The Cover Article of the Proceedings of the National Academy of Sciences of the United States! A Chinese Team Released an AI-adaptive Micro-spectrometer That Can Be Produced at the Wafer Level
The wavelength detection of light plays an important role in scientific research and industrial applications, and optical spectrometers are indispensable analytical tools. Nowadays, the bulky traditional spectrometers can no longer meet the growing demand for spectral detection technology.Miniaturization has become the only way for the development of spectrometers.It has great application potential in various fields such as machine vision, environmental monitoring, and medical diagnosis.
There are many different technical routes for miniaturization of spectrometers. In recent years, computational reconstruction spectrometers that rely on artificial intelligence algorithms have attracted much attention in the industry. This type of spectrometer uses high-speed computing to partially replace the workload of physical spectrometers, which can further reduce the size and weight of the instrument.
However, due to the diversity of spectral morphologies and the assumption of signal sparsity, previously reported reconstruction micro-spectrometers usually require manual calibration of algorithm parameters, otherwise the restoration results of the measured spectrum may be distorted. At the same time, the ability of such spectrometers to be directly mass-produced through integrated circuit technology has not yet been verified.
In this context, the research group of Professor Mei Yongfeng from the Department of Materials Science and the International Institute of Intelligent Nanorobots and Nanosystems at Fudan University published a research result entitled "CMOS-Compatible Reconstructive Spectrometers with Self-Referencing Integrated Fabry-Perot Resonators" in the Proceedings of the National Academy of Sciences of the United States of America.This achievement was also selected as the cover article of this issue.
The team proposed a new miniaturized reconstruction spectrometer design that combines the advantages of traditional spectrometers and computational reconstruction spectrometers through an integrated self-reference narrowband filter channel.This allows artificial intelligence algorithms to search spectral and algorithmic parameters simultaneously in a higher dimensional parameter space.Furthermore, the spectrometer can be manufactured at the wafer level through mature integrated circuit processes and has a millimeter-level size, which is sufficient to meet most miniaturized spectral testing needs.
Research highlights:
* This study proposes a new design of miniaturized reconstruction spectrometer, which exhibits accurate spectral reconstruction capability in the entire visible light band (400-800 nm), with a resolution of about 2.5 nm, an average wavelength deviation of about 0.27 nm, and a resolution of up to 5,806. * This study provides a new idea for realizing a universal and highly robust miniature reconstruction spectrometer, and is expected to promote the integration of miniature spectral detection systems into CIS image modules with the help of mature CMOS integrated circuit technology.
* The performance of this spectrometer is close to that of commercial fiber optic spectrometers, but the cost and size are greatly reduced
Paper address:
https://www.pnas.org/doi/10.1073/pnas.2403950121
The open source project "awesome-ai4s" brings together more than 100 AI4S paper interpretations and provides massive data sets and tools:
https://github.com/hyperai/awesome-ai4s
Datasets: Apply different spectral derivation methods to different datasets
The researchers divided the current data obtained from the micro-spectrometer into two data sets: narrowband channel currents and total channel currents, and then applied different spectral derivation methods to each data set.
* Narrowband dataset
The researchers used the data set to directly derive the point-by-point spectral curves of each channel by dividing the response current of each channel by its responsivity, which they call self-referenced spectroscopy.
* All channel data sets
The researchers compared the algorithm-reconstructed spectrum with the self-reference spectrum (derived from the narrowband channel) and iteratively optimized the algorithm parameters to achieve the optimal reconstruction and restoration of the spectral curve.
Working principle: Reconstruct accurate and stable spectrum by introducing self-reference spectrum
Figure A below shows the working principle of a conventional spectrometer.It uses narrow bandpass filters to distinguish different wavelengths, and the intensity of each wavelength is directly measured based on the amount of light passing through the corresponding filter. This process can be described as a "point-to-point" mapping. Although the resulting spectrum is rough, the position of each filter corresponding to the wavelength is relatively accurate.
Figure B above describes the working principle of a typical reconstructive spectrometer.The spectrometer encodes the unknown spectrum into collected data, and then reconstructs the data into a spectrum through a supervised algorithm with parameter Φ. The algorithm seeks the minimum cost function in the spectral parameter space S, usually implemented by regularization methods such as Tikhonov or total variation. Although this reconstruction can obtain high-resolution spectra, the results may be unstable because different choices of parameter Φ may lead to different minimum cost functions.
Figure C below shows the working principle of the self-adaptive spectrometer proposed in this study.In addition to encoding the spectrum as data for the algorithm, the spectrometer also provides a rough self-reference spectrum in the traditional way. This self-reference allows a two-level optimization in the spectral parameter space S and the algorithm parameter space Φ, so that the search for the minimum cost function covers higher dimensions. This enables the identification of the global minimum cost function by automatically selecting the optimal parameters, thereby reconstructing an accurate and stable spectrum.
The figure below further illustrates the reconstruction process of the adaptive spectrum, i.e. the adaptive algorithm.
Specifically, the miniaturized spectrometer has a set of narrow-band channels for conventional spectral measurements, so it provides two sets of current data for spectral sensing.The first group is the current of the narrow spectrum response channel.It can be viewed as the scalar product of the spectral intensity of a particular band and the response of the channel responsible for that band, from which an explicit but rough spectrum can be easily obtained.The second group is the current from all channels (including narrowband channels),It is the integrated result of multiplying the spectrum of each wavelength by the channel response (Scalar-product response).
The researchers introduced the spectral results obtained from the first set of data as a self-reference for the solution calculated from the second set of currents.The algorithm can adjust various parameters by itself and obtain stable results close to the real spectrum through inherent iteration.
