Miscellaneous

Reading out a superconducting qubit means listening to a faint microwave tone for a few hundred nanoseconds and deciding: ground or excited? The standard recipe integrates the noisy signal with fixed weights and compares the result to a threshold. That works — until the qubit decays in the middle of the measurement, which is exactly what happens to short-lived qubits. For these T1-limited devices, the integrated signal lands between the two clouds and the assignment goes wrong.

This project treats readout as a real-time inference problem instead. The full measurement chain — the resonator dynamics conditioned on the qubit state, the amplifier chain, and the noise — is modeled as a set of stochastic differential equations, and a Bayesian filter runs over the raw record, updating the probability of each qubit state nanosecond by nanosecond. A decay event mid-record is no longer a failure mode: the filter watches the evidence flip and tracks it.

On experimental data the filter reached single-shot fidelities of 0.992 and 0.983 for the two initial states, beating boxcar integration precisely in the regime where qubit decay dominates — and it keeps improving with longer records where integration saturates or gets worse. It requires no hardware changes: it runs as pure post-processing on the same IQ records the experiment already collects.

A superconducting-qubit lab is also a small industrial plant: dilution refrigerators report pressures, gas flows, and valve states; LakeShore controllers log millikelvin temperatures; Labrad servers drive the DC and microwave sources. Each instrument speaks its own protocol and writes to its own corner, so questions like "what was fridge 2's still pressure last Tuesday night?" used to mean digging through log files machine by machine.

I built a lab-wide monitoring platform on the Elastic stack to bring all of it into one searchable place. Custom Python collectors scrape the fridge ValveControl telnet servers, Filebeat and Logstash ingest the LakeShore temperature logs, and a Labrad client reads the source outputs — everything lands in an Elasticsearch database under a metadata schema designed for the lab's instruments, with index lifecycle management keeping storage bounded as data accumulates. Kibana dashboards on top give every fridge and instrument a live, browsable view.

The system ran across multiple dilution refrigerators and measurement hosts, giving the lab real-time monitoring and a searchable history of its operating conditions — the kind of infrastructure that turns "the fridge did something odd overnight" from a mystery into a query.

Long before the dilution refrigerators: watch a kite on a steady breeze and it often refuses to hover — instead it traces a big, repeating figure-8 across the sky. Why that shape? This high-school project (with teammates at Beijing No. 2 Middle School) set out to answer the question with real mechanics and a home-built experiment.

On the theory side we built two models of a kite on a string — first a wind-force model in spherical coordinates, then a refined deflection-force model — and derived the condition under which the center of the figure-8 is an equilibrium point, along with analytic trajectory families that reproduce both circular and figure-8 flight. On the experimental side we built our own measurement rig: the joystick potentiometer from a modified game controller tracked the string angles, a serial force gauge logged line tension, and a Bluetooth attitude sensor rode on the kite itself. Outdoor flights with several kite types produced data that matched the model's predictions.

The project won a Bronze Medal at the St. Yau High School Science Award in physics (2017). It was my first taste of the full research loop — model, build, measure, revise — and in hindsight, the same loop I now run on quantum hardware, just with better thermometers.

More entries are on their way: background-component analysis for the MAJORANA and LEGEND neutrinoless double-beta-decay experiments (UNC / Triangle Universities Nuclear Laboratory), and further projects from Tencent Quantum Lab — waveguide QED, benchmarking protocols, and two-qubit gates.