Todays networks carry electrons and photons (both electromagnetic fields but with different characteristics). The advent of optical fibres, to carry photons, has multiplied the networks capacity and decreased their cost (particularly the ones related to increase the capacity since once you have laid a fibre you can increase its capacity several folds just changing its termination points as electron to photon conversion technology progresses.
The bottle neck is this conversion of signals carried by electrons into signals carried by photons and viceversa. This conversion is a must because we have tools to switch electrons but we don't have tools to effectively switch photons. For this we need photonics and in order to make it affordable we need silicon photonics, that is using silicon to switch photons.
We already have in our networks plenty of optical signals management capability, like in optical add drop multiplexers or in GPON (Gigabit Passive Optical Network) but in all these cases we are managing an optical flow (imagine a beam of light that thanks to some mirrors of a sort gets multiplied or split).
We have also seen some photonics switches but they are (expensive) labs prototypes.
Now IBM has announced the development of a photonics tranceivers -industrial grade- that can switch 100 Gbps flows. This is exactly what we need to transition from current networks to full optical networks where electrical signals are substituted by light signals.
The transceivers have been manufactured with a sub100nm technology and can process up to four wavelengths at a time (like processing signals coming in four light colours). The throughput (speed) of the transceiver is such that it can process a full HD movie in 2 seconds.
The first applications of these transceivers are likely to be in big data centres, helping the processing of huge batch of data (Big Data) and only on a longer time frame will be used in the big networks. The mail reason is the different evolution speed of data centre vs networks. The latter evolve over long time period of time, requiring massive investment, whilst the former represents instances that can be evolved asynchronously.
IBM has reported on a trial where these transceivers have been provided signalling over 2km stretches of fibres, plenty for the distance involved within a data centre.