Evaluating Photonic Quantum Memristors in Noisy Environments.

While photonic quantum memristors (PQMs) offer promising avenues for neuromorphic computing, their performance is inherently affected by hardware noise, particularly photon loss and phase fluctuations. This study systematically investigates the impact of photon loss and phase fluctuations on PQM dynamics by employing the noisy gates approach, which integrates dissipative effects directly into the device evolution. At the device level, we demonstrate that photon loss alters the dynamic trajectory of individual PQMs. It induces evident deformations in the characteristic pinched hysteresis loops, with the degradation of non-Markovian memory effects being particularly pronounced at shorter integration times. To further evaluate system-level implications, we construct a two-PQM network to execute the NARMA2 time-series prediction task. Under noiseless conditions, the network exhibits strong representation capabilities with a normalized mean square error (NMSE) of 0.0448. However, performance degrades markedly under incoherent evolution; the NMSE increases to 0.1552, 0.2567, and 0.3056 for photon loss probabilities of 0.2, 0.4, and 0.5, respectively. Furthermore, at a high photon loss probability of 0.5, extending the integration time fails to compensate for the degradation and instead exacerbates the prediction error. These findings indicate that photon loss impairs both individual device dynamics and network-level processing, emphasizing the critical need for loss-tolerant architectures in deploying PQM networks.