Wired Article a Glimpse into IBM's Cell?

in General Discussion edited January 2014
[attr: MacSurfer.]

The article, on adiabatic logic (or "reversible computing"), describes a technology first developed at IBM in the 1960s. Instead of grounding the charge used in the previous calculation, the chip would store the charge and reuse it for the next one.

The result: The chip uses far, far less power and dissipates far less heat. The tradeoff is slower serial logic (i.e., one instruction after another after another - the sort of thing the Pentium 4 excels at), but as one of the technology's developers notes, "processors will be faster overall because they will be able to run thousands more computations in parallel than today's chips can" [quoting the article, not the developer].

Now, the kicker: Intel isn't interested, of course, but "[a]n IBM spokeswoman confirmed that researchers at the company are currently working on reversible computing technology."

Ultra-low-power computing best suited to highly parallel systems. Cell, anyone? I wondered at some of IBM's bolder claims for the technology, but this seems to solve many of them.


  • Reply 1 of 3
    wmfwmf Posts: 1,164member
    Wow, even more speculative than usual.
  • Reply 2 of 3
    alcimedesalcimedes Posts: 5,486member
    isn't heat why they're coming up with diamond based chips?

    although low wattage would be excellent for laptops.
  • Reply 3 of 3
    amorphamorph Posts: 7,112member
    The diamond-based technology is another route, although I think it's a bit further away - last I heard, they were far from growing large enough, flawless enough diamond fast enough. They're only far enough along to scare the crap out of DeBeers, but then the sort of rocks you find on rings would be far, far cheaper if the supply was not artificially (and severely) bottlenecked. Those diamonds are nothing special, but if you want diamonds suitable to make wafers, that's a whole 'nother story.

    This involves a design change, not a fabrication change, which should make it more feasible (if not easy). It's also been around a while, and IBM's already on it.

    Very-low-energy chips are also well suited to massively parallel implementations. If you're going to throw a lot of cores at the problem, even 20W CPUs add up quickly. And it sounds like you could make them with the same basic tech and on the same fabs that you make conventional throw-electrons-at-the-problem CPUs, which would also make them more attractive than diamond-based chips in the near term.
Sign In or Register to comment.