Intelligent statistical method and system for laser processing carbon nanotube wafer level characterization

Abstract

The application provides a kind of laser processing carbon nanotube wafer level characterization intelligent statistical method and system.The method of the application comprises the following steps:S1: growing carbon nanotube array or network structure on a silicon wafer;S2: laser processing carbon nanotube array or network structure using femtosecond laser;S3: obtaining silicon wafer surface data after laser processing using multi-scale characterization technology;S4: constructing a multi-modal deep learning model according to the silicon wafer surface data;S5: performing intelligent statistical analysis and generating wafer level characterization report according to the output results of the multi-modal deep learning model;S6: optimizing laser processing parameters according to the wafer level characterization report.The application realizes accurate processing and comprehensive characterization of carbon nanotubes at wafer level, which is beneficial to improve the manufacturing precision and yield of carbon nanotube devices.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new generation information technology, in particular to a laser processing carbon nanotube wafer level characterization intelligent statistical method and system. BACKGROUND

[0002] With the continuous development of semiconductor technology, traditional silicon-based devices are approaching their physical limits. Carbon nanotubes (CNTs) as a kind of nanomaterial with unique electrical and mechanical properties, show great potential in the field of future electronic devices, sensors and energy storage. However, there are still many challenges in successfully integrating carbon nanotubes into large-scale semiconductor manufacturing processes, the most critical of which is how to accurately control and characterize the growth and processing of carbon nanotubes at the wafer level.

[0003] Currently, the growth of 2-5nm thick carbon nanotube array or network film on 4-8 inch silicon wafers poses a great challenge to its characterization and processing. Traditional processing methods are difficult to achieve nanoscale precision, and conventional characterization techniques cannot evaluate the structure and performance of individual nanotubes. Laser processing, especially femtosecond laser technology, has become an ideal choice for processing carbon nanotubes due to its high precision and low heat-affected zone; however, to achieve uniform and controllable processing at the wafer level, it is still necessary to deeply understand the interaction mechanism between laser and carbon nanotubes and accurately control the laser parameters. In addition, defects and structural changes may be introduced during processing, which need to be evaluated and quantified by advanced characterization techniques; however, how to effectively integrate multi-scale, multi-modal data and extract meaningful statistical information from them has become a problem to be solved.

[0004] In view of this, the present application is proposed. SUMMARY

[0005] The purpose of the present application is to provide a laser processing carbon nanotube wafer level characterization intelligent statistical method and system, which realizes the accurate processing and comprehensive characterization of carbon nanotubes at the wafer level, and is beneficial to improve the manufacturing precision and yield of carbon nanotube devices.

[0006] The present application provides a laser processing carbon nanotube wafer level characterization intelligent statistical method, comprising the following steps:

[0007] S1: growing carbon nanotube array or network structure on a silicon wafer;

[0008] S2: using femtosecond laser to process the carbon nanotube array or network structure;

[0009] S3: using multi-scale characterization technology to obtain silicon wafer surface data after laser processing;

[0010] S4: constructing a multi-modal deep learning model according to the silicon wafer surface data;

[0011] S5: Intelligent statistical analysis based on the output results of the multi-modal deep learning model and generation of wafer-level characterization report;

[0012] S6: Optimization of laser processing parameters based on the wafer-level characterization report.

[0013] In step S1, the size of the silicon wafer is 4-8 inches; the silicon wafer surface is provided with a SiO2 layer with a thickness of 200-300 nm; the carbon nanotube array or network structure has a thickness of 2-5 nm; the density of the carbon nanotube array or network structure is 100-1000 roots / μm 2 ; the average diameter of the carbon nanotube array or network structure is 1-2 nm.

[0014] In step S2, the wavelength of the femtosecond laser is 400-800 nm; the pulse width is less than 500 fs; the power density is 5×10 12 -8×10 12 W / cm 2 ; the adjustable range of the repetition frequency is 1 Hz to 1 MHz; the adjustable range of the spot diameter is 1-100 μm.

[0015] In step S3, the multi-scale characterization technology is used to obtain the silicon wafer surface data, including:

[0016] a) Using a field emission scanning electron microscope (FE-SEM) with a resolution better than 1 nm to obtain large-area topography information;

[0017] b) Using a transmission electron microscope (TEM) with a resolution better than 0.5 nm to analyze the atomic structure of a single carbon nanotube;

[0018] c) Using a confocal Raman spectrometer with a spatial resolution less than 1 μm to evaluate the defect density and lattice changes of the carbon nanotube;

[0019] d) Using an atomic force microscope (AFM) with a vertical resolution better than 0.5 nm to measure the height and surface roughness of the carbon nanotube.

