Deiva Kumar K, Mathivanan Ponnambalam
Digital deformities and diabetic-foot risk in DPN were associated with combined metabolic, bioimpedance-derived, and systemic neuromuscular alterations. Handgrip strength and bioimpedance-derived parameters showed exploratory associations with risk and deformities; however, their role as complementary assessment measures requires confirmation in larger longitudinal studies with adjusted models.
Halide perovskite memdiodes have coupled ionic-electronic dynamics and are promising candidates for artificial synapses in neuromorphic computing. We provide an in silico neuromorphic circuit that includes a comprehensive perovskite memdiode model and confirm its synaptic plasticity repertoire through simulation-based validation. The proposed model recapitulates analog long-term potentiation/depression (LTP/LTD), spike-timing-dependent plasticity (STDP), spike-rate-dependent plasticity (SRDP), and paired-pulse facilitation/depression (PPF/PPD). We also describe pulse-amplitude- and pulse-width-dependent conductance modulation, pinched hysteresis I-V curves, and statistical robustness to device-to-device and cycle-to-cycle fluctuations through Monte Carlo simulation. This memdiode is implemented into a continuous-time neuromorphic predictive coding framework to minimize Variational Free Energy (VFE). In this architecture, we directly map four clinical hallmarks of diabetic peripheral neuropathy (DPN)-plasticity impairment, homeostatic failure, afferent attenuation, and conduction delay-to localized circuit parameters ([Formula: see text]). One severity parameter α is used to continuously tune the network between healthy predictive coding and computational collapse. Healthy baseline parameters minimize VFE quickly. Moderate DPN (α = 0.5) results in permanently elevated oscillatory VFE, a hypothetical counterpart of allodynia. Severe DPN (α ≥ 0.8) paralyzes the homeostatic plasticity, and the system collapses to a frozen maladaptive state that is similar to sensory ataxia. The noise resilience analysis and hyperparameter sensitivity analysis indicate that the architecture is highly resilient to stochasticity within the simulated parameter space and structurally robust to hyperparameter perturbations in all dynamical regimes. The primary contribution of this work is a computational framework demonstrating how second-order, BCM-capable memristive device dynamics can be used to model systems-level predictive-coding failure; the behaviorally validated perovskite memdiode model and the DPN severity mapping serve as a concrete, illustrative instantiation of this framework.