High-frequency water-immersion point-focus ultrasound imaging method and system for semiconductor wafer bonding defect detection

Abstract

The application provides a high-frequency water immersion point focusing ultrasonic imaging method and system for semiconductor wafer bonding defect detection, relates to the technical field of ultrasonic imaging, and calibrates detection parameters by combining wafer bonding parameters with preset detection parameter data, controls a probe to emit high-frequency ultrasonic signals through target detection conditions, and the high-frequency ultrasonic signals are incident to a wafer bonding interface after focusing through an acoustic lens, the reflected echo of defects is received and converted into an electric signal, the original defect signal is analyzed to extract a signal, two-dimensional imaging and three-dimensional size reconstruction information are obtained, defect characteristic analysis and judgment are performed through three-dimensional data and two-dimensional imaging of defects, and wafer bonding quality evaluation results are obtained. The application can effectively make up for the shortcomings of traditional detection technology, and significantly improve the accuracy and automation level of wafer bonding defect detection.

Description

Technical Field

[0001] This invention proposes a high-frequency water immersion point focusing ultrasonic imaging method and system for detecting bonding defects in semiconductor wafers, relating to the field of ultrasonic imaging technology, specifically to the field of high-frequency water immersion point focusing ultrasonic imaging technology for detecting bonding defects in semiconductor wafers. Background Technology

[0002] Semiconductor wafer bonding is a core process in semiconductor manufacturing, and its bonding quality directly determines the performance and reliability of semiconductor devices. However, the bonding interface is prone to micro-defects such as delamination, voids, and microcracks, which seriously affect the lifespan of devices. Currently, wafer bonding defect detection mainly uses traditional ultrasonic testing methods, but these methods have many shortcomings: poor compatibility between detection parameters and wafer bonding characteristics; scattered signal emission and inaccurate focusing; weak defect echoes are easily interfered with by noise; incomplete processing of the original signal, making it difficult to extract micro-defect features, leading to missed detections and false detections; fragmented testing processes, incomplete scanning, and strong subjectivity in quality judgment, making it difficult to meet the requirements of high-precision and automated testing. In addition, existing testing technologies largely rely on imports, resulting in high testing costs and high technical barriers. Summary of the Invention

[0003] This invention provides a high-frequency water immersion point focusing ultrasonic imaging method and system for detecting bonding defects in semiconductor wafers, in order to solve the above-mentioned problems:

[0004] The present invention proposes a high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers, the method comprising:

[0005] S1. By connecting the high-frequency water immersion point focusing ultrasonic probe to the ultrasonic testing system and calibrating it according to the preset detection parameters, the target detection conditions are obtained.

[0006] S2. The probe emits a high-frequency ultrasonic signal by controlling the target detection conditions. After being focused by the acoustic lens, the signal is incident on the wafer bonding interface. The reflected echo of the defect is received and converted into an electrical signal to obtain the original defect signal. The original defect signal is then analyzed for defect features to extract the signal and obtain the micro-defect identification signal.

[0007] S3. By using the micro-defect identification signal in conjunction with the automatic scanning device, the wafer is fully scanned and a two-dimensional image is converted to obtain a two-dimensional image. Then, the three-dimensional dimensions of the defect are reconstructed to obtain the three-dimensional data of the defect. The defect feature is analyzed and judged by the three-dimensional data of the defect and the two-dimensional image to obtain the wafer bonding quality evaluation result.

[0008] Furthermore, the system includes:

[0009] The detection configuration module is used to connect the high-frequency water immersion point focusing ultrasonic probe to the ultrasonic detection system, calibrate it according to preset detection parameters, and obtain the target detection conditions.

[0010] The defect identification module is used to control the probe to emit high-frequency ultrasonic signals according to the target detection conditions. After being focused by the acoustic lens, the signals are incident on the wafer bonding interface. The module receives the defect reflection echoes and converts them into electrical signals to obtain the original defect signals. The module then performs defect feature analysis on the original defect signals to extract the signals and obtain the micro-defect identification signals.

[0011] The analysis and judgment module is used to perform full-area scanning and two-dimensional imaging of the wafer by using the micro-defect identification signal in conjunction with the automatic scanning device to obtain a two-dimensional imaging image. Then, the three-dimensional dimensions of the defect are reconstructed to obtain the three-dimensional data of the defect. The defect feature is analyzed and judged by the three-dimensional data of the defect and the two-dimensional imaging image to obtain the wafer bonding quality evaluation result.

[0012] The beneficial effects of this invention are as follows: This method solves the technical problems of poor adaptability, insufficient ability to identify minute defects, and fragmented detection process of traditional ultrasonic testing methods; it realizes fully automated detection of semiconductor wafer bonding defects, completing parameter calibration, signal acquisition, imaging analysis, and quality judgment without manual intervention, thus improving detection efficiency and consistency; through high-frequency focusing and signal feature extraction technology, it enhances the ability to identify minute defects at the wafer bonding interface, avoiding the problems of missed or false detection of minute defects and improving detection accuracy; the constructed closed-loop detection process realizes the integration of defect identification, location, and quality judgment, improving the comprehensiveness and reliability of detection results; at the same time, it eliminates dependence on imported testing methods, reducing the technical threshold and application cost of semiconductor wafer bonding detection. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of a high-frequency water immersion point focused ultrasound imaging method for detecting bonding defects in semiconductor wafers. Detailed Implementation

[0014] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0015] Example 1

[0016] In one embodiment of the present invention, a high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers is proposed, the method comprising:

[0017] S1. By connecting the high-frequency water immersion point focusing ultrasonic probe to the ultrasonic testing system and calibrating it according to the preset detection parameters, the target detection conditions are obtained.

[0018] S2. The probe emits a high-frequency ultrasonic signal by controlling the target detection conditions. After being focused by the acoustic lens, the signal is incident on the wafer bonding interface. The reflected echo of the defect is received and converted into an electrical signal to obtain the original defect signal. The original defect signal is then analyzed for defect features to extract the signal and obtain the micro-defect identification signal.

[0019] S3. By using a micro-defect identification signal in conjunction with an automatic scanning device to perform a full-area scan and convert a two-dimensional image into a two-dimensional image, a three-dimensional size reconstruction of the defect is performed to obtain three-dimensional defect data. Defect feature analysis and judgment are then performed using the three-dimensional defect data and the two-dimensional image to obtain the wafer bonding quality evaluation result. Figure 1 As shown.

[0020] The working principle and technical effects of the above-mentioned technical solution are as follows: This method adapts the probe to the detection system, confirms and calibrates parameters to ensure precise matching between the detection parameters and the wafer bonding characteristics; by utilizing the calibrated target detection conditions, the probe is controlled to emit high-frequency ultrasonic signals, and the acoustic lens focusing characteristics are used to precisely apply the sound beam to the wafer bonding interface, receiving defect reflection echoes and completing signal conversion and feature extraction to filter out effective signals that can identify minute defects; finally, by combining with an automatic scanning device to achieve full-area wafer detection, the effective signals are converted into two-dimensional images and three-dimensional size reconstruction is completed. Defects are comprehensively judged based on industry standards, and the final wafer bonding quality evaluation result is output. The entire process revolves around the minute and concealed characteristics of semiconductor wafer bonding defects, achieving fully automated and precise detection from parameter adaptation to quality judgment.