[0020] In step S4, the multi-modal deep learning model has the following functions:

[0021] a) Identifying and classifying the structural features of carbon nanotubes, including diameter (accuracy ±0.1 nm), length (accuracy ±1 nm), chirality, and defect type;

[0022] b) Evaluating the impact of laser processing on the crystal structure of carbon nanotubes, identifying changes in defect density (accuracy ±1%);

[0023] c) analyze the uniformity and consistency of carbon nanotube distribution on the silicon wafer surface, and generate a distribution map with a spatial resolution of 1 μm x 1 μm.

[0024] The training data set of the multi-modal deep learning model contains at least 5000 samples; the accuracy of the multi-modal deep learning model is not less than 95%.

[0025] In step S5, the intelligent statistical analysis step includes:

[0026] a) evaluate the growth density (accuracy ± 1 root / μm 2 ) and orientation consistency (accuracy ± 1°) of nanotubes in different areas of the silicon wafer;

[0027] b) analyze the effect of laser processing on the structural integrity of carbon nanotubes, including defect type identification and density estimation (accuracy ± 0.1 / nm);

[0028] c) calculate the length and diameter distribution of carbon nanotubes after laser processing, and generate a histogram (bin width is 0.1 nm);

[0029] d) estimate the defect density and type on the surface of the silicon wafer, with a spatial resolution of 1 mm x 1 mm.

[0030] In step S6, the laser processing parameters are optimized according to the wafer-level characterization report, realizing closed-loop control, so that the processing accuracy reaches ± 1 nm.

[0031] The present application also provides a laser processing carbon nanotube wafer-level characterization intelligent statistical system for realizing the above method, comprising:

[0032] Carbon nanotube growth module: growing carbon nanotube array or network structure on silicon wafer;

[0033] Femtosecond laser processing module: using femtosecond laser to process carbon nanotube array or network structure;

[0034] Multi-scale characterization module: using multi-scale characterization technology to obtain silicon wafer surface data after laser processing;

[0035] Multi-modal deep learning module: processing silicon wafer surface data after laser processing and outputting results;

[0036] Intelligent statistical analysis module: intelligent statistical analysis according to the output results of the multi-modal deep learning model and generating wafer-level characterization report;

[0037] Parameter optimization module: optimizing laser processing parameters according to wafer-level characterization report.

[0038] Specifically, the carbon nanotube growth module can grow carbon nanotube array or network structure on 4-8 inch wafer with a density of 100-1000 roots / μm2 The laser repetition frequency of the femtosecond laser processing module can be adjusted in the range of 1 Hz-1 MHz, and the spot diameter can be adjusted in the range of 1-100 mu m; the multi-scale characterization module includes FE-SEM with a resolution better than 1 nm, TEM with a resolution better than 0.1 nm, Raman spectrometer with a spatial resolution less than 1 mu m, and AFM with a vertical resolution better than 0.5 nm; the processing speed of the multi-modal deep learning module is not less than 1 GB / s, and the data from different characterization devices can be processed simultaneously, and the atomic scale carbon nanotube structure defects as small as 0.1 nm can be identified and classified; the intelligent statistical analysis module can complete the data analysis and report generation of the whole wafer within 1 hour, and can generate a wafer surface nanotube distribution heat map with a spatial resolution of 10 mu m*10 mu m; the parameter optimization module can optimize the laser processing parameters to the best state within 10 iterations.

[0039] The laser processing carbon nanotube wafer level characterization intelligent statistical system of the application further comprises an automatic sample processing module, which can automatically complete the loading, unloading and transmission between different characterization devices of the wafer, and the processing time of a single wafer is not more than 30 minutes.

[0040] The laser processing carbon nanotube wafer level characterization intelligent statistical system of the application further comprises an in-situ monitoring module, which uses a high-speed camera (frame rate not less than 10,000 fps) to monitor the structural changes of carbon nanotubes in the laser processing process in real time.

[0041] The laser processing carbon nanotube wafer level characterization intelligent statistical system of the application further comprises a database module, which stores nanotube characterization data under different growth conditions and laser processing parameters, and the storage capacity is not less than 10 TB.

