High-spectral-uniformity micro-rings with artificial gauge fields for scalable photonic tensor computing

Wavelength division multiplexing is a crucial technique for scalable photonic tensor computation, imposing extremely stringent requirements on spectral uniformity. Based on the dispersionless coupling characteristics of artificial gauge fields, we theoretically derive the high spectral uniformity and weighted robustness of micro-rings based on artificial gauge fields. We present a micro-ring resonator (MRR) design utilizing the artificial-gauge-field method to achieve near-zero dispersion across the C-band and L-band, enabling full optical linewidth utilization and multi-band weighted sharing. Introducing periodic curved waveguides into the coupling architecture, coupling coefficient fluctuation is within 1 dB for the gap from 290 to 330 nm. Experimental characterization across 20 frequency bands yielded a linewidth of 36.70±2.38 GHz and an extinction ratio of 22.40±0.54 dB , significantly mitigating the frequency-dependent limitations inherent in conventional MRRs. The designed micro-ring facilitates multi-band parallel processing without incurring additional area or power overhead. Validation via simulation confirms that a photonic convolutional neural network implementing physical noise awareness and 4-bit nonlinear weighting achieves 96.8% accuracy on the MNIST recognition task.