Silicon waveguide detection method

Abstract

The invention discloses a silicon waveguide detection method, which comprises the following steps of: focusing measurement laser by using a focusing lens to obtain focused laser; the focused laser is irradiated to the to-be-measured silicon waveguide, the focused laser is refracted from the surface of the to-be-measured silicon waveguide, then enters the to-be-measured silicon waveguide, is reflected for multiple times on the inner wall of the to-be-measured silicon waveguide and then is emitted from the interior of the to-be-measured silicon waveguide, emergent laser is obtained, and the emergent laser carries information of the to-be-measured silicon waveguide; moving the to-be-measured silicon waveguide in a preset direction along the optical axis of the focusing lens, and acquiring images formed by emergent laser when the to-be-measured silicon waveguide is located at different positions, the different positions including the focus position of the focusing lens and the positions of the two ends of the focus; and obtaining the contrast ratio, the self-correlation function and the light intensity probability density of each image, and obtaining the roughness information of the silicon waveguide to be detected according to the contrast ratio, the self-correlation function and the light intensity probability density of each image.

Description

Technical Field

[0001] At least one embodiment of the present invention relates to the detection of waveguides, and in particular to a detection method suitable for silicon waveguides. Background Art

[0002] Photon integration technology is an integrated circuit formed by connecting integrated optical devices with optical waveguides, and is considered to be one of the most promising methods to break through the limit of Moore's Law. Different from integrated circuits that use electrons to transmit information, photon integration uses photons as the carrier of information transmission, with larger bandwidth, higher communication capacity, lower power consumption and higher reliability. Compared with silicon dioxide cladding or air cladding, silicon materials have a higher refractive index difference, which is beneficial to realizing waveguides and bending radii with smaller sizes, reducing the size of devices, and improving the chip integration density. Photon integration based on silicon-based materials can be compatible with CMOS processes, integrating a large number of photon devices densely on the same chip, thereby reducing the area and cost of the optical interconnection system and promoting its industrialization process.

[0003] As the most basic element of photon integration, silicon waveguides are not only responsible for the transmission of optical signals, but also the basic structure for constructing various passive devices, such as micro-rings, optical splitters, optical couplers, etc. The transmission loss and transmission mode of optical waveguides are important parameters for evaluating the performance of optical waveguides, and are also one of the important parameters for measuring the capabilities of silicon photonics process platforms. For silicon waveguide structures, the scattering loss caused by sidewall roughness is the main source of transmission loss. For example, when the sidewall roughness of the waveguide is 1 nm to 2 nm, its transmission loss is lower than 1 dB / cm; while when the sidewall roughness is 10 nm, its transmission loss will reach 30 dB / cm. In addition, geometric dimensions such as the height, width, and sidewall angle of silicon waveguides will also affect the light transmission performance of the waveguides. For example, the sidewall angle and width will affect the transmission loss, and thus determine the transmission distance of optical signals; the height and width determine the transmission mode of the waveguide, and thus affect the inter-mode crosstalk situation during optical signal transmission. Therefore, studying a high-precision non-destructive measurement method for silicon waveguide geometric parameters is of great significance for optimizing silicon photonics processes and improving the performance of photon chips.

[0004] Currently, the commonly used methods for measuring the geometric parameters of silicon waveguides include scanning electron microscopy, atomic force microscopy, laser confocal microscopy, optical interferometry, and laser speckle method. Scanning electron microscopy can obtain clear images with nm-level resolution and obtain geometric parameters such as waveguide edge roughness through the waveguide cross-section. However, this method must damage the sample, and the measurement accuracy is limited by the conductivity of the sample. Atomic force microscopy can obtain the surface profile of the sample by point-by-point scanning and can achieve nanometer-level accuracy measurement of the external profile. When using atomic force microscopy to measure the geometric dimensions of the waveguide, it is necessary to remove the silica cladding and silicon substrate around the waveguide core and protect the waveguide core from damage, and the sample preparation is complex. Laser confocal microscopy also obtains the sample profile by point-by-point scanning to obtain geometric parameters, but the accuracy can only reach dozens of nm, and the sample angle needs to be adjusted when measuring the sidewall roughness. Based on optical interference methods, such as white light interference or spectral interference, the waveguide depth can be measured with an accuracy in the nm order, but the width and sidewall angle information cannot be obtained. The laser speckle method can measure roughness, but the accuracy is only dozens of nm. In addition to the above-mentioned methods, there is also a commonly used roughness detection method, that is, by measuring the output power of waveguides with different lengths and fitting the transmission loss, and then calculating the sidewall roughness. This indirect measurement method is affected by various factors and cannot ensure the consistency of the coupling efficiency.

