New Perspectives on Quantum Technologies: Progress on Quantum Sensing and Quantum Computation

This dissertation reports three studies in quantum technologies, addressing one practical obstacle on each of three fronts through theoretical modelling and numerical simulation: hardware design for elevated-temperature operation of superconducting processors, a hybrid quantum-classical machine-learning architecture characterized against matched classical controls and validated on physical hardware, and a model of a single-photon detector intended for high-magnetic-field environments. The first contribution presents a theoretical analysis of an asymmetric Superconducting Quantum Interference Device (SQUID) transmon qubit based on NbN/AlN/NbN Josephson junctions, designed for operation at 4 K rather than at conventional millikelvin temperatures. Through circuit-level modelling and decoherence analysis, the study shows that the asymmetric SQUID supports two flux-insensitive sweet spots, and projects a two-qubit gate fidelity of approximately 98% under an echo cross-resonance protocol at the elevated-temperature operating point. Operation at 4 K would substantially reduce cryogenic complexity, although whether the projections survive a fabrication and measurement campaign remains an open experimental question. The second contribution develops, trains and physically validates a hybrid quantum-classical convolutional neural network (Q-CNN) for binary satellite-image classification on a subset of the EuroSAT dataset. A multi-seed campaign at R = 10 seeds compares the hybrid against two purpose-built classical controls — a structurally isomorphic matched-capacity baseline (within 0.07% of the hybrid's parameter count) and a high-capacity reference — and quantifies, through paired Wilcoxon exact tests, a ~2.6 pp asymptotic-accuracy deficit of the hybrid against the matched control alongside a ~29 pp training-efficiency advantage at the first training epoch. The contribution also derives a runtime-aware credit-cost model calibrated against the IQM Emerald production campaign, and closes the loop with a credit-bounded fine-tuning of the simulator-trained checkpoint on the IQM Emerald 54-qubit superconducting processor, where parameter-shift gradients restore the simulator's full validation accuracy in two on-device epochs. A complementary paired multi-seed study on a reduced 4-qubit version of the architecture under a calibrated IBM Heron r2 noise model further indicates that the hybrid pipeline is noise-resilient in the studied regime, with final classification accuracy statistically indistinguishable from the noiseless reference. The third contribution models a Rydberg-atom avalanche detector for single-photon sensitivity, motivated by the readout requirements of axion-haloscope searches. Numerical simulations across 0–5 T external magnetic fields predict amplification factors exceeding 3, dark-count rates below 0.01 Hz at 4 K, and a stable avalanche timescale of approximately 9–13 µs across the field range. The model provides quantitative performance targets for an experimental implementation that has not yet been attempted.