[0021] This method addresses the technical challenges of poor adaptability, insufficient ability to identify minute defects, and fragmented testing processes inherent in traditional ultrasonic testing methods. It achieves fully automated testing of semiconductor wafer bonding defects, enabling parameter calibration, signal acquisition, imaging analysis, and quality assessment without manual intervention, thus improving testing efficiency and consistency. Through high-frequency focusing and signal feature extraction techniques, it enhances the identification of minute defects at the wafer bonding interface, avoiding missed or false detections and improving testing accuracy. The constructed closed-loop testing process integrates defect identification, location, and quality assessment, enhancing the comprehensiveness and reliability of testing results. Furthermore, it eliminates reliance on imported testing methods, lowering the technical barriers and application costs of semiconductor wafer bonding testing.

[0022] In one embodiment of the present invention, S1 includes:

[0023] By connecting a high-frequency water immersion point focusing ultrasonic probe to a high-frequency ultrasonic testing system;

[0024] Confirm the core detection parameters of the probe and obtain the preset data of the detection parameters;

[0025] The detection parameters are calibrated by combining the preset detection parameter data with the core parameters of wafer bonding, and the detection parameter calibration data is obtained.

[0026] The compatibility between the preset detection parameters and wafer bonding parameters is verified and corrected multiple times using a parameter matching method to optimize the parameter combination. At the same time, the repeatability test of the detection parameter calibration data is carried out to obtain the target detection conditions.

[0027] Among them, by connecting the high-frequency water immersion point focusing ultrasonic probe to the high-frequency ultrasonic testing system, the following is included:

[0028] By precisely connecting the signal interface of the high-frequency water immersion focused ultrasonic probe to the signal input terminal of the high-frequency ultrasonic testing system, the sealing and contact stability of the interface connection are checked to eliminate signal transmission faults caused by poor contact and obtain probe access connection data.

[0029] By connecting the probe to the data, the signal transmission link between the probe and the detection system is initially debugged to ensure normal communication between the probe and the system, no delay or interference in signal transmission, and to complete the stable connection between the high-frequency water immersion point focusing ultrasonic probe and the high-frequency ultrasonic detection system.

[0030] Specifically, the detection parameters are calibrated by combining preset detection parameter data with core wafer bonding parameters to obtain detection parameter calibration data, including:

[0031] By combining the preset data of the detection parameters with the core parameters of wafer bonding, the matching relationship between the detection parameters and the wafer bonding parameters is confirmed, and parameter matching confirmation data is obtained.

[0032] By confirming the data through parameter matching, the calibrated detection parameters are subjected to fixed tests, and the detection signal sensitivity, stability, and defect identification accuracy data corresponding to the parameters are recorded to obtain detection parameter calibration data that meet the requirements of wafer bonding defect detection.

[0033] The working principle and technical effect of the above technical solution are as follows: This method precisely connects the probe and the detection system interface, checks the interface sealing and contact stability, eliminates signal transmission faults, and ensures smooth communication between the probe and the system; then, the core detection parameters of the probe are confirmed, and preset parameter data is obtained; combined with the core parameters of wafer bonding, the matching relationship between the detection parameters and the wafer bonding parameters is confirmed, and the matching effect is verified through fixed tests to obtain the detection parameter calibration data; multiple rounds of verification are performed through parameter matching methods to eliminate mismatched parameter combinations, and the stability of the calibration parameters is ensured through repeatability tests, thereby obtaining the target detection conditions suitable for wafer bonding defect detection and ensuring the rationality and reliability of parameters in subsequent detection processes.

[0034] This method addresses the technical problems of unstable probe-system interface, susceptibility to signal transmission interference, poor compatibility between probe parameters and wafer bonding characteristics, and inaccurate parameter calibration in traditional testing. Through interface interface checks and link verification, it improves the stability of the probe-system connection, reduces signal transmission interference, and lowers detection errors caused by connection issues. Through parameter verification, calibration, and multiple rounds of validation, it achieves precise matching between probe parameters and wafer bonding parameters, improving the accuracy and stability of parameter calibration and avoiding missed or false detections due to parameter incompatibility. Repeatability testing ensures the reliability of the target detection conditions, further enhancing the stability and accuracy of the entire testing process and reducing time costs caused by improper parameter settings.

[0035] In one embodiment of the present invention, the step of confirming the core detection parameters of the probe and obtaining preset detection parameter data includes:

[0036] The core detection parameters preset during probe manufacturing are read directly to obtain the preset parameter data;

[0037] Based on the preset parameter data, record the corresponding parameter values, signal feedback intensity, and detection stability data to obtain complete preset detection parameter data.

[0038] The working principle and technical effect of the above technical solution are as follows: This method confirms the power-on status of the probes already connected to the system, confirming that the probes themselves are working normally, and providing a benchmark for parameter calibration; based on the read preset parameters, the preset values ​​of the core detection parameters are recorded, and at the same time, the signal feedback, detection stability and parameter adaptability data of the probes in working state are collected in real time to ensure the comprehensiveness of the data; the data of each parameter are integrated to form complete preset data of detection parameters, ensuring the traceability and standardization of the parameter data.

[0039] This method solves the technical problems of non-standard parameter settings, blind acquisition of core parameters, and incomplete and untraceable parameter preset data in traditional probes. By confirming the status and directly reading preset parameters, it avoids on-site debugging, improves the stability of probe operation, and reduces detection deviations caused by manual parameter settings. Through direct recording of preset parameters and real-time data acquisition, it achieves rapid and accurate acquisition of core detection parameters, avoids uncertainties in the parameter acquisition process, and improves the accuracy of parameter calibration. By integrating and forming complete preset detection parameter data, it improves the efficiency and accuracy of parameter adaptation, and achieves traceability of parameter data, further enhancing the standardization and reliability of the entire detection process.

[0040] In one embodiment of the present invention, the method of repeatedly verifying and correcting the compatibility between preset detection parameter data and wafer bonding parameters through parameter matching, optimizing parameter combinations, and simultaneously performing repeatability tests on detection parameter calibration data to obtain target detection conditions includes:

[0041] The compatibility between the preset detection parameters and the core parameters of wafer bonding is verified multiple times using a parameter matching algorithm to obtain multiple verification data.

[0042] By comparing the sensitivity and stability of the detection signal under different parameter combinations based on multiple verification data, suitable comparison data is obtained.

[0043] Based on the adaptation comparison data, parameters with insufficient adaptability are corrected to obtain optimized parameter combination data;

[0044] By optimizing the parameter combination data, multiple rounds of repeatable tests were conducted on the detection parameter calibration data to obtain multiple rounds of test data;

[0045] By using multiple rounds of test data, the parameter deviations that occurred during the test were corrected a second time to obtain the target detection conditions.

[0046] Among them, by comparing the sensitivity and stability of the detection signal under different parameter combinations based on multiple verification data, suitable comparison data is obtained, including:

[0047] By verifying the data multiple times, different combinations of detection parameters were selected. For each combination of parameters, the sensitivity and stability of the detection signal were tested. The signal amplitude, signal-to-noise ratio, detection response speed and long-term working stability data of each combination of parameters were recorded to obtain the test data of each combination of parameters.

