Optical CPU?

[Elledan]

Regarding this article, what are the chances that we'll see optical CPU's appear in computers anytime soon?

From what I've read I can only conclude that optical CPU's would be a real breakthrough. Not only would they be faster, they would also consume less energy and produce very little heat. The ideal CPU, thus

Any insights?

 

[CStroman]

It's not likely that it will be soon. The copper 3GIO system won't be implemented until 2004, and then it'll be a bit longer before the optical version comes out.

 

[Sohcan]

That article is not about optical CPUs, but optical interconnects, replacing, for example, the electrical interconnects between components on a motherboard (CPU to Northbridge, Northbridge to Southbridge, etc). The challenge with optical interconnects is making the electrical -> optical conversion process on each end as fast as possible. Optical interconnects will most certainly start becoming prevalent in the next few years...there was a presentation at the recent MPF about this topic, let me see if I can find it...

Optical CPUs are a completely different story...I remember reading about silicon-based optical transistors a few years ago, but I haven't heard anything since. The challenge here is the miniaturization of the optical transistors....honestly, I don't think optical transistors will become commonplace anytime in the near future, assuming that the current work in molecular and atomic transistors leads to the replacement of semiconducting transistors in around 10 years or so.

 

[Elledan]

That's another thing I was wondering about: does an optical transistor work in a different way than those found in current CPU's? If so: how?

 

[StandardCell]

<< That's another thing I was wondering about: does an optical transistor work in a different way than those found in current CPU's? If so: how? >>



Ummm, well, an optical transistor (and I use that term very loosely) is basically an LED with a controlling input, but on a small scale. In a traditional CMOS setup, there are two sets of transistors, one PMOS and one NMOS, that either ground out or maintain a voltage at the output. That voltage is applied to the next gate to act in a certain way, and so on. In an optical setup, the difference is that the output (light) is sent along some type of optical pathway and received by a light-sensitive transducer.

After having seen the worst of the worst in ASIC design with traditional transistors and metal interconnect, I can guarantee you will NOT see on-chip optical interconnect for on-chip circuits for a LONG time. The process of producing optical components and interconnect small enough and reliable enough isn't even CLOSE mature for production. The design software and tools that exist today don't support an optical interconnect methodology. Finally, there are other technologies, like quantum dots, that are promising and may usurp hopes of on-chip optical devices and interconnect.

That doesn't mean that optical interconnect isn't used anywhere. For example, Lucent developed an all-optical broadband switch that basically directly connects different fiber ends optically. TI has an all-optical infrared Ethernet switch that uses micromirrors for a switch fabric. But even in these cases, there are underlying electronic devices that manipulate the switches. You won't get away from them any time soon. Unfortunately, making an all-optical CPU is not going to happen for a long time, if ever.

 

[rimshaker]

I believe it was Motorola recently that announced a breakthrough concerning the possible future of optics mixing in with IC's. They found a stable way of merging GaAs with Silicon. This is significant because optical and RF devices are usually made using GaAs material. The main problem was that combining Si with GaAs was nearly impossible due to lattice mismatch. Now that GaAs can be successfully grown on top of silicon.... it's very possible in the near future that optical and RF devices can be implemented with IC's.

 

[Noj]

The ideal would be to produce an optical version of the molecular transistors produced at Bell Labs. I?d pay for that.

 

[CStroman]

<< The ideal would be to produce an optical version of the molecular transistors produced at Bell Labs. I?d pay for that. >>



You'd pay for the research and production of that? You must have deep pockets.

 

[rimshaker]

You will see optical and RF subsections on a cpu within this decade, as well as magnetic ram (MRAM). But as for getting molecular transistors to work in such a way as to create something as complex as a cpu... truly star trek stuff. You won't see that sort of thing for at least few decades.

 

[jamarno]

I thought the main advantage to an optical CPU was extremely high fanout - one output could drive millions of inputs and reduce the need for buffer stages.

 

[StandardCell]

<< I thought the main advantage to an optical CPU was extremely high fanout - one output could drive millions of inputs and reduce the need for buffer stages. >>



In theory, this is true. You need to split the optical output with the same number of photons on each branch. That I'm not sure how you do, but it follows and would be GREAT for skew and insertion delay, particularly on clock trees. But, again, how you make an economical production-worthy process with a purely optical or hybrid strategy? Sadly, it's neither easy nor will we see it any time soon.