Integrated NIR spectral sensor based on an array of resonant-cavity enhanced detectors

Abstract

Spectral sensing in the near-infrared range is of increasing interest for application areas ranging from industrial processes to the agri-food sector, as it allows measuring the chemical composition of organic materials in a fast and non-destructive manner. On-field applications demand spectral sensors with a large spectral range, low angular dependence, robust geometry and small footprint, while being manufacturable on large scale at low cost. To meet these requirements, we developed a multi-pixel array of resonant-cavity enhanced detectors using an InP-membrane-on-silicon platform, where each of the 16 pixels provides an individual photoresponse with a number of resonances, which together cover the wavelength range of 800-1700 nm. The pixels consist of two metal mirrors, which enclose an InP p-i-n photodiode with an InGaAs absorber layer and a tuning layer, which varies in thickness to create the individual photoresponses of different pixels. The multi-pixel arrays are fabricated by adhesive wafer bonding followed by a series of lithography steps to define the tuned pixels and metal contacts for the photocurrent read-out. Despite the limited number of measurement channels with broad spectral response, information on the chemical composition of the sample can be directly retrieved using chemometrics. The sensing capabilities for our spectral sensors are demonstrated with two practical application cases: determination of the nutritional information in raw milk and the classification of different plastic types. For both experiments, the photocurrent measurements from the sensor were used directly in regression or classification algorithms based on partial least squares analysis, leading to convincing prediction performance. This shows that the fabricated spectral sensor can retrieve chemical information for a broad set of sensing problems. Simulations have shown that for a specific sensing problem the prediction performance of the spectral sensor can be further improved by an optimized selection of the available spectral responses.