[0005] In the related art, the applicant found that the above-mentioned prior art has the following technical defects: Scanning electron microscopy and atomic force microscopy can achieve nm-level measurement of waveguide geometric parameters, but they need to damage the sample, and the measurement accuracy of scanning electron microscopy is limited by the conductivity of the sample; Laser confocal microscopy and laser speckle method can achieve non-destructive measurement, but the measurement accuracy is only dozens of nm, which cannot meet the requirements; Optical interference methods can only measure the depth value and cannot obtain the width and sidewall angle information. Summary of the Invention

[0006] To solve the above technical problems, an embodiment of the present invention provides a detection method for a silicon waveguide, including:

[0007] Focusing the measurement laser by using a focusing lens to obtain focused laser;

[0008] Irradiating the focused laser onto the silicon waveguide to be measured, wherein the focused laser enters the silicon waveguide to be measured after refracting from the surface of the silicon waveguide to be measured, and after multiple reflections on the inner wall of the silicon waveguide to be measured, exits from the inside of the silicon waveguide to be measured to obtain an exit laser, and the exit laser carries the information of the silicon waveguide to be measured;

[0009] Move the silicon waveguide to be measured along the optical axis of the focusing lens in a preset direction, and obtain images formed by the emitted laser when the silicon waveguide to be measured is at different positions, where the different positions include the focal position of the focusing lens and the positions at both ends of the focal point;

[0010] Obtain the contrast, autocorrelation function, and light intensity probability density of each of the images, and obtain the roughness information of the silicon waveguide to be measured based on the contrast, autocorrelation function, and light intensity probability density of each of the images.

[0011] According to an embodiment of the present invention, obtaining the roughness information of the silicon waveguide to be measured based on the contrast, autocorrelation function, and light intensity probability density of each of the images includes:

[0012] Obtain the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured based on the contrast of each of the images;

[0013] Obtain the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured based on the autocorrelation function of each of the images;

[0014] Obtain the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured based on the light intensity probability density of each of the images;

[0015] Obtain the roughness information of the silicon waveguide to be measured based on the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured, the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured, and the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured.

[0016] According to an embodiment of the present invention, obtaining the roughness information of the silicon waveguide to be measured based on the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured, the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured, and the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured includes:

[0017] Input the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured, the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured, and the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured into a trained first convolutional neural network model to obtain the roughness information of the silicon waveguide to be measured.

[0018] According to an embodiment of the present invention, the training method of the first convolutional neural network model includes:

[0019] Obtain training samples, where the training samples include the relationship between the contrast and the position of the silicon waveguide sample in the image of the silicon waveguide sample, the relationship between the autocorrelation function and the position in the image of the silicon waveguide sample, and the relationship between the light intensity probability density and the position in the image of the silicon waveguide sample;

[0020] Input the training samples into an untrained first neural network model to output a prediction result of the roughness of the silicon waveguide sample;

[0021] Analyze the prediction result and adjust the parameters of the untrained first neural network model to obtain a trained first neural network model.

[0022] According to an embodiment of the present invention, the detection method further includes:

[0023] Obtain an overfocus scanning microscopic image of the silicon waveguide to be measured according to the images formed by the emitted laser when the silicon waveguide to be measured is at different positions;

[0024] Obtain the geometric parameters of the silicon waveguide to be measured according to the overfocus scanning microscopic image of the silicon waveguide to be measured.

[0025] According to an embodiment of the present invention, obtaining an overfocus scanning microscopic image of the silicon waveguide to be measured according to the images formed by the emitted laser when the silicon waveguide to be measured is at different positions includes:

[0026] Obtain the target pixel rows of each image to obtain a plurality of target pixel rows, where the plurality of pixel rows correspond to the same position of the silicon waveguide to be measured;

[0027] Stack the plurality of target pixel rows according to the arrangement rule of the images formed by the emitted laser when the silicon waveguide to be measured is at different positions to obtain an overfocus scanning microscopic image of the silicon waveguide to be measured.

[0028] According to an embodiment of the present invention, the geometric parameters of the silicon waveguide to be measured include the height, width, and sidewall angle of the silicon waveguide to be measured.

