Photonic switching is already a major player in today's telecommunication networks. There are thousands of photonics add-drop multiplexers in today's network to route optical data flows and you can also claim the GPONs (Gigabit Passive Optical Networks) are photonics switches of a sort (they actually split the optical flow in pairs duplicating it, they don't really switch it).
Photonics switching has become a sort of holy grail in telecommunications networks for its promises of high bandwidth and low cost.
The problem so far has been that real photonics switching requires relatively bulky equipment (and in electronics size equates to cost/price) and increasing the switch matrix (the number of input output channels) increases the power loss of the signals switched hence requiring their amplification, thus further increasing cost.
This is where the news from Berkleys comes handy. In a presentation at OFC2015 the researchers have presented a photonics switch on a chip, 1cm per 1 cm size, containing 50 input channels and 50 output channels sustained by a switching matrix of 2,500 points.
The breakthrough is the technology used that made it possible to cram all those switching points in just one square cm and the use of adiabatic switches that are not affecting the signal power (no power loss - almost) and do not require any power (almost) to work.
"Adiabatic" processes are not invalidating the second law of thermodynamics (for any transformation you've got to give something away, that is thermal energy), they are part of the first principle of thermodynamics where there can be processes that are so fast that no actual exchange of energy occurs (hence they don't get caught in the second law). Of course having a 100% adiabatic process is not practically feasible but you can get pretty close to it.
With an ingenious use of silicon grating and MEMS (Micro ElectroMechanical Systems) mirrors (also created by silicon grating) the researchers have been able to switch an incoming flow of light only in the point where it is needed (whilst current photonics switches will take the optical flow through many switching points till it reaches the one where it gets actually switched).
This has resulted in a chip that can switch optical signals at a third of the power budget of today's optical switches and at a rate that is a thousand times faster (down to microseconds rather than milliseconds).
These chips will eventually flatten out the network and will support full optical interconnection end to end, providing high bandwidth at lower cost. More and more telecommunications networks will look like data networks with big data centres and small (TB order of magnitude) data centres interconnected. Of course the small ones will be in our homes and in our hands, we call them cell phones, tablets, televisions, home computers....