Artificial intelligence enhanced strong-field terahertz spectroscopy detection imaging apparatus and method

Abstract

The present application relates to the technical field of terahertz wave detection imaging, and particularly relates to an artificial intelligence enhanced strong field terahertz spectroscopy detection imaging device and method, which comprises a femtosecond laser, a beam splitter, a chopper, a laser spot beam expander module, a spin terahertz emitter, a transmission-reflection integrated automatic scanning detection module, a terahertz detection device, and an artificial intelligence enhanced thickness measurement module and an artificial intelligence enhanced defect detection imaging module; the laser emitted by the femtosecond laser generates pump light and detection light by the beam splitter, the pump light sequentially passes through the laser spot beam expander module, the spin terahertz emitter and the transmission-reflection integrated automatic scanning detection module to enter the terahertz detection device; the detection light passes through the transmission-reflection integrated automatic scanning detection module to enter the terahertz detection device, the terahertz detection device outputs a terahertz pulse shape, and finally the sample thickness or defect profile is obtained; the present application can improve the sample defect detection efficiency and reduce the cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of terahertz wave detection imaging, in particular to an artificial intelligence enhanced strong field terahertz spectroscopy detection imaging device and method. BACKGROUND

[0002] The terahertz wave band is located in the frequency range of 0.1-10 THz, as an electromagnetic wave band between microwaves and infrared, it has the strong penetration of microwaves and the high resolution of infrared. Compared with traditional detection technology, terahertz imaging has unique advantages in the detection of metal defects, coatings, packaging materials and the like. Its penetration depth can reach centimeter level, and the spatial resolution can reach sub-millimeter level, so that the internal structure of the material can be finely characterized. Compared with microwave detection, the wavelength of terahertz electromagnetic wave is shorter, which can significantly improve the interface defect recognition ability; compared with infrared detection, terahertz wave can penetrate infrared shielding medium such as composite material, and avoid the influence of heat radiation and high energy on the detection object. At present, the nondestructive detection and spectroscopy imaging technology of terahertz wave has great application potential and broad development prospect in the basic scientific researches such as security screening, cultural relic protection, paint thickness measurement, internal structure detection of composite material, and nanometer scale imaging of condensed state.

[0003] With the new generation of technological revolution and industrial upgrading, the technologies of vehicle manufacturing, aerospace and the like are developing vigorously, and the demand for the thickness measurement of non-metallic materials such as paint and thermal barrier coating and the detection of metal surface defects is increasing. Benefiting from the penetration and high spatial resolution advantages of terahertz for many non-polar materials, terahertz spectral analysis and spectroscopy imaging have become common means for composite material thickness measurement and nondestructive detection. Time-of-flight method is a traditional method for measuring the thickness of penetrable material and CT imaging, which has excellent characteristics such as convenience and intuitiveness. However, in order to achieve more accurate measurement, the refractive index of the material to be measured needs to be accurately known. In addition, the time-of-flight method for measuring thickness is usually based on the time difference of the reflection waves of the front and back surfaces, and for samples with very small thickness, multiple reflection peaks may be overlapped, which is difficult to calculate directly. Existing researches have shown that by introducing artificial intelligence method and using the thickness measurement signal database for iteration, the trained results can effectively identify the reflection peak time difference corresponding to the mixed signal. At present, the research based on CNN convolutional neural network can realize the thickness measurement of 20 microns, and it is very important to improve the effectiveness and stability of the training results and the generalization in different scenarios. In addition to this, the imaging of material defects is also an important part of nondestructive detection. Single-point scanning is a general method for defect testing, which can effectively scan the detailed information of each part of the material. However, for larger size samples, point-by-point scanning is slightly time-consuming.

[0004] In summary, the existing terahertz nondestructive testing technology has problems such as unknown material refractive index parameters, time-domain pulse width too large to cause echo signal aliasing, low defect detection efficiency of large-size samples, high cost, and the like, which hinder the further development and breakthrough of terahertz nondestructive testing and spectral imaging technology. Therefore, it is very important to study a technology that can realize accurate thickness measurement of micrometer-thick samples and defect imaging of large-size samples, and requirements for flexibility and high efficiency of the detection system are also put forward. SUMMARY

[0005] In view of the above problems, the present application provides an artificial intelligence enhanced strong field terahertz spectral detection imaging device and method, which solves the technical problems of low sample defect detection efficiency and high cost in the prior art.

