CADAIT: a code for automatic design and AI training of microbeam systems

Abstract

A focused microbeam system with ion beams at MeV energies is a unique tool for material science, biomedical applications, and space risk evaluation. Microbeam system design traditionally relies on experienced knowledge of microbeam optics and many elaborate calculation procedures. In this work, an ion optics design code, CADAIT, is developed to design microbeam systems automatically. For a given microbeam layout, it allows for the automatic optimization of focusing conditions, the calculation of optical parameters, and the size of the focused beam through ray tracing. CADAIT enables the automatic optical design of microbeam layouts under input parameters and the selection of microbeam layouts with high performance. The accuracy of the CADAIT is verified with ion optics software packages (WinTRAX, Zgoubi, and FANM), which show good agreement. The evaluation of the performance of existing microbeam facilities with CADAIT and the application of CADAIT in the automatic design of a 12 MeV proton microbeam system are discussed. Thanks to its high efficiency in the optical design of microbeam systems, the CADAIT code is used to train artificial intelligence (AI) models for the intelligent design of microbeam systems with tremendous CADAIT-generated data. The artificial intelligence trained model, Artificial Intelligence Microbeam Producer (AIMP), is demonstrated to be capable of generating microbeam systems with superior performance and robust layouts within one minute. The above results show that CADAIT can significantly decrease the complexity and duration of microbeam optical design and prove the feasibility of intelligent microbeam design.