Performance differences of programmable multimode interferometer platform for different targets in inverse design

Recently, researchers have proposed a kind of programmable multimode interferometer (MMI) platform. By leveraging the inverse design, this kind of compact device is very promising for next generation of silicon-based photonic processors. However, the design capabilities of the programmable MMI have not yet been investigated. On the other hand, the optical performance of the inverse-designed device is usually determined by both the optimization algorithm and the platform itself, and one may wonder how much the algorithm or the platform affects the design results. In this paper, we distinguish between algorithm optimization capability and platform capability by introducing the random structure targets and the random function targets. The results show that for optical field amplitude targets, the device performance in inverse design is excellent, largely due to the algorithm's optimization, which has a more significant impact. However, for transmission matrix targets, the MMI platform itself becomes the primary limiting factor. This work provides a unique way to investigate the design capabilities of the programmable MMI, and, to some extent, also provides guidance for the further development of the programmable MMI.