Phase-limited quantization-aware training for diffractive deep neural networks

, effectively reducing the difficulty of grating etching while maintaining good network performance.