[0042] The laser processing carbon nanotube wafer level characterization intelligent statistical system of the application further comprises a human-computer interaction interface, which can display the laser processing and characterization process in real time, and allow the operator to adjust the laser processing parameters according to the intermediate results, and the response time is less than 100 ms.

[0043] The implementation of the application has at least the following advantages:

[0044] 1) The application can realize the accurate laser processing of 2-5 nm thick carbon nanotube layer on 4-8 inch wafer, and the processing accuracy reaches ±1 nm;

[0045] 2) The application provides multi-scale and all-around characterization capability from atom to wafer level, which is beneficial to deeply understand the structure and performance of carbon nanotube;

[0046] 3) The application realizes the rapid processing and analysis of large-scale data through artificial intelligence algorithm, which is beneficial to generate high-precision wafer level statistical report;

[0047] 4) The present application significantly improves the manufacturing efficiency and yield of carbon nanotube devices, accelerating their large-scale application in the semiconductor industry;

[0048] 5) The present application provides a comprehensive platform for the study of carbon nanotubes and other nanomaterials, promoting the development of nanotechnology. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can obtain other drawings without creative labor on the basis of these drawings.

[0050] Figure 1 Structure diagram of laser processing carbon nanotube wafer level characterization intelligent statistical system.

[0051] Explanation of reference signs:

[0052] 1: Carbon nanotube growth module; 2: Femtosecond laser processing module; 3: Multi-scale characterization module; 4: Multi-modal deep learning module; 5: Intelligent statistical analysis module; 6: Parameter optimization module; 7: In-situ monitoring module; 8: Database module; 9: Automatic sample processing module; 10: Man-machine interface. DETAILED DESCRIPTION

[0053] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0054] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form also includes the plural form unless the context clearly indicates otherwise, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component and / or combination thereof.

[0055] The technical solutions of the present application will be described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0056] Example 1

[0057] The laser processing carbon nanotube wafer level characterization intelligent statistical method of the embodiment takes a single-walled carbon nanotube (SWCNT) network structure grown on a 6-inch Si wafer as an object, and the steps are as follows:

[0058] 1. Sample preparation:

[0059] First, a high-purity (99.999%) 6-inch p-type Si wafer is selected as the substrate.

[0060] A 280 nm thick SiO2 layer is grown on the surface of the Si wafer by thermal oxidation method as an insulating layer and buffer layer for subsequent carbon nanotube growth; the uniformity of the SiO2 layer thickness is controlled within ±2 nm to ensure the consistency of subsequent carbon nanotube growth.

[0061] Subsequently, a 2-3 nm thick SWCNT network structure is grown on the surface of the SiO2 layer by an improved chemical vapor deposition (CVD) method; the specific steps are as follows:

[0062] a) Spin 0.01 mol / L Fe(NO3)3 ethanol solution on the surface of the SiO2 layer as a catalyst precursor;

[0063] b) Place the sample in a CVD furnace, introduce Ar gas (500 sccm) at 850°C for 30 minutes to reduce the formation of nanoscale Fe catalyst particles;

[0064] c) Switch the gas source to a mixture of CH4 (200 sccm) and H2 (50 sccm), grow at 900°C for 15 minutes;

[0065] d) Cool to room temperature to obtain a uniformly distributed SWCNT network structure.

[0066] Through the above-mentioned optimized growth parameters, an average diameter of 1.2±0.2 nm of the SWCNT network structure is obtained, and the density of the network structure reaches 200±20 roots / μm 2 .

[0067] Using Raman spectroscopy to confirm that more than 90% of the SWCNTs are semiconducting, which lays the foundation for subsequent device applications.

[0068] 2. Laser processing:

[0069] A titanium sapphire femtosecond laser system is used for precise processing of the SWCNT network structure, and the main parameters of the system are as follows:

[0070] Center wavelength: 800 nm;

[0071] Pulse width: 100 fs;

[0072] Repetition rate: adjustable range 1 Hz - 1 MHz, in this example set to 1 kHz;

[0073] Maximum output power: 5 W;

[0074] Spot diameter: adjustable range 1 - 100 pm, in this example set to 10 pm.

[0075] During laser processing, the power density is precisely controlled in the range of 5 x 10 12 - 8 x 10 12 W / cm 2 This range enables selective cutting and modification of the SWCNT network structure without introducing excessive defects.

