On-chip spectrometers with photonic integrated circuits

Abstract

The miniaturization of optical spectrometers, one of the most essential characterization tools for analyzing light and extracting material information, has long been a primary research focus in both academia and industry. In recent years, this trend is further accelerated by the rapidly growing market demand for in situ, in vitro, and in vivo spectroscopic sensing devices, which require compact, high-performance, and low-cost spectrometers. Leveraging the compatibility with well-developed CMOS technologies, lithography-based photonic integration platforms offer a cost-effective solution to develop chip-scale spectrometers with mass producibility. To date, numerous on-chip spectrometers have been reported, based on various working principles that fall into four main categories: dispersion, narrowband filtering, Fourier transform, and computational reconstruction. The on-chip implementation of the first three categories typically involves replicating traditional free-space optical components—such as dispersive prisms or Michelson interferometers—on photonic integrated circuits. However, the physical constraints of on-chip space limit the achievable optical path differences in waveguide-based dispersive elements and interferometers, resulting in inherent performance ceilings. Reconstructive spectrometers (RSs), on the other hand, employ computational algorithms based on compressive sensing to achieve superior performance while maintaining a compact footprint. Yet, the existing RS designs often suffer from a limited number of sampling channels and poor channel decorrelation. As a result, the bandwidth and resolution of current integrated spectrometers are generally confined to below 100 nm and on the nanometer scale, respectively. Furthermore, many reported RSs must also contend with issues such as optical losses, limited accuracy, and temperature tolerance. This leaves a clear performance gap in addressing practical spectroscopy needs for biomedical sensing or industrial chemical detection. To tackle these challenges, this thesis proposes a range of novel on-chip spectrometer designs based on photonic integrated circuits. The author first introduces an active RS, employing a reconfigurable photonic network with distributed broadband filters to generate a large number of highly-decorrelated sampling channels. Experimental results demonstrate a bandwidth-to-resolution ratio of approximately 4000, which surpasses the record at the time by an order of magnitude. To further enhance the device simplicity, another powerful reconstructive design is also developed, featuring a programmable photonic sampling circuit composed of multiple stages of tunable interferometers. This design enables an exponential increase in the number of high-performance channels and experimentally achieves an ultra-high resolution of less than 10 pm across a broad bandwidth exceeding 200 nm. This translates to a bandwidth-to-resolution ratio of over 20,000, once again significantly advancing the performance record for miniaturized spectrometers. Furthermore, to overcome the bandwidth limitations imposed by dispersion effects in integrated platforms, an ultra-broadband RS is proposed. This design employs a cascade of dispersion-engineered micro-ring resonators, achieving an experimental bandwidth exceeding 325 nm along with an exceptional resolution below 20 pm, setting a new benchmark for bandwidth performance in the field. Besides these active RS designs, a universal design scheme for passive RSs is also proposed to meet the application scenarios with stringent requirements for high sampling speed and low power consumption. This scheme is based on multi-resonant cavities with partially reflective mirrors. As a proof-of-concept, a single-shot, dual-band RS is demonstrated, achieving a total bandwidth of 270 nm and a resolution of 0.5 nm with only 15 sampling channels per band. Lastly, this thesis introduces a brand-new spectrometer category, namely, convolutional spectrometer. This is grounded on one of the most foundational mathematical principles—the convolution theorem—thus fundamentally distinguishing it from all existing spectrometer types. Specifically, a convolutional spectrometer necessitates only a few basic optical components to generate a periodic overlaid system response, which can then be linearly shifted across the spectral domain by applying temporal phase modulations. This waveform shifting enables the circular convolution operation on an arbitrary input spectrum, thereby achieving its flawless recovery. The recovery process involves only (inverse) Fourier transforms, thus resulting in minimal computational load. Extensive theoretical derivation and simulations illustrate that the convolutional spectrometer offers significant design flexibility in bandwidth and resolution, along with excellent noise tolerance and robust temperature stability. As a proof-of-concept, an integrated convolutional spectrometer was implemented, showcasing an ultra-broad operational wavelength range of around 500 nm with fine resolution. In addition, the high-precision recovery of a variety of incident spectra, as well as exceptional temperature tolerance exceeding 100°C, are experimentally validated. In summary, this thesis leverages photonic integrated circuits to develop a series of powerful on-chip spectrometer designs, overcoming numerous technical bottlenecks and setting multiple records across key performance metrics. Furthermore, these designs offer innovative insights and promising directions for future researches, featuring the potential to revolutionize miniaturized spectroscopic sensing devices and may find wide applications across diverse fields, such as chemical engineering, medical diagnostics, biological research, and astronomy.