We report the redox switching of ferrocene moieties embedded in a doubletunnel barrier plasmonic cavity fabricated from a click-chemistryself-assembled monolayers of ferrocenyl-alkylthiol on ultra flat gold surfaces,connected to a single poly(vinylpyrrolidone) capped silver nanocube, AgNC,which is contacted by the tip of a conductive-AFM to study the electrontransport properties in the dark and under light irradiation at the plasmonicresonance wavelengths. We observe a dual behavior in the current-voltage (I-V)characteristics in the dark: a large hysteresis loop at positive voltages andan hysteretic negative differential conductance (NDC) at negative voltages, dueto the redox switching of ferrocene between its oxidized (Fc+) and neutral(Fc0) states. The I-V curves are analyzed by a generalized combinedMarcus-Landauer model. We determine the highest occupied molecular orbital ofthe Fc+ and Fc0 states at 0.54 and 0.42 eV below the Fermi energy,respectively, with a weak reorganization energy < 0.1 eV upon switching. Underplasmonic excitation, the hysteresis and NDC behaviors are no longer observedand the I-V characteristics of the Au-ferrocenyl-alkylthiol/AgNC junctionsbecome similar to Au-ferrocenyl-alkylthiol SAMs. A virtual molecular orbitaldue to the plasmon-induced coupling (fast electron transfer) between the tworedox states of the Fc is determined at 0.46 eV. This dynamic behavior opensperspectives in artificial synaptic devices for neuromorphic computing with theadditional function to turn on/off this synaptic behavior on-demand by light.
Spin-based semiconductor qubits hold promise for scalable quantum computing,yet they require reliable autonomous calibration procedures. This studypresents an experimental demonstration of online single-dot charge autotuningusing a convolutional neural network integrated into a closed-loop calibrationsystem. The autotuning algorithm explores the gates' voltage space to localizecharge transition lines, thereby isolating the one-electron regime withouthuman intervention. This exploration leverages the model's uncertaintyestimation to find the appropriate gate configuration with minimal measurementswhile reducing the risk of failures. In 20 experimental runs, our methodachieved a success rate of 95% in locating the target electron regime,highlighting the robustness of this approach against noise and distributionshifts from the offline training set. Each tuning run lasted an average of 2hours and 9 minutes, primarily due to the limited speed of the currentmeasurement. This work validates the feasibility of machine learning-drivenreal-time charge autotuning for quantum dot devices, advancing the developmenttoward the control of large qubit arrays.