[0076] The laser processing mechanism is mainly based on the synergistic effect of photothermal and photochemical effects:

[0077] a) Photothermal effect: femtosecond laser pulses transfer energy to SWCNTs in a very short time, causing a rapid increase in local temperature (estimated to be 2000-3000 K), leading to C-C bond rupture;

[0078] b) Photochemical effect: high-energy photons directly cause the rupture and recombination of chemical bonds in SWCNTs, especially at defect sites and tube ends.

[0079] By precisely controlling the above two effects, local modification of the network structure can be achieved while maintaining the integrity of most SWCNTs. For example, metal-type SWCNTs can be selectively cut, improving the semiconductor properties of the network; or controlled defects can be introduced at specific locations to adjust the carrier transport properties.

[0080] To achieve uniform processing over a large area, a grid scanning strategy is adopted. A 6-inch wafer is divided into 1 cm x 1 cm regions, and a serpentine scan is performed in each region. The overlap rate between adjacent scan lines is set to 30% to ensure uniformity of processing. The processing time for the entire wafer is about 2 hours.

[0081] 3. Multi-scale characterization:

[0082] Through various high-resolution characterization techniques, comprehensive analysis from atomic to wafer scale is achieved:

[0083] a) Field emission scanning electron microscope (FE-SEM):

[0084] The whole wafer was scanned using FE-SEM with a resolution of 0.8 nm. Low acceleration voltage (1 kV) was used to reduce the damage to SWCNTs. High-resolution imaging of the whole wafer was completed within 12 hours by using an automated stage and image stitching technique. These images were used to analyze the large-area morphology, density distribution, and connectivity of SWCNT networks.

[0085] b) Transmission Electron Microscopy (TEM):

[0086] The atomic structure of individual SWCNTs was analyzed using TEM with a resolution of 0.2 nm. Sample preparation used a direct transfer method: a layer of PMMA was applied on the wafer surface, and then the SWCNT network was transferred to a TEM copper grid by etching the SiO2 layer with HF. The SWCNTs near the laser-processed region were analyzed in detail to observe their wall structure, defect types, and distribution.

[0087] c) Confocal Raman Spectrometer:

[0088] A confocal Raman spectrometer with a spatial resolution of 0.5 μm was used, with 532 nm laser excitation. By analyzing the intensity ratio of G peak (~1590 cm-1) and D peak (~1350 cm-1) (IG / ID), the defect density variation of SWCNTs was evaluated. Meanwhile, the diameter distribution of SWCNTs was analyzed using the RBM mode (100-300 cm -1 ). 100 points on the wafer were selected for measurement to obtain statistically significant Raman spectral changes.

[0089] d) Atomic Force Microscopy (AFM):

[0090] AFM with a vertical resolution of 0.1 nm was used to measure the height distribution and surface roughness of SWCNT networks. Tapping mode was used to reduce the impact on the sample. Within each 1 cm x 1 cm area, 5 100 μm x 100 μm scan areas were randomly selected to construct a three-dimensional topographic map of the SWCNT network.

[0091] 4. Multimodal Deep Learning Analysis:

[0092] A complex multimodal deep learning model was constructed, including the following components:

[0093] Convolutional Neural Network (CNN) for SEM image analysis;

[0094] Graph Neural Network (GNN) for TEM atomic structure identification;

[0095] Recurrent Neural Network (RNN) for Raman spectral data processing;

[0096] 3D-CNN for AFM data analysis;

[0097] Multimodal fusion networks integrate data from different sources.

[0098] The model was trained using 8000 pre-labeled samples, including SWCNT samples under various growth conditions and laser processing parameters. The training set was further expanded using data augmentation techniques such as rotation, flipping, and adding noise. The final model achieved a recognition accuracy of 96.5% on the validation set.

[0099] The above-mentioned multimodal deep learning model has the following functions:

[0100] a) Identify the diameter (accuracy ±0.1 nm), length (accuracy ±1 nm), and chirality of SWCNTs; chirality identification is based on a combined analysis of TEM images and Raman spectroscopy;

[0101] b) Evaluate the impact of laser processing on the SWCNT network structure, including changes in defect density (accuracy ±1%) and defect types (such as quinary rings, septums, vacancies, etc.);

[0102] c) Analyze the uniformity and connectivity of the SWCNT network to generate a distribution map with a resolution of 1 μm × 1 μm; connectivity analysis is crucial for predicting the performance of electronic devices.

