3 ms·

https://en.wikipedia.org/wiki/Superconducting_computing https://en.wikipedia.org/wiki/Superconducting_computing Apparently. Whole article sounds like science-f

https://en.wikipedia.org/wiki/Superconducting_computing https://en.wikipedia.org/wiki/Superconducting_computing Apparently. Whole article sounds like science-fiction. 500x efficiency increase while scaling operating frequencies 250x.

jcranmer 6y ago
Drilling down into the sources suggests that the top-level claims are a bit of a reach: > Specifically, while superconducting technology eliminates the cost of moving bits within the superconducting domain, the cost of transferring data from the outside into the superconducting domain does dissipate energy at 4 K. Consequently, superconducting is expected to perform poorly on, for example, big data applications where the bulk of data is streamed from the outside into the processor. What is more, the energy cost of logic in superconducting circuits is expected to be similar to that of CMOS.
- Whirl 6y ago
  This is true for now. The IARPA SuperCables program exists to tackle this specific problem. Probably the solution will require converting microwave pulses into optical frequencies to get the signals out of the cold spaces with fiber-optic infrastructure. If you read the broad agency announcement, they even have explicit energy dissipation targets (per bit). I think it’s something like attoJoules or femtoJoules? https://www.iarpa.gov/index.php/research-programs/supercables/supercables-baa https://www.iarpa.gov/index.php/research-programs/supercable...
- rsfern 6y ago
  You seem to be quoting [0]. This isn't really my area of expertise, but the 4K referenced here seems to be the operating temperature. I imagine the thermodynamics of moving data around changes if you have room-temperature superconducting materials and aren't constrained to cryogenic computing, but I am not really sure what the impact would be on mitigating (or exacerbating) this kind of data transfer cost. [0]: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.1021.1180&rep=rep1&type=pdf https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.10...

jcranmer 6y ago

Drilling down into the sources suggests that the top-level claims are a bit of a reach: > Specifically, while superconducting technology eliminates the cost of moving bits within the superconducting domain, the cost of transferring data from the outside into the superconducting domain does dissipate energy at 4 K. Consequently, superconducting is expected to perform poorly on, for example, big data applications where the bulk of data is streamed from the outside into the processor. What is more, the energy cost of logic in superconducting circuits is expected to be similar to that of CMOS.

Whirl 6y ago
This is true for now. The IARPA SuperCables program exists to tackle this specific problem. Probably the solution will require converting microwave pulses into optical frequencies to get the signals out of the cold spaces with fiber-optic infrastructure. If you read the broad agency announcement, they even have explicit energy dissipation targets (per bit). I think it’s something like attoJoules or femtoJoules? https://www.iarpa.gov/index.php/research-programs/supercables/supercables-baa https://www.iarpa.gov/index.php/research-programs/supercable...
rsfern 6y ago
You seem to be quoting [0]. This isn't really my area of expertise, but the 4K referenced here seems to be the operating temperature. I imagine the thermodynamics of moving data around changes if you have room-temperature superconducting materials and aren't constrained to cryogenic computing, but I am not really sure what the impact would be on mitigating (or exacerbating) this kind of data transfer cost. [0]: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.1021.1180&rep=rep1&type=pdf https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.10...