A corrective motion was then performed, in the direction of the observed gradient, the step size perhaps proportional to the error signal and subject to limits to prevent overshoot. The process is then repeated. As (the condition of alignment) the system naturally stops when alignment is optimized, yet it can realign when necessary due to drift or other disturbances.
PI’s LightLine alignment system implemented a novel piezo stepping mechanism to achieve similar aims, and in 1997 PI introduced its H-206 photonics alignment hexapod, a micro-robot that implemented its own hill-climbing algorithms and scans and could perform alignments and precision closed-loop motions in six degrees of freedom, with rotations cast by software to any desired center point, such as a beam waist or focal point. This proved essential for the packaging applications that emerged in the industry’s first great blossoming from 1997-2001. Many packaging applications of that era involved angle-sensitive devices, such as confocal optical trains, small lensed assemblies that required alignment in multiple degrees of freedom.
These options served the market well through its original boom phase and then through its long doldrums. But with the advent of Silicon Photonics (SiP), new challenges necessitated new solutions.
SiP Devices Are Different
Silicon Photonics is the fabrication of planar optical devices and circuits, often alongside microelectronics, using materials and processes familiar from integrated circuit microlithography. The dozen years of industry doldrums that began in 2002 allowed the patient development of enabling materials, such as germanium-on-silicon and high-index-contrast films and the requisite building-blocks like couplers, resonators, interferometers, detectors, and the all-important laser integration approaches.
As with microelectronics, SiP devices are formed in layered fashion on the surface of a silicon substrate. The resulting waveguide channels are typically rectangular in format, presenting substantially different optical physics than the cylindrically symmetric geometries familiar from single-mode fibers and most free-space laser applications. Device characteristic dimensions can also be much smaller than was typical back in the halcyon days of photonics’ youth. Often, diffractive couplers are used to kick light out of the wafer plane, making in-wafer devices accessible without dicing and facilitating some packaging approaches for the final devices [7]. Coupling profiles are often asymmetric, astigmatic, and (especially in the case of short waveguide devices) multi mode. It’s common to see coupling cross-sections that have a narrow, knife-edge profile in one dimension and a much broader profile in the orthogonal dimension. And the inputs and outputs of short waveguide devices can interact, with slight changes of alignment position at each of the N inputs potentially steering the aligned position at the M outputs. So for these devices, overall alignment is a moving target on top of it all.
These all add up to significant alignment-process challenges for traditional techniques and mechanisms. The small dimensions and odd, non-circularly-symmetric profiles are problematic for legacy alignment mechanisms and algorithms. And even the simplest waveguide device presents an input and an output, both of which need to be aligned to source and receiving elements such as a laser or fiber. Since input and output can steer each other in many such situations, conventional alignment mechanisms often need to be actuated in an iterative, looping sequence until a consensus alignment is achieved. Add the angular optimization required of arrayed devices and many others, and the alignment process times quickly escalate in an economically prohibitive way. Then there are drift processes, which necessitate similarly complicated and time-consuming sequential realignments.