[0048] By using test data of various parameter combinations, the sensitivity and stability of the detection signals of different parameter combinations are quantitatively compared, and the parameter combination range with the best fit is selected to obtain the fit comparison data.

[0049] The working principle and technical effect of the above technical solution are as follows: This method adopts a parameter matching algorithm to perform multiple rounds of adaptation and verification between the preset detection parameter data and the core parameters of wafer bonding, obtain multiple sets of verification data, and initially judge the parameter compatibility; then, based on the verification data, different parameter combinations are screened, and signal sensitivity and stability tests are performed on each set of parameter combinations, and relevant test data are recorded. Through quantitative comparison, the parameter combination range with the best compatibility is screened to obtain the adaptation comparison data; based on the adaptation comparison data, parameters with insufficient compatibility are corrected to form optimized parameter combinations; finally, based on the optimized parameter combinations, multiple rounds of repeatability tests are performed on the detection parameter calibration data to check parameter deviations and perform secondary corrections, ultimately obtaining target detection conditions with strong compatibility and high stability, ensuring the reliability and consistency of parameters in subsequent detection processes.

[0050] This method addresses the technical problems of insufficient parameter adaptation and verification cycles, inaccurate parameter combination optimization, and poor stability and susceptibility to deviations in traditional calibration parameters. Through multiple rounds of adaptation and verification and parameter combination comparison, it improves the adaptability of parameter combinations, avoiding the decrease in detection accuracy caused by insufficient parameter adaptation. Quantitative comparison and selection of the optimal parameter combination reduces the blindness of parameter optimization, improving its efficiency and accuracy. Multiple rounds of repeatability testing and secondary deviation correction ensure the stability of the target detection conditions, reducing the impact of parameter fluctuations on the detection results and improving the consistency and reliability of the results. Simultaneously, it eliminates invalid adjustment data, reducing data redundancy, improving parameter processing efficiency, and further reducing time and labor costs in the detection process.

[0051] In one embodiment of the present invention, S2 includes:

[0052] The high-frequency water immersion point focusing probe is controlled by the target detection conditions to emit a high-frequency ultrasonic signal at a preset frequency. The signal is then processed by the concave acoustic lens built into the probe to obtain the acoustic beam focusing data.

[0053] The lens curvature parameters are adjusted according to the acoustic beam focusing processing data to control the ultrasonic beam focusing point to fall on the wafer bonding interface, and the reflected echo generated by the defects of the bonding interface is received to obtain the ultrasonic reflected echo signal.

[0054] The ultrasonic reflected echo signal is converted into a processable electrical signal by a high-frequency piezoelectric transducer built into the probe.

[0055] The processable electrical signal is amplified to obtain the original defect signal;

[0056] Defect feature analysis and separation data are obtained by performing defect feature analysis and separation on the original defect signal.

[0057] Defect feature extraction data is obtained by separating the defect feature analysis data and then performing defect feature extraction data.

[0058] The validity of the defect feature extraction data is verified to obtain the identification signal of minor defects.

[0059] Specifically, defect feature analysis and separation are performed on the original defect signal to obtain defect feature analysis and separation data, including:

[0060] The original defect signal is denoised using an adaptive noise reduction algorithm to filter out invalid interference components and obtain the denoised original signal.

[0061] By using the denoised original signal, a bandpass filtering method is used to filter the signal frequency bands and retain the signal frequency bands related to wafer bonding defects. At the same time, the amplitude and phase characteristics of the defect signal are analyzed to accurately separate the effective defect signal and the remaining noise signal in the original signal, and obtain defect feature analysis and separation data.

[0062] The working principle and technical effect of the above technical solution are as follows: Based on the target detection conditions, this method controls the probe to emit a high-frequency ultrasonic signal at a preset frequency. The sound beam is focused by a concave acoustic lens built into the probe, and the lens curvature parameter is adjusted to ensure that the focal point of the sound beam accurately falls on the wafer bonding interface, maximizing the intensity of the defect echo signal. Subsequently, the reflected echo signal generated by the bonding interface defect is received, and the ultrasonic signal is converted into a processable electrical signal by a high-frequency piezoelectric transducer and amplified to obtain the original defect signal containing defect information, noise, and interference signals. Next, an adaptive noise reduction algorithm is used to filter out invalid interference. Bandpass filtering is used to select signal frequency bands related to the defect. Combining signal amplitude and phase characteristics, the effective defect signal and the remaining noise signal are accurately separated to obtain defect feature analysis separation data. Finally, defect features are extracted from the separated data, and through validity verification, false feature signals are eliminated to obtain a micro-defect identification signal that can accurately identify minute defects.

[0063] This method addresses the technical challenges of traditional ultrasonic signal emission dispersion and inaccurate focusing, weak and easily interfered defect echo signals, incomplete original signal processing, and difficulty in extracting minute defect features. By precisely controlling the acoustic beam focusing and focal point, it improves the utilization rate of the ultrasonic signal and the intensity of the defect echo signal, enhancing the detection capability of minute defects at wafer bonding interfaces. Through piezoelectric transducer signal conversion and amplification, it solves the problem of capturing weak defect signals, improving signal processability. Adaptive noise reduction and bandpass filtering effectively filter noise and interference signals, achieving precise separation of valid defect signals from noise and improving the purity of signal processing. Feature extraction and validity verification eliminate false feature signals, improving the accuracy of minute defect identification, avoiding missed or false detections of minute defects, and further enhancing the accuracy of the entire detection process.

[0064] In one embodiment of the present invention, the high-frequency water immersion point focusing probe, controlled by target detection conditions, emits a high-frequency ultrasonic signal at a preset frequency. The signal is then processed by a concave acoustic lens built into the probe to obtain acoustic beam focusing data, including:

[0065] A high-frequency water immersion point focusing probe emits high-frequency ultrasonic signals to obtain high-frequency ultrasonic transmission signals;

[0066] The high-frequency ultrasonic signal is transmitted to the concave acoustic lens built into the probe, and the ultrasonic signal is focused to obtain focused data.

[0067] By adjusting the focusing angle and range of the lens through focusing data processing, sound beam focusing data is obtained.

[0068] The working principle and technical effect of the above technical solution are as follows: This method controls the high-frequency water immersion point focusing probe to emit a stable high-frequency ultrasonic signal according to the preset frequency parameters in the target detection conditions, ensuring that the frequency and power of the emitted signal are consistent with the target detection conditions, and obtaining a high-frequency ultrasonic emission signal that meets the requirements; then, the emitted ultrasonic signal is transmitted to the concave acoustic lens built into the probe, and the focusing characteristics of the concave lens are used to converge the ultrasonic signal, and the relevant data in the focusing process are recorded to obtain the focusing processing data; finally, based on the focusing processing data, the focusing angle and range of the lens are adjusted to optimize the sound beam focusing effect, ensuring that the sound beam focus point can accurately fall on the wafer bonding interface, and the adjusted focusing parameters are recorded to obtain complete sound beam focusing processing data.

