Knowledge Distillation-Assisted Lightweight Neural Networks for Superconducting Multi-Qubit Readout

Superconducting qubits are among the most promising candidates for building quantum information processors. Yet, they are often limited by slow and error-prone qubit readout—a critical factor in achieving high-fidelity operations. Current methods, including deep neural networks, enhance readout accuracy but usually require large networks and lack support for mid-circuit measurements essential for quantum error correction. We introduce an independent qubit readout architecture based on lightweight neural networks optimized through knowledge distillation, achieving a 95% reduction in model size with comparable qubit-state-discrimination accuracy. By assigning a dedicated, compact neural network for each qubit, our approach enables rapid, independent qubit-state readouts that support mid-circuit measurements. This work demonstrates that compressed neural networks can maintain high-fidelity readout while addressing scalability challenges, advancing practical quantum computing.