The objective of the proposed research is to  enhance conventional topology optimization in three areas: (1) demonstrate topology optimization for devices with nonvertical etches and layers, (2) devolop topology optimization routines to maximize nonlinear effects such as second harmonic generation, and (3) train machine learning enhancements on initial design runs to improve device performance and decrease optimization runtime of all future devices. The contribution of these aims is to make topology optimization applicable to a larger class of design problems. The topology optimization improvements for geometries with nonvertical etches and layers are a realistic, real-world challenge. They are commonly used fabrication techniques and by developing algorithms for these types of devices, the scope of topology optimization is improved. Nonlinear effects have seen little work done in a topology optimization context. For this reason, nonlinear devices typically implement intuitive device geometries such as waveguides, rings, and cavities. Extending topology optimization to these problems can drastically improve the performance or decrease the footprint of such devices, as has already been demonstrated in linear structures. Training reusable filters seeks to mitigate the computation complexity of topology optimization by preconditioning optimizers to quickly find high-performing devices, without the need to start from zero for every device. This will dramatically improve the utility of topology optimization for problems such as designing a process design kit, or designing multiple devices with slightly different purposes (e.g. arbitrary ratio splitters or devices operating at different wavelengths).