plan is to build proof-of-concept superconducting logic and memory — and if that’s a success, to move onto phase 2 of C3, which will see those new technologies actually fashioned into a working, usable superconducting computer. Early research suggests superconducting logic can switch at speeds north of 770GHz, and provide around 100 petaflops of performance while consuming just 200 kilowatts — orders of magnitude more efficient than current CMOS-based supercomputers.

In the mean time, of course, IBM and Nvidia are all set to build two 150-petaflop supercomputers for the US Department of Energy — they’ll be by far the most efficient supercomputers in the world, but they’ll still consume enough electricity that could’ve powered thousands of homes.

Now read: IBM shows off quantum computing advances, says practical qubit computers are close



Source Article from http://www.extremetech.com/extreme/195322-us-intel-agency-is-developing-a-superconducting-exascale-computer-and-cryogenic-memory http://www.extremetech.com/wp-content/uploads/2014/08/meissner-effect-superconducting-levitation-magnet-640x353.jpg
US intel agency is developing a superconducting exascale computer and cryogenic memory

No comments:

Post a Comment