[0103] 5. Intelligent statistical analysis:

[0104] Based on the output of a multimodal deep learning model, data analysis of the entire 6-inch wafer is completed within 45 minutes, generating a comprehensive characterization report; the report includes:

[0105] a) SWCNT network density distribution map (accuracy ±1 root / μm) 2 Presented in the form of a heatmap;

[0106] b) SWCNT defect types and density distribution in the laser-processed area (accuracy ±0.1 defects / nm), including pie charts of defect types and density gradient charts;

[0107] c) Calculate the histogram of SWCNT diameter and length distribution (bin width 0.1 nm) and the mean and standard deviation;

[0108] d) Evaluation of SWCNT network uniformity on wafer surface, with a spatial resolution of 1 mm × 1 mm, quantified using coefficient of variation (CV).

[0109] e) Network conductivity prediction map, based on a comprehensive analysis of SWCNT density, orientation, and defect distribution.

[0110] 6. Closed-loop control optimization:

[0111] Laser processing parameters are automatically adjusted based on characterization results; optimization goal is set to maximize electrically. Optimization process uses Bayesian optimization algorithm, considering the interaction of multiple parameters. After 7 iterations, the following optimized parameters are obtained:

[0112] Laser power density: 6.5 x 1012 W / cm 2 ;

[0113] Scan speed: 2 mm / s;

[0114] Spot overlap ratio: 30%;

[0115] Pulse repetition frequency: 2 kHz.

[0116] After optimization, the conductivity of SWCNT network structure is improved by 20%, while maintaining a low defect density (IG / ID > 100). Through the local annealing effect, a reduction in contact resistance between SWCNTs is also observed.

[0117] 7、In-situ monitoring:

[0118] A high-speed CMOS camera with a frame rate of 20,000 fps is used to monitor the laser processing process in real time. The camera is triggered synchronously with the laser, capturing transient changes caused by each laser pulse. Through real-time image processing algorithms, the system can detect structural changes in the SWCNT network during processing, such as sudden brightness changes or topography changes, which are used to adjust laser parameters in real time.

[0119] In addition, an integrated photoconductivity measurement module is used to monitor changes in the conductivity of the SWCNT network during processing by applying a microelectrode array on the sample surface, thereby providing direct feedback on changes in material performance.

[0120] 8、Data storage and management:

[0121] All experimental data and analysis results are stored in a 15TB capacity high-performance solid state hard disk array. A distributed database system is used to ensure fast data read / write and retrieval. The data organization uses a multi-level structure:

[0122] First layer: raw data (SEM, TEM, Raman, AFM, etc.);

[0123] Second layer: processed data (image enhancement, spectral analysis results, etc.);

[0124] Third layer: deep learning model output;

[0125] Fourth layer: statistical analysis report and optimization suggestions.

[0126] The system automatically classifies and indexes data and implements data version control. Through machine learning algorithms, potential process rules can be mined from historical data to guide future experimental design.

[0127] Embodiment 2

[0128] In combination Figure 1 As shown in the figure, the embodiment provides an intelligent statistical system for laser processing carbon nanotube wafer-level characterization, which is used to implement the method of embodiment 1, and includes:

[0129] 1) Carbon nanotube growth module:

[0130] The carbon nanotube growth module 1 is used to grow nanotube arrays or network structures with a density of 100-1000 roots / μm on 4-8 inch wafers. 2

[0131] 2) Femtosecond laser processing module:

[0132] The femtosecond laser processing module 2 uses femtosecond laser to process carbon nanotube arrays or network structures. A femtosecond laser system with a wavelength of 400-800 nm and a pulse width less than 500 fs is used, and the power density can be accurately adjusted in the range of 5×10 12 -8×10 12 W / cm 2

[0133] The femtosecond laser processing module 2 can selectively process 2-5 nm thick nanotube arrays or network structures, achieving nanoscale precision cutting, modification or functionalization. The laser repetition frequency can be adjusted in the range of 1 Hz-1 MHz, and the spot diameter can be adjusted in the range of 1-100 μm to adapt to different processing needs.

[0134] 3) Multi-scale characterization module:

[0135] The multi-scale characterization module 3 uses multi-scale characterization technology to obtain silicon wafer surface data after laser processing. The multi-scale characterization module 3 integrates a variety of high-resolution characterization equipment, including:

[0136] Field emission scanning electron microscope (FE-SEM) with resolution better than 1 nm;

[0137] Transmission electron microscope (TEM) with resolution better than 0.1 nm;

[0138] Confocal Raman spectrometer with spatial resolution less than 1 μm;

[0139] Atomic force microscope (AFM) with vertical resolution better than 0.1 nm.

