Abstract
The selection of the capacitor bank presents a significant challenge in pulsed modulators. Pulsed-film capacitors are used commonly, but they show high cost and low capacitance. Alternatives that satisfy performance requirements while minimizing voltage droop and cost are required. This work presents the design of a solid-state modulator rated at 80 kV and 110 A, using only aluminum electrolytic capacitors. In this study detailed schematics and PCB layouts are developed, which ensures fabrication readiness. The capacitor bank configuration is formulated as a constrained multi-objective optimization problem with objectives to reduce voltage droop, ESR losses, and cost. NSGA-II is applied to generate Pareto-optimal solutions for six capacitor candidates. A surrogate deep neural network (DNN) is trained to approximate the mapping between design parameters and performance, achieving a coefficient of determination (R-squared) close to one on validation data. The optimized design achieves a cost reduction of thirty percent, voltage droop below two percent, and ESR losses under one kilowatt. Key contributions include the replacement of pulsed-film capacitors with electrolytics for the DC-link and the integration of evolutionary optimization with a surrogate model, which provides a scalable and fabrication-ready method for medical and industrial Linear Accelerator (LINAC) modulators.