Terahertz wave-based nondestructive testing device for defects of lower substrate of copper bar coating

Abstract

The invention relates to the technical field of nondestructive testing, in particular to a copper bar coating lower substrate defect nondestructive testing device based on terahertz waves, which comprises a terahertz scanning acquisition module used for controlling a transmitting and receiving unit to perform two-dimensional scanning so as to acquire an original waveform; the flight time dynamic correction module is used for resolving the local coating thickness and carrying out dynamic time axis alignment; the substrate roughness deconvolution module is used for calling the reference model to carry out deconvolution filtering so as to strip background scattering noise; the interface dielectric feature analysis module is used for extracting a dielectric feature vector reflecting an interface dielectric constant mutation state; the defect risk assessment imaging module is used for constructing a three-dimensional chromatography map and outputting a detection result; according to the method, the physical characteristic decoupling of the signal source is realized, the coating thickness fluctuation and the real defect are effectively distinguished, the misjudgment rate is obviously reduced, and the evaluation accuracy is ensured.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, specifically to a nondestructive testing device for defects in a copper-plated substrate based on terahertz waves. Background Technology

[0002] In the application scenario of electrical connection reliability assessment of power equipment, copper conductive components rely on high-quality surface plating process to ensure contact resistance stability and corrosion resistance. The detection system usually needs to combine terahertz wave detection data and signal processing algorithm to perceive the bonding state of the plating and substrate interface in real time.

[0003] For the detection of hidden defects under coatings, existing solutions generally adopt a direct time-domain waveform analysis architecture. This involves acquiring reflected echoes using a terahertz time-domain spectroscopy system, extracting signal features using the peak amplitude threshold method or a simple pulse echo localization method, and directly using changes in echo intensity as a criterion to assess the presence of delamination or foreign matter at the interface. While this approach is feasible under conditions of uniform coating thickness and ideally smooth substrate surfaces, its over-reliance on the absolute amplitude of the original signal and lack of decoupling mechanisms for physical feature dimensions makes it highly susceptible to interference from coating thickness fluctuations and substrate roughness common in actual industrial production. A single waveform analysis algorithm can easily misinterpret geometric thickness differences. Phase jitter caused by roughness or background scattering caused by roughness can be misidentified as defect signals, leading to erroneous detection conclusions. In addition, purely time-domain driven analysis methods are unable to resolve weak spectral fingerprints reflecting abrupt changes in the dielectric properties of the interface, and are unable to eliminate system background noise that does not conform to the layered reflection law. This makes the detection process prone to masking effects or false alarms, making it difficult to support high-sensitivity and high-confidence quantitative assessment of the interface quality of copper products. Therefore, how to establish a detection mechanism with physical feature decoupling capabilities, effectively eliminating thickness noise and roughness interference while improving the robustness and accuracy of identifying small air gaps and oxidation defects, has become an urgent technical problem to be solved. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides a non-destructive testing device for substrate defects under copper plating based on terahertz waves. Specifically, the technical solution of the present invention includes:

[0005] The terahertz scanning acquisition module is used to control the terahertz time-domain spectral emission and receiving unit to perform two-dimensional point-by-point scanning of the copper bar surface under test, and to acquire the reflected echo signals of each scanning point to obtain the original terahertz time-domain waveform sequence.

[0006] The time-of-flight dynamic correction module is used to analyze the first echo peak of the air-coating interface and the second echo peak of the coating-substrate interface in the original terahertz time-domain waveform sequence. Based on the time difference between the first echo peak and the second echo peak, the local coating thickness of each scanning point is calculated, and the local coating thickness is used to dynamically align the original terahertz time-domain waveform sequence with the time axis to obtain a thickness-normalized waveform sequence.

[0007] The substrate roughness deconvolution module is used to call a preset defect-free surface scattering reference model and use the defect-free surface scattering reference model to perform deconvolution filtering on the thickness normalized waveform sequence to remove the background scattering noise component caused by substrate roughness, thereby obtaining the interface feature enhancement signal.

[0008] The interface dielectric feature analysis module is used to perform time-domain phase polarity analysis and frequency-domain spectral attenuation analysis on the interface feature enhancement signal to extract dielectric feature vectors that reflect the abrupt change in the interface dielectric constant.

[0009] The defect risk assessment imaging module is used to construct a three-dimensional tomographic map reflecting the integrity of the interface bonding based on the dielectric feature vector and the local coating thickness, and output the defect detection results under the copper plating.

[0010] Preferably, the terahertz scanning acquisition module includes:

[0011] A femtosecond laser pumping unit is used to generate femtosecond laser pulses to excite a terahertz emission source to generate broadband terahertz waves.

[0012] A reflective confocal optical path unit is used to focus the broadband terahertz wave onto the interface between the coating and the substrate of the copper bar under test through a preset polyethylene lens, and to collect the reflected echo.

[0013] A two-dimensional displacement control unit is used to drive the scanning mechanism to move the reflective confocal optical path unit or the copper bar surface to be measured relative to each other in order to achieve full coverage scanning.

[0014] Preferably, the time-of-flight dynamic correction module is used for:

[0015] The two peaks with the largest absolute amplitude in the original terahertz time-domain waveform sequence are identified as the first echo peak and the second echo peak.