[0069] This method solves the technical problems of unstable ultrasonic signal transmission, inaccurate beam focusing, and uncontrollable focusing effect in traditional ultrasonic methods. By controlling the probe to emit high-frequency ultrasonic signals that meet the requirements, the stability of signal transmission is improved, ensuring the consistency of signal frequency and power, and reducing the impact of signal fluctuations on detection. Through focusing with a concave acoustic lens and adjustment of focusing parameters, precise beam focusing is achieved, improving the energy concentration of the beam and enhancing the penetration ability and detection sensitivity of ultrasonic signals to wafer bonding interface defects. By acquiring complete beam focusing processing data, the focusing process is made traceable and optimizable, facilitating the adjustment of focusing parameters according to different wafer bonding parameters, improving the adaptability and flexibility of detection, and reducing the problem of missed defects caused by inaccurate beam focusing, further improving the accuracy of detection.

[0070] In one embodiment of the present invention, the step of performing defect feature analysis and separation on the original defect signal to obtain defect feature analysis and separation data includes:

[0071] Defect feature analysis data is obtained by using an adaptive noise reduction algorithm and a bandpass filtering method to analyze the original defect signal.

[0072] By analyzing defect feature data, defect feature information in the original signal is identified, and original signal feature analysis data is obtained.

[0073] The noise interference defect signal of the original signal is separated by analyzing the original signal feature data to obtain defect feature analysis separation data.

[0074] Specifically, defect feature analysis is performed on the original defect signal using an adaptive noise reduction algorithm and a bandpass filtering method to obtain defect feature analysis data, including:

[0075] Using the original defect signal, an adaptive noise reduction algorithm is used to adaptively filter out random noise and environmental interference in the original signal. The noise reduction threshold is dynamically adjusted according to the signal amplitude change, and weak defect signal components are retained to obtain the noise-reduced signal.

[0076] After noise reduction, the signal is filtered using a bandpass filter to set a frequency range that matches the wafer bonding defect signal, filtering out invalid signals outside the frequency band. At the same time, the amplitude, phase, propagation time and other characteristics of the signal are analyzed to extract the characteristic parameters related to the defect and obtain defect characteristic analysis data.

[0077] The working principle and technical effects of the above-mentioned technical solution are as follows: This method is based on the original defect signal and adopts an adaptive noise reduction algorithm. The noise reduction threshold is adjusted according to the dynamic changes in the signal amplitude, which can effectively filter out invalid components such as random noise and environmental interference, while retaining weak defect signals and avoiding the false filtering of weak defect signals. Subsequently, a bandpass filtering method is applied to the denoised signal to set a frequency band range that matches the wafer bonding defect signal and filter out invalid signals outside the frequency band, further improving the signal purity. At the same time, the amplitude, phase, propagation time and other characteristics of the denoised and filtered signal are analyzed to extract defect-related feature parameters and obtain defect feature analysis data. Based on the defect feature analysis data, defect feature information in the original signal is identified to distinguish between valid defect signals and noise interference signals. Finally, the noise interference and valid defect signals in the original signal are accurately separated to obtain defect feature analysis separation data.

[0078] This method addresses the technical challenges of poor noise reduction, difficulty in separating noise from valid defect signals, and the easy filtering of weak defect signals and inaccurate defect feature extraction in traditional signal processing. By dynamically adjusting the threshold in an adaptive noise reduction algorithm, a balance is achieved between noise filtering and the preservation of weak defect signals, improving the flexibility and accuracy of noise reduction processing. Bandpass filtering further filters out invalid signals, enhancing signal purity. Multi-dimensional feature analysis and extraction clarify the core features of defect signals. Precise separation of noise from valid defect signals avoids the impact of noise interference on defect identification, improving the accuracy of defect feature analysis and further enhancing the accuracy and reliability of the entire detection process, while simultaneously reducing the cost of manual intervention in signal processing.

[0079] In one embodiment of the present invention, S3 includes:

[0080] By using minute defect identification signals in conjunction with a high-precision automatic scanning device, the wafer is scanned according to a preset spiral or matrix scanning path to obtain scanning process data.

[0081] The feature extraction of minute defect identification signals is performed on the scanning process data to obtain scanning feature extraction data, and the intensity change and spatial location information of minute defect identification signals are collected in real time.

[0082] Two-dimensional imaging images are obtained by converting the scan feature extraction data into two-dimensional imaging images using ultrasound imaging processing methods.

[0083] The defects at the wafer bonding interface are reconstructed in three dimensions by combining two-dimensional imaging with three-dimensional reconstruction algorithms to obtain reconstruction data.

[0084] Error correction is performed on the reconstructed data to obtain three-dimensional data of the defects;

[0085] By combining 3D defect data and 2D imaging with semiconductor wafer bonding industry acceptance standards, a comprehensive analysis and judgment of defect characteristics is conducted to obtain wafer bonding quality evaluation results.

[0086] The working principle and technical effects of the above-mentioned technical solution are as follows: This method uses minute defect identification signals as a basis, and in conjunction with a high-precision automatic scanning device, performs a full-area, no-dead-angle scan of the wafer according to a preset spiral or matrix scanning path to ensure that no area is missed. Simultaneously, relevant data during the scanning process is collected to obtain scanning process data. Subsequently, feature extraction of minute defect identification signals is performed on the scanning process data, and signal intensity changes and spatial location information are collected simultaneously to obtain scan feature extraction data, clarifying the spatial distribution and signal characteristics of defects. Through ultrasonic imaging processing, the scan feature extraction data is transformed into an intuitive two-dimensional image, clearly presenting the location, distribution range, and morphology of defects. Then, combined with a three-dimensional reconstruction algorithm, based on the two-dimensional image, the defects are reconstructed in three dimensions to obtain spatial dimension data. Error correction is used to improve the accuracy of the reconstructed data, obtaining three-dimensional defect data. Finally, in conjunction with semiconductor wafer bonding industry acceptance standards, the three-dimensional defect data and the two-dimensional image are comprehensively analyzed to determine the defect type and severity, forming a complete quality evaluation result and achieving comprehensive control of wafer bonding quality.

[0087] This method addresses the technical problems of incomplete scanning, inaccurate defect localization, poor imaging, difficulty in accurately acquiring defect dimensions, and the lack of unified standards and strong subjectivity in quality judgment in traditional inspection methods. Through high-precision automatic scanning and preset scanning paths, it achieves full-area, blind-spot-free scanning of the wafer, avoiding omissions and deviations inherent in manual scanning and improving scanning efficiency and comprehensiveness. By using two-dimensional imaging and three-dimensional reconstruction, abstract signals are transformed into intuitive images and precise three-dimensional data, enabling accurate defect localization and size acquisition, thus enhancing the intuitiveness and accuracy of defect detection. Error correction further improves the accuracy of three-dimensional defect data. By combining industry acceptance standards for comprehensive judgment, the quality judgment process is standardized, subjectivity is reduced, and the objectivity and authority of quality evaluation results are enhanced. Simultaneously, it automates the entire process from scanning to evaluation, improving inspection efficiency and reducing labor costs.

[0088] In one embodiment of the present invention, the step of comprehensively analyzing and judging defect characteristics by combining three-dimensional defect data, two-dimensional imaging images, and semiconductor wafer bonding industry acceptance standards to obtain wafer bonding quality evaluation results includes:

[0089] By combining 3D defect data and 2D imaging images with semiconductor wafer bonding industry acceptance standards, a comprehensive defect feature analysis is conducted on defect anomaly information to obtain comprehensive defect analysis data.

[0090] By analyzing the comprehensive defect data, classifying the defect levels according to industry acceptance standards and quality requirements, clarifying the degree of impact of defects on wafer bonding quality, generating a complete analysis report containing detailed defect information, judgment criteria, and quality level, and obtaining wafer bonding quality evaluation results.

