Ultra-compact and ultra-broadband arbitrary-order silicon photonic multi-mode converter designed by an intelligent algorithm.

Multi-mode converters, which can achieve spatial mode conversion in multimode waveguide, play a key role in multi-mode photonics and mode-division multiplexing (MDM). However, rapid design of high-performance mode converters with ultra-compact footprint and ultra-broadband operation bandwidth is still a challenge. In this work, through combining adaptive genetic algorithm (AGA) and finite element simulations, we present an intelligent inverse design algorithm and successfully designed a set of arbitrary-order mode converters with low excess losses (ELs) and low crosstalk (CT). At the communication wavelength of 1550 nm, the footprint of designed TE0-n (n = 1, 2, 3, 4) and TE2-n (n = 0, 1, 3, 4) mode converters are only 1.8 × 2.2 µm2. The maximum and minimum conversion efficiency (CE) is 94.5% and 64.2%, and the maximum and minimum ELs/CT are 1.92/-10.9 dB and 0.24/-20 dB, respectively. Theoretically, the smallest bandwidth for simultaneously achieving ELs ≤ 3 dB and CT ≤ -10 dB exceeds 70 nm, which can be as large as 400 nm for the case of low-order mode conversion. Moreover, the mode converter in conjunction with a waveguide bend allows for mode-conversion in ultra-sharp waveguide bends, significantly increasing the density of on-chip photonic integration. This work provides a general platform for the realization of mode converters and has good prospect in application of multimode silicon photonics and MDM.