Computational Spectrometers Based on Multilayer Thin Films

Abstract

A spectrometer is a device that measures the intensity of light emitted or absorbed by an object as a function of wavelength. It is used in various research fields and industries, such as chemical analysis, remote sensing, and color inspection. Recently, spectral technology has attracted researchers’ attention because of its potential to improve the quality of life of ordinary citizens. However, spectrometers face several limitations that they are expensive, heavy, large in size, and require sophisticated manipulation. One approach to overcome these limitations is computational spectroscopy using multilayer thin films (MTF). The MTF can be realized to have unique transmission characteristics by controlling the thickness and number of layers and can be made small in size. In this dissertation, we present a computational spectrometer using multilayer thin films In the first half of this dissertation, we present the design, fabrication, and implementation of an MTF filter array for computational spectroscopy. In general, an MTF filter used in a filter-based spectrometer is designed in a bandpass type that only transmits a specific wavelength. To use a bandpass type filter as a spectrometer, a set of filters that continuously transmits signals of different wavelengths is required, and plenty of filters are needed to cover a wide wavelength range. In addition, to achieve high resolution, it is necessary to fabricate a sophisticated filter with a narrow transmission width. Unlike band-pass filters, the proposed MTF filters were designed to transmit a signal in unique transmission patterns over a wide wavelength range. The MTF filters were fabricated all at once in the form of an array using a combinatorial deposition technique. By attaching the fabricated filter array to a complementary metal-oxide-semiconductor (CMOS) image sensor, we built the device for computational spectroscopy. In the second half of this dissertation, we present the reconstruction of unknown spectra using computational approaches. An unknown incident spectrum is modulated by the fabricated MTF filter array and measured by a CMOS image sensor. The measured intensities are signals spectrally modulated by the unique transmissions of the MTF filters. The relation among the unknown incident spectrum, transmissions of MTF filters, and the measured intensities can be expressed in linear equations. Using a small number of linear equations, it is possible to reconstruct the unknown incident spectrum in high resolution from the measured intensities by applying computational approaches. We implement two computational approaches: numerical optimization based on L1 norm and deep learning based on convolutional neural network. Optical experiments were conducted to demonstrate the reconstruction performances of these two approaches using various kinds of light sources. Finally, the reconstruction performances of these two approaches in the noisy environments were comparatively analyzed through simulations.