[0006] In one aspect, the present application provides an artificial intelligence enhanced strong field terahertz spectral detection imaging device, which comprises a femtosecond laser, a beam splitter, a chopper, a laser spot beam expander module, a spin terahertz emitter, a transmissive and reflective integrated automatic scanning detection module, a terahertz detection device, and an artificial intelligence enhanced thickness measurement module and an artificial intelligence enhanced defect detection imaging module.

[0007] The laser emitted by the femtosecond laser generates pump light and probe light by the beam splitter, the pump light sequentially passes through the chopper, the laser spot beam expander module, the spin terahertz emitter, and the transmissive and reflective integrated automatic scanning detection module to enter the terahertz detection device; the probe light passes through the transmissive and reflective integrated automatic scanning detection module to enter the terahertz detection device, and the terahertz detection device outputs a terahertz pulse shape.

[0008] The artificial intelligence enhanced thickness measurement module and the artificial intelligence enhanced defect detection imaging module process based on the terahertz pulse shape to obtain the sample thickness or the defect profile.

[0009] Preferably, the laser spot beam expander module comprises a concave lens and a convex lens, and the pump light sequentially passes through the concave lens and the convex lens to complete beam expansion.

[0010] Preferably, the spin terahertz emitter comprises a one-dimensional photonic crystal enhanced nanometer ferromagnetic thin film, and the expanded pump light is injected into the nanometer ferromagnetic thin film to generate a terahertz wave.

[0011] Preferably, the nanometer ferromagnetic thin film is composed of W, CoFeB and Pt layers, and the single-layer thickness of the W, CoFeB and Pt layers is 2 nm.

[0012] The nanometer ferromagnetic thin film is grown on a Si substrate by a magnetron sputtering method, and between the Si substrate and W, there are 40 groups of periodically arranged SiO2 with a thickness of 136 nm and HfO2 with a thickness of 92 nm, forming a one-dimensional photonic crystal with a total thickness of 9 microns.

[0013] Preferably, the transmissive and reflective integrated automatic scanning detection module comprises a placement table, a two-dimensional translation table, a perforated metal plate fixed on the two-dimensional translation table, an off-axis parabolic mirror, a high-density polyethylene plate and a one-dimensional delay line, wherein the placement table is a transmissive detection function mirror placement table or a reflective detection function sample placement table.

[0014] The terahertz wave generated by the terahertz emitter sequentially passes through the placement table, the perforated metal plate, the off-axis parabolic mirror and the high-density polyethylene plate to complete the transmission or reflection of the measured sample; and the detection light passes through the one-dimensional delay line for delay.

[0015] Preferably, the terahertz detection device comprises, in sequence, a detection light focusing lens, ITO glass, an electro-optic crystal zinc telluride, a detection light collimating lens, a 1 / 4 wave plate, a Wollaston prism and a balanced detector.

[0016] The detection light output by the one-dimensional delay line passes through the detection light focusing lens, is incident to the ITO glass together with the pump light output by the high-density polyethylene plate, and then passes through the electro-optic crystal zinc telluride, the detection light collimating lens, the 1 / 4 wave plate, the Wollaston prism and the balanced detector to finally output a terahertz waveform signal.

[0017] In one aspect, the present application provides a detection method of an artificial intelligence enhanced strong-field terahertz spectroscopy detection imaging device, comprising the following steps:

[0018] In the thickness detection, the reflective detection is used, and if the front surface reflection wave and the back surface reflection wave of the terahertz waveform signal are mixed, the terahertz waveform signal is preprocessed and input into the artificial intelligence enhanced thickness measurement module, and a trained thickness measurement neural network model is used to output a predicted thickness result.

[0019] In the defect detection of a large-size sample, the reflective detection is used, the measured sample reflects a terahertz echo signal, and the existence of a defect is judged according to the terahertz waveform signal.

[0020] In the defect profile extraction imaging, the transmissive detection is used, the terahertz waveform is input into the artificial intelligence enhanced defect detection imaging module, and a defect profile of the sample is output.