[0091] Specifically, by combining 3D defect data and 2D imaging with semiconductor wafer bonding industry acceptance standards, a comprehensive defect feature analysis is conducted on defect anomaly information to obtain comprehensive defect analysis data, including:

[0092] By using three-dimensional defect data (including parameters such as defect length, width, depth, volume, and burial depth) and two-dimensional imaging (including defect location, distribution range, defect morphology, and signal intensity distribution), combined with semiconductor wafer bonding industry acceptance standards (such as SEMI standards), defect anomaly information is classified and sorted to clarify defect types (delamination, voids, microcracks, impurities, etc.).

[0093] By obtaining defect classification information and comparing it with the judgment thresholds for defect size, distribution density, and severity in industry standards, various types of defects are quantitatively analyzed, and the specific parameters, exceedances, and potential impacts of the defects are recorded to obtain comprehensive defect analysis data.

[0094] The working principle and technical effect of the above technical solution are as follows: This method integrates three-dimensional defect data and two-dimensional imaging images to extract core defect information, including defect size, location, distribution, morphology, signal strength, and burial depth. Combined with semiconductor wafer bonding industry acceptance standards (such as SEMI standards), all defect anomaly information is systematically classified and sorted to clarify the specific types of defects (such as delamination, voids, microcracks, impurities, etc.), and to distinguish the characteristic differences of different types of defects. Subsequently, based on industry standards, and comparing with the judgment thresholds for defect size, distribution density, and severity in the standards, various types of defects are quantitatively analyzed to clarify the specific parameters of each defect, whether it exceeds the standard, and the degree of exceeding the standard. Simultaneously, the potential impact of defects on wafer bonding quality is analyzed, and all analytical information is integrated to obtain comprehensive defect analysis data. Finally, based on the comprehensive defect analysis data, defect levels are classified, the degree of impact of defects on wafer bonding quality is clarified, and a complete analysis report containing detailed defect information, judgment criteria, and quality level is generated, ultimately obtaining objective, comprehensive, and traceable wafer bonding quality evaluation results.

[0095] This method addresses the technical problems of traditional defect analysis, such as incompleteness, unclear defect classification, lack of unified standards for quality judgment, strong subjectivity, and incomplete and untraceable evaluation results. By integrating 3D defect data with 2D imaging, it achieves comprehensive extraction of defect information, improving the comprehensiveness of defect analysis. By combining industry acceptance standards for defect classification and quantitative analysis, it clarifies defect types and exceedance situations, improving the accuracy and standardization of defect analysis and avoiding biases caused by subjective judgment. By analyzing the potential impact of defects, it enhances the scientific rigor of quality evaluation results. By generating complete analysis reports, it achieves traceability of quality evaluation results, facilitating quality review and process optimization. Simultaneously, it standardizes the quality judgment process, reduces the subjectivity of human intervention, and enhances the objectivity and authority of quality evaluation results.

[0096] According to one embodiment of the present invention, the system includes:

[0097] The detection configuration module is used to connect the high-frequency water immersion point focusing ultrasonic probe to the ultrasonic detection system, calibrate it according to preset detection parameters, and obtain the target detection conditions.

[0098] The defect identification module is used to control the probe to emit high-frequency ultrasonic signals according to the target detection conditions. After being focused by the acoustic lens, the signals are incident on the wafer bonding interface. The module receives the defect reflection echoes and converts them into electrical signals to obtain the original defect signals. The module then performs defect feature analysis on the original defect signals to extract the signals and obtain the micro-defect identification signals.

[0099] The analysis and judgment module is used to perform full-area scanning and two-dimensional imaging of the wafer by using the micro-defect identification signal in conjunction with the automatic scanning device to obtain a two-dimensional imaging image. Then, the three-dimensional dimensions of the defect are reconstructed to obtain the three-dimensional data of the defect. The defect feature is analyzed and judged by the three-dimensional data of the defect and the two-dimensional imaging image to obtain the wafer bonding quality evaluation result.

[0100] The working principle and technical effects of the above-mentioned technical solution are as follows: This system adapts the probe to the detection system, and through parameter adjustment and calibration, ensures that the detection parameters are precisely matched with the wafer bonding characteristics, providing a stable and compatible foundation for subsequent detection. Then, using the calibrated target detection conditions, the system controls the probe to emit high-frequency ultrasonic signals. Utilizing the focusing characteristics of an acoustic lens, the sound beam is precisely applied to the wafer bonding interface. Defect reflection echoes are received, and signal conversion and feature extraction are performed to filter out effective signals that can identify minute defects. Finally, by combining with an automatic scanning device, full-area wafer detection is achieved, converting the effective signals into a two-dimensional image and completing three-dimensional size reconstruction. Defects are comprehensively judged based on industry standards, and the final wafer bonding quality evaluation result is output. The entire process revolves around the minute and concealed characteristics of semiconductor wafer bonding defects, achieving fully automated and precise detection from parameter adaptation to quality judgment.

[0101] This system solves the technical problems of poor adaptability, insufficient ability to identify minute defects, and fragmented testing processes in traditional ultrasonic testing methods. It achieves fully automated testing of semiconductor wafer bonding defects, completing parameter calibration, signal acquisition, imaging analysis, and quality assessment without manual intervention, thus improving testing efficiency and consistency. Through high-frequency focusing and signal feature extraction technologies, it enhances the ability to identify minute defects at the wafer bonding interface, avoiding missed or false detections of minute defects and improving testing accuracy. The constructed closed-loop testing process integrates defect identification, location, and quality assessment, improving the comprehensiveness and reliability of testing results. At the same time, it eliminates dependence on imported testing methods, reducing the technical threshold and application cost of semiconductor wafer bonding testing.

[0102] Example 2

[0103] In another embodiment of the present invention, multi-feature fusion discrimination processing is further performed on the original defect signal to further improve the authenticity and anti-interference ability of the micro-defect identification signal, and the post-processing discrimination method of the original defect signal is optimized. Specifically, this embodiment includes the following steps:

[0104] S1. By connecting the high-frequency water immersion point focusing ultrasonic probe to the ultrasonic testing system and calibrating it according to the preset detection parameters, the target detection conditions are obtained.

[0105] S2. The probe emits a high-frequency ultrasonic signal under the control of the target detection conditions. After being focused by the acoustic lens, the signal is incident on the wafer bonding interface. The reflected echo of the defect is received and converted into an electrical signal to obtain the original defect signal. The original defect signal is then subjected to adaptive noise reduction and bandpass filtering, and the initial defect feature signal is extracted.

[0106] S3. The wafer is fully scanned by an automatic scanning device. According to the spatial position information during the scanning process, the initial defect feature signal is associated with its corresponding scanning sampling point to obtain candidate defect signal data with scanning position information.

[0107] S4. For each candidate defect signal corresponding to a scanning sampling point, the echo amplitude response feature, phase change feature and propagation time matching feature are introduced to comprehensively judge whether the candidate defect signal conforms to the time domain characteristics of the real defect echo of the wafer bonding interface, and obtain the confidence value of the candidate defect echo.

