> I also remember something rather puzzling. Apparently somebody
> discovered that if you etch silicon with a certain kind of acid, it
> produces a special microscopic structure which has an unexpected
> property: it can transform electricity into like, and the reverse. This
> was hailed as the future of IC technology. In the future, we were told,
> interconnects on an IC would work using light rather electricity. For
> light has one really critical advantage: beams of light can pass through
> each other.
>
> By contrast, if you want to move signals around using wires, you either
> have to have extremely long and convoluted wire routes to get around all
> the obstacles in your way, or complex multiple-layer wiring designs. But
> with light, a signal can just go straight from A to B, intersecting as
> many other signal paths as you like.
>
> So if this technology is the future... where is it? How come it's
> completely vanished off the face of existence?
>
> There seemed to be some suggesting that the entire IC might work by
> processing light instead of electricity. I'm sceptical about whether
> that could work. I'm not aware of any light-based switching technology.
>
> On the other hand, just using light for implementing long-range
> connections? That could *totally* work! By strategically using optical
> signals in place of electronic ones, you might be able to drastically
> reduce signal path lengths, which reduces propagation delays. More to
> the point, if I'm understanding this right, long traces have the problem
> of high capacitance too, which an optical signal path would seemingly
> also avoid.
>
> So why is absolutely nobody using this stuff? I can only imagine that
> the answer is the same as for the 3D IC. In other words, "it's too
> expensive" combined with "we haven't reached the hard limits of current
> methods yet".
>

For optical computers, there are several working prototypes in several labs.

The good:
Multiple parallel use of the same switch/circuit by using light of 
different colours, over 100 simultaneous computing channels on 1 
component at the same time! Brings up realy massive parallel 
multiprocessing.
Digital circuits having 5 discreet stable signal level. Base 5 computing 
anyone?...
No cross interference.
No capacitance nor inductance.
Capable of theoretical clock speed in the THz.
Capable of mixed digital and analogic processing.
Dirrect intercomponents data paths.

The bad:
Most of those components are cm sized, or monstrously huge compared to 
electronic parts.
Mostly organic material based can compromise longevity and stability.
Non-organic based solutions tend to be very brittle and devlop faults 
and minute cracks.
There are problems with the RAM side. Low capacity and retention time. 
Refreshing dynamic RAM is difficult and static RAM not realy functional. 
Storage loops have huge latency or very low capacity.
Nobody can yet take advantage of 5 digital states.




Alain

Post a reply to this message