Research results: Accurate spectral reconstruction capability across the entire visible light band
Wavelength resolution is an important parameter of a spectrometer, especially in applications such as wavelength meters or high-precision material identification. In tests to test performance, the spectrometer demonstrated accurate spectral reconstruction capabilities across the entire visible light band (400-800 nm). The following figure shows the comparison between the input peak wavelength and the output reconstructed peak wavelength.Shows good consistency.
The researchers further analyzed the deviations of the miniaturized spectrometer, as shown below, and calculated the resolution at a given input peak wavelength: Rλ = λ/Δλ,An average wavelength deviation of about 0.27 nm and a resolution of up to 5,806 were achieved.
The researchers also applied the resolution test of traditional spectrometers to their miniature spectrometers: they irradiated two monochromatic light peaks onto the spectrometer at the same time and gradually reduced the distance between them to study the minimum distance at which the miniaturized spectrometer can still distinguish the two spectral lines. As shown in the figure below,Two peaks separated by 2.5 nm and located around 518 nm can be resolved.
These results indicate thatThe performance of the miniaturized spectrometer designed in the study is comparable to that of commercial fiber-optic spectrometers and other small spectrometers, but at a significantly reduced cost and size.
On this basis, the research team further demonstrated the performance of the adaptive micro-spectrometer in common laboratory applications such as transmission, absorption and photoluminescence spectroscopy measurements after combining microfluidics and mechanical scanning systems. The results were basically consistent with those of commercial fiber optic spectrometers, as shown in Figures AF below.
(A) Schematic diagram of micro transmission-absorption spectroscopy test;
(BC) Reconstruction results of the transmission spectrum (B) and absorption spectrum (C) of vitamin B;
(D) Schematic diagram of micro-photoluminescence spectrum test;
(E) Reconstruction results of the photoluminescence spectrum of Rhodamine B;
(F) Reconstruction results of the photoluminescence spectrum of graphene quantum dots
In addition to excellent performance, more importantly,The spectrometer can be manufactured at the wafer level using mature integrated circuit processes and has a millimeter-scale size.It is sufficient to meet most miniaturized spectral testing needs.
In summary, this research provides a new idea for realizing a miniature reconstruction spectrometer with universality and high robustness. It is expected to promote the integration of miniature spectral detection systems into CIS image modules with the help of mature CMOS integrated circuit technology, so as to give priority to their application in mobile portable measurement, vehicle-mounted machine vision and distributed monitoring systems.
Continue to deepen the basic science fields such as materials
The research mentioned above was funded and supported by the National Key R&D Program, the National Natural Science Foundation, the Shanghai Science and Technology Commission and other projects. Some experiments were carried out in the Micro-Nano Processing and Devices Public Laboratory of Fudan University. Professor Mei Yongfeng is the corresponding author of the paper.
As a professor of materials science at Fudan University, Mei Yongfeng has always been a practitioner and advocate of basic research. He once said: "Basic research aims to understand phenomena, discover and open up new areas of knowledge. It seems to be very far from life and seems to have no practical use, but in fact,Basic research is the most fundamental driving force for social development.Just like the bricks needed to build a house, although you don’t know what a certain brick is used for, if you remove this brick, the house will collapse.”
With this concept, the research team led by Professor Mei Yongfeng has made many outstanding contributions in basic research and materials science, published more than 300 academic papers in Science Robotics, Science Advances, Nature Communications, Advanced Materials, etc., cited more than 10,000 times, and authorized more than 20 invention patents.
As one of the typical achievements, in January 2023, Mei Yongfeng's research group published an article titled "Self-rolling of vanadium dioxide nanomembranes for enhanced multi-level solar modulation" in "Nature Communications".
The research team was inspired by Venetian blinds.The strained vanadium dioxide film on the glass is desorbed and rolled into a "leaf" array smart window using self-rolling technology.The smart window can be modulated into fully curled (open), half curled (half open) and flat (closed) states by changes in ambient temperature, and self-responsive intelligent switching can be achieved, thereby greatly improving the transmittance in the fully open state and achieving multi-level light transmittance modulation with different opening degrees.
This work creatively combines the thermo-induced deformation ability and thermochromic ability of smart vanadium dioxide thin film materials.It breaks through the difficulty of traditional flat films in taking into account light transmittance, energy-saving efficiency and multi-environment adaptability.It provides a new feasible idea for the new generation of efficient smart windows.
"Two-dimensional nanofilms self-assemble into three-dimensional microstructures" is considered an important way to manufacture the next generation of microelectronic devices, which is crucial for the upcoming advanced electronic and optoelectronic applications. However, the formation of the final geometric shape of the two-dimensional nanofilm is affected by etching tracks, chemical reactions, aspect ratios, and other complex factors, making it difficult to improve the yield and finished product rate of self-assembled devices during the manufacturing process, which seriously hinders its real transition from laboratory to industrial application.
In response to this, in June this year, Professor Mei Yongfeng's research group published a research result titled "Multilevel design and construction in nanomembrane rolling for three-dimensional angle-sensitive photodetection" in "Nature Communications".
This study proposed a multi-level quasi-static finite element analysis method, and based on this design, constructed six types of silicon (Si) /chromium (Cr) nanofilm assembled three-dimensional microstructures and corresponding three-dimensional optical detectors, fully verifying the good versatility and industrial practicability of this technology.
In the future, Professor Mei Yongfeng's research group will also delve deeper into micro-nano mechanics, nano-optics, nanoelectronics, micro-nano robotics, micro-nano fluidics, micro-energy storage, surface plasmons and metamaterials, and continue to promote the progress of basic science.
References:
1.https://news.fudan.edu.cn/2024/0820/c5a141853/page.htm
2.https://www.memstraining.com/news-41.html
3.https://www.sohu.com/a/634625615_12