Ultra-low-loss microfiber knot resonator photonic processor for cost-effective feedback-free reservoir computing

High-performance photonic computing fundamentally demands physical memory units that are both low-loss and cost-effective. In this work, four sets of microfiber knot resonators (MKRs) were fabricated using a three-electrode tapering technique, featuring insertion loss below 0.2 dB and quality factors ranging from 10⁴ to 10⁵. The devices are subsequently packaged with polydimethylsiloxane (PDMS), maintaining a total loss less than 1.0 dB. Utilizing these packaged MKRs, a feedback-free photonic reservoir computing (FF-PRC) system is constructed. Experimental results demonstrate that as the Q-factor increases, the normalized mean square errors (NMSEs) for the Mackey-Glass and Santa Fe time-series prediction tasks dropped from 0.390 and 0.420 to 0.022 and 0.034, respectively, while the system achieves a classification accuracy of 97.5% on the MNIST dataset. The proposed FF-PRC system, enabled by ultra-low-loss MKRs, exhibits low-error and high-accuracy performance in processing complex temporal tasks. This eliminates the need for external feedback loops specifically for computing tasks whose memory requirements match the intrinsic photon lifetime of the MKRs, offering a standalone and cost-effective memory unit solution for neuromorphic photonic edge nodes.