Chongfan Technology
News
02
2024
-
12
Computationally Reconstructed On-Chip Spectrometer Based on Reconfigurable Silicon Photonic Filters
Author:
Spectral analysis technology is widely used in fields such as biomedicine, materials science, and remote sensing. However, traditional benchtop spectrometers based on spectroscopic devices have complex diffraction or interference setups, are bulky, and expensive, making them unsuitable for emerging demands such as wearable health monitoring and Lab-on-Chip applications. Compared to traditional benchtop spectrometers, integrated spectrometers, which lack mechanical moving parts, offer higher feasibility, compactness, and field application capabilities, thus broadening the application areas of spectral measurements.
According to a report by MEMS Consulting, a research team composed of Chang'an University, the Institute of Semiconductors of the Chinese Academy of Sciences, and the University of the Chinese Academy of Sciences has designed an on-chip spectrometer based on a silicon photonic platform. This device consists of a silicon optical filter with a reconfigurable transmission spectrum. By changing the transmission spectrum of the filter, multiple and different samplings of the input spectrum can be achieved. Combined with artificial neural network algorithms, the incident spectrum can be reconstructed from the sampled signals. The reconfigurable silicon optical filter is composed of mutually coupled Mach-Zehnder interferometers and micro-ring resonators. An integrated thermo-optic phase shifter introduces phase changes, allowing for the reconstruction of the filter's transmission spectrum. This device can achieve precise spectral computation reconstruction in the wavelength range of 1500-1600 nm, with a resolution of up to 0.2 nm. The reconfigurable characteristics of the silicon optical filter provide additional degrees of freedom for the spectral detection capabilities of this device, enabling users to change the number of sampling channels based on application scenarios and flexibly configure characteristics such as resolution, relative error, and computational complexity. The related research was published in the journal "Acta Physica Sinica" under the title "On-chip Spectrometer Based on Reconfigurable Silicon Optical Filter for Computational Reconstruction."
The on-chip spectrometer designed in this study mainly consists of a waveguide-type silicon optical filter with a reconfigurable transmission spectrum. A schematic diagram of the device structure is shown in Figure 1. By applying control signals to the phase shifters, the phase changes introduced by the phase shifters can be altered to reconstruct the filter's transmission spectrum. Each time the transmission spectrum of the filter is changed (the working state of the filter), a sampling of the input optical signal can be completed. By reasonably setting the control signals, the transmission spectra of the reconfigurable filters in different states can have unique spectral characteristics, resulting in a set of optimized low-correlation filter transmission spectra. Based on this set of transmission spectra, diversified sampling of the incident spectrum can be achieved. The sampled spectral signals are converted into response vectors at the output end of the filter by a photoelectric detector. The spectral reconstruction process is illustrated in Figure 2.
The designed reconfigurable silicon optical filter consists of a Mach-Zehnder interferometer (MZI) and two sets of second-order cascaded micro-ring resonators (MRR). The input and output ends of the second-order MRR are coupled to the upper and lower phase shifters of the MZI, respectively. The introduction of MRR can provide wavelength-dependent phase shifts near the resonant wavelength and enhance the diversity of the optical signal's propagation path within the device. The radii of the MR can be designed with different parameters to introduce a cursor effect, further enhancing the diversity of the filter's transmission spectrum. Additionally, both sets of second-order MRR are designed with over-coupling, resulting in a wide spectral characteristic of the obtained MRR transmission spectrum, which can also improve the overall sampling efficiency of the filter for the incident spectrum.
The filter contains five thermo-optic phase shifters, with PS1 used to adjust the phase difference between the two arms of the MZI, and PS2, PS3, PS4, and PS5 used to adjust the phases of the four micro-rings (MR). By controlling the voltages of the five electrodes separately, the phase change amounts introduced by the phase shifters (Δφ1, Δφ2, Δφ3, Δφ4, Δφ5) can be altered through the thermo-optic effect, allowing for flexible adjustments to the filter's transmission spectrum.
