"The first approach is to take a CMOS process and try and adapt it for photonics. The second approach is to develop a process optimized for photonics from the ground up and which therefore has little resemblance to established processes."
Rickman said he understood why major foundries were trying to do the former. The less they have to change their established processes the more they can leverage the money they have invested in R&D and specific manufacturing equipment. "TSMC, Globalfoundries, STMicroelectronics are in the first camp, building waveguides on adapted CMOS. In principle, it is a good vision. You can push for integration of photonics with the microelectronics."
However, Rickman said that ultimately vision fails because of the physics of photon behaviour in silicon as opposed to that of electrons in silicon. Ultimately it comes down to the fact that the wavelength of the photon is of the order of micron while the wavelength of the electron is of the order of a nanometer.
Rickman said there are specific disadvantages to using the nanometer-scale of modern CMOS for optical integration, including increased waveguide loss, wavelength sensitivity, fiber coupling loss, polarization sensitivity. Similarly, while it may be possible to build modulation circuitry in silicon, indium-phosphide is superior and being open to III-V materials provides broader options and capabilities. Finally, Rickman pointed out that use of trailing-edge silicon manufacturing capability at the micron dimension brings reduced manufacturing cost although with some additional manufacturing steps. He added: "There will always be some packaging."
Rickman added that it is already accepted in the microelectronics field that for many applications multi-die components that use different processes can be the optimum production method. MEMS provide a good example. And with those lessons learned in MEMS and other applications, the same packaging infrastructure can be available for the production of silicon photonics ICs.
Rockley's first application is the 400G capable data center. In 2018 Rockley demonstrated its platform could produce the world’s first single ASIC Level 3 datacenter routing switch with integrated 100G network ports using single-mode optical fiber.
And the first customer is Hengtong Optic-Electric Suzhou, an optical fiber and cable provider, based in Suzhou, Jiangsu, China. Rockley formed a joint venture with Hengtong to make 400G DR4 transceivers using Rockley's LightDriver 1310nm wavelength optical engines. Hengtong also participated in a $30 million investment round in Rockley. The technology is good for the new breed of datacenters that supporting cloud computing, artificial intelligence and machine learning. "It's a great route into the Chinese market for us," said Rickman.
Next: Sensing, China