Current servers may use up to 32 silicon photonics transceivers at the 100-Gb/s rate for the traffic between servers. The next-generation interconnects will need the same number of 400-Gb/s chips to achieve a quadrupling of capacity. After that, if data capacity increases fourfold as expected, today’s approach of using individual pluggable photonic transceivers with separate microelectronic switches will not work. Instead, small photonic chips, which Blum called “chiplets” or “photonic engines,” will be placed next to the electrical networking chip to integrate photonics and electronics in a single multichip package.

https://www.photonics.com/Articles/Data_Centers_and_More_for_Silicon_Photonics/a64879