William Tang

An optical tweezer traps a single neutral atom in a tightly focused laser beam. To move the atom, you move the beam — so a galvanometer mirror does the steering, and the mirror's stability becomes the atom's stability. Datasheets do not answer that question, so I built a bench to measure it: a quadrant photodiode at the end of a 300 mm propagation path, turning where the spot lands into a voltage I could log.

I calibrated voltage-to-angle for three candidates, then characterized each one with the mirror off, holding at DC, and under square-wave drive at 100 Hz, 1 kHz and 10 kHz. A single-element photodiode ran in parallel to confirm the differences were real and not artifacts of the quadrant-photodiode processing.

Galvo	DC jitter	Overshoot	Settling
Scanlab dynAXIS 421	< 15 µrad	~400 µrad	~5 ms
Thorlabs QS15Y	~50 µrad	~600 µrad	~30 ms
Cambridge 6215H	~150 µrad	~4000 µrad	~0.3 ms

Overshoot and settling measured on a 10 Hz, 20 mrad peak-to-peak square step.

The Cambridge unit settles more than an order of magnitude faster than anything else in the table, which makes it the obvious pick until you look at the other two columns — 150 µrad of jitter and 4 mrad of overshoot would throw an atom out of the trap long before the speed helped. I selected the Scanlab, then built and aligned the optical rig around it (fiber collimation, cabling, heat sinking) and integrated it for transport of 500 µm or more in the science plane, using the objective's 41.3 mrad ↔ 400 µm mapping to convert commanded angle into real displacement.

Two things I did not expect. A signal generator was injecting a 24 kHz spur into the drive, which vanished once I moved to a DAC. And ramping the steps rather than stepping hard cut jitter from roughly 900 to 500 µrad at 100 Hz — but each added step lengthens the ramp, and the firmware froze past about 660 of them, so the operating envelope is a genuine trade rather than a setting to max out.

I led the galvo characterization subproject, defined the test protocols, and drove the selection. Supported by an HCRP award, summer 2025.

The Harvard Undergraduate Aerospace Collective and its UAV team were founded at the same time; I co-founded the UAV team, which is where I have spent my time since. The goal was to build an aircraft from the ground up rather than assemble a kit. The program is sponsored by Zipline, whose markings the airframe carries.

I lead power and controls across a team of more than ten engineers: flight controller and ESC integration, a dual-battery power architecture, telemetry, and the safety-critical side of the electrical design. Then the testing, which is where you find out what you actually built.

An early flight test. It ends in a crash — which is what early flight tests are for.

HUCSat is a 2U CubeSat — about the size of two stacked Rubik's cubes — selected by NASA's CubeSat Launch Initiative. Its payload asks a narrow question: can you point a solar panel in space using an alloy instead of a mechanism?

Nitinol is a nickel-titanium alloy that can be trained to a shape and returns to it when heated. The payload puts four nitinol coils around a solar platform, one per edge, and heats them selectively. Heating them unevenly makes the contraction asymmetric, which rotates the platform toward the sun. There is no motor and no gearbox in the loop, and power is drawn only in the moments when it is actually moving.

I designed the nitinol spring actuator system in SolidWorks and supported prototype testing — furnace heat-cycling to see whether the trained shape held up under repeated actuation — then documented the results. We characterized the payload in-house before delivering it to NASA.

The satellite deployed from the International Space Station in April 2026.

Harvard SEAS on the launch · Chief engineer's project write-up

Satellite photograph by Christopher Prainito.