[0021] Preferably, in the thickness detection, the following steps are further included:

[0022] If the front surface reflection wave and the back surface reflection wave of the terahertz waveform signal are not mixed, the thickness of the sample is calculated based on the distance between the reflection waves and the refractive index, and the calculation expression of the refractive index is:

[0023] (1)

[0024] (2)

[0025] (3)

[0026] wherein, 、 、 Ei, E2 and E3 represent the electric field intensity of the front surface reflected wave, the back surface reflected wave and the back surface twice reflected wave respectively, is the incident angle of the sample surface, represents the ratio of the three electric field intensities, represents the field ratio enhancement factor, represents the refractive index.

[0027] Preferably, the training completed thickness measurement neural network is obtained in the following manner:

[0028] A terahertz signal dataset is established based on the measured terahertz echo signal and the simulated terahertz echo signal obtained by applying Snell's law and Fresnel equation;

[0029] The thickness measurement neural network model is trained using the terahertz signal dataset, Gaussian noise is added to the data in the dataset during the training process, and a periodic learning rate is used, and finally a training completed thickness measurement neural network is obtained.

[0030] Compared with the prior art, the present application has at least the following beneficial effects:

[0031] (1) The non-destructive testing function of the present application is comprehensive and flexible, and the optical path design of the device has both transmission and reflection detection functions, which can simultaneously meet the detection needs of various materials such as metal surface defect detection, non-conductive material internal structure detection imaging, etc., and has a wide range of application scenarios. The present technology can also realize the determination of sample parameters such as refractive index calculation and coating thickness measurement, has scalability for different types of materials, and the detection function can be switched according to the detection needs;

[0032] (2) The terahertz emission source used in the present application has excellent performance and is easy to prepare. The single photonic crystal enhanced spin terahertz radiation source has the advantages of high emission efficiency, wide frequency band, controllable polarization, stable and reliable, simple structure, etc. It can emit high-intensity, ultra-short pulse width terahertz signals. Compared with traditional methods, it can realize direct thickness measurement of thinner coating samples and higher signal-to-noise ratio and higher resolution defect detection imaging. The terahertz emission source can be realized using various commonly used devices and relatively mature magnetron sputtering devices, without the need for complex and high-cost manufacturing processes of various super materials, overcoming the shortcomings of complex structure and high cost of various terahertz emission sources;

[0033] (3) The terahertz emission source used in this invention is larger in size and can emit terahertz radiation with a beam diameter of 4 inches. Compared with the point-by-point scanning of the sample by a small terahertz spot, this invention can cover a circular area with a diameter of 4 inches in a single scan, which improves the scanning efficiency per unit area and realizes rapid detection of the presence of defects in decimeter-level samples.

[0034] (4) The present invention can use the reflective detection function to calculate the refractive index of a single-layer unknown material in the terahertz band by using the front surface reflected wave, the back surface reflected wave and the back surface secondary reflected wave of the sample under small angle incident conditions; this breaks through the normal incident condition of the existing calculation method and improves the scalability of the present invention for the detection of different materials.

[0035] (5) Based on Snell's law and Fresnel's equation, this invention establishes a dataset and builds a network model based on the Conformer architecture, which realizes the thickness prediction of the superposition signal of ultrathin samples of 1 to 1000 micrometers. The accuracy reaches 99.8% in the ±8 micrometer range. Compared with the existing thickness measurement of samples of 20 micrometers based on the convolutional neural network model, it has better effect and higher accuracy.

[0036] (6) The present invention has low cost, the optical path design does not require the use of parabolic mirror combination, and the characteristics of spectroscopic imaging mean that the detection does not require expensive terahertz cameras and other equipment. Compared with traditional terahertz imaging schemes, it reduces the cost of device construction, use and maintenance. Attached Figure Description

[0037] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.

[0038] Figure 1 A schematic diagram of the artificial intelligence-enhanced strong-field terahertz spectroscopy detection and imaging device provided by the present invention.

[0039] Figure 2 The diagram illustrates the transmission detection function and the reflection detection function provided by this invention.

[0040] Figure 3 A schematic diagram of the artificial intelligence-enhanced thickness measurement neural network model provided by this invention.

