Mid-infrared photonics covers the sources, detectors and optics for roughly the 2-12 µm band where molecules absorb, and the current wave of work sidesteps poor infrared sensors by nonlinearly converting mid-IR light up into the near-infrared where silicon cameras and single-photon detectors already work.
Summary
The mid-infrared spans the wavelengths at which molecular bonds vibrate, so it is the natural band for chemical identification: gas sensing, combustion diagnostics, biomedical spectroscopy, materials inspection and thermal imaging all live there. The problem has never been the physics of absorption but the hardware. Mid-IR detectors are noisy, and array formats are limited by readout electronics and sensitivity, which caps frame rates for large pixel counts. Parallel spectrometers suffer the same penalty because every pixel in the array contributes noise.
The dominant technical answer in the recent literature is nonlinear upconversion. A mid-IR signal is mixed with a strong pump beam in a nonlinear crystal, producing a near-infrared replica that carries the same spectral or spatial information but can be read out with a silicon camera or a silicon single-photon detector. Because the conversion is low-noise and works at room temperature, the sensitivity penalty of the infrared sensor disappears. The same nonlinear step can be engineered to do optical work in passing: phase matching acts as a spatial-frequency filter, so edge enhancement and higher-order image differentiation can be imposed on the image during conversion rather than in software.
The parameters that decide the field are conversion efficiency, noise equivalent power, spectral bandwidth and resolution, and the pump power needed to get there. Recent work reports 22% peak conversion efficiency and a noise equivalent power of 0.3 fW/Hz^(1/2) using an external cavity to enhance a 1064 nm pump by a factor of 36 to 55 W intracavity average power. Spectrometers built on this route reach 0.5 cm^-1 resolution at illumination levels of 0.14 photons per nm per pulse across 2.4-4.2 µm, and hyperspectral imagers acquire 100 spectral bands over 2600-4085 cm^-1 in 10 ms.
Evidence base
Open questions
- Can the 0.3 fW/Hz^(1/2) noise equivalent power be retained without a 55 W intracavity pump, or does sensitivity collapse when the cavity enhancement of 36 is removed?
- Do any of the upconversion results reproduce outside the originating groups, and at what performance penalty?
- How far can the demonstrated bands, currently 2.4-4.2 µm and 2600-4085 cm^-1, be extended into the longer-wavelength fingerprint region where most chemical discrimination lives?
- Is there a credible waveguide-integrated path for the nonlinear conversion stage, given that the demonstrated foundry-compatible mode-hybridisation work was not done at mid-IR wavelengths?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.