Deep-Learning-Designed AlGaAs Interface Linking Trapped Ions to Telecom Quantum Networks

Abstract

The realization of a scalable quantum internet requires efficient light-matter interfaces that map stationary qubits onto photonic carriers for long-distance transmission. A central challenge is the generation of entangled photons simultaneously compatible with single-emitter transitions and low-loss telecom fiber infrastructure. Spontaneous parametric down-conversion in integrated photonic platforms offers a promising route toward this goal. Among available material systems, AlGaAs is particularly attractive due to its large second-order nonlinearity and strong potential for monolithic integration. However, engineering the spectral and spatial properties of the generated quantum states requires the simultaneous optimization of numerous geometric and material parameters, a task remaining computationally demanding for conventional numerical approaches. To address this challenge and enable rapid and high-fidelity modeling of complex nonlinear photonic devices, we develop an inverse-design framework based on neural network surrogate models. Using this readily extendable method, we design a transversely pumped AlGaAs waveguide microcavity that produces polarization-entangled photon pairs in distinct spatial modes and frequency channels, one at 1092 nm, resonant with a $^{88}\text{Sr}^{+}$ transition, and the other at 1550 nm in the telecom C-band. This device establishes a direct photonic interface between trapped-ion qubits and long-haul fiber networks, providing a scalable pathway toward hybrid quantum network architectures.