A method for failure analysis of an optically pumped laser chip

Abstract

The application discloses a kind of optical pumping laser chip failure analysis methods, it is related to laser detection technical field, comprising: establish uniform coordinate system under the fixture alignment reference, the coordinate mapping relationship of calculation acquisition module is recorded calibration residual index, within the analysis excitation dose budget acquisition scattering microscopic imaging data, heat distribution data and electro-optical parameter data;Data is registered to uniform coordinate system, generates candidate failure point and calculates coordinate uncertainty;Under the error budget constraint, extract interpretable features, form evidence chain and output risk score, failure category and secondary verification instruction;According to instruction, secondary data is collected and back registration, if necessary, destructive mechanism confirmation is sampled and back value label is reproduced. Minute level output is realized, cross-station traceable coordinates and category, reduce sample making pseudo-defects and thermal diffusion error, reduce verification occupation and improve reproducibility.

Description

Technical Field

[0001] This invention relates to the field of laser detection technology, specifically a method for failure analysis of optically pumped laser chips. Background Technology

[0002] In industrial applications, high-power pumped lasers (LDs) are commonly used in systems such as fiber lasers. These systems operate with high current and heat flux, often employing TO, C-Mount, or ceramic packaging for heat dissipation and interconnection. Production line screening and rework analysis frequently require initial screening, classification, and localization within a limited timeframe to improve yield and reliability. Existing processes begin with electro-optical parameters and microscopic observation, incorporating thermal anomaly imaging or luminescence detection when necessary. After sample preparation including unpacking, grinding, substrate removal, or cross-sectioning, the process proceeds to failure localization and mechanism confirmation.

[0003] Chinese patent document CN114252319A (March 29, 2022) discloses a "fault analysis method, sample preparation method and system for semiconductor lasers". This document takes a semiconductor laser containing a chip and a TO bracket as the object. After the encapsulant is cured, the curing agent is sealed with the TO bracket. The device is fixed from the opposite side of the chip on the die bonding platform and ground / polished. The bracket and carrier are removed to expose the chip back. The grinding level is controlled, and images of the grinding process are collected. The grinding state is identified based on a pre-trained image classification model to guide the switching of grinding processes. After the back is exposed, hot spot location analysis is performed on the ground device based on the back to locate the failure site. Chinese patent document CN114486926A (May 13, 2022) discloses a scheme for removing the semiconductor laser chip substrate to expose the epitaxial layer and analyzing the failure site by observing the epitaxial layer in dark field mode of a microscope.

[0004] While the aforementioned existing technologies can achieve sample preparation and positioning of packaged devices, their application in high-power pumped LD scenarios remains limited. For example, taking CN114252319A as an example, encapsulation curing shrinkage, thermal expansion mismatch, and grinding loads can introduce microcracks or delamination between the chip, solder, and carrier, as well as alter the surface morphology, leading to abnormal signals caused by sample preparation. Due to the influence of thermal diffusion and the packaging thermal path, hotspot positioning on the back of the chip is prone to widening and shifting, and the hotspot center may differ from the actual defect location. Failure analysis often involves switching between multiple devices, and alignment errors will prevent subsequent verification or cross-sectional operations from reproducing the positioning results, leading to repeated sample preparation and extended cycle time. Furthermore, for the substrate removal and dark-field observation path in CN114486926A, grinding / etching causes uneven thickness or surface damage, and dark-field imaging reacts strongly to scattering and contamination, reducing interpretation stability and potentially leading to positioning errors and increased costs.

[0005] Therefore, the technical problems of the above-mentioned existing technologies can be summarized as follows: how to achieve reproducible location and mechanism identification of failure sites in packaged high-power semiconductor lasers with minimal damage or controllable sample preparation defects. Summary of the Invention

[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for failure analysis of optically pumped laser chips. The method involves acquiring scattering microscopy data, thermal distribution data, and electro-optic parameter data within the excitation dose budget; registering the data to a unified coordinate system, generating candidate failure points, and calculating coordinate uncertainties; extracting interpretable features under error budget constraints, forming a chain of evidence, and outputting a risk score, failure category, and secondary verification instructions; acquiring secondary data according to the instructions and re-registering the data; and, if necessary, sampling to confirm destructive mechanisms and rewriting true value labels. This method achieves minute-level output, cross-station traceability of coordinates and categories, reduces sample preparation false defects and thermal diffusion errors, and solves the technical problems described in the background art.