The ANN model used in this spectrometer includes one convolutional layer, three fully connected layers, and one residual connection from input to output. The convolutional layer is used to extract nonlinear features from the mapping, while the fully connected layers are used for spectral reconstruction. The convolutional layer contains a set of one-dimensional learnable convolutional kernels with a specific window size. The number of kernels and the window size are shown in Figure 3. After each convolutional layer, a rectified linear unit (ReLU) is used as the activation function, followed by non-overlapping max pooling to downsample the output of the activation function, with both the pooling window and stride set to 2. The downsampled output is flattened and then input into the subsequent three fully connected layers. After the first two layers, ReLU and dropout are applied. Dropout is introduced to reduce overfitting in the network. The output of the last fully connected layer is added to the input of the network and fed into the regression output layer. The number of units in each fully connected layer is shown in Figure 3. The output layer has 1001 nodes, representing the reconstructed spectral intensity, with wavelengths ranging from 1500 to 1600 nm, and a spacing of 0.1 nm between each wavelength point. The mean squared error (MSE) is used as the loss function during network training.
Subsequently, the researchers conducted simulation and modeling based on the design of the spectrometer. The spectrometer is intended to use a silicon-on-insulator (SOI) process platform, with a top silicon thickness of 220 nm and a buried oxide layer thickness of 2 μm. The waveguide is a single-mode ridge waveguide with a height of 220 nm, a width of 450 nm, and a thickness of 70 nm for the side plate regions. This study utilized Ansys Lumerical Interconnect to perform circuit-level simulations of the device and employed an optical network analyzer module to obtain the spectral response of the device. The wavelength range of the light source is 1500-1600 nm, with a resolution of 0.1 nm.
The reconfigurable filter is the core unit of the on-chip spectrometer designed in this study. The spectral response of the reconfigurable filter is altered by the thermo-optic phase shifters to obtain multiple sampling results. Completing a full sampling requires traversing 32 sampling states of the filter. This sampling strategy requires a certain amount of sampling time, mainly including the response time of the thermo-optic phase shifters and control circuits, with the response time of the thermo-optic phase shifters in the microsecond range. For the control circuits, the delay for each sampling channel is less than 1 ms, so it is estimated that this on-chip spectrometer can complete a scan of the incident spectrum within 32 ms. It is important to note that in some applications where real-time performance is critical, such as the detection of non-repetitive pulse signals and real-time optical coherence tomography, spectrometers based on multi-channel filters (such as waveguide array gratings) remain the preferred solution. Although the resolution of such solutions may be relatively lower, they can capture the optical intensity information of all wavelength channels simultaneously without the need for scanning or computational reconstruction. The performance of this on-chip spectrometer is compared with other filter-based computational reconstruction spectrometers, as shown in Figure 4.
In summary, this study presents an on-chip spectrometer based on a reconfigurable silicon optical filter. By controlling the signal to change the transmission spectrum of the filter, multiple and different samplings of the input spectrum can be achieved. Combined with the ANN algorithm, the incident spectrum can be reconstructed from the sampled signals. This device achieves diversified sampling of the incident spectrum using a single reconfigurable filter, significantly reducing the area of the spectral detection chip, and offering more obvious advantages in integration and hardware costs. Simulation results show that this device can accurately compute and reconstruct different types of spectra in the wavelength range of 1500-1600 nm. On a test set composed of synthetic spectra, the average reconstruction error RMSE of the computed spectra is 0.0075, and the average relative error ε is 0.0174. The resolution for double narrow peak spectra reaches 0.2 nm. Thanks to the reconfigurable characteristics of the silicon photonic filter, this chip can flexibly change the number of sampling channels, allowing users to configure the chip according to application scenarios. Limited by the microsecond-level response time of thermal optical modulation, the sampling speed of this device is difficult to further enhance. In the future, the sampling speed of the on-chip spectrometer can be improved by introducing thermal isolation structures or using electro-optic phase shifters based on active processes. This computationally reconstructed spectral chip based on a reconfigurable filter provides a new solution for chip-level spectrometers and has broad application prospects in fields such as miniaturized spectrometers and Lab-on-Chip.
LATEST NEWS
Thank you for visiting the official website of Chongfan Technology. If you have cooperation intentions or suggestions, please contact us through the following methods, and we will reply as soon as possible, thank you!
Address: Room 403, Building 6, Phase III of R&D, No. 36 Xiyong Avenue, High tech Zone, Chongqing, China.
Telephone: +86-13658337211
E-mail: Sales@cfkeji.net
Website: www.cfkeji.net
Mobile Version