[0140] ​​The combination of the above-mentioned multiple high-resolution characterization devices enables the system to comprehensively characterize the structure and performance of carbon nanotubes from atomic to wafer level.

[0141] 4) Multi-modal deep learning module:

[0142] The multi-modal deep learning module 4 processes the data of the silicon wafer surface after laser processing and outputs the results, which can simultaneously process data from different characterization devices, with a processing speed not less than 1GB / s.

[0143] The multi-modal deep learning module 4 has the following functions:

[0144] Identify and classify the structural features of carbon nanotubes, including diameter (accuracy ±0.1 nm), length (accuracy ±1 nm), chirality and defect type;

[0145] Evaluate the impact of laser processing on the crystal structure of carbon nanotubes, identify the change in defect density (accuracy ±1%);

[0146] Analyze the uniformity and consistency of the distribution of carbon nanotubes on the silicon wafer surface, and generate a distribution map with a spatial resolution of 1μm×1μm.

[0147] 5) Intelligent statistical analysis module:

[0148] The intelligent statistical analysis module 5 performs intelligent statistical analysis based on the output results of the multi-modal deep learning model and generates a wafer-level characterization report, which can complete data analysis and report generation for the entire wafer within 1 hour, including:

[0149] Evaluate the growth density (accuracy ±1 root / μm 2 ) and orientation consistency (accuracy ±1°) of carbon nanotubes in different areas of the silicon wafer;

[0150] Analyze the impact of laser processing on the structural integrity of carbon nanotubes, including defect type identification and density estimation (accuracy ±0.1 / nm).

[0151] Calculate the length and diameter distribution of carbon nanotubes after laser processing, and generate a histogram (bin width 0.1 nm);

[0152] Estimate the defect density and type on the surface of the silicon wafer with a spatial resolution of 1mm×1mm.

[0153] 6) Parameter optimization module:

[0154] The parameter optimization module 6 optimizes the laser processing parameters based on the wafer-level characterization report for closed-loop control and parameter optimization, which can optimize the laser processing parameters to the best state within 10 iterations, making the processing accuracy reach ±1nm.

[0155] 7) In-situ monitoring module:

[0156] The in-situ monitoring module 7 uses a high-speed camera (frame rate not less than 10,000 fps) to monitor the structural changes of nanotubes in real time during laser processing, providing real-time feedback for process control.

[0157] 8) Database module:

[0158] The storage capacity of the database module 8 is not less than 10 TB, and it stores the characterization data of carbon nanotubes under different growth conditions and processing parameters, providing data support for long-term process optimization.

[0159] 9) Automatic sample processing module:

[0160] The automatic sample processing module 9 can automatically complete the loading, unloading and transmission between different characterization devices of the wafer, and the processing time of a single wafer is not more than 4 hours, greatly improving the work efficiency.

[0161] 10) Human-computer interaction interface:

[0162] The response time of the human-computer interaction interface 10 is less than 100 ms, which can display the processing and characterization process in real time and allow the operator to adjust the parameters according to the intermediate results.

[0163] The system of the embodiment integrates carbon nanotube growth, femtosecond laser processing, multi-scale characterization, multi-modal deep learning, intelligent statistical analysis and parameter optimization modules, realizes the accurate laser processing of 2-5 nm thick carbon nanotube layer on 4-8 inch wafers, provides multi-scale and all-around characterization capability from atomic to wafer level, realizes the rapid processing and analysis of large-scale data through artificial intelligence algorithm, improves the manufacturing precision and yield of carbon nanotube devices, and lays a foundation for its application in large-scale integrated circuits.