[0007] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: A failure analysis method for optically pumped laser chips is provided for packaged optically pumped high-power semiconductor laser devices under test. The method includes: clamping the device in a fixture with an alignment reference; identifying the alignment reference; calculating the coordinate mapping relationship of the navigation imaging module, scattering microscopy module, and thermal imaging module to a unified coordinate system and recording the calibration residual index; acquiring scattering microscopy imaging data, thermal distribution data, and electro-optic parameter data under controlled excitation; registering the data to a unified coordinate system; generating candidate failure points based on thermal anomalies, scattering anomalies, and electro-optic parameter anomalies; and calculating the coordinate uncertainty of the candidate failure points based on the calibration residual index and the effect of thermal diffusion. Thermal features, scattering features, and electro-optic features are extracted to generate risk scores and failure categories. Secondary verification commands are triggered based on risk scores and coordinate uncertainties. In response to secondary verification commands, secondary data is collected and registered to a unified coordinate system, and suspected coordinates, coordinate uncertainties, risk scores, and failure categories are output.

[0008] Furthermore, when acquiring scattering microscopy imaging data, thermal distribution data, and electro-optic parameter data, the device clamping posture, acquisition timestamp, excitation conditions, and module status are recorded simultaneously. The records are associated with the coordinate mapping relationship and calibration residual index and written into the process metadata. The process metadata is output along with the registration results.

[0009] Furthermore, the controlled stimulus sets the analysis stimulus dose budget based on the preset test formula and monitors the temperature rise and power during the acquisition of thermal distribution data; when the analysis stimulus dose budget is exceeded, the stimulus conditions are adjusted, including reducing the stimulus intensity and shortening the stimulus time, and the data is reacquired, and the over-budget event, the adjusted stimulus conditions, and the confidence degradation flag are written into the process metadata.

[0010] Furthermore, when scattering microscopy imaging data, thermal distribution data and electro-optic parameter data are registered to a unified coordinate system, a coordinate version identifier is generated and associated with the clamping posture; when the clamping posture changes or the sample preparation stage changes, a new coordinate version record is established and the historical coordinate version record is retained.

[0011] Furthermore, generating candidate failure points includes: extracting hotspot regions from thermal distribution data and determining hotspot centers and hotspot boundaries; extracting scattering anomaly connected domains from scattering microscopy data and determining scattering centers; calculating the thermal dissipation distance between the hotspot center and the scattering center in a unified coordinate system and generating spatial correlation weights; and filtering candidate failure points based on the spatial correlation weights.

[0012] Furthermore, the coordinate uncertainty is calculated by the error budget, which includes the calibration residual index, the pixel quantization error corresponding to the grid spacing of the unified coordinate system, and the hot spot widening term characterized by the area and perimeter of the hot spot region. The coordinate uncertainty is output along with the candidate failure point.

[0013] Furthermore, after extracting thermal features, scattering features, and electro-optic features, credibility grading is performed based on the credibility downgrade marker in the process metadata, and credibility weights are generated. The thermal features, scattering features, and electro-optic features are written into the feature record. When the feature record meets the conflict conditions of the conflict rule base, the corresponding candidate failure point is marked as requiring secondary verification, and a secondary verification instruction is generated based on the risk score and coordinate uncertainty, with a verification code attached.

[0014] Furthermore, in response to the secondary verification command, the device is partially exposed while maintaining the identifiable alignment reference. Subsequently, lock-in thermal imaging lightweight acquisition data, photoluminescence observation acquisition data, and photon emission microscopy acquisition data are collected as secondary data. The secondary data are then registered to a unified coordinate system to update the risk score, failure category, and coordinate uncertainty.

[0015] Furthermore, when the risk score is high and the coordinate uncertainty does not converge after secondary verification, devices are selected for destructive mechanism confirmation according to the preset sampling strategy. The destructive mechanism confirmation process includes cross-sectional sample preparation, electron beam induced current observation and acquisition, and focused ion beam cutting. The obtained truth labels are then written back to update the trigger threshold, confidence grading rules, and thermal diffusion uncertainty parameters.

[0016] Furthermore, the thermal anomaly extraction employs dual-threshold segmentation to identify hotspot regions in the thermal distribution data and output the hotspot center and hotspot boundary. The second threshold of the dual-threshold segmentation is higher than the first threshold. The secondary verification instruction is generated based on the trigger threshold set by the preset test formula. The trigger threshold includes the threshold of risk score and the threshold of coordinate uncertainty, and the risk score and coordinate uncertainty are jointly determined.

