Hands-on research in cryogenics, ultra-high vacuum, superconductivity, Josephson Field-Effect Transistors and VLSI device fabrication. At the NEST Lab of Scuola Normale Superiore I designed, fabricated and characterised hybrid superconducting devices together with my tutor Dr. Alessandro Paghi. Each project below opens an interactive bay with a 3D model of the setup and a live physics simulation.
Looking for the peer-reviewed papers behind this work? They live on the Publications page.
Each panel opens a full horizontal bay containing an interactive 3D rendering of the experimental setup and a live numerical demo of the physics. Drive the sliders — the sims run in real-time in your browser.
A real 1-D CNN, trained offline in PyTorch and run live in your browser through ONNX Runtime, racing the matched filter that every readout chain already computes. The honest headline: on an isolated qubit it ties — the matched filter is already optimal — and the neural gain only appears once multiplexed readout crosstalk enters.
Why silicon became the foundation of semiconductor electronics: a covalent diamond-cubic crystal with a perfectly engineered 1.12 eV indirect gap, compatible thermal oxide, and ultra-clean Czochralski wafers.
A hands-on gallery of the main Bravais lattices that build the materials I work with every day — from Al superconducting pads to InAs nanowires and Si substrates.
Top-gated diffusive Josephson Field-Effect Transistor on InAs/Al heterostructure. Real experimental parameters: ξ₀ = 651 nm, W_JJ = 6 µm, R_N ≈ 550 Ω. Dual gate-insulator comparison (HfO₂ vs Al₂O₃) with I_c(V_GS), R_N(V_GS), I_c(L_JJ), non-sinusoidal CPR and dI/dV subgap MAR.
A monolayer of graphene deposited on hexagonal boron nitride: a small lattice mismatch and an in-plane twist generate a long-period moiré superlattice. The carrier density is tuned by a back gate, and an out-of-plane magnetic field collapses the Dirac cone into a √B Landau fan.
An interactive study of Wigner-function tomography. The live demo races two reconstructors on an identical shot budget: the classical filtered back-projection of homodyne marginals, and a browser reimplementation of the sparse-regression method of Tang et al. (arXiv:2607.06232), which fits W from sparse displaced-parity measurements with provable O(log d) sample complexity. The note below derives both and credits the paper for the theory.
Numerical simulation of continuous-variable non-classical states emitted by Josephson Parametric Amplifiers: Fock, Schrödinger-cat, squeezed vacuum under parametric pump, and finite-energy GKP grid states — each with its live Wigner function and negativity witness.
A physics-informed neural network solves the Bogoliubov-de Gennes equations for hybrid InAs/Al nanowires. The live demo diagonalises a 100-site Kitaev chain and overlays the Δ_bulk(μ,Δ) phase diagram — the same map that the Topological Gap Protocol (Aasen et al., PRB 2023) uses to certify Majorana-carrying regions.
An interactive simulator of the full superconducting qubit dispersive readout chain — from the Jaynes-Cummings Hamiltonian to the IQ plane measurement result. Send a probe pulse and watch the resonance curve shift, the IQ point rotate, and T₁ decay in real time.
A physics-informed solver for the 1-D time-independent Schrödinger equation. The user paints V(x) on a canvas; a shifted inverse-iteration loop refines four eigenfunctions and their eigenvalues live, while the residual ‖Hψ − Eψ‖ is plotted per iteration — the convergence diagnostic every PINN paper reports.
Given a noisy synthetic JoFET dataset, an on-the-fly symbolic regression loop fits five candidate functional forms in parallel. The winner — A · exp(−L / ξ) — emerges as the χ² curves bifurcate, recovering the law I_c = I_c0 · exp(−L / ξ_N) from nothing but data.
A Gaussian-process surrogate models the single-shot readout fidelity F(g) as a function of qubit-resonator coupling g. Expected-Improvement acquisition chooses the next measurement; clicking a point or pressing 'auto' executes it and updates the posterior live.