Application-Specific Optimization of Integrated Spectral Sensors.

Near-infrared spectral sensing serves as a powerful technique for nondestructive analysis of material composition in a wide field of applications. A typical spectral sensor comprises an array of detectors, each with a response in a certain spectral band. We demonstrate an algorithm capable of tailoring these sensors for a specific near-infrared spectroscopy application by optimizing for all possible combinations of spectral bands. This approach outperforms manually selected designs, achieving high sensing performance even with just a few pixels. The results are confirmed by experiments on fabricated four-pixel devices, which feature a sensing accuracy exceeding the one of general-purpose sensors for a problem of practical relevance. This approach may enable cost-effective spectral sensors with simple read-out mechanisms for industrial and consumer applications.