[0017] (III) Beneficial Effects This invention provides a method for failure analysis of optically pumped laser chips, which has the following advantages: A unified coordinate system is established based on the fixture alignment reference. The coordinate mapping relationship between each acquisition module and the unified coordinate system is calculated. The residual index and process metadata are calibrated. Within the analysis excitation dose budget, scattering microscopy data, thermal distribution data, and electro-optic parameter data are acquired to ensure that the analysis is source-consistent, controlled, and traceable. Multimodal data are registered to the unified coordinate system to generate a set of candidate failure points. Coordinate version identifiers are used to manage clamping posture or stage changes. Coordinate uncertainty is output for candidate failure points, and the candidate coordinates are repositioned on the workstation and written back to the cross-sectional positioning chain.

[0018] By incorporating calibration residuals, pixel quantization errors, and thermal diffusion effects into the same coordinate uncertainty propagation framework under the constraint of error budget, thermal features, scattering features, and electro-optic features are extracted to form an interpretable chain of evidence. Risk scores and failure categories are output, and consistency classification is performed on evidence conflicts, making the judgment a reproducible piece of evidence rather than a single modality.

[0019] By configuring risk scores and coordinate uncertainties together on the secondary verification instructions, and ranking candidates based on the expected benefits and action costs of convergence of candidate verification actions and increase in confidence of failure categories, high-cost verification is only used for high-risk or high-uncertainty samples, thereby matching the production line screening rhythm.

[0020] By responding to secondary verification commands, collecting secondary data and re-registering it to the same coordinate system to update candidate coordinates, coordinate uncertainty, risk score and failure category, and confirming the destructive mechanism only for a small number of samples when closure is still not possible, generating truth label write-back thresholds and rules, so that the process forms a root cause closed loop and consistent engineering reusability. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the structure of a failure analysis system for an optically pumped laser chip according to the present invention; Figure 2 This is a schematic diagram illustrating the establishment of alignment references and cross-module coordinate mapping in this invention; Figure 3 This is a schematic diagram illustrating the unified coordinate registration and candidate failure point generation of the present invention. Figure 4 This is a schematic diagram illustrating the local evidence alignment and feature record generation for candidate failure points in this invention. Figure 5 This is a confidence level and trigger-based two-level verification decision diagram for the present invention; Figure 6 This is a schematic diagram illustrating the typical failure mode discrimination of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1-6 This invention provides a method for failure analysis of optically pumped laser chips, comprising: Step 1: Under the condition that the packaging boundary and thermal path of the packaged optically pumped high-power semiconductor laser device are significantly different, first establish the alignment reference and cross-module coordinate mapping relationship, and then complete the synchronous acquisition of scattering microscopy imaging data, thermal distribution data and electro-optic parameter data under the constraint of excitation dose budget analysis.

[0024] In this approach, geometric consistency and excitation consistency are treated as the two ends of the same chain: geometric consistency unifies the coordinates of the fields of view of each module to a unified coordinate system through alignment references, and solidifies the residuals after mapping solution into calibration residual indices; excitation consistency constrains the driving waveform within a repeatable range by analyzing the excitation dose budget, and writes the dose records and sampling timing sequences into the process metadata. Geometric consistency ensures that different modes see the same location, while excitation consistency ensures that different samples are observed under the same excitation framework. The synergy between the two lies in the fact that the calibration residual index determines the initial magnitude of the subsequent error budget, the analysis of the excitation dose budget determines the usable boundary of thermal distribution data and the comparable boundary of electro-optic parameter data, and the process metadata binds these two types of boundaries to the same acquisition event.

[0025] Specifically, the failure analysis system includes: a fixture equipped with an alignment reference, a navigation imaging module, a scattering microscopy module, a thermal imaging module, an electro-optic parameter acquisition module, and a data processing unit in the form of an industrial computer. The data processing unit is electrically connected to each module and to a power supply / pulse source. The data processing unit outputs synchronous trigger signals to the scattering microscopy module, thermal imaging module, and electro-optic parameter acquisition module via a trigger distributor, ensuring that the three types of sampling are performed around the same excitation window. The fixture is fixed to the stage, and the stage's limiting structure constrains the device's attitude, ensuring the alignment reference remains repeatable within a single acquisition cycle. The stage remains locked during acquisition to prevent micro-displacement during module switching.

