Thesis title: Enabling Artificial Intelligence in Photonic Integrated Circuits Manufacturing
The consistently growing demand for robust automated assembly, testing and packaging of Photonic Integrated Circuits (PICs) is increasingly oriented towards high volume and continuously sets new challenges to overcome concerning throughput and cost-efficiency. Production processes' intrinsic complexity, combined with short product life cycle and the necessity of quickly ramping up those to high volume, require smarter solutions to guarantee high yield as well as low cycle time. Robust automated solutions for manufacturing and testing of PICs, demand for complex systems capable of realizing high speed repeated movements over variably large ranges with nanoscale precision. Motion specifications are of a fundamental importance to enable optical microscopy for high precision passive positioning and active positioning based on optical feedback from assemblies which features size are below the wavelength of light. Research of innovative motion solutions that reduce positioning errors, is therefore at the foundation of state of the art machines that implement production processes of PICs. Among the predominant positioning error sources that directly impact on the quality and feasibility of manufacturing processes, thermally induced position drift and the machine tool tip positioning repeatability, are identified as a major cause of deviation from required operational conditions. Changing temperature of the workspace as well as the excess heat generated by the machine itself, can induce position drifts upwards hundreds nanometer per minute that prevents the axis systems from meeting positional requirements for optimal functioning, eventually imposing the integration of expensive temperature control systems to minimize the impact of thermally induced drift. Similarly, motion error components introduced by the execution of motion curves optimized for speed, pose at risk the quality of assemblies and high precision fiducial measurements performed with the vision system; residual mechanical vibrations and instability phenomenon have a strong impact on both position errors and measurements precision. Fiducials repeatability and robustness to variability in the assemblies, workspace and mode of operation, is critical to achieve fully automated reproducible systems, thereby ideal working conditions are preserved by slowing down motion, ending up trading cycle time for precision. Abundance of heterogeneous data, such as images, measurements, sensor readings, motion controller data and process logs, contain valuable information useful for the optimization of important key performance indicators, such as overall equipment effectiveness, throughput and yield. In this work, data driven modeling methodologies of positioning inaccuracies are proposed and framed within a model-based optimization fashion that opens up the possibility of enabling the use of artificial intelligence in key tasks required for PICs manufacturing.