Photonic computers...

[BrianHG]

I wonder if the day will come when I can have this in my home...


I wonder how fast a dedicated photonic bitcoin miner would be.  Whoever generates the first dedicated photonic SHA-256 algorithm engine will screw all the ASIC miners in speed and even more so on power consumption by leaps and bounds.

[ejeffrey]

Photonic computer will never ever replace transistors.

There are a ton of technical challenges to make photonic computers that are anything like general purpose -- constructing gates is hard, constructing registers/memory is hard, attenuation is high, and integrating sources (LEDs or lasers) is hard.  Most photonic stuff is actually pretty slow.

But the insurmountable problem is that photons are just too big.  Intel is having a ton of trouble moving from their 14 nm process to a 10 nm process.  They have a 7 nm process on their roadmap.  The wavelength of 405 nm light in silicon is 120 nm, while in SiO2 it is 270 nm.  So it just won't be possible to reach the same level of integration as silicon chips already do -- the photonic structures are just going to be too big.

Photonic processing is interesting for applications like optical networking switches if you can avoid the transition from optical -> electronic -> back.  An CPU replacing a modern Intel/AMD is crazy talk.

A lot of the stuff they talk about with "rethinking the way computer work" and making them parallel, pipelined, and so on are all things that we could be doing (or are doing) with silicon processors but it is actually really hard to program for and only useuful in certain applications --  particularly when you have very little data dependence. 

[raptor1956]

Optical communications will become more important and some day dominant for computer networks and even peripheral communications but optical processors -- not so much.  In order to make it work and do so economically you'd need to have standardized connectors with fiber-optics that fit the task and are small enough to connect even small components.  And, those connectors and cables will need to be cheap enough and that's pretty hard at this point.

Come up with a connector that integrates power and ground wires with one or more optical fibers that are small and cheap and we're off to the races.  Building a standard around it for component interconnection that specifies permissible noise limits on the power lines so that component makers know what level of power conditioning they need in there components to stay within those limits and the comm protocols is a tough ask but sooner or later optical comm will dominate even at the component interconnect level.


Brian