The auxiliary power supply for medium voltage converters requires high insulation capability between the source and the load. Inductive power transfer technology, with an air gap between the primary and secondary coil, offers such high insulation capability, making it a potential candidate for auxiliary power supply in medium voltage converters. However, the large air gap between the primary and secondary coil typically results in a loosely coupled inductive power transfer system, necessitating optimization of the inductive power transfer system to achieve high efficiency and power density. This thesis focuses on the coil link optimization. It introduces a novel design of the coil link structure, models the coil pairs, and presents an optimization flow to design optimal winding geometry based on given electric specifications. The research of this thesis revolves around PCB coils, which are favored for their easily controlled manufacturing process. To maintain modularity in medium voltage converters, the inductive power transfer system consists of multiple coil pairs, each supplying one power electronics building block. The primary side of each coil pair is connected to a common power source. To ensure independent operation of the secondary coils without closed-loop control, the LCL-S compensation network is utilized. The advantage of this compensation network is analyzed and its design process is discussed in this thesis. The characteristics of a coil, including self and mutual inductance and coil losses, are defined by its winding geometry. Therefore, by optimizing the winding geometry, high coil link efficiency can be achieved. This thesis develops a model to calculate the magnetic field inside each winding turn when no ferrite is present behind the winding. The brute-force optimization is conducted on the PCB coil pairs with the proposed model, resulting in a Pareto front showcasing the trade-off between high efficiency and high power density. When ferrite is added to the backside of a winding, it alters its inductance and resistance, and therefore has the potential to increase power transfer efficiency. However, the modeling of the coil pair becomes more complex due to the crowding field around ferrite edges. In this thesis, different ferrite shapes are compared, and a magnetic model with 5 mm thick round shape ferrite is proposed, based on a database from finite element simulation together with an artificial neural network, predicting the coil pair characteristics. The validity of the finite element simulation data is pre-verified with impedance analyzer and power tests, and the accuracy of the magnetic model is confirmed with simulation test datasets and characteristic tests on one coil prototype. The design of the coil link requires thermal coordination considering the temperature limit of each component inside coils. This thesis proposes a thermal model to predict the temperature rise in coil pairs. The proposed electric circuit model with LCL-S compensation network, the magnetic model based on artificial neural network, and the thermal model based on thermal network are all independent from external simulations and easily integrated into the optimization flow, ensuring fast optimization time. After exploring all degrees of freedom of winding geometries, one coil pair on the Pareto front is selected and tested under various load conditions.