[0164] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A smart statistical method for wafer-level characterization of laser-processed carbon nanotubes, characterized in that, Includes the following steps: S1: Growing carbon nanotube arrays or network structures on silicon wafers; S2: Laser processing of carbon nanotube arrays or network structures using femtosecond lasers; S3: Multi-scale characterization techniques were used to obtain surface data of the silicon wafer after laser processing; S4: Construct a multimodal deep learning model based on silicon wafer surface data; S5: Perform intelligent statistical analysis based on the output of the multimodal deep learning model and generate a wafer-level characterization report; S6: Optimize laser processing parameters based on wafer-level characterization reports; In step S3, obtaining silicon wafer surface data using multi-scale characterization techniques includes: a) Use a field emission scanning electron microscope with a resolution better than 1 nm to obtain large-area morphology information; b) The atomic structure of a single carbon nanotube was analyzed using a transmission electron microscope with a resolution better than 0.5 nm; c) Use a confocal Raman spectrometer with a spatial resolution of less than 1 μm to evaluate the defect density and lattice variation of carbon nanotubes; d) The height and surface roughness of carbon nanotubes were measured using an atomic force microscope with a vertical resolution better than 0.5 nm; In step S4, the multimodal deep learning model includes the following components: Convolutional neural networks for SEM image analysis; Graph neural networks for TEM atomic structure identification; Recurrent neural networks for Raman spectroscopy data processing; 3D-CNN for AFM data analysis; Multimodal fusion networks integrate data from different sources.

2. The intelligent statistical method for wafer-level characterization of laser-processed carbon nanotubes according to claim 1, characterized in that, In step S1, the silicon wafer is 4-8 inches in size; the surface of the silicon wafer has a SiO2 layer with a thickness of 200-300 nm; the carbon nanotube array or network structure has a thickness of 2-5 nm; and the density of the carbon nanotube array or network structure is 100-1000 nanotubes / μm. 2 .

3. The intelligent statistical method for wafer-level characterization of laser-processed carbon nanotubes according to claim 1, characterized in that, In step S2, the femtosecond laser has a wavelength of 400-800 nm, a pulse width of less than 500 fs, and a power density of 5 × 10⁻⁶. 12 -8×10 12 W / cm 2 The repetition frequency is adjustable from 1 Hz to 1 MHz; the spot diameter is adjustable from 1 to 100 μm.

4. The intelligent statistical method for wafer-level characterization of laser-processed carbon nanotubes according to claim 1, characterized in that, In step S4, the multimodal deep learning model has the following functions: a) Identify and classify the structural features of carbon nanotubes, including diameter, length, chirality, and defect type; b) Evaluate the impact of laser processing on the crystal structure of carbon nanotubes and identify changes in defect density; c) Analyze the uniformity and consistency of carbon nanotube distribution on the silicon wafer surface to generate a distribution map with a spatial resolution of 1 μm × 1 μm.

5. The intelligent statistical method for wafer-level characterization of laser-processed carbon nanotubes according to claim 1, characterized in that, In step S5, the intelligent statistical analysis steps include: a) Evaluate the growth density and orientation consistency of carbon nanotubes in different regions of a silicon wafer; b) Analyze the impact of laser processing on the structural integrity of carbon nanotubes, including defect type identification and density estimation; c) Calculate the length and diameter distribution of carbon nanotubes after laser processing and generate a histogram; d) Estimate the density and type of defects on the silicon wafer surface with a spatial resolution of 1 mm × 1 mm.

6. A smart statistical system for wafer-level characterization of laser-processed carbon nanotubes for implementing the method of any one of claims 1-5, characterized in that, include: Carbon nanotube growth module: grows carbon nanotube arrays or network structures on silicon wafers; Femtosecond laser processing module: Uses femtosecond lasers to perform laser processing on carbon nanotube arrays or network structures; Multi-scale characterization module: Employs multi-scale characterization techniques to acquire surface data of laser-processed silicon wafers; Multimodal deep learning module: processes surface data of silicon wafers after laser processing and outputs results; Intelligent statistical analysis module: Performs intelligent statistical analysis based on the output of the multimodal deep learning model and generates a wafer-level characterization report; Parameter optimization module: Optimizes laser processing parameters based on wafer-level characterization reports.

7. The intelligent statistical system for wafer-level characterization of laser-processed carbon nanotubes according to claim 6, characterized in that, It also includes an in-situ monitoring module, which uses a high-speed camera to monitor the structural changes of carbon nanotubes in real time during laser processing.

8. The intelligent statistical system for wafer-level characterization of laser-processed carbon nanotubes according to claim 6, characterized in that, It also includes a database module to store nanotube characterization data under different growth conditions and laser processing parameters.

9. The intelligent statistical system for wafer-level characterization of laser-processed carbon nanotubes according to claim 6, characterized in that, It also includes a human-machine interface that can display the laser processing and characterization process in real time and allow the operator to adjust the laser processing parameters based on intermediate results.

Images