At the forum, 孫銘賢 showed X-ray images of a circuit board, an ear thermometer and a mobile phone. Images produced with the perovskite conversion layer developed by the 睿生光電 team looked sharper than those from a commonly used GOS scintillator, but they were less sensitive. The result did not resemble a conventional product launch—and became the most important part of the talk.

Flat-panel X-ray detectors can follow two routes. A direct-conversion material absorbs X-rays and creates electrical charge, which is collected by the pixel circuitry. An indirect detector first uses a scintillator to convert X-rays into visible light and then sends that light to the photodiode below. Common CsI(Tl) uses a columnar structure to restrict light spreading sideways, while GOS offers mature manufacturing and cost advantages. Comparing these materials means trading among absorption, brightness, resolution and manufacturability.

Perovskites have been explored along both routes. A 2017 Nature paper printed a polycrystalline perovskite directly onto a thin-film-transistor backplane. A separate 2018 Nature paper used all-inorganic perovskite nanocrystals as scintillators. The first converts X-rays directly into charge; the second converts them into visible light first. The material family is the same, but the signal follows a different path.

睿生光電 presented the indirect route. The team fabricated a roughly 400 μm CsPbBr3 conversion layer on glass or a flexible substrate and combined it with an existing flat-panel detector for readout. The layer is far thicker than a film in a typical light-emitting device because it must first stop enough X-rays. Once thickness increases, how far visible light travels inside the material—and in which direction—also appears in the image.

In the comparison shown by 孫銘賢, the perovskite layer allowed less lateral light spreading, so image detail was preserved more readily than with GOS. Yet its emission did not align with the wavelength range where the photodiode below was most sensitive. The circuit-board traces became clearer, while conversion into an electrical signal remained inefficient. Dark current and long-term reliability also still require device testing; one sharp image cannot answer every question.

The team then shifted the materials toward better spectral matching. Cs3CuI5 first generated blue light, and CdSe converted it into green light that the photodiode could receive more effectively. The next version rearranged how the scintillator, color-conversion layer and readout device handed off the signal, bringing brightness back into the design of the full stack.

A useful comparison comes from another team. A 2026 Nature Communications paper also treated nanocrystal self-absorption, light yield, thick-film fabrication and fast imaging as one problem. The study demonstrated a resolution of 27.6 line pairs per millimeter (27.6 lp/mm) and imaging at 7,680 frames per second. It used a different material and process, but confronted the same question: how light is generated, transported and read out inside a thick film.

By acknowledging the loss in sensitivity on stage, 孫銘賢 made the spectral mismatch the starting point for the team's next material design. In an indirect flat-panel detector, what light the conversion layer emits and how much of it the photodiode below can receive are both part of the final image.

Sources and further reading

Talk and local industry context

Representative primary papers