[0029] According to an embodiment of the present invention, obtaining the geometric parameters of the silicon waveguide to be measured according to the overfocus scanning microscopic image of the silicon waveguide to be measured includes:

[0030] Input the overfocus scanning microscopic image of the silicon waveguide to be measured into a trained second convolutional neural network model to obtain the geometric parameters of the silicon waveguide to be measured.

[0031] According to an embodiment of the present invention, before focusing the measurement laser using a focusing lens, the detection method further includes:

[0032] Collimate and expand the beam of the initial laser to obtain the measurement laser.

[0033] In the embodiment of the present invention, the over-focus scanning microscopy imaging method is used to obtain the laser speckle images formed by the emitted laser when the silicon waveguide to be measured is located at the focal position and at both ends of the focal point. By collecting the statistical characteristic parameters of the laser speckle images at different positions, the contrast, autocorrelation function and light intensity probability density of each laser speckle image can be obtained, and the roughness information of the silicon waveguide to be measured can be obtained according to the contrast, autocorrelation function and light intensity probability density of each laser speckle image. Compared with the traditional laser speckle method, the method of the embodiment of the present invention can obtain more information related to the roughness of the silicon waveguide sidewall, and can effectively improve the measurement accuracy. Description of the Drawings

[0034] Figure 1 Shows a flowchart of a method for detecting a silicon waveguide according to an embodiment of the present invention;

[0035] Figure 2 Shows a schematic diagram of a detection device for a silicon waveguide according to an embodiment of the present invention.

[0036] Description of the Reference Numerals:

[0037] 1: Light source;

[0038] 2: Beam shaping module;

[0039] 3: Beam splitter;

[0040] 4: Focusing lens;

[0041] 5: Displacement stage;

[0042] 6: High-speed camera;

[0043] 7: Computer;

[0044] 8: Silicon waveguide to be measured. Detailed Embodiments

[0045] In the process of implementing the present invention, it is found that through-focus scanning optical microscopy (TSOM) is an image-based micro-nano measurement method. In traditional image-based measurement methods, the sample needs to be adjusted to the focal point to obtain a clear image of the sample. While in the TSOM method, the sample needs to be scanned along the optical axis and pass through the focal point. During the scanning process, the camera will collect a series of sample images, arrange the images according to the scanning positions and perform vertical profile processing to generate a TSOM image reflecting the trend of the light intensity distribution changing with the sample position. Since any change in the geometric parameters of the sample will affect the intensity distribution in the TSOM image, the TSOM method can be used to measure multiple geometric dimension parameters such as the height, width, and sidewall angle of a silicon waveguide. At the same time, compared with traditional imaging methods, the TSOM method utilizes more information and can break through the optical diffraction limit to obtain higher measurement accuracy.

[0046] Laser speckles are generated by the superposition of light waves scattered from a rough surface. Therefore, the roughness information of the irradiated surface is carried in the speckle image. The statistical characteristic parameters of the laser speckle image, such as contrast, autocorrelation function, and light intensity probability density function, can all reflect the roughness information of the silicon waveguide sidewall to a certain extent. However, the highest measurement accuracy of the laser speckle method can only reach dozens of nanometers. Therefore, the present invention proposes a roughness measurement method combining through-focus scanning optical microscopy and laser speckles.

[0047] To make the objectives, technical solutions, and advantages of the present disclosure clearer and more understandable, the following further elaborates on the present disclosure in detail with reference to specific embodiments and the accompanying drawings.

[0048] Figure 1 The flowchart of the detection method for a silicon waveguide according to an embodiment of the present invention is shown.

[0049] As Figure 1 shown, the detection method includes operations S1 to S4.

[0050] In operation S1, a focusing lens is used to focus the measurement laser to obtain a focused laser.

[0051] In operation S2, the focused laser is irradiated onto the silicon waveguide to be measured. Among them, after the focused laser is refracted from the surface of the silicon waveguide to be measured and enters the interior of the silicon waveguide to be measured, it undergoes multiple reflections on the inner wall of the silicon waveguide to be measured and then exits from the interior of the silicon waveguide to be measured to obtain an output laser, and the output laser carries the information of the silicon waveguide to be measured;

[0052] In operation S3, the silicon waveguide to be measured is moved along the optical axis of the focusing lens in a preset direction to obtain images formed by the output laser when the silicon waveguide to be measured is at different positions, and the different positions include the focal position of the focusing lens and the positions at both ends of the focal point;

[0053] In operation S4, obtain the contrast, autocorrelation function, and light intensity probability density of each image, and obtain the roughness information of the silicon waveguide to be measured based on the contrast, autocorrelation function, and light intensity probability density of each image.

