A Bright Idea in Imaging

Both our eyes and image sensors are capable of distinguishing between red, green, and blue wavelengths of light, but that is where the similarities end. Human eyes have specialized types of cells that are able to recognize each of these colors, but image sensors have only silicon. And silicon is not picky — it absorbs all wavelengths of visible light. So, in order to zero in on each color individually, a separate pixel is needed for each. These pixels have filters in front of them that are designed to block all wavelengths of light aside from the range that they are meant to detect.

This arrangement works, however, the filters limit the amount of light (including the target color) that reaches the sensors. As such, image sensors are not nearly as sensitive to light as they could be. Due to the filters, image sensors have problems under low-light conditions, and are also susceptible to certain artifacts, like demosaicing and the moiré effect. In most situations, this will not be noticeable in photos we snap with our phones, but for more sensitive applications, like machine vision and medical imaging, it is a big problem.

A more human-like sensing method developed by researchers at ETH Zurich and Empa may soon overcome these issues. Rather than silicon, the team has used a material called perovskite to construct a new type of image sensor. Like silicon, perovskite is also a semiconductor, but it is much easier to process. That made it possible for additional chemicals to be added to the material, altering the wavelengths of light it can absorb without a filter.

By carefully adjusting the chemical makeup of the perovskite — such as adding iodine for red sensitivity, bromine for green, and chlorine for blue — the researchers created pixel layers that naturally absorb specific colors. Furthermore, these layers remain transparent to other wavelengths, allowing them to be stacked vertically. That stacking ability means that a single pixel location can detect red, green, and blue without filters and without wasting light.

In addition to consumer electronics, perovskite sensors also hold promise for hyperspectral imaging. Unlike RGB-based sensors, hyperspectral systems can detect many distinct wavelength bands. By fine-tuning each perovskite layer to absorb specific ranges, custom sensors could be built for everything from analyzing crop health to detecting disease.

Though the technology is still in the early stages, the researchers have already built two working prototypes using industrial thin-film processing techniques. These early models have pixel sizes between 0.5 and 1 millimeter. This is large by commercial standards, but a meaningful step toward miniaturization. The next challenge is to scale down the pixels to the micrometer level used in current image sensors.