Q&A: Fulfilling A Vision For High-speed Interconnects
In 2017, Sicoya -- a Germany-based developer of highly-integrated silicon photonics and spin-out of TU Berlin -- launched its first products into the data centre market. Sven Otte, the firm's CEO, explains why we need Photonic Integrated Circuits (PICs) to deliver high-speed transceivers in the large quantities that customers are calling for. Interview by James Tyrrell.
Q1 - What do you see as the big benefits of PICs and how is Sicoya applying this technology to deliver attractive solutions to its customers?
We are now going into an era in which we need ultra-high speed transceivers. Huge demand is being driven by both machine-to-machine communication and mobile communication, but applying discrete manufacturing approaches to 100x higher volumes is impractical - the cost doesn't scale with volume any further and quality can be affected. Photonic integrated circuit (PIC) technology addresses these challenges - the volume constraints quality and cost.
Co-integration solution: Sicoya's transceiver chips feature analog electronics and optics fabricated, using SiGe-BiCMOS process technology.
There is some debate in the industry about which PIC technology to use "“ for example, are silicon (Si) or indium phosphide (InP) PICs better suited for future solutions? But we see different PIC platforms becoming commercially available and that's good because each technology addresses a specific problem. Silicon is the better material system from a manufacturing point of view. It is, practically speaking, available in abundance and offers outstanding material properties when, for example, looking at the number of crystal dislocations or the surface quality. As a result, the cost of Si is 10x lower than InP. In addition epitaxial growth and lithography processes in Si outperform their InP counterparts. From here we can easily see why processing 300mm Si wafers with very high yields is achievable while the size of InP wafers is limited to 80mm and yields are lower.
Silicon devices, both the electrical and the photonic components, tend to be much smaller because of the lithography processes and the material properties - for example, we can build very high-speed modulators that are less than 100 micrometers long in Si while in InP the same modulator is around a centimetre in length. Overall, this leads to a situation where all building blocks that can be achieved in Si will be made using Si eventually. Obviously, Si is an indirect semiconductor and hence cannot generate photons. So, consequently, we will continue to use InP, but only where needed - for example, to create the laser.
However, even lasers can be broken down into subparts such as the gain medium and the grating. Gratings can be made in silicon and for the gain medium a thickness of a few hundred nanometres is sufficient, which can be integrated in a CMOS process. Indium phosphide may therefore become a complementary material in an overall Si wafer manufacturing flow just like germanium is in today's BiCMOS processes, for example. This however is in the future. Today InP and silicon photonics co-exist.
At Sicoya, we bring the integration approach to the boil and we are combining entire transceivers into a single Si chip, which we call "˜Module-in-Chip' (MiC) technology. All electrical transceiver functions such as clock and data recovery, control loops, amplifiers, modulator drivers and all optical functions including modulators, multiplexers and photodetectors are integrated in one small chip.