[0016] Calculate the time-of-flight delay between the first echo peak and the second echo peak;

[0017] By combining the preset coating refractive index, the flight time delay is converted into the local coating thickness;

[0018] Using the time point of the second echo peak as a reference, the original terahertz time-domain waveform sequence is shifted on the time axis to eliminate phase jitter caused by the local coating thickness difference.

[0019] Preferably, the substrate roughness deconvolution module is used for:

[0020] The Wiener filtering algorithm or wavelet transform algorithm is used to take the thickness-normalized waveform sequence as the input signal and the good surface scattering reference model as the system response function.

[0021] In the frequency domain, a deconvolution operation is performed to subtract the scattering component related to the system response function from the input signal, so as to retain the reflection component caused by the change in the dielectric properties of the interface as the interface feature enhancement signal.

[0022] Preferably, the interface dielectric feature analysis module is used for:

[0023] Detect the phase polarity of the interface feature enhancement signal at the second echo peak position;

[0024] If the phase polarity is a negative polarity reversal, a first numerical identifier is generated as a first dielectric characteristic component to characterize good metal contact.

[0025] If the phase polarity is positive or without reversal, a second numerical identifier is generated as a second dielectric characteristic component to characterize the presence of an air gap or delamination.

[0026] The first dielectric feature component or the second dielectric feature component is encoded into the dielectric feature vector.

[0027] Preferably, the interface dielectric feature analysis module is further configured to:

[0028] The interface feature enhancement signal is subjected to a fast Fourier transform to obtain a frequency domain spectrum.

[0029] Calculate the amplitude attenuation rate of the frequency domain spectrum in a specific oxide absorption frequency range;

[0030] The amplitude attenuation rate is compared with a preset oxidation threshold;

[0031] If the amplitude attenuation rate is greater than the oxidation threshold, a third dielectric characteristic component representing the presence of oxide inclusions is generated.

[0032] If the amplitude attenuation rate is less than or equal to the oxidation threshold, then a normal characteristic component characterizing the absence of oxidation inclusions is generated.

[0033] The generated feature components are appended to the dielectric feature vector.

[0034] Preferably, the defect risk assessment imaging module is used for:

[0035] Generate the three-dimensional tomographic map containing the first layer and the second layer;

[0036] The first layer is a coating uniformity distribution map generated based on the local coating thickness;

[0037] The second layer is an interface-combined quality heatmap generated based on the dielectric feature vector, where the highlighted areas correspond to the coordinates of the abnormal values ​​displayed by the dielectric feature vector.

[0038] Preferred options also include:

[0039] The process closed-loop feedback module is used to calculate the thickness variance in the coating uniformity distribution diagram and the defect area ratio in the interface bonding quality thermogram.

[0040] If the thickness variance is greater than the upper limit of the preset tolerance range, or the defect area ratio is greater than the preset safety threshold, an electroplating process parameter adjustment instruction is generated and sent to the production line control system.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. This system effectively solves the problem of misjudgment caused by phase jitter due to coating thickness fluctuations in traditional time-domain analysis by constructing a dynamic time-of-flight correction mechanism. Unlike existing technologies that directly rely on the original waveform amplitude, this solution analyzes the echo peak time difference between the air-coating interface and the coating-substrate interface, and uses a frequency-domain phase rotation algorithm to dynamically align the original waveform sequence on the time axis. This processing method logically constructs a virtual plane of uniform thickness, eliminating signal misalignment caused by differences in geometric thickness. This allows the detection device to focus on the physical property analysis of the interface itself when dealing with products with large surface process tolerances, significantly improving its adaptability to the detection of non-flat surfaces.

[0043] 2. This system introduces a substrate roughness deconvolution strategy based on a reference model to overcome background scattering noise interference caused by rough substrates in industrial environments. By calling a defect-free surface scattering reference model optimized by a time-domain window function and performing deconvolution operations using Wiener filtering or wavelet transform algorithms with energy fingerprint protection, this method can accurately remove random scattering components caused by substrate roughness while completely preserving the deep interface reflection characteristics caused by changes in dielectric properties. This physical mechanism-based signal purification method effectively solves the problem of weak signal extraction under strong scattering backgrounds and significantly improves the detection sensitivity of the device for small layered defects in low signal-to-noise ratio environments.

[0044] 3. This system establishes a time-frequency joint interface dielectric feature analysis system, realizing accurate classification and quantification of air gap delamination and oxide inclusions; by using Fresnel reflection principle through time-domain phase polarity analysis, it distinguishes between metal contacts and air gaps, and by combining frequency-domain spectral attenuation analysis to identify the absorption fingerprint of specific oxides. This scheme makes up for the shortcoming of single-dimensional analysis in being unable to identify defect types; in particular, the introduced thickness attenuation decoupling mechanism uses local thickness data to compensate and correct the spectral amplitude, eliminating the absorption loss interference caused by thickness differences, thereby ensuring extremely high confidence in the assessment of hidden oxide defects.