[0108] S5. For each candidate defect signal corresponding to each scanning sampling point, based on the continued use of echo amplitude response characteristics and phase change characteristics, target frequency band energy concentration characteristics and neighborhood space consistency characteristics are introduced to comprehensively judge whether the candidate defect signal has the continuous response characteristics of the real defect in the frequency domain and space, and obtain the frequency-space consistency discrimination value.

[0109] S6. Based on the candidate defect echo confidence value and frequency-space consistency discrimination value, the candidate defect signal is finally fused and screened to remove isolated noise spikes, wafer edge reflection interference and local scattering interference, to obtain the optimized micro-defect identification signal, and the optimized micro-defect identification signal is input into the two-dimensional imaging image conversion and three-dimensional size reconstruction process in Example 1.

[0110] In S4, for any scanning sampling point obtained by the automatic scanning device, a candidate echo window corresponding to the wafer bonding interface is extracted from the original defect signal based on the wafer bonding interface depth, probe focal position, and sampling time threshold determined in the target detection conditions. This candidate echo window is used to limit the time range in which the current scanning sampling point may have reflected echoes from micro-voids, delamination edges, or micro-cracks, thereby avoiding the direct inclusion of wafer edge reflections, water-immersed surface reflections, or water-immersed coupled stray signals that are clearly not part of the bonding interface into the defect judgment.

[0111] Within the candidate echo time window, echo amplitude response features, phase change features, and propagation time matching features are extracted.

[0112] The echo amplitude response feature is used to characterize the degree of anomalousness of the echo intensity at the current scan sampling point within the candidate echo time window relative to the local reference echo. Specifically, the peak echo amplitude, effective amplitude, or envelope amplitude at the current scan sampling point within the candidate echo time window can be normalized and compared with the interface reference echo amplitude of adjacent non-obvious regions in the same wafer to obtain the degree of anomalousness of the echo amplitude at the current scan sampling point. The higher the degree of anomalousness of the echo amplitude, the higher the probability that there is reflection enhancement, reflection weakening, or local echo anomaly at the current scan sampling point caused by minor defects.

[0113] The phase abrupt change feature is used to characterize the degree of abrupt change in the echo phase of the current scanning sampling point within the candidate echo time window relative to the echo phase of the adjacent normal bonding interface. Specifically, the phase abrupt change degree of the current scanning sampling point can be obtained by normalizing the phase difference, phase change slope, or phase reversal characteristics between the current candidate echo and the adjacent reference echo. The higher the degree of phase abrupt change, the more significant the change in interfacial acoustic impedance at the current scanning sampling point, which is more consistent with the phase perturbation characteristics generated by microcrack edges, delamination boundaries, or micro-void interfaces.

[0114] The propagation time matching feature is used to characterize the degree of matching between the arrival time of the current candidate echo and the theoretical echo time gate of the wafer bonding interface. Specifically, based on the water immersion sound path, wafer thickness, sound velocity parameters, and bonding interface position determined in the target detection conditions, the theoretical echo time range corresponding to the wafer bonding interface is determined. When the arrival time of the current candidate echo falls within or is close to this theoretical echo time range, its propagation time matching degree is determined to be high. When the current candidate echo is significantly earlier or later than the bonding interface time gate, its propagation time matching degree is determined to be low, thereby reducing the possibility of wafer edge reflection, water immersion surface reflection, or non-target interface reflection being misjudged as defects.

[0115] After obtaining the echo amplitude response characteristics, phase abrupt change characteristics, and propagation time matching characteristics, these three characteristics are weighted and fused to obtain the candidate defect echo credibility value. Specifically, in the weighted fusion process, the echo amplitude response characteristics, phase abrupt change characteristics, and propagation time matching characteristics are assigned corresponding influence weights, and a candidate defect echo credibility discrimination benchmark is set. When the comprehensive result of the weighted fusion of the three characteristics is higher than the candidate defect echo credibility discrimination benchmark, it indicates that the current candidate defect signal is more consistent with the time-domain characteristics of the real defect echo at the wafer bonding interface; when the comprehensive result of the weighted fusion of the three characteristics is lower than the candidate defect echo credibility discrimination benchmark, it indicates that the current candidate defect signal is more likely to originate from non-target reflection or random interference.

[0116] The echo amplitude response characteristics, phase abrupt change characteristics, and propagation time matching characteristics can all be normalized to enable the fusion of features with different dimensions under the same discrimination scale. The influence weights corresponding to each feature and the credibility discrimination benchmark for candidate defect echoes can be determined using repeatability test data from the target detection condition establishment stage in Example 1, reference data of defect-free wafers, and calibration data of known micro-voids, delaminations, and micro-cracks.

[0117] With the above settings, S4 does not simply judge based on the echo amplitude, but rather couples the echo amplitude, phase abrupt change, and propagation time location for judgment. For isolated amplitude spikes formed by water immersion coupled noise, even if their echo amplitude response characteristics are strong, their candidate defect echo confidence value will not be significantly increased because their phase abrupt change characteristics and propagation time matching characteristics are usually unstable. For weak echoes generated by real small defects, even if their amplitude is not high, if their phase abrupt change characteristics are obvious and their propagation time matches the time gate of the wafer bonding interface, a high candidate defect echo confidence value can still be obtained, thereby reducing the situation where real weak defect signals are mistakenly filtered out.

[0118] In S5, based on the confidence value of the candidate defect echo obtained in S4, the frequency band energy distribution and the spatial consistency of adjacent scanning points are further combined to perform frequency-space consistency discrimination on the candidate defect signal.

[0119] Specifically, for each scanning sampling point, the echo amplitude response characteristics and phase change characteristics in S4 are continued to be used, while the target frequency band energy concentration characteristics and neighborhood spatial consistency characteristics are extracted.

[0120] The target frequency band energy concentration feature is used to characterize whether the energy of the current candidate defect signal is concentrated in the target frequency band related to wafer bonding defects. Specifically, the signal within the current candidate echo time window is subjected to frequency domain transformation to obtain its frequency band energy distribution; based on the center frequency and bandwidth of the high-frequency water immersion focused ultrasonic probe and the defect-sensitive frequency band determined by the bandpass filtering method in Example 1, the energy proportion within the target frequency band is statistically analyzed. If the main energy of the current candidate signal is concentrated in the defect-sensitive frequency band, its target frequency band energy concentration is determined to be high; if the energy distribution of the current candidate signal is discrete, or mainly concentrated in the non-target frequency band corresponding to water immersion coupling noise, mechanical disturbance, or edge reflection, its target frequency band energy concentration is determined to be low.

[0121] The neighborhood spatial consistency feature is used to characterize the continuity of defect response between the current scanning sampling point and its neighboring scanning points. Specifically, based on the spiral or matrix scanning path of the automatic scanning device, neighboring scanning sampling points within a preset neighborhood range around the current scanning sampling point are determined; the consistency between the current scanning sampling point and its neighboring scanning sampling points in terms of echo amplitude variation trend, phase change direction, propagation time gate position, and target frequency band energy concentration state is compared. If there are multiple scanning sampling points with similar echo characteristics and similar time gate positions near the current scanning sampling point, its neighborhood spatial consistency is determined to be high; if the current scanning sampling point only exhibits a single-point sudden spike, and the surrounding scanning sampling points do not show a corresponding continuous response, its neighborhood spatial consistency is determined to be low.