[0041] Figure reference numerals: 1-Femtosecond laser, 2-Beam splitter, 3-Chopper, 4-Concave lens, 5-Convex lens, 6-Spin terahertz emitter, 7-Stage, 8-Perforated metal plate, 9-Off-axis parabolic mirror, 10-High-density polyethylene plate, 11-One-dimensional delay line, 12-Probe light focusing lens, 13-ITO glass (indium tin oxide glass), 14-Electro-optic crystal zinc telluride, 15-Probe light collimating lens, 16-1 / 4 wave plate, 17-Wollaston prism, 18-Balance detector. DETAILED DESCRIPTION

[0042] In order to enable a more clear understanding of the above-mentioned objects, features and advantages of the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict. In addition, the present application can also be implemented in other ways different from those described herein, and therefore the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0043] The present application is based on a spin electron terahertz emitter and a transmission-reflection integrated automatic scanning detection device, which ingeniously fuses a terahertz radiation source with ultra-wideband, large size and high signal-to-noise ratio, a spectral detection imaging device and a deep learning neural network technology, to realize thickness measurement of micron-level samples and efficient defect detection and profile extraction imaging of large-size targets. The present application is low in cost, simple to build, firm and reliable in performance, has scalability for different types of materials, and the detection function can be switched according to the detection requirements, which is high in efficiency and strong in flexibility.

[0044] In order to illustrate the effectiveness of the method of the present application, the above technical solutions of the present application will be described in detail below through a specific embodiment, as shown in Figure 1 A kind of artificial intelligence enhanced strong field terahertz spectral detection imaging device is disclosed, including: femtosecond laser 1, beam splitter 2, chopper 3, laser spot beam expander module, spin terahertz emitter 6, transmission-reflection integrated automatic scanning detection module, terahertz detection device and artificial intelligence enhanced thickness measurement module and artificial intelligence enhanced defect detection imaging module.

[0045] The laser emitted by femtosecond laser 1 generates pump light and probe light by beam splitter 2, the pump light passes through chopper 3, laser spot beam expander module, spin terahertz emitter 6 and transmission-reflection integrated automatic scanning detection module in turn and enters terahertz detection device; the probe light passes through transmission-reflection integrated automatic scanning detection module and enters terahertz detection device, and terahertz detection device outputs terahertz pulse shape;

[0046] The artificial intelligence enhanced thickness measurement module and the artificial intelligence enhanced defect detection imaging module perform thickness measurement, defect detection and profile extraction based on the terahertz pulse shape.

[0047] The laser spot beam expander module of the present application includes concave lens 4 and convex lens 5, and the pump light passes through concave lens 4 and convex lens 5 in turn to complete beam expansion.

[0048] The spin terahertz emitter 6 of the present application includes a one-dimensional photonic crystal enhanced nanometer ferromagnetic film, and an external magnetic field is provided, and the expanded pump light is injected into the one-dimensional photonic crystal enhanced nanometer ferromagnetic film to generate terahertz waves.

[0049] In some embodiments, the nanometer ferromagnetic film of the present application is a heterojunction film with a multilayer film structure composed of a W / CoFeB / Pt layer, wherein the thickness of the W, CoFeB, and Pt single layers is 2 nm, and the total thickness of the nanometer ferromagnetic film is 6 nm. The nanometer ferromagnetic film is grown on a Si substrate by a magnetron sputtering method, and the thickness of the Si substrate is 136 nm. The spin nanometer film can emit broadband terahertz waves with an effective bandwidth greater than 5 THz under laser pumping, serving as a new technology of terahertz emission source, which can significantly improve the performance of the system in measuring the thickness of micron-level ultrathin samples compared with traditional emission sources.

[0050] In some embodiments, between the Si substrate and the heavy metal W, 40 groups of periodically arranged SiO2 with a thickness of 136 nm and HfO2 with a thickness of 92 nm form a multilayer film one-dimensional photonic crystal structure with a total thickness of about 9 μm, which can improve the utilization rate of the pump laser energy by the sample, realize terahertz wave coherent enhancement, increase the field strength of terahertz radiation, and improve the signal-to-noise ratio of the radar system, thereby improving the defect detection performance and the quality of profile extraction imaging.

[0051] In some embodiments, the W / CoFeB / Pt heterojunction film as the emission source has a diameter of 4 inches, and can directly generate large-size terahertz wave radiation with a diameter of 4 inches under laser pumping with a diameter of 4 inches, without the need to use an off-axis parabolic mirror to combine and expand the terahertz wave, thereby reducing the complexity of the system and making it easier to integrate while achieving efficient detection of defects in large-size samples.

