Abstract
Optical spectroscopy—the measurement of electromagnetic spectra—is fundamental to various scientific domains and serves as the building block of numerous technologies. Computational spectrometry is an emerging field that employs an array of photodetectors with different spectral responses or a single photodetector device with tunable spectral response, in conjunction with numerical algorithms, for spectroscopic measurements. Compact single photodetectors made from layered materials are particularly attractive, since they eliminate the need for bulky mechanical and optical components used in traditional spectrometers and can easily be engineered as heterostructures to optimize device performance. However, compact tunable photodetectors are typically nonlinear devices and this adds complexity to extracting optical spectra from the device response. Here, we report on the training of an artificial neural network (ANN) to recover the full nonlinear spectral photoresponse of a nonlinear problem of high dimensionality of a single GeSe-InSe p-n heterojunction device. We demonstrate the functionality of a calibrated spectrometer in the spectral range of 400-1100 nm, with a small device footprint of ~25X25 micrometers, and we achieve a mean reconstruction error of 0.0002 for the power-spectrum at a spectral resolution of 0.35 nm. Using our device, we demonstrate a solution to metamerism, an apparent matching of colors with different power spectral distributions, which is a fundamental problem in optical imaging.
Introduction
Standard optical spectrometers are typically bulky and expensive, relying on moving parts like gratings or prisms to spatially separate light into its constituent wavelengths. While "computational spectrometry" offers a path toward miniaturization by using algorithms to reconstruct spectra from tunable detectors, it faces a significant hurdle: nonlinearity. Most semiconductor photodetectors exhibit nonlinear behavior at the nanoscale, where the output current is not strictly proportional to input light power. Traditionally, researchers have avoided these nonlinearities because the mathematical inversion required was considered intractable. This work demonstrates that nonlinearity can be embraced as an information-rich feature when paired with machine-learning tools.
Device Architecture and Physics
The researchers fabricated a vertical van der Waals heterostructure consisting of p-type Germanium Selenide (GeSe) and n-type Indium Selenide (InSe), encapsulated in hexagonal Boron Nitride (hBN). The active area of the device is incredibly compact, measuring approximately $25 \times 25\ \mu m$.
As shown in the characterization data, the device operates as a p-n heterojunction with distinct rectification behavior. The "secret sauce" of this spectrometer is its voltage-tunable responsivity. By changing the bias voltage, the researchers can modify the band alignment and interfacial charge transfer at the junction. This allows the device to produce unique photocurrent responses for different wavelengths across the visible and near-infrared spectrum (400–1100 nm).
Computational Reconstruction via Neural Networks
To decode the complex, nonlinear photoresponse, the team moved beyond standard linear equations to a general nonlinear model. They trained a Multi-Layer Perceptron (MLP) neural network with four hidden layers (2048, 1024, 512, and 128 neurons) using ReLU activation. This ANN learns the complex inverse mapping required to transform a vector of photocurrents measured at 101 different voltages back into a high-resolution power spectrum.
The training dataset included seven different LED sources at various intensities and a Laser-Driven Light Source (LDLS) used with 50 different filters. The authors argue that the high-dimensional parameter space created by nonlinear coefficients actually provides more information for the ANN to work with, allowing for higher reconstruction accuracy than linear models might achieve in such a small footprint.
Key Results and Applications
- Ultra-Compact Footprint: The active area is only $\sim 25 \times 25\ \mu m$, ideal for on-chip integration.
- High Precision: The device achieved a spectral resolution of 0.35 nm with a mean reconstruction error of only $2 \times 10^{-4}$.
- Solving Metamerism: A major highlight is the device's ability to solve the problem of metamerism—where two different spectral distributions appear as the same color to the human eye or standard cameras. Because this device captures the full underlying spectrum, it can distinguish between these "metameric" pairs easily.
Conclusion and Future Outlook
While the nonlinear reconstruction involves roughly $4 \times 10^6$ coefficients and is computationally more intensive than linear methods, the performance gains are substantial. Future iterations of this technology could lead to single-element, on-chip spectrometers for smartphones or wearable tech, enabling rapid and inexpensive optical sensing for applications ranging from food safety to medical diagnostics.
Summary updated