First off, the reason you want light for transmission on chips isn't because it is "faster", it is because it is less susceptible to electromagnetic interference. The speed of light in a medium is the speed electronic signals travel, and the speed light waves travel. However, electronic signals bleed interference, and light waves don't. You also can switch light waves on and off at theoretically higher rates, with less power, so optical signals can break the 4 GhZ barrier we are currently stuck at.
Secondly, they can't bend light without loss. Just with less loss. But the interesting part of the paper is the fact that the turning is selective depending on polarization (one polarization turns, and one doesn't). They are also able to turn light very quickly, compared to the wavelength of the light. So the radius of curvature divided by the wavelength of the light is around 6.4 for this paper.
I believe the most important part of the article is:
"The team then ran light beams through the lattices and confirmed that they could flow light (without loss) through turns that are twice as tight as any done previously."
If you can bend light without loss, I imagine that is very valuable. All that said, IANAP (I am not a physicist) so if someone else can correct me or add more information, I'm sure that would be helpful.
It's not useful to transmit light through air; instead waveguides are used[0]. I don't know all the benefits, but I imagine trying to keep a path through space for light beams to travel is harder than using a waveguide. In that case, mirrors are useless. You can think of a waveguide as being a highly engineered mirror, really. However, the turning radius of a fiber-optic beam isn't great, and apparently this is an improvement on that.
One reason is because mirrors (freespace optics in general), are limited in how narrowly confined the beam can be. Generally speaking, waveguides can have cross sectional area of ~ λ^2 or less, while freespace optics can only get below that for some small focus volume. So a chip with mirrors on it would have to be much much bigger to handle the beam width of all the laser light.
First off, the reason you want light for transmission on chips isn't because it is "faster", it is because it is less susceptible to electromagnetic interference. The speed of light in a medium is the speed electronic signals travel, and the speed light waves travel. However, electronic signals bleed interference, and light waves don't. You also can switch light waves on and off at theoretically higher rates, with less power, so optical signals can break the 4 GhZ barrier we are currently stuck at.
Secondly, they can't bend light without loss. Just with less loss. But the interesting part of the paper is the fact that the turning is selective depending on polarization (one polarization turns, and one doesn't). They are also able to turn light very quickly, compared to the wavelength of the light. So the radius of curvature divided by the wavelength of the light is around 6.4 for this paper.