[0122] After obtaining the echo amplitude response characteristics, phase change characteristics, target frequency band energy concentration characteristics, and neighborhood spatial consistency characteristics, these characteristics are fused to obtain a frequency-space consistency discrimination value. Specifically, in the fusion process, the echo amplitude response characteristics and phase change characteristics are used as the basic abnormal characteristics of the candidate defect signal, and the target frequency band energy concentration characteristics and neighborhood spatial consistency characteristics are used as the constraint characteristics for suppressing noise interference. When the candidate defect signal simultaneously possesses obvious echo amplitude response, phase change, target frequency band energy concentration, and continuous response at adjacent scan points, a higher frequency-space consistency discrimination value is obtained. When the candidate defect signal only shows an amplitude spike at a single scan sampling point, or its frequency band energy distribution does not conform to the defect sensitive frequency band characteristics, or its surrounding scan sampling points do not show continuous response, a lower frequency-space consistency discrimination value is obtained.

[0123] The echo amplitude response characteristics and phase abrupt change characteristics are simultaneously involved in the reliability judgment of candidate defect echoes in S4 and the frequency-space consistency judgment in S5. This subjectes the same echo amplitude anomaly and the same phase abrupt change characteristic to two different mechanisms: on the one hand, in S4, it is used to determine whether the signal conforms to the time-domain reliability characteristics of defect echoes; on the other hand, in S5, it is used to determine whether the signal has defect response characteristics consistent with the frequency band energy and spatial neighborhood continuity. The target frequency band energy concentration characteristics and the neighborhood spatial consistency characteristics together form a frequency-space consistency constraint, which can effectively reduce the interference of single-point noise spikes, wafer edge reflections, and local material scattering on the defect identification results.

[0124] Through the above settings, S5 can avoid relying solely on a single scan point signal for judgment. For real micro-cavities or layered edges, their defect response usually exhibits a certain continuity between adjacent scan sampling points and has a relatively concentrated energy distribution within the defect-sensitive frequency band, thus resulting in a high frequency-space consistency discrimination value. For water immersion coupled noise or random electronic noise, even if it produces a high amplitude at a certain sampling point, it usually lacks stable frequency band concentration and spatial continuity, thus resulting in a low frequency-space consistency discrimination value.

[0125] In S6, the candidate defect signals are finally fused and screened based on the candidate defect echo confidence value obtained in S4 and the frequency-space consistency discrimination value obtained in S5.

[0126] Specifically, if a scanning sampling point has a high candidate defect echo confidence value but a low frequency-space consistency discrimination value, it indicates that although the signal exhibits certain defect echo characteristics within the current echo time window, it may be an isolated noise spike, edge reflection, or local scattering signal, and should not be directly output as a micro-defect identification signal. If a scanning sampling point has a high frequency-space consistency discrimination value but a low candidate defect echo confidence value, it indicates that the area may have continuous changes in background texture or normal interface reflection changes, which is still insufficient to confirm it as a real micro-defect. Only when both the candidate defect echo confidence value and the frequency-space consistency discrimination value reach a high level, and the difference between them is small, is the candidate defect signal retained as a real micro-defect signal.

[0127] During the fusion screening process, the confidence value of the candidate defect echo and the frequency-space consistency discriminant value are jointly judged, and a consistency constraint is introduced between the two. This consistency constraint means that when both the confidence value and the frequency-space consistency discriminant value of the candidate defect echo are high and the difference between them is small, the final identification priority of the candidate defect signal is increased; when either one is significantly low, or the difference between them is large, the final identification priority of the candidate defect signal is decreased. In this way, the final retained micro-defect identification signals must not only conform to the time-domain characteristics of wafer bonding interface defect echoes, but also meet the constraints of target frequency band energy distribution and spatial continuity of adjacent scan points.

[0128] During the final screening, a threshold for optimizing the identification of minor defects can be preset. For each scanning sampling point, the confidence value of its candidate defect echo, the frequency-space consistency discrimination value, and the consistency constraint result of the two are comprehensively processed to obtain the optimized identification result of the minor defect for that scanning sampling point. When the optimized identification result of the minor defect reaches the threshold for optimizing the identification of minor defects, the candidate defect signal corresponding to that scanning sampling point is retained as the optimized minor defect identification signal. When the optimized identification result of the minor defect does not reach the threshold for optimizing the identification of minor defects, the candidate defect signal is discarded as an interference signal or a low-confidence signal. The threshold for optimizing the identification of minor defects can be determined based on the test results of known defective samples and defect-free samples, or it can be set during the repeatability test of the target detection conditions in Example 1 based on the balance between the false negative rate and the false positive rate.

[0129] After obtaining the optimized micro-defect identification signal, this optimized micro-defect identification signal is used as the input data for the two-dimensional imaging image conversion in Example 1. Subsequently, following the method in Example 1, the optimized micro-defect identification signal is processed into a two-dimensional imaging image using ultrasonic imaging processing to obtain a two-dimensional imaging image; then, the two-dimensional imaging image is combined with a three-dimensional reconstruction algorithm to reconstruct the three-dimensional dimensions of the defects at the wafer bonding interface to obtain reconstructed data; error correction is performed on the reconstructed data to obtain three-dimensional defect data; finally, a comprehensive analysis and judgment of defect characteristics is performed in conjunction with semiconductor wafer bonding industry acceptance standards to obtain the wafer bonding quality evaluation result.

[0130] The working principle and technical effect of the above technical solution are as follows: Based on the calibration of detection parameters, acquisition of original defect signals, adaptive noise reduction and bandpass filtering already completed in Example 1, this embodiment further couples and analyzes the echo amplitude, phase change, propagation time, frequency band energy distribution and spatial consistency of adjacent scanning points of the candidate defect signal. S4 determines whether the candidate signal conforms to the time domain characteristics of the real defect echo at the wafer bonding interface through echo amplitude response characteristics, phase change characteristics and propagation time matching characteristics; S5 determines whether the candidate signal has the continuous response characteristics of the real defect in the frequency domain and space through the target frequency band energy concentration characteristics and neighborhood spatial consistency characteristics; S6 retains signals that simultaneously satisfy the consistency of multiple features by using the consistency constraints of the candidate defect echo confidence value and the frequency-space consistency discrimination value, and eliminates interference signals that only show abnormalities in a single feature.

[0131] This embodiment addresses the technical problem that conventional adaptive noise reduction and bandpass filtering can easily filter out genuine weak defect signals or misidentify isolated noise spikes as defects. Through time-domain reliable discrimination in S4, interference from unbonded interface reflections and edge reflections can be reduced; through frequency-space consistency discrimination in S5, interference from water immersion coupling noise, random electronic noise, and local material scattering can be reduced; and through fusion screening in S6, the ability to detect minute defects can be improved while reducing the false detection rate, making the obtained minute defect identification signals more suitable for subsequent two-dimensional image conversion, three-dimensional size reconstruction, and wafer bonding quality evaluation.

