Optical spectroscopy is a tool that has already been used for decades to determine materials’ properties by analyzing the interaction between light and the material. When shining light on an object, the light spectrum that is reflected back or transmitted by the object represents the spectral “fingerprint” containing useful information on the object’s composition. The spectrum is usually measured with spectrometers, which can provide a high resolution, but are bulky and contain moving parts, making them unsuitable for a handheld solution. Recently there has been a trend in miniaturizing the spectral sensors where the small size and portability together with the low costs are more important. This trend can potentially extend their application to the consumer level and ultimately be embedded into smartphones. To achieve this goal chip-level integration of the sensor is required, which is an exceedingly difficult task, especially in the near-infrared range.
To circumvent this issue, we noted that the goal is not the measurement of the spectrum, but characterization of the material under investigation. Human eyes are excellent spectral sensors, providing essential information on food, health condition and material properties, based on the signals from only three types of cone cells, each measuring a part of the spectrum. These three signals define the “color”, which our brain uses to make predictions based on experience (for example a red strawberry is sweet, a green one not). More evolved color vision systems exist too, for example the mantis shrimp which has up to 16 filters. The design of our sensor is inspired by this. We have created an array of detectors which all have a distinct response and of which the outputs can directly be used to create a prediction model without the need for reconstructing the reflected/transmitted spectrum (see Fig. 1). An important difference with respect to human eyes is that the detectors are sensitive in the near-infrared, beyond the visible range, a part of the spectrum where many materials have their fingerprint.
Each pixel in the array is based on a resonant-cavity-enhanced photodetector operating in the near-infrared range, where both the absorbing layer and the tuning element are integrated inside a cavity system (see Fig. 1a). Using this approach provides a robust design, without the need for any moving elements. This creates a filter response with multiple peaks in the near-infrared range (see Fig. 1b), also known as generalized filters, which strongly simplifies the on-chip integration. As a proof of principle, the sensor is used in several sensing experiments. We have for example used it to determine the nutritional properties of milk, which is a sensing problem of practical relevance as it impacts the economic value of milk and it helps monitoring the cow’s health. We also used it to discriminate between different plastic types, which is beneficial for the environment since the improvement of the recycling process is key for the optimization of the waste sorting process. We expect that similar sensing performance can be obtained for a wide range of organic and inorganic materials. Most components that can be detected by commercially available spectrometers such as sugar, starch, fat, protein etc., present relatively broad spectral signatures and can also be detected using an array-based spectral sensor optimized and tailored to the application. The shown results enable a major step in the miniaturization of spectral sensors. The sensing platform provides a solution for a broad range of quantification and classification problems when minimizing the complexity, size and cost of the sensors is critical.
If you would like to know more about this work, please read the complete study in Nature Communications, “Integrated near-infrared spectral sensing”.