[0045] 4. This system constructs a nonlinear risk assessment imaging and process closed-loop feedback system, which improves the intuitiveness of the detection results and the level of intelligent production control. By using a risk quantification function based on probability union, this method avoids the masking effect that may occur when linear weighted models fuse multidimensional features, ensuring that any risk factor, such as physical layering or chemical oxidation, can be highlighted. At the same time, the system can automatically generate adjustment instructions for electroplating current or acid concentration based on thickness variance and defect ratio, transforming nondestructive testing data into a basis for real-time correction of production parameters, realizing a leap from passive quality screening to proactive process optimization. Attached Figure Description

[0046] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0047] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0049] Example 1:

[0050] Please see Figure 1 A non-destructive testing device for defects in a copper-plated substrate based on terahertz waves, comprising:

[0051] The terahertz scanning acquisition module is used to control the terahertz time-domain spectral emission and receiving unit to perform two-dimensional point-by-point scanning of the copper bar surface under test, and to acquire the reflected echo signals of each scanning point to obtain the original terahertz time-domain waveform sequence.

[0052] The time-of-flight dynamic correction module is used to analyze the first echo peak of the air-coating interface and the second echo peak of the coating-substrate interface in the original terahertz time-domain waveform sequence. Based on the time difference between the first echo peak and the second echo peak, the local coating thickness at each scanning point is calculated, and the local coating thickness is used to dynamically align the original terahertz time-domain waveform sequence with the time axis to obtain a thickness-normalized waveform sequence.

[0053] The substrate roughness deconvolution module is used to call a preset defect-free surface scattering reference model. The defect-free surface scattering reference model is used to perform deconvolution filtering on the thickness normalized waveform sequence to remove the background scattering noise component caused by substrate roughness, thereby obtaining the interface feature enhancement signal.

[0054] The interface dielectric feature analysis module is used to perform time-domain phase polarity analysis and frequency-domain spectral attenuation analysis on the interface feature enhancement signal in order to extract dielectric feature vectors that reflect the abrupt change in the interface dielectric constant.

[0055] The defect risk assessment imaging module is used to construct a three-dimensional tomographic map reflecting the integrity of the interface bonding based on the dielectric feature vector and the local coating thickness, and output the defect detection results under the copper plating.

[0056] This embodiment constructs a detection system with decoupled physical features. The terahertz scanning acquisition module is configured to control the terahertz time-domain spectrum emission and reception unit, which acquires the original terahertz time-domain waveform sequence. This sequence originates from a two-dimensional point-by-point scanning process and physically represents the electric field intensity change curve with complete time information acquired for each spatial coordinate point. It not only contains the amplitude information of the reflected echo but also implicitly includes phase and spectral fingerprints. The time-of-flight dynamic correction module aims to eliminate geometric thickness noise. It resolves the first echo peak generated when the terahertz wave enters the coating from the air and the second echo peak generated at the interface between the coating and the substrate after the terahertz wave penetrates the coating. Based on this, the module calculates the local coating thickness at each point by solving the time difference between the two echo peaks and uses this as a reference to dynamically translate the waveform to generate a thickness-normalized waveform sequence.

[0057] The substrate roughness deconvolution module calls upon a defect-free surface scattering reference model, which is derived from the pre-acquired terahertz frequency domain response of a good copper bar surface. This module performs deconvolution filtering to subtract the background scattering component caused by roughness from the signal, thereby outputting an interface feature enhancement signal. Subsequently, the interface dielectric feature analysis module performs dual analysis on the enhancement signal, using time-domain phase polarity analysis to distinguish between metal contacts and air gap stratification, and using frequency-domain spectral attenuation analysis to identify oxide inclusions, ultimately outputting a dielectric feature vector that quantitatively describes the physical state of the interface. Based on the above vector and thickness data, the defect risk assessment imaging module constructs a three-dimensional tomographic map, intuitively displaying the location, type, and potential risks of defects.

[0058] This embodiment achieves signal source decoupling through the synergistic effect of the above modules. That is, in complex industrial environments, it effectively distinguishes signal fluctuations caused by changes in coating thickness from signal anomalies caused by actual defects, and further separates roughness scattering from defect reflection. This decoupling design based on physical mechanisms enables the device to maintain an extremely low misjudgment rate when dealing with copper products with large surface process tolerances, ensuring the accuracy of electrical connection reliability assessment.

[0059] Example 2:

[0060] The terahertz scanning acquisition module includes:

[0061] A femtosecond laser pumping unit is used to generate femtosecond laser pulses to excite a terahertz emission source to generate broadband terahertz waves.

[0062] The reflective confocal optical path unit is used to focus broadband terahertz waves onto the interface between the coating and the substrate on the copper bar surface under test through a preset polyethylene lens, and to collect the reflected echo.

[0063] The two-dimensional displacement control unit is used to drive the scanning mechanism to move the reflective confocal optical path unit or the copper bar surface to be measured relative to each other in order to achieve full coverage scanning.