[0132] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers, characterized in that, The method includes: S1. By connecting the high-frequency water immersion point focusing ultrasonic probe to the ultrasonic testing system and calibrating it according to the preset detection parameters, the target detection conditions are obtained. S2. The probe emits a high-frequency ultrasonic signal by controlling the target detection conditions. After being focused by the acoustic lens, the signal is incident on the wafer bonding interface. The reflected echo of the defect is received and converted into an electrical signal to obtain the original defect signal. The original defect signal is then analyzed for defect features to extract the signal and obtain the micro-defect identification signal. S3. By using the micro-defect identification signal in conjunction with the automatic scanning device, the wafer is fully scanned and a two-dimensional image is converted to obtain a two-dimensional image. Then, the three-dimensional dimensions of the defect are reconstructed to obtain the three-dimensional data of the defect. The defect feature is analyzed and judged by the three-dimensional data of the defect and the two-dimensional image to obtain the wafer bonding quality evaluation result.

2. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 1, characterized in that, S1 includes: By connecting a high-frequency water immersion point focusing ultrasonic probe to a high-frequency ultrasonic testing system; Confirm the core detection parameters of the probe and obtain the preset data of the detection parameters; The detection parameters are calibrated by combining the preset detection parameter data with the core parameters of wafer bonding, and the detection parameter calibration data is obtained. The compatibility between the preset detection parameters and wafer bonding parameters is verified and corrected multiple times using a parameter matching method to optimize the parameter combination. At the same time, the repeatability test of the detection parameter calibration data is carried out to obtain the target detection conditions.

3. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 2, characterized in that, The process of confirming the core detection parameters of the probe and obtaining preset detection parameter data includes: The core detection parameters preset during probe manufacturing are read directly to obtain the preset parameter data; Based on the preset parameter data, record the corresponding parameter values, signal feedback intensity, and detection stability data to obtain complete preset detection parameter data.

4. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 2, characterized in that, The method involves multiple verifications and corrections of the compatibility between the preset detection parameters and the wafer bonding parameters using a parameter matching approach, optimizing the parameter combination, and simultaneously performing repeatability tests on the detection parameter calibration data to obtain the target detection conditions, including: The compatibility between the preset detection parameters and the core parameters of wafer bonding is verified multiple times using a parameter matching algorithm to obtain multiple verification data. By comparing the sensitivity and stability of the detection signal under different parameter combinations based on multiple verification data, suitable comparison data is obtained. Based on the adaptation comparison data, parameters with insufficient adaptability are corrected to obtain optimized parameter combination data; By optimizing the parameter combination data, multiple rounds of repeatable tests were conducted on the detection parameter calibration data to obtain multiple rounds of test data; By using multiple rounds of test data, the parameter deviations that occurred during the test were corrected a second time to obtain the target detection conditions.

5. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 1, characterized in that, S2 includes: The high-frequency water immersion point focusing probe is controlled by the target detection conditions to emit a high-frequency ultrasonic signal at a preset frequency. The signal is then processed by the concave acoustic lens built into the probe to obtain the acoustic beam focusing data. The lens curvature parameters are adjusted according to the acoustic beam focusing processing data to control the ultrasonic beam focusing point to fall on the wafer bonding interface, and the reflected echo generated by the defects of the bonding interface is received to obtain the ultrasonic reflected echo signal. The ultrasonic reflected echo signal is converted into a processable electrical signal by a high-frequency piezoelectric transducer built into the probe. The processable electrical signal is amplified to obtain the original defect signal; Defect feature analysis and separation data are obtained by performing defect feature analysis and separation on the original defect signal. Defect feature extraction data is obtained by separating the defect feature analysis data and then performing defect feature extraction data. The validity of the defect feature extraction data is verified to obtain the identification signal of minor defects.

6. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 5, characterized in that, The high-frequency water immersion point focusing probe, controlled by target detection conditions, emits a high-frequency ultrasonic signal at a preset frequency. The signal is then focused by a concave acoustic lens built into the probe to obtain focused acoustic data, including: A high-frequency water immersion point focusing probe emits high-frequency ultrasonic signals to obtain high-frequency ultrasonic transmission signals; The high-frequency ultrasonic signal is transmitted to the concave acoustic lens built into the probe, and the ultrasonic signal is focused to obtain focused data. By adjusting the focusing angle and range of the lens through focusing data processing, sound beam focusing data can be obtained.

7. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 5, characterized in that, The step of performing defect feature analysis and separation on the original defect signal to obtain defect feature analysis and separation data includes: Defect feature analysis data is obtained by using an adaptive noise reduction algorithm and a bandpass filtering method to analyze the original defect signal. By analyzing defect feature data, defect feature information in the original signal is identified, and original signal feature analysis data is obtained. The noise interference defect signal of the original signal is separated by analyzing the original signal feature data to obtain defect feature analysis separation data.

8. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 1, characterized in that, S3 includes: By using minute defect identification signals in conjunction with a high-precision automatic scanning device, the wafer is scanned according to a preset spiral or matrix scanning path to obtain scanning process data. The feature extraction of minute defect identification signals is performed on the scanning process data to obtain scanning feature extraction data, and the intensity change and spatial location information of minute defect identification signals are collected in real time. Two-dimensional imaging images are obtained by converting the scan feature extraction data into two-dimensional imaging images using ultrasound imaging processing methods. The defects at the wafer bonding interface are reconstructed in three dimensions by combining two-dimensional imaging with three-dimensional reconstruction algorithms to obtain reconstruction data. Error correction is performed on the reconstructed data to obtain three-dimensional data of the defects; By combining 3D defect data and 2D imaging with semiconductor wafer bonding industry acceptance standards, a comprehensive analysis and judgment of defect characteristics is conducted to obtain wafer bonding quality evaluation results.

9. The high-frequency water immersion point focusing ultrasonic imaging method for detecting bonding defects in semiconductor wafers according to claim 7, characterized in that, The process involves a comprehensive analysis and judgment of defect characteristics using 3D defect data, 2D imaging, and semiconductor wafer bonding industry acceptance standards to obtain wafer bonding quality evaluation results, including: By combining 3D defect data and 2D imaging images with semiconductor wafer bonding industry acceptance standards, a comprehensive defect feature analysis is conducted on defect anomaly information to obtain comprehensive defect analysis data. By analyzing the comprehensive defect data, classifying the defect levels according to industry acceptance standards and quality requirements, clarifying the degree of impact of defects on wafer bonding quality, generating a complete analysis report containing detailed defect information, judgment criteria, and quality level, and obtaining wafer bonding quality evaluation results.

10. A high-frequency water immersion point focusing ultrasonic imaging system for detecting bonding defects in semiconductor wafers, characterized in that, The system includes: The detection configuration module is used to connect the high-frequency water immersion point focusing ultrasonic probe to the ultrasonic detection system, calibrate it according to preset detection parameters, and obtain the target detection conditions. The defect identification module is used to control the probe to emit high-frequency ultrasonic signals according to the target detection conditions. After being focused by the acoustic lens, the signals are incident on the wafer bonding interface. The module receives the defect reflection echoes and converts them into electrical signals to obtain the original defect signals. The module then performs defect feature analysis on the original defect signals to extract the signals and obtain the micro-defect identification signals. The analysis and judgment module is used to perform full-area scanning and two-dimensional imaging of the wafer by using the micro-defect identification signal in conjunction with the automatic scanning device to obtain a two-dimensional imaging image. Then, the three-dimensional dimensions of the defect are reconstructed to obtain the three-dimensional data of the defect. The defect feature is analyzed and judged by the three-dimensional data of the defect and the two-dimensional imaging image to obtain the wafer bonding quality evaluation result.

Images