Prediction of photonic wire bonding waveguide parameters based on convolutional neural networks

Traditional photonic wire bonding (PWB) design relies on iterative electromagnetic simulations, suffering from low efficiency and high computational costs when dealing with multi-structural parameter coupling. To address these limitations, we present a convolutional neural network (CNN)-based key-parameter prediction model. Our end-to-end model intelligently predicts core structural parameters—including taper dimensions, bending strategies, and waveguide diameters—across the entire PWB chain. Results show that the CNN accurately establishes the implicit relationship between geometry and optical transmission characteristics. Compared to conventional methods, this approach dramatically reduces computation time while maintaining high prediction accuracy, thus providing an efficient pathway for lightweight design and global optimization of high-speed photonic interconnects.