[0064] This embodiment specifies the hardware architecture of the terahertz scanning acquisition module. The femtosecond laser pumping unit uses a titanium-doped sapphire femtosecond laser to generate femtosecond laser pulses with a center wavelength of 800 nm and a pulse width of less than 100 fs. These pulses are used to excite a photoconductive antenna or a nonlinear crystal, thereby generating a broadband terahertz wave covering a frequency range of 0.1 THz to 4.0 THz. This broadband characteristic is designed to meet the requirements for resolving micron-level coating thicknesses. The reflective confocal optical path unit adopts an optical path design where the transmitting and receiving focal points coincide, and a polyethylene lens is selected. This lens material has extremely low absorption in the terahertz band. The absorption coefficient and dispersion characteristics maximize signal energy retention. The lens focal length is set to 50mm to focus the terahertz wave onto the interface between the coating and the substrate on the copper bar surface under test, forming a spot with a diameter of approximately 0.5mm. It should be noted that this spot diameter corresponds to the diffraction limit of high-frequency components in broadband terahertz waves, such as above 2.5THz, thus ensuring lateral resolution for minute defects, while low-frequency components mainly contribute to the penetration depth. The two-dimensional displacement control unit uses a high-precision XY linear module to drive the reflective confocal optical path unit for grating scanning, with a step accuracy set to 0.1mm to ensure full-coverage scanning without blind spots.

[0065] This embodiment employs femtosecond laser pumping combined with a reflective confocal optical path, which significantly improves the temporal resolution and spatial focusing capability of terahertz waves, enabling the device to detect minute air gap defects. The use of polyethylene lenses effectively reduces optical path loss and improves the signal-to-noise ratio of the echo signal. This high signal-to-noise ratio and high resolution hardware foundation provides a physical guarantee for extracting weak interface defect signals from strong background noise.

[0066] Example 3:

[0067] The time-of-flight dynamic correction module is used for:

[0068] Locate the two peaks with the largest absolute amplitude in the original terahertz time-domain waveform sequence as the first echo peak and the second echo peak;

[0069] Calculate the time delay between the first echo peak and the second echo peak;

[0070] By combining the preset coating refractive index, the time-of-flight delay is converted into local coating thickness;

[0071] Using the time point of the second echo peak as a reference, the original terahertz time-domain waveform sequence is shifted on the time axis to eliminate phase jitter caused by local coating thickness differences.

[0072] This embodiment details the specific algorithm logic for dynamic time-of-flight correction. The module extracts the envelope of the original waveform sequence, locates the two peak moments with the largest absolute amplitude, and marks them as the first echo peak moment at the air-coating interface and the second echo peak moment at the coating-substrate interface, respectively. The system calculates the time-of-flight delay between the two peak moments. The formula is as follows: ;

[0073] in, The value is derived from real-time sensor data acquisition and peak-finding algorithm, and its physical meaning is the arrival time of the second echo peak, measured in ps. The data is obtained from real-time sensor data acquisition and peak-finding algorithm. Its physical meaning is the arrival time of the first echo peak, and the unit is ps.

[0074] Calculate the local coating thickness at each scan point using time-of-flight delay. To correct the calculation errors caused by the inconsistency between the units of time (ps), speed (m / s), and length (μm), a unit conversion factor is introduced. The revised formula is as follows: ;

[0075] in, Derived from a physical constant, its physical meaning is the speed of light in a vacuum, and its unit is m / s; : Derived from a preset value, its physical meaning is the refractive index of the coating material, with a unit of 1; : Derived from the requirement for dimensional unification, its physical meaning is the unit conversion factor, with a value of This coefficient is a comprehensive conversion factor, which includes a factor for converting picoseconds (ps) to seconds (s). And the factor that converts the calculation results from meters (m) to micrometers (μm). ;

[0076] To eliminate phase jitter caused by thickness variations and achieve alignment accuracy at the sub-sampling level, the module does not employ simple integer index shifting. Instead, it utilizes the shifting properties of the Fourier transform to perform operations in the frequency domain. The system constructs a phase correction factor based on the second echo peak time. The original waveform's spectrum is phase-rotated and then transformed back to the time domain to obtain the corrected waveform. The calculation formula is: ;

[0077] in, Sourced from all scan points The statistical average value, with the physical meaning of the preset global reference time, is expressed in ps. and : These represent the Fast Fourier Transform and the Inverse Fourier Transform, respectively;

[0078] In this embodiment, the frequency domain shifting algorithm described above is used to force the alignment of the originally uneven time-domain waveforms at the coating-substrate interface. This dynamic correction mechanism avoids waveform distortion caused by time-domain interpolation, eliminates time-domain misalignment caused by uneven coating thickness, and constructs a virtual plane of uniform thickness. This allows subsequent feature extraction algorithms to focus on the physical properties of the interface itself, rather than being disturbed by changes in geometric position.

[0079] Example 4:

[0080] The substrate roughness deconvolution module is used for:

[0081] The Wiener filtering algorithm or wavelet transform algorithm is used, the thickness-normalized waveform sequence is used as the input signal, and the surface scattering reference model of the good product is used as the system response function.

[0082] In the frequency domain, a deconvolution operation is performed to subtract the scattering component related to the system response function from the input signal, so as to retain the reflection component caused by the change in the dielectric properties of the interface as the interface feature enhancement signal.