[0052] The transmissive and reflective integrated automatic scanning detection module comprises a placing table 7, a two-dimensional translation table, a metal plate with holes 8 carried on the two-dimensional translation table, an off-axis parabolic mirror 9, a high-density polyethylene plate 10, and a one-dimensional delay line 11.

[0053] The terahertz wave generated by the terahertz emitter 6 passes through the placing table 7, the metal plate with holes 8, the off-axis parabolic mirror 9, and the high-density polyethylene plate 10 in sequence to complete the transmission or reflection of the measured sample. The probe light generated by the beam splitter 2 is delayed by the one-dimensional delay line 11. The placing table 7 is a transmissive detection function mirror placing table or a reflective detection function measured sample placing table.

[0054] Specifically, as shown in Figure 2 Fig. 1, the transmissive detection function or the reflective detection function is selected according to the detection requirements. The reflective detection function is used when measuring the thickness of micron-level samples, and the one-dimensional delay line moves under program control to reflect the terahertz echo signal of the measured sample. The reflective detection function is used when detecting defects in large-size samples, and the one-dimensional delay line moves under program control to reflect the terahertz echo signal of the measured sample.

[0055] In some embodiments, the one-dimensional delay line and two-dimensional translation stage are cooperatively moved under program control to achieve automated scanning without manual operation under various imaging functions through the control of a connected computer.

[0056] The terahertz detection device includes a detection light focusing lens 12, ITO glass 13 (indium tin oxide glass), an electro-optic crystal zinc telluride 14, a detection light collimating lens 15, a 1 / 4 wave plate 16, a Wollaston prism 17, and a balanced detector 18 connected in sequence. The detection light output by the one-dimensional delay line 11 passes through the detection light focusing lens 12, is incident on the ITO glass 13 together with the pump light output by the high-density polyethylene plate 10, and then passes through the electro-optic crystal zinc telluride 14, the detection light collimating lens 15, the 1 / 4 wave plate 16, the Wollaston prism 17, and the balanced detector 18 to finally output a terahertz waveform signal.

[0057] The artificial intelligence enhanced thickness measurement module and the artificial intelligence enhanced defect detection imaging module can process the terahertz waveform signal based on the trained thickness measurement neural network model to obtain the sample thickness, and can obtain the defect profile based on the contour extraction neural network model.

[0058] In some embodiments, the artificial intelligence enhanced thickness measurement module and the artificial intelligence enhanced defect detection imaging module are computers having a processor, a memory, and an input and output device, the memory records the trained thickness measurement neural network model and the contour extraction neural network model, the computer reads the terahertz waveform signal through the input and output device, processes the terahertz waveform signal using the neural network model through the processor, and obtains the sample thickness or the defect profile.

[0059] The application also provides a detection method of the artificial intelligence enhanced strong-field terahertz spectroscopy detection imaging device, which includes the following steps.

[0060] In the micron-level sample thickness measurement, the reflection detection function is used, the terahertz waveform signal is preprocessed and input into the artificial intelligence enhanced thickness measurement module, and the trained thickness measurement neural network model is used to output a predicted thickness result; in the large-size sample defect detection, the reflection detection is used, the sample reflection terahertz echo signal is detected, and the terahertz detection device outputs a time-domain waveform, and the existence of the defect can be determined according to the waveform; in the defect profile extraction imaging, the transmission detection is used, the terahertz waveform is input into the artificial intelligence enhanced defect detection imaging module, and the defect profile of the sample is output.

[0061] In some embodiments, the present application is based on the measured terahertz echo signal, and the Snell's law and the Fresnel equation are applied to simulate the terahertz echo signal to establish a terahertz signal dataset for training and testing. The R-square value is selected to verify the similarity of the simulated signal and the actual signal, and the availability of the simulated signal as the dataset is confirmed. This method not only ensures the correct and solid theoretical basis, but also reduces the workload and cost of establishing the dataset by collecting a large number of repetitive experimental terahertz measured signals.

[0062] In some embodiments, the thickness measurement neural network model of the present application is as shown in Figure 3 The input of the thickness measurement neural network model can be a terahertz waveform signal belonging to a plurality of thickness categories. The thickness measurement neural network model includes two groups of convolution layers, activation functions, and maximum pooling layers connected in sequence. The maximum pooling layers of the second group are respectively coupled to a feature coupling unit through one input, and the other input is coupled to the feature coupling unit through position encoding and a Transformer layer. The feature coupling unit includes a linear layer, and the result of the linear layer is processed through multi-layer perception fusion, residual connection, and normalization. The final predicted thickness result is output through a fully connected layer.