[0054] According to an embodiment of the present invention, the image formed by the emitted laser is the laser speckle image of the silicon waveguide. The embodiment of the present invention uses the defocus scanning microscopy imaging method to obtain the laser speckle images formed by the emitted laser when the silicon waveguide to be measured is at the focal position and at both ends of the focal point. By collecting the statistical characteristic parameters of the laser speckle images at different positions, the contrast, autocorrelation function, and light intensity probability density of each laser speckle image can be obtained, and the roughness information of the silicon waveguide to be measured can be obtained based on the contrast, autocorrelation function, and light intensity probability density of each laser speckle image. Compared with the traditional laser speckle method, the method of the embodiment of the present invention can obtain more information related to the roughness of the silicon waveguide sidewall and can effectively improve the measurement accuracy.

[0055] According to an embodiment of the present invention, in operation S4, obtaining the roughness information of the silicon waveguide to be measured based on the contrast, autocorrelation function, and light intensity probability density of each image includes operations S41 to S44.

[0056] In operation S41, obtain the relationship between the contrast of the image and the position of the silicon waveguide to be measured according to the contrast of each image.

[0057] In operation S42, obtain the relationship between the autocorrelation function of the image and the position of the silicon waveguide to be measured according to the autocorrelation function of each image.

[0058] In operation S43, obtain the relationship between the light intensity probability density of the image and the position of the silicon waveguide to be measured according to the light intensity probability density of each image.

[0059] In operation S44, obtain the roughness information of the silicon waveguide to be measured according to the relationship between the contrast of the image and the position of the silicon waveguide to be measured, the relationship between the contrast of the image and the position of the silicon waveguide to be measured, and the relationship between the light intensity probability density of the image and the position of the silicon waveguide to be measured.

[0060] According to an embodiment of the present invention, the traditional laser speckle method for obtaining roughness information only uses the contrast, autocorrelation function, and light intensity probability density of one image. The method provided by the embodiment of the present invention uses the relationship between the contrast and the position of the silicon waveguide to be measured, the relationship between the contrast and the position of the silicon waveguide to be measured, and the relationship between the light intensity probability density and the position of the silicon waveguide to be measured to obtain the roughness of the silicon waveguide, which is more accurate than the measurement result of the traditional speckle method.

[0061] According to an embodiment of the present invention, in operation S44, the roughness information of the silicon waveguide to be measured obtained from the relationship between the contrast of the image varying with the position of the silicon waveguide to be measured, the relationship between the autocorrelation function of the image varying with the position of the silicon waveguide to be measured, and the relationship between the probability density of the light intensity of the image varying with the position of the silicon waveguide to be measured includes:

[0062] Input the relationship between the contrast of the image varying with the position of the silicon waveguide to be measured, the relationship between the autocorrelation function of the image varying with the position of the silicon waveguide to be measured, and the relationship between the probability density of the light intensity of the image varying with the position of the silicon waveguide to be measured into the trained first convolutional neural network model to obtain the roughness information of the silicon waveguide to be measured.

[0063] According to an embodiment of the present invention, the training method of the first convolutional neural network model includes: steps A - C.

[0064] Step A: Obtain training samples, where the training samples include the relationship between the contrast in the image of the silicon waveguide sample varying with the position of the silicon waveguide sample, the relationship between the autocorrelation function in the image of the silicon waveguide sample varying with the position in the image of the silicon waveguide sample, and the relationship between the probability density of the light intensity in the image of the silicon waveguide sample varying with the position of the image of the silicon waveguide sample.

[0065] Step B: Input the training samples into the untrained first neural network model and output the prediction result of the roughness of the silicon waveguide sample.

[0066] Step C: Analyze the prediction result and adjust the parameters of the untrained first neural network model to obtain the trained first neural network model.

[0067] According to an embodiment of the present invention, a group of silicon waveguide samples with different geometric parameters are selected, the sidewall profile of the waveguide sample is measured by an atomic force microscope, and its roughness is extracted as the true value of the sample roughness. The three statistical characteristic parameter change curves corresponding to the waveguide sample constitute the training data set. The first neural network model is used to establish the mapping relationship between the three statistical characteristic parameter change curves and the sidewall roughness of the waveguide. The trained first neural network model can be used for the prediction of the sidewall roughness of the silicon waveguide.

[0068] According to an embodiment of the present invention, the above detection method further includes: operations S5 - S6.