[0083] This embodiment details the implementation process of substrate roughness deconvolution, aiming to solve the challenge of weak signal extraction under strong scattering backgrounds; the system defines a defect-free surface scattering reference model as the system response function. This model is based on known surface roughness conforming to standards. Furthermore, the frequency domain response was obtained by averaging multiple scans of a gold sample with copper without internal defects. To overcome the technical bias that directly using the full waveform of the gold sample would lead to the implicit reflection at the coating-substrate interface, i.e., the second echo, being mistakenly filtered out as background noise, this embodiment constructs... A time-domain window function truncation step was introduced: the system identifies the first echo peak in the gold sample's time-domain waveform, i.e., the air-coating interface, applies a Hanning window with a width of approximately 20 ps for truncation, and forcibly sets the time period containing the second echo to zero; thus obtaining... It contains only the convolutional features of the incident pulse and surface roughness, but not the substrate interface information, thus ensuring that the subsequent deconvolution operation only strips away surface scattering and system blur, while fully preserving the reflection features of the deep interface.

[0084] The time-domain signal of the thickness-normalized waveform sequence is transformed to the frequency domain, denoted as . ;

[0085] If the Wiener filtering algorithm is used, the system performs Wiener deconvolution operation, as shown in the following formula: ;

[0086] in, Derived from the calculation results, its physical meaning is the frequency domain representation of the estimated interface feature enhancement signal, and the unit is dimensionless; : It originates from the output of the preceding module after FFT transformation, and its physical meaning is the frequency domain representation of the input thickness normalized waveform; Derived from mathematical operations, its physical meaning is the system response function. The complex conjugate of is used for phase matching and correction; The peak value of the power spectrum of the system response function is used for the magnitude balance of the regularization term; Derived from system settings, its physical meaning is a signal-to-noise ratio adjustment constant, used to prevent numerical instability caused by an excessively small denominator; specifically, this constant... The selection is based on the system's average signal-to-noise ratio, and the calculation formula is approximately as follows: In this embodiment, the noise power is calculated by pre-collecting dark noise data of the detector under no-light conditions. The value range is to ;

[0087] As another optional implementation, if a wavelet transform algorithm is used, an energy fingerprint protection strategy is adopted in this embodiment to avoid erroneous deletion of useful signals; the module selects a mother wavelet, such as Symlet4, which has a high similarity to the terahertz pulse waveform, for the defect-free surface scattering reference model. Corresponding time-domain form That is, the surface reflection signal after the above-mentioned truncated processing is subjected to multi-scale discrete wavelet decomposition to calculate the level of each scale. Energy percentage: ;

[0088] in, For the first Wavelet decomposition coefficients of the layer; The total energy of the signal across all scales; the system identifies Several scale levels with a cumulative percentage exceeding 85% are labeled as signal-dominant scale levels, while the remaining scale levels are labeled as noise-dominant scale levels; during this process, the noise standard deviation required for soft thresholding is determined. The system specifically selected The leading baseline region, i.e., the arrival time of the first echo peak. Previous ps is a relative time interval, used as a noise estimation window; where... To The first echo peak time obtained from signal peak finding; the reference time here. It is not an undefined variable, but rather a variable obtained through... The fixed value determined by performing the peak-finding operation is preferably a fixed value in this embodiment. The time corresponding to the maximum value of the envelope is used to ensure that the noise calculation window remains constant throughout all operations. The standard deviation of the signal samples within this window is then calculated as... Perform family-based wavelet decomposition on the thickness-normalized waveform sequence: 1. For the noise-dominated level, perform hard thresholding or directly set the threshold to zero to filter out high-frequency roughness scattering; 2. For the signal-dominated level, perform soft thresholding, with the threshold set to... ,in, The total number of sampling points in the thickness-normalized waveform sequence is used to preserve the interface reflection characteristics while suppressing superimposed small disturbances.

[0089] The signal is reconstructed by inverse wavelet transform to obtain the interface feature enhancement signal. In this embodiment, the above algorithm is used to effectively suppress the random scattering noise caused by substrate roughness by using the gold sample model optimized by time domain truncation as prior knowledge, and significantly improves the detection sensitivity of the device for small layered defects in low signal-to-noise ratio environment.

[0090] Example 5:

[0091] The interface dielectric characteristic analysis module is used for:

[0092] The phase polarity of the enhanced signal at the second echo peak position is detected.

[0093] If the phase polarity is reversed to negative, a first numerical identifier is generated as the first dielectric characteristic component to characterize good metal contact.

[0094] If the phase polarity is positive or there is no flip, a second numerical identifier is generated as a second dielectric characteristic component to characterize the presence of an air gap or delamination.

[0095] The first dielectric feature component or the second dielectric feature component is encoded into the dielectric feature vector.

[0096] This embodiment illustrates the defect determination logic based on phase polarity analysis. The module focuses on the vicinity of the reference time after time axis alignment and detects the peak polarity of the interface feature enhancement signal. This step is based on the Fresnel reflection principle: when a terahertz wave enters a high-refractive-index copper substrate from a low-refractive-index coating, the reflection coefficient is negative, and the waveform should undergo a negative polarity reversal; conversely, if an air gap exists, the terahertz wave enters the air from the coating, the reflection coefficient is positive, and the waveform will maintain positive polarity. Based on this, in response to the detection of a negative polarity reversal and the trough amplitude exceeding a preset threshold, the preset threshold is calculated in real-time using the root mean square value of the background noise in the current scanning environment, and then... to As a lower limit for judgment to prevent misjudgment caused by random thermal noise, the system generates a first numerical identifier set to 0, indicating good metal contact; in response to the detection of a positive polarity peak or no obvious negative peak, the system generates a second numerical identifier set to 1, indicating the presence of an air gap or delamination; here, 1 serves as a defect indicator bit, ensuring logical consistency with the subsequent defect risk assessment module where high values ​​represent high risks; this numerical identifier is encoded into the first dimension of the dielectric feature vector.

