Abstract
On-chip spectrometers have the potential to offer dramatic size, weight, and power advantages over conventional benchtop instruments for many applications such as spectroscopic sensing, optical network performance monitoring, hyperspectral imaging, and radio-frequency spectrum analysis. Existing on-chip spectrometer designs, however, are limited in spectral channel count and signal-to-noise ratio. Here we demonstrate a transformative on-chip digital Fourier transform spectrometer that acquires high-resolution spectra via time-domain modulation of a reconfigurable Mach-Zehnder interferometer. The device, fabricated and packaged using industry-standard silicon photonics technology, claims the multiplex advantage to dramatically boost the signal-to-noise ratio and unprecedented scalability capable of addressing exponentially increasing numbers of spectral channels. We further explore and implement machine learning regularization techniques to spectrum reconstruction. Using an 'elastic-D1' regularized regression method that we develop, we achieved significant noise suppression for both broad (>600 GHz) and narrow (<25 GHz) spectral features, as well as spectral resolution enhancement beyond the classical Rayleigh criterion.
Introduction
Spectrometers are essential tools for measuring light intensity across wavelengths, with applications spanning chemical sensing to telecommunications. However, researchers have long faced a "size vs. performance" deadlock. Conventional benchtop spectrometers are bulky, while existing on-chip versions often rely on dispersive elements that spread light thin, degrading the signal-to-noise ratio (SNR). Furthermore, traditional on-chip designs scale linearly, meaning higher resolution requires a proportional increase in chip area. This paper presents a digital Fourier transform (dFT) architecture that overcomes these limitations through exponential scaling and machine learning-enhanced reconstruction.
Digital Fourier Transform Architecture
The core innovation is a "digital" approach to path length modulation. Instead of relying on weak thermo-optic effects to slightly alter the refractive index, the dFT architecture uses optical switches to physically reroute light through discrete waveguide paths.
As shown in the schematic, the device consists of a Mach-Zehnder Interferometer (MZI) where each arm contains cascaded switches. These switches direct light into waveguides with lengths that follow a binary sequence ($\Delta L, 2\Delta L, 4\Delta L, \dots, 2^k \Delta L$). This allows the number of spectral channels ($N$) to scale as $2^j$, where $j$ is the number of switches. This exponential scaling allows for massive channel counts on a tiny footprint; for instance, adding just two switches quadruples the resolution while only increasing the chip area by approximately 40%.
Device Implementation
The researchers fabricated a 64-channel dFT spectrometer using industry-standard silicon-on-insulator (SOI) technology. The device integrates thermo-optic switches and a germanium photodetector on a single chip, achieving over two orders of magnitude larger optical path length modulation per unit length compared to traditional methods.
The micrograph illustrates the compact layout of the six switches required to address 64 channels. This high-density integration is key to the device's performance, allowing for high-resolution spectroscopy without the bulk of conventional instruments.
Machine Learning-Enhanced Reconstruction
Because the system is mathematically underconstrained, the team developed a specialized machine learning regularization technique called "elastic-D1." This algorithm acts as a computational lens by combining three distinct penalties:
- L1-norm: Promotes "sparsity," which is essential for resolving sharp laser lines.
- L2-norm: Bounds the overall magnitude of the signal.
- First-derivative penalty (D1): Ensures "smoothness," allowing the system to accurately reconstruct broad, continuous spectral features.
By leveraging prior knowledge of physical spectra, the software can resolve features that exceed the hardware's classical limits, effectively beating the Rayleigh criterion.
Key Results and Performance
The dFT spectrometer demonstrated significant performance leaps across several metrics:
- Super-Resolution: Achieved a spectral resolution of 0.2 nm, successfully surpassing the classical Rayleigh criterion of 0.4 nm for this device.
- Accuracy: Reproduced laser wavelengths with an accuracy of $\pm 0.025$ nm.
- Robustness: The elastic-D1 method significantly outperformed standard techniques like LASSO or Ridge regression, particularly when reconstructing complex spectra with both sharp and broad features.
- Scalability: The architecture is highly scalable; a device with just 10 switches could address 1,024 spectral channels, rivaling high-end benchtop instruments while remaining on a single chip.
Conclusion
The on-chip dFT spectrometer represents a significant advancement in miniaturized optical sensing. By combining exponential hardware scaling with machine learning-driven super-resolution, this technology provides a path toward high-performance spectroscopy in portable, low-power formats. Future applications include space-borne spectroscopy, optical network monitoring, and chemical "lab-on-a-chip" sensors.
Summary updated