[0063] In some embodiments, the terahertz signal dataset is used to train the thickness measurement neural network model. The neural network training program adds Gaussian noise to improve the generalization ability of the network, and uses a periodic learning rate to avoid training from falling into a local optimum, but to find a global optimum or a better solution, thereby improving the final performance. The trained thickness measurement neural network model uses residual connection and layer normalization, which helps to improve the training efficiency and stability of the model, and improves the generalization ability of the model.

[0064] The following describes various devices on the optical path of the present application, as shown in Figure 1 The artificial intelligence enhanced strong field terahertz spectroscopy detection imaging device includes a femtosecond laser 1, a beam splitter 2, a chopper 3, a concave lens 4, a convex lens 5, a spin terahertz emitter 6, a placement table 7, a perforated metal plate 8 mounted on a two-dimensional translation table, an off-axis parabolic mirror 9, a high-density polyethylene plate 10, a one-dimensional delay line 11, a probe light focusing lens 12, ITO glass 13 (indium tin oxide glass), an electro-optic crystal zinc telluride 14, a probe light collimating lens 15, a 1 / 4 wave plate 16, a Wollaston prism 17, and a balanced detector 18. The two-dimensional translation table is an electrically controlled translation table, which can move in two dimensions under the control of a computer program. The perforated metal plate 8 mounted on the translation table moves with it. The one-dimensional delay line 11 is an electrically controlled translation table, which can move in a single axis under the control of a computer program.

[0065] The transmissive and reflective integrated automatic scanning detection device is as shown in Figure 2As shown, two subgraphs respectively describe the system of transmission detection and reflection detection two functions. Transmission detection will be detected sample placed in the light path, terahertz by the transmitter 6 normal incidence on the sample, and using the mirror placed in the table 7 terahertz to the receiver (terahertz receiving module is Figure 1 Labels 14-18). Reflection detection uses the sample to reflect the terahertz directly, the terahertz is emitted by the emitter 6 and enters the sample placed in the table 7 at 15°, and is reflected by the sample or substrate to the receiver at 15°. Both detection methods can be used in the light path with a two-dimensional movement of the metal plate 8 with holes to achieve single point scanning. Considering that different samples need different detection methods: non-conductive composite materials and low density materials are suitable for transmission detection; and metal substrate coating materials are suitable for reflection detection.

[0066] The diameter of 0.5 inch of femtosecond laser 1 is emitted from one side into the system, and is divided into two paths by a beam splitter 2, wherein 90% of the energy of the femtosecond laser enters the pump path, and the remaining 10% of the energy of the femtosecond laser enters the detection path.

[0067] The chopper 3 connected with and synchronized with the laser pulse is placed in the pump path, which works at 500 Hz to suppress noise and improve the signal-to-noise ratio of the system. Then the pump laser passes through the beam expander system composed of a concave lens 4 with a focal length of-50 mm and a convex lens 5 with a focal length of 500 mm, and is irradiated on the 4-inch spin terahertz emitter 6 to generate a large-size ultra-wideband strong-field terahertz wave with a diameter of 4 inches. The terahertz wave enters the transmission and reflection integrated automatic scanning detection device, and the signal emitted after detection by the transmission / reflection detection method is collected and focused on the electro-optic crystal zinc telluride 14 by the off-axis parabolic mirror 9 with a focal length of 203.2 mm. The function of the high-density polyethylene plate 10 is to block the propagation of the laser while transmitting the terahertz wave.

[0068] The probe laser passes through a one-dimensional delay line 11 and several mirrors, and is focused by a probe light focusing lens 12 with a focal length of 150 mm to coincide with the terahertz wave in the electro-optic crystal zinc telluride 14. Based on the electro-optic effect, the terahertz electric field changes the refractive index of the electro-optic crystal zinc telluride 14, thereby changing the polarization state of the probe laser, and the probe light collimation lens 15 adjusts the beam state again. A 1 / 4 wave plate 16 and a Wollaston prism 17 are used to adjust the two orthogonal polarization components of the probe laser, and a balanced detector 18 is used to collect them respectively, which is processed by a lock-in amplifier and plotted in the self-programming software to draw the terahertz waveform signal. The computer controls the movement of the delay line to change the optical path of the probe laser, which is used to map the terahertz electric field strength at each time, and draw the shape of the terahertz pulse.

