Very cool indeed.
There has been talk of using light paths instead of electrical
paths inside of high-speed chips for at least a couple of decades.
It's nice to see someone finally pursuing it in earnest.
It will be interesting to see how this develops over the next few
years, especially as we think about the per-unit production costs,
and how they go about keeping those down without sacrificing
performance.
Thanks for the great post!
There’s a big bottleneck in front of Moore’s Law, whether you apply it to chips or the networks that carry data.
Electricity. (Picture from Wikipedia.)
This has been known for decades, since the old AT&T first started spinning optical fibers from glass at a plant in Norcross, Ga., near my home.
I got a tour of the plant while working for a local newspaper and the great unanswered question was how you connect the fibers together.
For a long time the answer was a simple switch. Data would be turned from light to electricity, run through the switch, then get turned back into light.
Today we have all-optical switches, so even though fibers no longer send a single beam of white light, but a stream of signals in a rainbow of colors (seen and unseen) our networks can handle the load.
There remains the problem of scaling that down. How do we move data between chips, within systems, at the speed of light rather than the speed of electricity?
Scientists at UC San Diego say they now have the answer — a laser measuring one micron in size, built with super-thin layers of silica and aluminum, that works at room temperature and emits a beam whose wavelength is 1.4 microns, larger than the laser itself.
These nanolasers can be packed onto a wafer with the aluminum acting as a heatsink. Lasers can be run as close together as circuit lines on current chips. The pulsing of the light can represent 1s and 0s, read on either side of the junction by a diode.
Nanolasers like those made at UC San Diego could hit commercial production in just a few years, built into optical networking gear or, perhaps, directly onto chips, reducing that bottleneck between electricity and light still further.
So no matter how fast you think you think tomorrow’s chips and networks will think faster, because our chemical brains are still running on electricity while tomorrow’s networks will run on light.
