Andrew Mobbs (mobbsy) wrote,
Andrew Mobbs

Superhuman machines

Ars Technica's coverage of the story about the simulation of a cat's cerebral cortex contains a reference to this being about 4.5% of the capacity of a human cerebral cortex.

Just running with that figure for a second, and assuming Moore doubling continues every 18 months, that gives just 4.42 doublings in capacity, or a little under 7 years, to simulate a brain with the capacity human cortex. The cat simulation ran at 1% real time, which gives another 6.64 doublings, or 10 years, to get up to full speed.

Will we really be able to run a real-time simulation of a brain with more capacity than a human cerebral cortex in 17 years time? I can't find a definitive reference, but I'd guess that the PowerPC 450 CPUs used in the Dawn BlueGene/P system were fabricated at 65nm. Current CPUs are 45nm with 32nm being available early next year. Intel claim to have a "clear way" to 11nm, which would get 5 doublings over the current simulation, or enough to do the simulation of the full human cerebral cortex at 1% speed using equivalent hardware. This presentation mentions in brief 8nm and 5nm nodes, and talks about 3D lithography further increasing density.

I'm no expert on the physics involved, either for the lithography or whether there are quantum effects of having so few electrons in gates. It appears that people who know what they're talking about can see at least another 8 doublings of feature density as achievable, which is nearly enough. When I was an undergraduate, 15 years ago, I knew people who thought that Moore's Law was coming to an end at that point.

We might hit some issues, but I expect that in my lifetime machines based on something not dissimilar to current semiconductor technology will be able to simulate a human brain at faster than real-time (or with more capacity), that will probably happen within the next 30 years. However, that simulation will be on a massive computer, it probably won't achievable on something the equivalent of a desktop PC for another 30 years, and if that happens, it won't be anything like the computers we know today. Maybe the superhuman machines can help us design them?
  • Post a new comment


    default userpic

    Your IP address will be recorded 

    When you submit the form an invisible reCAPTCHA check will be performed.
    You must follow the Privacy Policy and Google Terms of use.