[0069] In the thickness measurement of the micron-level sample, the reflection detection is used, the obtained signal contains the front surface reflection wave, the back surface reflection wave and the back surface secondary reflection wave, the echo pulse electric field intensity is respectively , , , let be the incident angle of the sample surface, the unknown material refractive index in the terahertz band can be calculated by the following formula .

[0070]

[0071]

[0072]

[0073] wherein, , , the front surface reflection wave, the back surface reflection wave and the back surface secondary reflection wave are respectively represented by the electric field intensity, the ratio of the three electric field intensities is represented by, the field ratio enhancement factor is represented by, the refractive index is represented by.

[0074] For the signals of the front surface reflection wave and the back surface reflection wave without aliasing, the sample thickness can be directly calculated from the distance between the two reflection waves and the refractive index; for the signals of the front surface reflection wave and the back surface reflection wave with aliasing, the signals can be input into the trained thickness measurement neural network model, and the output value is proportionally operated with the refractive index to obtain the sample thickness.

[0075] In the defect detection and profile extraction imaging of the large-size target, the transmission detection or the reflection detection is used, the signal pulse number is directly observed to judge whether there is a defect, the two-dimensional scanning is carried out by using the metal plate 8 with holes mounted on the two-dimensional translation table, the terahertz signals of each scanning point position are obtained to carry out peak value imaging, and the imaging results are sequentially input into the spectral imaging and optimization processing program and the mature profile extraction neural network model to obtain the defect profile extraction result.

[0076] In the present application, unless otherwise clearly specified and limited, the terms such as "mounting", "connection", "connection", "fixing" and the like should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or it can be integrated; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication or interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0077] In the present application, unless otherwise explicitly specified and limited, "on" or "under" of a first feature to a second feature can include that the first and second features are in direct contact, or that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, "on", "above" and "over" of a first feature to a second feature includes that the first feature is directly above and obliquely above the second feature, or only indicates that the first feature is horizontally higher than the second feature. "Under", "below" and "underneath" of a first feature to a second feature includes that the first feature is directly below and obliquely below the second feature, or only indicates that the first feature is horizontally lower than the second feature.

[0078] In the present application, the terms "first", "second", "third", "fourth" are only for descriptive purpose, and should not be understood as indicating or implying relative importance. The term "a plurality of" means two or more, unless otherwise explicitly limited.

[0079] The above description is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this. Any changes or replacements within the technical scope disclosed by the present application, which can be easily thought by those skilled in the art, should be covered within the protection scope of the present application.

Claims

1. An artificial intelligence-enhanced strong-field terahertz spectroscopy detection and imaging device, characterized in that, include: Femtosecond laser (1), beam splitter (2), chopper (3), laser spot beam expander module, spin terahertz emitter (6), transmission and reflection integrated automated scanning and detection module, terahertz detection device, and artificial intelligence enhanced thickness measurement module and artificial intelligence enhanced defect detection imaging module. The laser emitted by the femtosecond laser (1) generates pump light and probe light by the beam splitter (2). The pump light passes through the chopper (3), the laser beam expander module, the spin terahertz emitter (6), and the transmission-reflection integrated automatic scanning and detection module in sequence before entering the terahertz detection device. The probe light enters the terahertz detection device through the transmission-reflection integrated automatic scanning and detection module. The terahertz detection device outputs the shape of the terahertz pulse. The AI-enhanced thickness measurement module and the AI-enhanced defect detection imaging module process terahertz pulse shapes to obtain sample thickness or defect contours. The integrated transmission and reflection automated scanning and detection module includes a placement stage (7), a two-dimensional translation stage, a perforated metal plate (8) fixed on the two-dimensional translation stage, an off-axis parabolic mirror (9), a high-density polyethylene plate (10), and a one-dimensional delay line (11). The placement stage (7) is a transmission detection function reflector placement stage or a reflection detection function sample placement stage. The terahertz wave generated by the spin terahertz transmitter (6) passes sequentially through the placement stage (7), the perforated metal plate (8), the off-axis parabolic mirror (9), and the high-density polyethylene plate (10) to complete the transmission or reflection of the sample under test; the probe light is delayed by a one-dimensional delay line (11). Transmission detection places the sample to be detected in the optical path. Terahertz is incident on the sample from the transmitter (6) and reflected to the receiver by the plane mirror at the stage (7). Reflection detection uses the sample to directly reflect the terahertz. The terahertz is emitted from the transmitter (6) and incident at 15° into the sample at the stage (7). It is reflected to the receiver by the sample or substrate at 15°. Both detection methods use a perforated metal plate (8) to move in two dimensions in the optical path to achieve single-point scanning.

