On 17 Nov 2011 at 0:03, Keith Monahan wrote:
While I'm not sure about the async multiplier you
link to on that
paper, for what it's worth, there is an async 32x32-bit multiplier
here
http://pastebin.com/NqtpNuDz
I think not quite what I had in mind. The self-timed design that
essentially outputs a "done" signal when the product is finally
formed. The verilog for this appears to reflect a fixed-time
multiplier--at least there's no "done" output.
Years ago, asynchronous computing was a hot topic--essentially,
instructions were unclocked and took however long it required to
signals to propogate through the logic to complete.
That was the gist of Achronix' idea--that you could get some
freakingly fast computation out of a clockless design.
I also recall that the Philco System 2000 was one of the few
commercial versions utilizing that idea. "Asynchronous" was part of
their advertising campaign--and, for a very short time, they had one
of the fastest transistorized machines.
--Chuck
P.S. I won't even broach the subject of reversible computation... :)