Fiber optic interconnects have enjoyed a relatively long history in long haul telecommunications due to their ability to transmit large quantities of data across great distances. As our society’s insatiable demand for communications bandwidth grows, we have seen the emerging necessity of optical solutions for ever shorter links, and at a much larger scale. Whereas conventional optical systems were used predominantly for links spanning thousands of kilometers, today’s state-of-the-art data centers require them for most links greater than 5 meters in length. With no signs of the rapid increase for bandwidth diminishing, within a few years the need for optics will take its next big step beyond the enterprise world and into the mainstream consumer markets.
Despite steady growth of the optical industry, the majority of the design practices have remained the same. Optical systems historically have been assembled out of a collection of discrete components, and integrated photonics solutions have seemed as elusive as the monolithic integrated circuits that appeared in the middle of the 20th century. In response to this limitation, we have seen the emergence of a CMOS photonics industry, which has revolutionized the high-speed interconnect industry through integrating optical functions into conventional CMOS processes. Beyond being able to integrate a family of optical components, CMOS photonics design platforms have enabled the design of full ElectroPhotonic Integrated Circuits (EPICs) and systems.
The benefits of EPICs are many and continuously unfolding, but their biggest impact is the ability to drive significant costs savings, both when employed in place of discrete optical solutions in a standard space, or when pulled into systems that otherwise could not have accommodated optical systems. By enabling fabrication in conventional CMOS foundries, EPICs can reap the scalability and cost benefits inherent to the semiconductor industry, as well as leverage all the advances realized in that industry to push the technologies ever further.
Fabrication of an EPIC
While many EPIC solutions have been demonstrated using post-processed silicon wafers, the most significant technologies enable full integration into a CMOS foundry flow. There are two general approaches for such solutions: front end of the line (FEOL) and back end of the line (BEOL) integration, although thus far only FEOL integration has been demonstrated at production scale. FEOL integration uses an SOI substrate to enable integration of optical components in the same active silicon as electronic components, where the SOI substrate is necessary to provide a bottom cladding for optical structures. The two major benefits of FEOL integration is that it can use the same doping steps employed in electronic component design to generate active optical components as well as enable very small scale integration between electronic and optical components (compared to the larger geometries required in BEOL routing). A cross section of FEOL CMOS Photonics integration can be seen in Figure 1.
The key optical components of a CMOS photonics EPIC are all based on the basic waveguide structure, which is etched into the active silicon layer within a given integrated circuit. Out of that, modulators and photodetectors can be built (the latter using a germanium deposition process), as well as vertically incident optical I/Os constructed using reflective gratings in the same active silicon layer. While not employed in all CMOS photonic processes, the last component is vital to ensure a full EPIC solution. It is the only way to enable wafer scale electro-photonic tests and the only solution that doesn’t violate long term IC reliability rules for a secure IC guard ring.
While simple in concept, CMOS photonics solutions have needed to overcome a number of challenges to become product worthy, most of them attributed to the requirement to maintain compatibility with a standard silicon foundry. While deviations in the fabrication modules do exist (most notably for germanium deposition to create photodiodes), every customization must be possible using the standard foundry tool sets and without interrupting the foundries ability to fabricate non-photonic technologies.