[0097] This embodiment utilizes the phase reversal characteristics of terahertz waves when reflected at different media interfaces to simplify complex waveform analysis into binary polarity feature extraction. This method can extremely accurately identify dry-attach defects, i.e., situations where there is physical contact but no adhesion forming a micron-level air gap, effectively filling the technical gap that traditional ultrasonic or X-ray detection cannot detect such hidden defects.

[0098] Example 6:

[0099] The interface dielectric characteristic analysis module is also used for:

[0100] A fast Fourier transform is performed on the interface feature enhancement signal to obtain the frequency domain spectrum;

[0101] Calculate the amplitude attenuation rate of the frequency domain spectrum in a specific oxide absorption frequency range;

[0102] The amplitude attenuation rate is compared with a preset oxidation threshold;

[0103] If the amplitude decay rate is greater than the oxidation threshold, a third dielectric characteristic component is generated, which indicates the presence of oxide inclusions.

[0104] If the amplitude attenuation rate is less than or equal to the oxidation threshold, then a normal characteristic component representing the absence of oxidation inclusions is generated.

[0105] The generated feature components are appended to the dielectric feature vector.

[0106] This embodiment further incorporates frequency domain spectral attenuation analysis to detect oxidation defects and introduces a thickness-attenuation decoupling mechanism to eliminate the interference of coating thickness variations on oxidation determination; the system performs a fast Fourier transform on the interface feature enhancement signal to obtain the amplitude spectrum; and targets the characteristic absorption frequency bands of cuprous oxide or copper oxide. In this embodiment, the copper oxide value is set to 2.1 THz. In order to eliminate the Beer-Lambert absorption loss caused by the local coating thickness difference, the module uses the local coating thickness obtained in the previous step. The spectral amplitude is compensated and corrected, and the corrected amplitude is calculated. : ;

[0107] in, Derived from real-time calculation, its physical meaning is the original spectral amplitude of the current scan point at the characteristic frequency; Sourced from a materials database, its physical meaning refers to the coating material at a certain frequency. Absorption coefficient at [location], in mm. ; The local coating thickness is calculated by the time-of-flight dynamic correction module and is expressed in μm. Derived from product calibration, its physical meaning is the reference standard thickness, and the unit is μm; Originating from the need for dimensional unification, its physical meaning is the unit conversion factor for converting micrometers to millimeters, with a value of [value missing]. ;

[0108] Calculate amplitude attenuation rate The formula is as follows: ;

[0109] in, This is the reference amplitude for a good quality, non-oxidizing product at the same frequency; the system will calculate... The sample is compared with a preset oxidation threshold, such as 15%. This preset oxidation threshold, such as 15%, is determined through statistical calibration: a group of gold and copper samples, such as N=50, verified by X-ray photoelectron spectroscopy to be clean and free of oxidation, are selected. The background attenuation rate distribution in the characteristic frequency range is calculated. Based on the assumption that the background noise of good products follows a normal distribution, the mean of this distribution is taken plus three times the standard deviation. ;

[0110] This threshold is used to ensure that false alarms caused by inherent system noise are excluded with a 99.7% confidence level; if If the value exceeds the threshold, it is determined that oxidation inclusions exist, and a third dielectric characteristic component quantifying the degree of oxidation is generated; if If the value is less than or equal to the threshold, normal feature components are generated.

[0111] To ensure that the subsequent risk assessment module can obtain complete physical information, this embodiment clearly defines the construction of the dielectric characteristic vector: the generated discrete characteristic components are compared with the calculated original attenuation rate. Joint encoding to construct structured vectors ;in The phase polarity identifier obtained in the previous step is 0 or 1. The oxidation discrimination flag generated in this step is 0 or 1. The floating-point amplitude attenuation rate calculated in this step This hybrid encoding method avoids the loss of intermediate process information caused by binarization processing, ensuring that the quantization accuracy of the spectrum analysis results is fully preserved in the imaging stage.

[0112] Example 7:

[0113] Defect risk assessment imaging module, used for:

[0114] Generate a three-dimensional tomographic map containing a first layer and a second layer;

[0115] The first layer is a coating uniformity distribution map generated based on the local coating thickness;

[0116] The second layer is an interface combined with a quality heatmap generated based on dielectric feature vectors. The highlighted areas in the heatmap correspond to the coordinates of the abnormal values ​​displayed by the dielectric feature vectors.