2. The artificial intelligence-enhanced strong-field terahertz spectroscopy detection and imaging device according to claim 1, characterized in that, The laser beam expansion module includes a concave lens (4) and a convex lens (5). The pump light passes through the concave lens (4) and the convex lens (5) in sequence to complete the beam expansion.

3. The artificial intelligence-enhanced strong-field terahertz spectroscopy detection and imaging device according to claim 2, characterized in that, The spin terahertz emitter (6) includes a one-dimensional photonic crystal-enhanced nano-ferromagnetic film, and the expanded pump light is injected into the nano-ferromagnetic film to generate terahertz waves.

4. The artificial intelligence-enhanced strong-field terahertz spectroscopy detection and imaging device according to claim 3, characterized in that, The nano-ferromagnetic thin film is composed of W, CoFeB and Pt layers, and the thickness of each W, CoFeB and Pt layer is 2 nm. Nanoscale ferromagnetic thin films are grown on Si substrates by magnetron sputtering. Between the Si substrate and W, there are 40 sets of periodically alternating SiO2 with a thickness of 136 nm and HfO2 with a thickness of 92 nm, forming a one-dimensional photonic crystal with a total thickness of 9 μm.

5. The artificial intelligence-enhanced strong-field terahertz spectroscopy detection and imaging device according to claim 4, characterized in that, The terahertz detection device includes a probe light focusing lens (12), ITO glass (13), electro-optic crystal zinc telluride (14), probe light collimating lens (15), quarter wave plate (16), Wollaston prism (17), and balanced detector (18) connected in sequence. The probe light output from the one-dimensional delay line (11) passes through the probe light focusing lens (12) and is incident on the ITO glass (13) together with the pump light output from the high-density polyethylene plate (10). Then, it passes through the electro-optic crystal zinc telluride (14), the probe light collimating lens (15), the quarter wave plate (16), the Wollaston prism (17), and the balanced detector (18) to finally output a terahertz waveform signal.

6. A detection method for an artificial intelligence-enhanced strong-field terahertz spectroscopy detection imaging device according to any one of claims 1-5, characterized in that, Includes the following steps: When performing thickness detection, a reflective detection method is used. If the front and rear surface reflected waves of the terahertz waveform signal are superimposed, the terahertz waveform signal is preprocessed and input into the AI-enhanced thickness measurement module. The trained thickness measurement neural network model is used to output the predicted thickness result. When performing defect detection on large-size samples, a reflective detection method is used. The sample under test reflects a terahertz echo signal, and the presence of defects is determined based on the terahertz waveform signal. During defect contour extraction imaging, a transmission-type detector is used to input the terahertz waveform signal into the AI-enhanced defect detection imaging module, which then outputs the defect contour of the sample.

7. The detection method according to claim 6, characterized in that, The thickness measurement process also includes the following steps: If the front and rear surface reflected waves of the terahertz waveform signal do not overlap, the sample thickness is calculated based on the distance between the reflected waves and the refractive index. The expression for calculating the refractive index is as follows: in, , , Let represent the electric field intensities of the wave reflected from the front surface, the wave reflected from the rear surface, and the secondary reflection from the rear surface, respectively. The incident angle at the sample surface. It represents the ratio of the three electric field intensities. Indicates the field ratio enhancement factor. It represents the refractive index.

8. The detection method according to claim 7, characterized in that, The trained thickness measurement neural network model is obtained as follows: A terahertz signal dataset was established based on measured terahertz echo signals and simulated terahertz echo signals obtained by applying Snell's law and Fresnel's equations. The thickness measurement neural network model was trained using the terahertz signal dataset. Gaussian noise was added to the data in the terahertz signal dataset during the training process, and a periodic learning rate was used to finally obtain the trained thickness measurement neural network model.

Images