[0069] In operation S5, obtain the defocus scanning microscopy image of the silicon waveguide to be measured according to the image formed by the emitted laser when the silicon waveguide to be measured is at different positions;

[0070] In operation S6, obtain the geometric parameters of the silicon waveguide to be measured according to the defocus scanning microscopy image of the silicon waveguide to be measured.

[0071] According to an embodiment of the present invention, a method for measuring geometric parameters of a silicon waveguide based on the principle of defocus scanning microscopy can achieve high-precision non-destructive measurement of multiple geometric parameters such as the height, width, and sidewall angle of the silicon waveguide.

[0072] According to an embodiment of the present invention, in operation S5, obtaining a defocus scanning microscopy image of the silicon waveguide to be measured based on the images formed by the emitted laser when the silicon waveguide to be measured is at different positions includes operations S51 to S52.

[0073] In operation S51, obtaining the target pixel rows of each image to obtain a plurality of target pixel rows, where the plurality of pixel rows correspond to the same position of the silicon waveguide to be measured;

[0074] In operation S52, stacking the plurality of target pixel rows according to the arrangement rule of the images formed by the emitted laser when the silicon waveguide to be measured is at different positions to obtain a defocus scanning microscopy image of the silicon waveguide to be measured.

[0075] According to an embodiment of the present invention, the geometric parameters of the silicon waveguide to be measured include the height, width, and sidewall angle of the silicon waveguide to be measured.

[0076] According to an embodiment of the present invention, obtaining the geometric parameters of the silicon waveguide to be measured based on the defocus scanning microscopy image of the silicon waveguide to be measured includes: inputting the defocus scanning microscopy image of the silicon waveguide to be measured into a trained second convolutional neural network model to obtain the geometric parameters of the silicon waveguide to be measured.

[0077] According to an embodiment of the present invention, a scanning electron microscope is used to obtain a cross-sectional view of the silicon waveguide sample and extract the numerical values of three geometric parameters, namely height, width, and sidewall angle, as the true values of the sample geometric parameters, and form a training data set with the TSOM image corresponding to the silicon waveguide sample. A mapping relationship between the TSOM image and the waveguide geometric parameters is established using a second convolutional neural network model. The trained second convolutional neural network model can predict the numerical values of the three geometric parameters of height, width, and sidewall angle.

[0078] According to an embodiment of the present invention, before focusing the measurement laser using a focusing lens, the above detection method further includes: collimating and expanding the beam of the initial laser to obtain the measurement laser.

[0079] According to an embodiment of the present invention, a database can be established by numerical simulation to improve the CNN model. The model is used to simulate the change curves of three statistical characteristic parameters of the laser speckle image and the TSOM image under different conditions (geometric dimensions of the silicon waveguide sample, sidewall roughness, incident angle of the illumination light), and the library matching method is used to extract the parameters to be measured. The measurement method proposed by the present invention can also be used to measure the surface and inner wall roughness of the inner wall of a hole or any other structure.

[0080] Figure 2Shows a schematic diagram of a detection device for a silicon waveguide provided according to an embodiment of the present invention.

[0081] As Figure 2 shown, the detection device for the silicon waveguide includes: a light source 1, a beam shaping module 2, a semi-transparent semi-reflective mirror 3, a focusing lens 4, a displacement stage 5, a high-speed camera 6, and a computer 7.

[0082] The initial laser output by the light source 1 is collimated and expanded by the beam shaping module 2. The expanded parallel laser is the detection laser. The detection laser is reflected by the semi-transparent semi-reflective mirror 3 and then focused by the focusing lens 4 and irradiated onto the silicon waveguide 8 to be measured. The silicon waveguide 8 to be measured is placed on the displacement stage 5 and is in an out-of-focus position. The laser beam reflected by the silicon waveguide 8 to be measured is detected by the high-speed camera 6 after passing through the focusing lens 4 and the semi-transparent semi-reflective mirror 3 in sequence. The displacement stage 5 and the high-speed camera 6 are linked and controlled by the computer 7 to realize the simultaneous scanning of the reflected laser by the high-speed camera and data acquisition. The collected data is saved and processed by the computer 7, and laser speckle patterns at different positions of the silicon waveguide 8 to be measured can be obtained. According to the laser speckle patterns at different positions of the silicon waveguide 8 to be measured, the TSOM map of the silicon waveguide 8 to be measured and the variation curves of three laser speckle parameters, namely the contrast, autocorrelation function, and light intensity probability density of the laser speckle image, with the sample position are obtained.

