Quantum Sensing Like a Bird

A few nights ago I was on a panel on Quantum Sensing for Defense. Preparing for this panel crystallized some ideas for me, which I think I should write down somewhere. So here it is!

When we talk about "quantum sensing," we often imagine a single, exquisite, hyper-sensitive system, device, or platform. But I’m starting to believe the most transformational advance won't come from a single "hero" sensor or approach. I think we should instead look to networks of tiny, low-power, and hyper-aware sensors, working in concert.

The Thesis

Here's the core argument:

• Quantum sensing is about how much decision-useful information can we extract per joule, per kilogram, per second.
• The future of sensing will involve dense networks of tiny, intelligent sensors — some quantum, some classical — that are precise, ultra-low power, cheap to deploy.
• The key is achieving Scaling, Intelligence, and Integration of Quantum Sensors: The real power comes from fusing data from diverse sensors into a single, precise, resilient understanding of the world. The relatively modest gains quantum sensing strategies provide at single sensor level will lead to an exponential quantum advantage in the performance of the network.

Learning from Migratory Birds

Apparently I answered almost every question on panel by asking the audience to “think about the birds”. I don’t regret this — birds are truly astonishing. Please read Ed Yong’s An Immense World.

Birds cross oceans and return to the same tree without GPS. How? The best evidence suggests use a collection of sophisticated low-power sensors fused together via intelligence. They combine multiple inputs:

• Magnetoreception: Specialized proteins allow them to "see" the Earth's magnetic field lines. There's even research suggesting that they operate at quantum limits on this front.
• Celestial Navigation: They use the sun's position and star patterns.
• Biochemical Sensing: An "odor landscape" provides crucial olfactory cues — even over areas like oceans that lack clear visual markers.
• Visual & Auditory Cues: They recognize landmarks and even detect the infrasound of ocean waves and mountain ranges.

Fusing these inputs improves sensitivity and resilience. When clouds obscure the stars, they lean on magnetic sensing. When magnetic sensing is disrupted, they shift to visual and olfactory cues. Most important, they don’t care if a sensor is "quantum" or "classical"; they just use whatever combination of signals gives them the most reliable information.

This is precisely the architecture we should be building for critical applications (like defense): not one perfect sensor that becomes a single point of failure, but dense networks that are robust, adaptive, and greater than the sum of their parts.

What does "Quantum" add?

So, where does the "quantum" advantage come in? Quantum mechanics defines the absolute physical limits of measurement. Employing quantum strategies allows us to approach those limits, in principle providing significant gains in information efficiency given SWaP (size, weight and power) and damage threshold constraints.

• Compounding gains: In practice the quantum enhancement on single sensors has been small. But even modest sensor gains can compound. If I have five sensors, and these five sensors are all now 3x better, the advantage in aggregate may be much greater than 3x — it may be closer to 3^5 > 200x. With larger number of sensors, exponential gains are possible.
• Integrated Squeezed Light Sensors: Any measurement with light is limited by "shot noise" due to the random arrivals of individual photons. Squeezed light is a quantum state of light where correlations between the photons allows us to suppress the noise without affecting the signal. This lets us make measurements that are quieter than the standard quantum limit would allow. While this has been used in LIGO (to obtain ~2-3x increase in sensing volume), a breakthrough for a deployable, scalable device would lead to appreciable wins in networks of low-power quantum sensors. In principle, squeezed light can improve chip-scale gyroscopes, magnetometers, biochemical sensors, and allow them to achieve sensitivities that are impossible under the same constraints using classical approaches.
• Processing the Field Before Detection: Today's cameras just count photons. Tomorrow's sensors may manipulate the optical field before it hits a detector. These cameras would then select the specific spatial, spectral, or temporal modes of light that carry the most information, and maybe even access information that’s just not possible with a standard intensity based imaging.
• Compute-as-a-Sensor: Our attempts to build quantum computers have inadvertently created some of the most sensitive detectors on the planet. As the superconducting quantum community and specifically Google has found, a quantum processor can act as detectors of high-energy particles. We may be able to use the stream of syndromes from error-correction data of a quantum computer not just to stabilize a computation, but as a direct, high-fidelity sensor for radiation, high energy phonons, neutrinos, or other hard to detect particles.
• Quantum computational sensing (Sensors as sub-routine): This is a new idea that I have been intrigued by since it started to emerge over the last couple of years. Instead of the classical "prepare → measure → post-process" approach, we can treat the sensor as a callable subroutine inside a quantum algorithm. This means we can use phase estimation to lock onto weak tones buried in noise. Or do a quantum search to winnow huge parameter spaces. Many applications involve detection and then classification under uncertainty. Quantum algorithmic approaches promise immense gains on this front — if we can make them practical, low-power, and deployable.

How do we get there?

Quantum sensors can be propelled forward by a convergence of other powerful technologies. The physical concepts on which these ideas are based are ahead of the demonstrations. Even after initial demonstrations, it may take time for real impact, as their true power may only come at larger scale, deployment, and integration with intelligence. Luckily industrial developments are aligning to help move this vision forward:

Scaling: The ability to build these tiny, powerful sensors relies on breakthroughs in advanced Photonics & heterogeneous integration. We're now creating light sources, cavities, and detectors on a chip using materials like thin-film lithium niobate (TFLN) and silicon nitride (SiN). This lets us leverage the massive new investments in data-center/telecom photonics to build sensors.

Intelligence: A swarm of sensors is useless without the AI and on-board inference to fuse the data. On-board, low-power AI and maybe even neuromorphic processing would be the brains of the operation, and lots of investment is going into this.

Packaging and I/O: The practical execution is key in quantum sensing. I believe that our limiting resource isn’t entanglement — it’s I/O bandwidth. For example, SPADs are too slow right now to benefit from the rate at which we can generate entangled photons. This aspect suffers from a bit of chick-or-egg problem in terms of investment. But I remain hopeful!

Related reading

Nice papers from Soonwon Choi and Peter McMahon's group on Quantum Computational Sensing: https://arxiv.org/abs/2501.07625, https://arxiv.org/abs/2507.16918

Mankei Tsang's paper on processing optical fields before detection to realize much better sensors (in the context of Rayleigh's curse):
https://journals.aps.org/prx/abstract/10.1103/PhysRevX.6.031033

Papers from our group at Stanford on integrated squeezed light sensors (particularly from Hubert Stokowski, Taewon Park) and improving power efficiency in such device (Devin Dean and Taewon Park).
https://www.nature.com/articles/s41467-023-38246-6 https://www.science.org/doi/full/10.1126/sciadv.adl1814
https://www.researchsquare.com/article/rs-7596899/v1