Purdue Tackles Three Big Roadblocks in Light-Powered Quantum Chips

How do you make one photon at a time?

In 2021, the team published a study in Science Advances showing how to emit single photons from silicon nitride, now the industry-standard material for quantum photonic platforms. The industry's usual method, called spontaneous parametric down-conversion, produces single photons inconsistently and inefficiently.

Purdue's approach works differently. The researchers engineered tiny flaws in the material's atomic structure — "defects" — that limit emission to exactly one photon at a time. Every time a laser beam passes through the platform, dozens of these locations each fire off a single photon at room temperature.

At the 2022 Conference on Lasers and Electro-Optics, the team went further, showing they could control the placement and number of defects by embedding them in nanosized pillars. That uniformity matters for manufacturing chips at scale.

What about switching photons on and off?

There was a second problem: single photons don't readily interact with each other, which makes them hard to steer with switches. Conventional computers use transistors to toggle bits between zero and one; a photonic quantum chip needs an equivalent.

In a Nature Nanotechnology paper last year, the team reported a proof-of-concept switch that operates at the single-photon level. It adapts a mechanism called optical nonlinearity, which the industry already uses to modulate light in fiber-optic communications — but normally at power levels far too high for single photons.

Here's the trick: one photon from a control beam strikes silicon and frees an electron, triggering an avalanche of electrons. That avalanche changes silicon's refractive index — the property governing how light propagates through it — so the probe beam passes through only when a single photon starts the process. One photon effectively flips the switch.

Can errors be predicted before they happen?

Even photon qubits can decohere — fall out of superposition — when disturbed, wrecking calculations. Purdue professor Alexander Kildishev and graduate student Pranshu Maan led a team, including Shalaev and Boltasseva, that published a way to anticipate decoherence in Nature Communications in May.

"Decoherence is seemingly random, but in fact there is a certain pattern that can be recognized," Boltasseva said. "This study takes an important step in identifying such a pattern using machine learning."

The team first found, through statistical-physics analysis, that decoherence over time follows a pattern. Machine learning then confirmed those patterns carry predictable information. They focused on spectral diffusion — when an emitted photon drifts across the spectrum because of disorder in its environment, changing its frequency and energy. Since superposition requires all photons to be indistinguishable, one shifted photon can break the whole system.

The researchers discovered these spectral shifts retain a memory of the photon's past frequency wandering. They built an algorithm with a built-in memory mechanism, trained it on data from individual quantum emitters, and showed that anticipatory control could reduce spectral shifts by as much as 20-fold in the systems studied.

The payoff could be dramatic. "Let's say you need a hundred qubits to run a quantum algorithm, but typically far more than that, say 10,000 qubits, to deal with errors caused by decoherence," Shalaev said. "If I minimize those errors somehow, then the number of qubits needed to run my algorithm would also be minimized."

Next, the team plans to use statistical physics to extract hidden properties of a quantum system's fluctuating environment and feed those insights into the machine-learning models. They also want to test whether the models can pinpoint the microscopic processes behind spectral shifts, with the long-term goal of suppressing decoherence before it disrupts the quantum state. These are early-stage results, and the 20-fold reduction was demonstrated only in the studied systems — but the direction is clear.