[0083] According to the measurement principle of over-focus scanning combined with the laser speckle method proposed in the embodiment of the present invention, non-destructive measurement of the sidewall roughness of the silicon waveguide to be measured can be achieved, and the measurement accuracy can reach the nm level.

[0084] The above specific embodiments have further detailed the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A detection method for a silicon waveguide, comprising: Focusing a measurement laser using a focusing lens to obtain a focused laser; Irradiating the focused laser onto the silicon waveguide to be measured, wherein the focused laser enters the interior of the silicon waveguide to be measured after refraction from the surface of the silicon waveguide to be measured, undergoes multiple reflections on the inner wall of the silicon waveguide to be measured, and then exits from the interior of the silicon waveguide to be measured to obtain an output laser, and the output laser carries information of the silicon waveguide to be measured; Moving the silicon waveguide to be measured along the optical axis of the focusing lens in a preset direction, and acquiring images formed by the output laser when the silicon waveguide to be measured is at different positions, where the different positions include the focal position of the focusing lens and positions at both ends of the focal point; Acquiring the contrast, autocorrelation function, and light intensity probability density of each image, and obtaining the roughness information of the silicon waveguide to be measured based on the contrast, autocorrelation function, and light intensity probability density of each image.

2. The detection method according to claim 1, wherein, Obtaining the roughness information of the silicon waveguide to be measured based on the contrast, autocorrelation function, and light intensity probability density of each image includes: Obtaining the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured based on the contrast of each image; Obtaining the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured based on the autocorrelation function of each image; Obtaining the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured based on the light intensity probability density of each image; Obtaining the roughness information of the silicon waveguide to be measured based on the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured, the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured, and the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured.

3. The detection method according to claim 2, wherein Obtaining the roughness information of the silicon waveguide to be measured based on the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured, the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured, and the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured includes: Inputting the relationship between the change in the contrast of the image and the position of the silicon waveguide to be measured, the relationship between the change in the autocorrelation function of the image and the position of the silicon waveguide to be measured, and the relationship between the change in the light intensity probability density of the image and the position of the silicon waveguide to be measured into a trained first convolutional neural network model to obtain the roughness information of the silicon waveguide to be measured.

4. The detection method according to claim 3, wherein, The training method of the first convolutional neural network model includes: Obtaining training samples, where the training samples include the relationship between the change in the contrast in the image of the silicon waveguide sample and the position of the silicon waveguide sample, the relationship between the change in the autocorrelation function in the image of the silicon waveguide sample and the position in the image of the silicon waveguide sample, and the relationship between the change in the light intensity probability density in the image of the silicon waveguide sample and the position of the image of the silicon waveguide sample; Inputting the training samples into an untrained first neural network model and outputting a prediction result of the roughness of the silicon waveguide sample; Analyzing the prediction result and adjusting the parameters of the untrained first neural network model to obtain a trained first neural network model.

5. The detection method according to claim 1, wherein, The detection method further includes: Obtain the defocused scanning microscopic image of the silicon waveguide to be measured based on the images formed by the emitted laser when the silicon waveguide to be measured is at different positions; Obtain the geometric parameters of the silicon waveguide to be measured based on the defocused scanning microscopic image of the silicon waveguide to be measured.

6. The detection method according to claim 5, wherein, Obtaining the defocused scanning microscopic image of the silicon waveguide to be measured based on the images formed by the emitted laser when the silicon waveguide to be measured is at different positions includes: Obtain the target pixel rows of each image to obtain a plurality of target pixel rows, where the plurality of pixel rows correspond to the same position of the silicon waveguide to be measured; Stack the plurality of target pixel rows according to the arrangement rule of the images formed by the emitted laser when the silicon waveguide to be measured is at different positions to obtain the defocused scanning microscopic image of the silicon waveguide to be measured.

7. The detection method according to claim 5, wherein, The geometric parameters of the silicon waveguide to be measured include the height, width and sidewall angle of the silicon waveguide to be measured.

8. The detection method according to claim 5, wherein, Obtaining the geometric parameters of the silicon waveguide to be measured based on the defocused scanning microscopic image of the silicon waveguide to be measured includes: Input the defocused scanning microscopic image of the silicon waveguide to be measured into a trained second convolutional neural network model to obtain the geometric parameters of the silicon waveguide to be measured.

9. The detection method according to any one of claims 1-8, wherein, Before focusing the measurement laser using a focusing lens, the detection method further includes: Collimate and expand the beam of the initial laser to obtain the measurement laser.

Images