Fibre-optic spiking sensors detect remote temperature, audio and turbulence

What the work claims

The authors claim a working neuromorphic photonic remote-sensing system. A fibre Bragg grating embedded in a standard telecom fibre reflects infrared light whenever temperature or strain shifts the grating's resonance. A photo-detecting resonant tunnelling diode, biased near its negative-differential-resistance region, converts the reflected light into electrical spikes. The spike rate encodes the stimulus amplitude, so the same physical channel performs sensing, analogue-to-spike conversion, and event-driven pre-processing in one step.¹

How it works

A fibre Bragg grating is a periodic refractive-index modulation written into the core of an optical fibre. It reflects a narrow wavelength band and transmits the rest. Temperature and mechanical strain shift that band, so a fixed laser wavelength couples more or less light into the reflected mode. The authors use a tuneable Santec laser near 1542 nm, an erbium-doped fibre amplifier, and a circulator to route reflected light to the detector.

The detector is a photo-detecting resonant tunnelling diode. Its active layer is a double-barrier quantum well: 1.7 nm AlAs barriers surrounding a 5.7 nm InGaAs well, with a 250 nm InGaAs photoabsorbing layer on top. Under reverse bias the device has a negative-differential-resistance region in its current-voltage curve. A small photocurrent shifts the operating point into or out of that region, switching the device between a quiet resting state and a continuous electrical spiking regime. Spike rates from a few megahertz to above 10 MHz are reported, with individual spikes at nanosecond width.

The authors test four regimes. For slow temperature changes they heat the grating from 18.22 C to 26.16 C and observe a transition from rest to bursting at 5.0 MHz and then continuous spiking at 11.0 MHz. For strain they tap the fibre manually and record spike bursts that rise from 3.2 MHz to 8.5 MHz as strain increases. For audio they attach the fibre to a loudspeaker subwoofer, play a 40 Hz tone and two clips from the ESC-50 sound dataset, and show that the spike pattern tracks the sound envelope. For fast turbulence they blow compressed air across the fibre and resolve frequency content up to about 2 kHz.

Where a skeptic should push

The most load-bearing assumption is that spike rate is a useful encoding for downstream processing. The paper demonstrates event detection and amplitude-to-rate mapping, but there is no classifier, no decoder, and no closed-loop actuation. A spike train that carries 11 MHz of raw temporal information still has to be read, buffered, and interpreted by something else.

Second, the experiments are hand-calibrated demonstrations rather than calibrated sensor characterisations. The temperature sweep spans only 8 C, the strain stimulus is a finger tap, and the turbulence source is a compressed-air duster. Sensitivity, linearity, and cross-sensitivity are not quantified. The useful dynamic range is therefore unclear.

Third, the system consumes continuous-wave laser power, erbium-doped fibre amplification, and a precision bias supply. Wall-plug energy per detected event is not reported. The event-driven advantage applies only after the optical front end, not necessarily to the whole platform.

Fourth, the device is a single spiking node, not a network. The authors cite future work on coupled RTD neurons and bistable spiking flip-flop memory, but those elements are not present here.

OI implication: event-driven sensing without biology

For organoid intelligence and biological computing, this paper is a warning about the narrowing event-driven niche. Biological neural tissue is often justified by its sparse, event-driven coding and its ability to combine sensing and computation. Here, a non-living device performs a comparable trick at megahertz rates on an installed telecom fibre, with no incubator, no perfusion, and no variability between batches.

The opportunity for the OI field is to study what the biological substrate can do that this device cannot. The RTD-FBG system detects amplitude changes and encodes them as spike rate. It does not learn, adapt, or rewire. It does not discriminate complex temporal patterns without a separate classifier. It does not self-repair after radiation or thermal stress. Those are precisely the properties that organoid intelligence advocates need to make concrete and measurable.

The threat is that many near-term applications of biological computing do not require learning or plasticity. If the goal is simply to detect motion, sound, or temperature at the edge with low data volume, a photonic spiking sensor is already closer to deployment than a living culture. The energy and maintenance overhead of keeping neural tissue alive becomes harder to justify when a diode on a fibre can do the sensory front end.

The honest conclusion is that event-driven spiking is no longer a unique selling point of living tissue. The burden of proof shifts. Future OI work must show that biological substrates win on tasks that need continuous adaptation, context-dependent coding, or closed-loop learning, not merely on tasks that can be solved by a biased quantum-well detector.

The bottom line

The paper experimentally demonstrates a photonic event-based sensor that converts infrared fibre-Bragg-grating reflections into electrical spikes. It covers temperature, strain, audio, and turbulence with nanosecond temporal resolution. That is a real advance in integrated photonic sensing. What remains unproven is whether the spike encoding can feed a useful decision layer, how efficient the full system is, and how it performs outside a laboratory bench. For organoid intelligence, the result narrows the defensible territory of biological computing to tasks where learning and plasticity matter, because event-driven sensing itself is now demonstrably achievable in silicon photonics.