Direct-detected spectroscopy based on a plasmonic Schottky photodetector and a deep neural network

Abstract

Computational algorithms have facilitated the miniaturization of spectrometers, which is essential for on-chip and portable applications. A plasmonic Schottky photodetector provides a filter-free and CMOS-compatible scheme for spectral measurement. In this study, we report on a direct-detected spectral analysis based on an integrated vertically coupled plasmonic nanostructure Schottky photodetector. We demonstrate that the plasmonic Schottky photodetector has a fast response with a –3 dB bandwidth of 600 kHz and a high peak detectivity of 8.65 × 1010 Jones. By designing a deep neural network (DNN), we demonstrate the reconstruction of the unknown spectrum with a mean square error (MSE) of 1.57 × 10−4 at a broad operating wave band of 450–950 nm, using only 20 distinct devices. Moreover, the spectral resolution of the 20 devices can reach to 7 nm. These findings provide a promising route for the development of chip-integrated spectrometers with high spectral accuracy and optical performance.