Slashdot Mirror
Intel Researchers Build Laser on Chip
Victor Ramen writes "Working with the basic material of computer chips, Intel Corp. researchers have constructed an all-silicon laser that could lead to computers one day harnessing light waves rather than electrical currents to shuttle data swiftly. 'Once you have silicon as an optical material, then you can take advantage of this enormous (silicon) infrastructure that exists around the world,' said Mario Paniccia, director of Intel's photonics lab. 'You can imagine starting to siliconize photonic devices, and maybe integrate photonics and electronics.'"
here
Many fellow /.ers seem to wonder why this is newsworthy since integrated photonics is not something new. That's true. But the introduction of solid-state silicon-based lasers is nothing short of revolutionary.
The discussion and research, thus far, on integrated electronics has hit a road block. Electronics is a silicon-based techology; photonics, for the most (and better part) is not. Specifically, photonic devices, and in particular laser emitters, are made out of a group of materials known as III-V (called three-five) materials, in reference to their position in the corresponding tables of the periodic table (consider, for example, gallium-arsenide GaAs).
Silicon is not a III-V material. It belongs to column II of the periodic table (notice that columnnar position refers to atomic properties and not to the actual column of the table. For example, column III in the periodic table is spread over actual columns number 3 and 13).
The fact that silicon and III-V materials do not share common chemical and crystalline properties, as implied by their different positions on the periodic table, is detrimental. The mismatch in their crystalline structure makes the monolithic integration of tiny laser emitters on top of silicon chips, impossible.
Yet we all agree that optical interconnections between computer components are the key for electronic computers to become better and faster.
Since monolithinc integration of lasers and CPUs was impossible, till now, because of the materials' mismatch we had to resort to the following limited ways of engaging photonics in computing:
(a) use of photonics for long-haul data transfer, ie, optical interconnects between entire computers, aka, optical networks; they are great and fast but we still face the bottlenect at the points of conversion between optical and electronic signals.
(b) hybrid optoelectronic chips; consider a silicon chip with pads on which a GaAs photonic chip rests. The two chips exchange signals thru these pads. The drawback here is the rather poor yields in fabrication and the high cost due to limited demand (and applications) for such devices.
(c) all optical computers. This was sort of a chimera for many researchers (myself included). While the idea and the concept are promising the implementation is extremely difficult and the promise of quantum computers, now, makes optical data processing a thing of the past.
Ideally we want a CPU chip made of silicon capable of emitting and receiving light. The photonic component was very difficult on silicon. Silicon is not an ideal material for coherent light emision, neither does it detect light easily. You need a larger area to sense light on silicon, than on GaAs, making silicon photodetectors rather large and thus affecting the scale of integration.
What Intel appears to have done now, is to introduce a way to monolithically integrate laser sources on silicon chips. They have solved a problem that has been open for years. Their solution will catalyze a field that has been waiting years for such a breakthrough. We knew what to do but we did not have the technology to do it. Intel just gave us the technology we've been expecting.
Quote from howstuffworks.com:
Laser light is very different from normal light. Laser light has the following properties:
* The light released is monochromatic. It contains one specific wavelength of light (one specific color). The wavelength of light is determined by the amount of energy released when the electron drops to a lower orbit.
* The light released is coherent. It is organized -- each photon moves in step with the others. This means that all of the photons have wave fronts that launch in unison.
* The light is very directional. A laser light has a very tight beam and is very strong and concentrated. A flashlight, on the other hand, releases light in many directions, and the light is very weak and diffuse.
I think the best answer here is that normal light source has all the power scattered around all wavelenghts. Laser on the opposite focus power on one wavelenght and has the above mentioned properties. So it can travel in fibers or in other substances/vacuum without scattering too much and losing its power too soon.
I want DLP micromirror devices combined with RGB silicon lasers now! One chip lightweight laser DLP projector!
As others have pointed out, yes, integrated optics has existed for some time, but that doesn't mean that the field is so mature that important breakthroughs can't occur. The field of integrated optics includes lots of things - III-V devices, (GaAs and InP), silica and silicon devices, polymers, etc. The thing they all have in common is that the devices are fabricated monolithically on a substrate.
You can generally break up the field into 2 catagories: materials that have great properties, but are a pain to process (GaAs, InP, LiNO3), and materials that are easy to process but aren't as great optically (Silica, silicon, polymers). Silicon is attractive because there is such a large amount of infrastructure available, and the hope is to be able to put CMOS and optics on the same chip. However, silicon has an indirect band-gap (this was sort of mentioned in the article by Because of silicon's crystalline makeup, energy from stimulated electrons is released as heat and vibration.). So, this means lasers can't be made by normal methods. In addition, modulators and detectors (for wavelengths longer than 1 um) are hard to do.
The solution for making a laser done by Intel here (and done earlier by Jalali at UCLA and even earlier by Osgood at Colombia) uses Raman scattering. Unfortunately, what the article leaves out is that you need another laser to pump the silicon laser. In addition, this laser is not just an ordinary diode laser, because you need very short pulses to get the peak power necessary for a non-linear effect such as Raman scattering. (The same limitation also occurs with the all optical switch done by Lipson at Cornell and mentioned here on Slashdot a few months ago). So this may be useful for some applications, but it's not a solution to the general problem of creating a light source in silicon.
No, they're typically made with other processes. One such method is with compound semiconductors using III-V materials like Gallium Arsenide. "III-V" refers to the general group of materials used on the periodic table - the materials are usually from those two groups. There are other methods of making lasers too, and I'm sure google can help you find information if you're intersted.
The big thing is that the processes are different from that which makes Silicon semiconductions, meaning that indutries that already have heavy investment in Silicon production technology would have to purchase new capital equipment. Compound semiconductors are generally more versitile than silicon in general, but are more expensive to make.
Its kind of a race though, since there have been some big pushes in the compound semiconductor equipment industry in recent years that are reducing some of the negatives from years past.
I think the immediate usefulness of this in regards to a cpu would be the IO. I think the chip core would still be traditional, but instead of having 1000 pins or whatever, you could change buses to be serial and reduce the pin count considerably.
The speed of electricity varies with voltage and current, but I think the generally accepted value is 1/3*c. So there would be first order speed gains.
But there is also bandwidth too, much more information can be encoded and sent over fiber than can be sent over metal. So now you could have very large parallel memory interfaces. The width might be measured in hundreds of Kbits instead of bits in the near future. The limit would be the speed of the shift registers on the silicon side of the cpu and memory.
There are also benefits from being electrically isolated... ground bounce could be a problem of the past with this.
And electromagnetic radiation could be less of a problem too (spies as well as cross talk).
Also, in the same amount of time, say a nanosecond, electricity only travels 10cm (approx), where as light travels 30cm. So you can make faster systems on a larger area (distributed or monolithic even)
So there is a lot more too this and an all light cpu.