What Does a Quantum Hardware Engineer Do in the Lab?

By Julia Berens · · 4 min read

Close-up of tangled blue network cables in an equipment rack, with red and yellow wires crossing gray metal hardware.

What does a quantum hardware engineer do when a device gives a different answer every time they test it? First, they work out whether the variation comes from the device, the equipment used to measure it, or the way the test is run. That distinction matters. Replacing a chip will not fix a loose connection in a room-temperature measurement rack. In the illustrative lab session below, an engineer tests a superconducting qubit, investigates an unstable signal, and leaves enough evidence for a teammate to check the fix.

What does a quantum hardware engineer do before a test?

The engineer starts with a specific goal: check whether the readout still reliably distinguishes two prepared qubit states. A superconducting qubit is a circuit designed to behave as a quantum system. It operates at very low temperatures inside a refrigerator, while much of the equipment that generates and records signals sits outside at room temperature. IBM Quantum is one example of a program that builds computers based on superconducting circuits.

Before interpreting anything from the chip, the engineer checks the settings against a previous working run. Are they using the same pulse timings, readout frequency, and amplifier settings? Is the script saving the individual measurements, or only their average? An average may look steady even when individual measurements are spreading out. Keeping the individual readings makes it possible to see that difference later.

For this test, a program repeatedly tries to prepare the qubit in one state or the other, then sends a microwave pulse to read it out. The qubit state affects the microwave response of a nearby circuit, and the instruments convert that response into numbers. A plot can show two clusters of points, one for each preparation. The points need not be identical within a cluster. The practical question is whether the clusters are distinct enough to classify individual readings reliably.

Track down the unstable signal

Suppose the clusters look less distinct than they did in the previous run. The engineer plots the measurements in time order instead of looking only at the two clusters. The new plot shows abrupt shifts that affect readings from both preparations. That pattern gives the engineer a clue: something shared by the two measurements may be changing. It is not proof that the qubit is fine, but it gives the engineer a better place to start than changing the qubit controls at random.

One way to narrow the search is to send a stable reference signal through a room-temperature section of the measurement chain, without using the chip. If the reference also jumps, the engineer has reproduced the symptom without the qubit. The fault could still be in the reference source, a cable, a connector, or the receiving instrument. The test has shortened the list, not identified the culprit.

In this example, the engineer swaps a suspect cable assembly for a known-good one while keeping the instrument settings fixed. The reference trace steadies. Then they repeat the original qubit test to see whether the two clusters separate again. Those checks support keeping the replacement in place. They do not show whether the old cable had an internal fault or whether reseating a connector during the swap restored a good contact. A useful diagnosis says exactly how far the evidence goes.

Make the change easy for a teammate to check

A lab note that says “fixed noisy readout” leaves the next person guessing. A useful entry identifies the cable by its label and names the ports it connects. It links to the reference and qubit data from before and after the swap, records the script version and relevant settings, and says what was left unchanged. If the engineer updated a wiring diagram, the note points to that revision too.

The comparison should be easy to inspect. If two plots show the spread of the readings, the axes should use the same scales and units. Otherwise, a change in the display can look like a change in the device. The engineer can add a number, such as the fraction of readings that the classifier assigned to the expected state, provided the comparison uses the same classification rule. That number describes the test as run. On its own, it does not separate mistakes in preparing the state from mistakes in reading it.

The note can also recommend a first check if the problem returns: run the reference test before adjusting the qubit pulses. This gives a teammate a fast way to see whether the same part of the setup is acting up again.

Practice with a sensor and a short program

You do not need a qubit or a refrigerator to practice this kind of work. Try a light sensor module and a low-voltage microcontroller that can send readings to a computer. Keep the sensor and a lamp in fixed positions. Log the readings with timestamps, then plot them in Python or a spreadsheet. Label the time axis and the measurement axis, using the module's output units rather than calling the values lux unless you have calibrated them.

Next, pick one possible source of variation. For example, take a second run with a cardboard shield positioned to block room light without blocking the lamp. Keep the sampling rate and sensor position the same. Compare the traces on matching axes. If the shield changes the mean reading but the fluctuations look similar, report both observations. An effect on the average is not necessarily an effect on the noise.

Save the raw data separately from the plotting code. Put the wiring, settings, and one sentence about the change in a short readme file. Ask a classmate to recreate the plot without your help. If they have to ask which file came first or how often the sensor was sampled, add that information. For the physics behind the devices, see our quantum mechanics learning resources. For more on related roles, browse Studies and Careers.

Frequently asked questions

Do quantum hardware engineers write code?

Yes, many write scripts to control instruments, record measurements, or analyze results. The mix of coding and hands-on work depends on the role.

Does a noisy measurement mean the qubit is broken?

No. The variation could come from the qubit, the measurement equipment, or the test procedure. A reference test can help narrow down the source.