Using the current method of characterizing waveguides manufactured for research in optical communications, it can take up to 8 hours to characterize the waveguides on a single silicon wafer. Using the automatic computer automation system developed, the characterization process has been reduced to less than an hour. After the silicon wafer containing the waveguides is manufactured, the waveguides must be evaluated to determine the optical characteristics of the various waveguides on the silicon wafer.
The figure shows the current setup used to measure the optical properties of the waveguides on a silicon wafer. The silicon wafer is approximately 1 inch square and can contain as many as 50 separate waveguides that need to be analyzed. In this setup, laser light is passed through an optical fiber into one end of the waveguide. At the other end of the waveguide, another optical fiber is used to capture the light exiting the waveguide and is measured using a power meter. Each fiber is attached to x-y-z stages, which allows for fine adjustment of the fiber along three directions. This fine adjustment is needed in order to get the laser light into the waveguide as well as into the fiber on the exit end of the waveguide. Thus, there are six stages used to maximize the laser light passed through the waveguide. In addition to the six stages, there are two stages that are used for coarse adjustments. The coarse adjustments are only along the x-direction, and are used for the initial alignment of the two fibers and for the displacement needed to move to the next waveguide.
When a new wafer is placed into the setup, the coarse adjustments are used to get the two ends of the fibers close to the desired location. Next, the six fine adjustments are used to maximize the power through the waveguide. Once the desired data has been gathered, the coarse adjustments are used to go to the next waveguide, and the adjustment process is repeated to maximize power through the waveguide.
The automation of the system to align the fibers to characterize the optical waveguides required the motion control and power measurements to be conducted through a single computer interface. LabVIEW™ software from National Instruments was used to design the computer interface.
The controllers contained a BNC input connection for each axis. The BNC connections allowed for a 0-10 volt input analog signal to be applied for the motion control of the x-y-z fine adjustment stages. Using a USB-3103 eight-channel analog voltage output device, it was possible to control all three fine adjustments for each of the fibers via LabVIEW. The optical power meter contained a USB interface that allowed it to be connected to the computer, and power measurements could then be made using LabVIEW. To optimize the power through the waveguide, two techniques were investigated: the hill-climbing method and the Simplex method.
The hill-climbing method is the most widely used algorithm in the current photonics automation industry. In the hill-climbing method, each movement is based on the comparison between current and previous output light intensity measurements. Motion continues along a particular direction as long as the measurement increases. When the measurement decreases, motion is stopped and then moved to the previous position where the measurement was a maximum. Then the process is repeated for the next direction.
A simplex is defined as a convex hull with N+1 vertices in an N-dimensional space. These vertices satisfy the non-degeneracy condition that the volume of the simplex hull is nonzero. Each dimension corresponds to a variable or factor in the optimization procedure. Thus, a two-dimensional simplex is seen to be a triangle while a three-dimensional simplex is seen to be a tetrahedron. The algorithm used in the simulation was then used to generate a LabVIEW virtual instrument (VI) for the simplex method.
The main difference between the hillclimbing method and the simplex method is that the hill-climbing method works with only one axis at a time while the simplex method uses two axes at once. Because of this, the simplex method should locate the desired fiber positions that resulted in maximum power quicker than the hill-climbing method. In testing the two methods, it was determined that the simplex method was very sensitive to the values of the parameters used to determine new points for the vertices of the triangles. By making small changes, it was possible for the method to locate the desired position, but when a new starting position was used, the method would often fail and the parameters had to be changed to find the desired location.
The hill-climbing method was more robust and more successful at locating the desired fiber positions for maximum power. In terms of time to converge, the simplex method was about twice as fast as the hill-climbing method in locating the desired position. The hill-climbing method was selected as the method to use for optimizing the power through the waveguide because of its robustness and ability to consistently locate the maximum power position.
This work was done by Ryan P. Lu, Chunyan Lin, and Ayax D. Ramirez of SSC Pacific, and Gabe V. Garcia of New Mexico State University. SPAWAR-0001