[0117] This embodiment describes the construction method of a three-dimensional tomographic map. The module generates a first layer, which is based on the local coating thickness data calculated in the previous steps. A grayscale mapping is used to generate a coating uniformity distribution map, where the grayscale depth intuitively reflects the coating thickness variation. Simultaneously, the module generates a second layer, which is based on dielectric feature vector data to generate an interface combined with a quality heatmap. In this process, in order to solve the logical defect of the traditional linear weighted model that may miss alarms due to the smoothing effect of weights when a single defect is extremely severe (i.e., the masking effect), this embodiment constructs a nonlinear risk quantification function based on probability union. The input variable of this function is the self-permeable eigenvector. The system extracts the first component of the vector. As an indicator of physical contact Extract the third component of the vector. As a variable of continuous decay rate Based on this, the risk calculation formula is as follows: ;

[0118] in, : Derived from vector components The value is 1 when it is determined to be an air gap or delamination, and 0 otherwise; physically it represents the certainty of physical contact failure. Physically, this represents the normalized probability of oxidation occurring, mapped using the Sigmoid function. ;

[0119] in, : Derived from vector components Real-time read values; : Derived from a preset oxidation threshold, such as 15%; The standard deviation derived from the statistics of good product samples, such as 0.02, is used to control the steepness of the probability mapping;

[0120] The physical meaning of this formula is: as long as either physical stratification or chemical oxidation risk factor tends towards certainty, i.e., tends towards 1, the total risk value... That is, it tends towards 1, thus forcing highlighting and effectively avoiding logical cancellation during multi-dimensional feature fusion; the system will calculate Mapping to a pseudo-color spectrum via linear interpolation, such as JetColormap, in response to If the risk exceeds a preset high-risk threshold, such as 0.8, the system marks the corresponding coordinate point as a bright red area on the heatmap. This high-risk threshold, such as 0.8, is not an empirical estimate, but is determined based on reverse anchoring from destructive physical analysis: after performing terahertz scanning on historical test samples and calculating the Risk value, microslicing and pull-out force tests are performed to establish a regression curve of Risk value-bonding force. The statistical lower limit of the Risk value corresponding to the bonding force being lower than the industry qualified standard, such as 5 N / mm², is set as this high-risk threshold, thus giving the non-destructive testing results a clear physical failure meaning.

[0121] These two layers are overlaid to form a complete three-dimensional tomographic map. This embodiment achieves equal sensitivity in capturing both latent oxidation and explicit stratification through the aforementioned nonlinear fusion algorithm, ensuring the absolute fidelity of the imaging results to potential failure risks.

[0122] Example 8:

[0123] Also includes:

[0124] The process closed-loop feedback module is used to calculate the thickness variance in the coating uniformity distribution diagram and the defect area ratio in the interface bonding quality thermal diagram.

[0125] If the thickness variance exceeds the upper limit of the preset tolerance range, or the defect area ratio exceeds the preset safety threshold, an electroplating process parameter adjustment instruction is generated and sent to the production line control system.

[0126] This embodiment introduces a process closed-loop feedback mechanism, whereby the module calculates the thickness variance in the coating uniformity distribution map. And the percentage of defect area in the interface bonding quality heatmap. The system executes closed-loop control based on the principle of negative feedback, and the specific logic is as follows:

[0127] For thickness uniformity adjustment: the system employs a combination strategy of dead-zone control and incremental PI control; responding to thickness variance. If the thickness exceeds the preset tolerance limit, the system will divide the thickness distribution map into left and right sides. ,middle and the right side The specific rules for region segmentation are as follows: obtain the total horizontal pixel width of the thickness distribution map. The image is cut along the horizontal coordinate axis according to a preset ratio, specifically defined as follows: Left side area , For the middle area , The right side area This spatial division corresponds to the physical positions of the three independently controlled auxiliary anodes (left, center, and right) in the electroplating tank; the deviation on the left side is calculated. Deviation from the right side To prevent frequent actuator oscillations caused by measurement noise, a system adjustment dead zone is set. For example, 0.5um; only when Only then is the incremental PI algorithm used to generate current adjustment commands: ;

[0128] in, for or ;

[0129] Meanwhile, to ensure production safety, the system applies a hard saturation constraint to the total current command. ;

[0130] in, These are the lower and upper limits of the rated current of the auxiliary anode power supply, respectively.

[0131] For defect rate adjustment: responding to the percentage of defect area Greater than the preset safety threshold Calculate the deviation If the principal component of the eigenvector indicates oxide inclusions, the system generates a command to adjust the concentration of the pretreatment pickling and activating solution. Based on sensitivity coefficient Defined as Typically a negative value, it indicates that the defect rate decreases as the concentration increases. The reverse compensation amount is calculated as follows: ;

[0132] in, Derived from system settings, its physical meaning is the minimum adjustment concentration step of the dosing pump, with units of... ;

[0133] Similarly, the calculated concentration setpoint must be verified by safety logic: if That is, saturated solubility, then forced clamping to And trigger a concentration alarm;

