Robust inverse design of digital photonic devices for photonic integrated circuits

Digital nanophotonic devices have achieved both an ultra-compact footprint and flexibly designed functions, which hold great potential for ultradense photonic integrated circuits. However, the performance deteriorates rapidly when the size of pixels varies due to fabrication errors. In this work, we propose an inverse design method called robust adjoint method to design fabrication-tolerant digital nanophotonic devices. In the proposed method, multiple patterns with identical pixel distribution but different pixel sizes are taken into account, ensuring that the optimized designs maintain high performance when the pixels suffer diameter variations. We experimentally demonstrate a three-mode (de)multiplexer and a three-mode broadcasting photonic integrated circuit, both designed by the robust adjoint method. For the three-mode (de)multiplexer, the device achieves insertion losses < 2.0 dB, and crosstalk < -17 dB across the wavelength range of 1530-1570 nm. When it suffers ±10 nm diameter variations, the device maintains insertion losses < 2.5 dB, and crosstalk < -17 dB across the wavelength range of 1530-1570 nm. Compared with the conventional adjoint method, the average insertion losses of TE₁ and TE₂ modes are reduced by 3.2 dB and 2.6 dB, respectively. For the three-mode broadcasting circuit, the average insertion losses of TE₁ and TE₂ modes are reduced by 6.1 dB and 5.9 dB respectively under ±10 nm diameter variations, in comparison to the conventional adjoint method. The proposed robust inverse design method offers significant promise for ultradense photonic integrated circuits.