One problem, however, is that Si is not a direct band-gap material, making it very difficult to use to create lasers, semiconductor optical amplifiers (SOAs), or other gain elements. Thus, the industry began searching for ways to integrate the best III-V lasers and gain elements with Si. The traditional solution to this problem, and how most of photonic systems are built today, has been to use discrete III-V laser chips and other optical components to create and amplify light and then bring that light into the Si PIC using optical fiber, usually by aligning the fiber with grating structures on the PIC to guide the light into the horizontal plane of the Si PIC.
While this approach works (and is used in all data centers today), it comes with its own share of problems. For example, roughly one-third of power is lost at every interface between laser, fiber, and PICs. To compound the issue, each photonic device on the PIC also has insertion loss and eventually, if you string together enough components, you run out of power. This limits design complexity that can be supported. In addition, the waveguides used in PICS are small and can only carry so much power, so you can’t simply turn up the laser power to accommodate for all the losses because this would damage the PIC.
Even with the cheaper silicon-based PICs, the cost of the overall photonic systems has not come down as desired since lasers, fibers and PICs must all be carefully aligned to transfer light from one to the other. This implies expensive manufacturing processes and a system that is difficult to use in harsh environments with large temperature swings, vibrations, or anything that could put the system out of alignment. Another problem is that individual lasers and fibers take up space in the system and each comes with their own probabilities of failure over time. The more discrete parts there are in a system, the higher the probability of failure caused by any one of them.
Efforts have been made to reduce cost and SWaP (size, weight, and power) of photonic systems by trying to integrate lasers more tightly with Si PICs. Epitaxial growth of III-V materials on Si has proven problematic due to different lattice structures and dislocations caused by defects in the materials.
Some companies have successfully etched holes in Si PICs and then bonded III-V laser chiplets into the holes, leaving a small air gap between the laser and the waveguides of the PIC. This works well, but the manufacturing process is complex since it requires very precise 3D alignment of the bonded chip to align the output of the laser with the Si waveguides. Designers must pay special attention to the interface facets of the laser and the PIC to avoid unwanted reflections and loss. Nonetheless, the method does replace two lossy transitions (laser to fiber and fiber to PIC) with one less lossy transition, and it removes optical fibers to simplify system assembly.