[0134] To meet the stability requirements of industrial control, this embodiment also clarifies the method for obtaining core control parameters: PID controller parameters. The sensitivity coefficient was determined during the production line commissioning phase using the step response method. This involved applying a 5% current step to the auxiliary anode, fitting the thickness response curve to obtain a first-order inertial pure time-delay model, and then calculating the sensitivity coefficient using the IMC internal model control rule. The statistical slope is calculated using the least squares linear regression method based on historical acid concentration-defect rate data from the past 30 batches in the MES system. This parameter traceability mechanism ensures the executability and robustness of the closed-loop control logic in the actual physical system.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A non-destructive testing device for defects in a copper-plated substrate based on terahertz waves, characterized in that, include: The terahertz scanning acquisition module is used to control the terahertz time-domain spectral emission and receiving unit to perform two-dimensional point-by-point scanning of the copper bar surface under test, and to acquire the reflected echo signals of each scanning point to obtain the original terahertz time-domain waveform sequence. The time-of-flight dynamic correction module is used to analyze the first echo peak of the air-coating interface and the second echo peak of the coating-substrate interface in the original terahertz time-domain waveform sequence. Based on the time difference between the first echo peak and the second echo peak, the local coating thickness of each scanning point is calculated, and the local coating thickness is used to dynamically align the original terahertz time-domain waveform sequence with the time axis to obtain a thickness-normalized waveform sequence. The substrate roughness deconvolution module is used to call a preset defect-free surface scattering reference model and use the defect-free surface scattering reference model to perform deconvolution filtering on the thickness normalized waveform sequence to remove the background scattering noise component caused by substrate roughness, thereby obtaining the interface feature enhancement signal. The interface dielectric feature analysis module is used to perform time-domain phase polarity analysis and frequency-domain spectral attenuation analysis on the interface feature enhancement signal to extract dielectric feature vectors that reflect the abrupt change in the interface dielectric constant. The defect risk assessment imaging module is used to construct a three-dimensional tomographic map reflecting the integrity of the interface bonding based on the dielectric feature vector and the local coating thickness, and output the defect detection results under the copper plating.

2. The non-destructive testing device for defects in copper-plated substrates based on terahertz waves according to claim 1, characterized in that, The terahertz scanning acquisition module includes: A femtosecond laser pumping unit is used to generate femtosecond laser pulses to excite a terahertz emission source to generate broadband terahertz waves. A reflective confocal optical path unit is used to focus the broadband terahertz wave onto the interface between the coating and the substrate of the copper bar under test through a preset polyethylene lens, and to collect the reflected echo. A two-dimensional displacement control unit is used to drive the scanning mechanism to move the reflective confocal optical path unit or the copper bar surface to be measured relative to each other in order to achieve full coverage scanning.

3. The non-destructive testing device for defects in copper-plated substrates based on terahertz waves according to claim 1, characterized in that, The time-of-flight dynamic correction module is used for: The two peaks with the largest absolute amplitude in the original terahertz time-domain waveform sequence are identified as the first echo peak and the second echo peak. Calculate the time-of-flight delay between the first echo peak and the second echo peak; By combining the preset coating refractive index, the flight time delay is converted into the local coating thickness; Using the time point of the second echo peak as a reference, the original terahertz time-domain waveform sequence is shifted on the time axis to eliminate phase jitter caused by the local coating thickness difference.

4. The non-destructive testing device for defects in copper-plated substrates based on terahertz waves according to claim 1, characterized in that, The substrate roughness deconvolution module is used for: The Wiener filtering algorithm or wavelet transform algorithm is used to take the thickness-normalized waveform sequence as the input signal and the good surface scattering reference model as the system response function. In the frequency domain, a deconvolution operation is performed to subtract the scattering component related to the system response function from the input signal, so as to retain the reflection component caused by the change in the dielectric properties of the interface as the interface feature enhancement signal.

5. The non-destructive testing device for defects in copper-plated substrates based on terahertz waves according to claim 1, characterized in that, The interface dielectric feature analysis module is used for: Detect the phase polarity of the interface feature enhancement signal at the second echo peak position; If the phase polarity is a negative polarity reversal, a first numerical identifier is generated as a first dielectric characteristic component to characterize good metal contact. If the phase polarity is positive or without reversal, a second numerical identifier is generated as a second dielectric characteristic component to characterize the presence of an air gap or delamination. The first dielectric feature component or the second dielectric feature component is encoded into the dielectric feature vector.

6. The non-destructive testing device for defects in copper-plated substrates based on terahertz waves according to claim 5, characterized in that, The interface dielectric feature analysis module is also used for: The interface feature enhancement signal is subjected to a fast Fourier transform to obtain a frequency domain spectrum. Calculate the amplitude attenuation rate of the frequency domain spectrum in a specific oxide absorption frequency range; The amplitude attenuation rate is compared with a preset oxidation threshold; If the amplitude attenuation rate is greater than the oxidation threshold, a third dielectric characteristic component representing the presence of oxide inclusions is generated. If the amplitude attenuation rate is less than or equal to the oxidation threshold, then a normal characteristic component characterizing the absence of oxidation inclusions is generated. The generated feature components are appended to the dielectric feature vector.

7. The non-destructive testing device for defects in copper-plated substrates based on terahertz waves according to claim 1, characterized in that, The defect risk assessment imaging module is used for: Generate the three-dimensional tomographic map containing the first layer and the second layer; The first layer is a coating uniformity distribution map generated based on the local coating thickness; The second layer is an interface-combined quality heatmap generated based on the dielectric feature vector, where the highlighted areas correspond to the coordinates of the abnormal values ​​displayed by the dielectric feature vector.

8. The non-destructive testing device for defects in copper-plated substrates based on terahertz waves according to claim 7, characterized in that, Also includes: The process closed-loop feedback module is used to calculate the thickness variance in the coating uniformity distribution diagram and the defect area ratio in the interface bonding quality thermogram. If the thickness variance is greater than the upper limit of the preset tolerance range, or the defect area ratio is greater than the preset safety threshold, an electroplating process parameter adjustment instruction is generated and sent to the production line control system.

Images