[0026] Metal reflections, solder overflow, ceramic textures, and cap structures in packaged devices can cause visible boundaries to differ across modes. If a random texture of a single mode is used directly as an anchor point, coordinate drift can easily occur after switching modes. Using a set of alignment reference points as the core object, the navigation imaging module, scattering microscopy module, and thermal imaging module establish coordinate mapping relationships around the same set of reference points. The mapping quality is recorded as a calibration residual index, providing a starting point for subsequent error estimation.

[0027] Therefore, after clamping and establishing a reference point set, alignment reference maps used for calibration are collected respectively, and the coordinate mapping relationship from the module to the unified coordinate system is solved; calibration residual records are generated, and it is determined whether the threshold is exceeded. If the threshold is exceeded, re-alignment and re-collection are performed until the calibration residual records meet the conditions for proceeding to the next step.

[0028] A reference point set is constructed using a combination of a fixture reference and reversible markings on the device surface to avoid using natural defects and textures as the sole anchor point. The fixture reference is preferably an etched geometric shape, with the etched area located on the plane of the fixture close to the device, forming a fixed angle with the fixture's limiting edge. The fixture material is preferably hard stainless steel or alumina ceramic to maintain sharp edges. The reversible markings on the device surface are preferably thin film labels, composed of a polyimide substrate layer and a soluble pressure-sensitive adhesive layer. Two orthogonal thin lines and a circular center mark are formed on the polyimide substrate layer, with the geometric center of the circular center mark serving as the reference point. To improve the anti-scaling capability of geometric recognition, the width of the thin lines is set to 1 / 10 to 1 / 5 of the diameter of the circular center mark, allowing the navigation imaging module to determine the center point through geometric proportions at different magnifications. The soluble pressure-sensitive adhesive layer uses a formulation that can be dissolved by electronic-grade isopropanol, allowing it to be wiped away after acquisition without leaving residue on the packaging surface. The film label's attachment area is limited to non-light-emitting areas and non-pad areas to avoid obstructing light emission and affecting electrical connections.

[0029] Taking TO package or C-mount package as an example: The operator places the device into the fixture limiting slot and tightens the pressure plate to make the package shell fit against the fixture plane; then, the non-light-emitting area is wiped twice with fibrous wiping paper soaked in electronic grade isopropyl alcohol and dried; then, the film label is attached to the non-light-emitting area with the fine line parallel to the fixture limiting edge; after attachment, a constant pressing force is applied with a limiting pressing head, the pressing force is set to 5 Newtons to 30 Newtons, and held for 30 to 90 seconds to fully wet the pressure-sensitive adhesive layer. After completion, the navigation imaging module acquires navigation images, and the data processing unit extracts a set of reference points from the images. The set of reference points includes at least the fixture corner point, the center point of the circular center mark, and the two endpoints of the package geometric boundary. After acquisition, the operator wets the pressure-sensitive adhesive layer along the edge of the film label with electronic grade isopropyl alcohol and slowly peels off the film label along the guide corner, then wipes it a second time with fibrous wiping paper to remove residual solvent, thereby restoring the package surface state.

[0030] In use, the reference point set serves as a common identification object for the three imaging modules, avoiding dependence on random textures. The thin-film tag is a reversible and invasive marker that can be removed after acquisition, avoiding long-term impact on the encapsulation surface. The reference point set serves as the determining input for subsequent coordinate mapping relationships, providing anchor points for cross-modal switching.

[0031] The preferred coordinate mapping relationship is a two-dimensional affine transformation, where the coordinate mapping relationship from the pixel coordinates of each acquisition module to the coordinates of the unified coordinate system is represented by a two-dimensional affine transformation: ; ; Affine parameters : Scaling / rotation contribution coefficient from pixel coordinates to uniform coordinates, with a finite range of real numbers, used to describe the linear effect on ; Affine parameter : Shearing contribution coefficient from pixel coordinates to uniform coordinates, with a finite range of real numbers, used to describe the linear effect on ; Affine parameter Translation term in unified coordinates, with a finite range of real numbers, used to describe the offset of the coordinate origin; Affine parameters : Shearing contribution coefficient from pixel coordinates to uniform coordinates, with a finite range of real numbers, used to describe the linear effect of ; Affine parameter : Scaling / rotation contribution coefficient from pixel coordinates to uniform coordinates, with a finite range of real numbers, used to describe the linear effect on ; Affine parameter Translation term in unified coordinates, with a finite range of real numbers, used to describe the offset of the coordinate origin; Pixel coordinates : The horizontal pixel coordinates of the image acquired by the acquisition module, with values ​​ranging from real numbers to integers, used to index pixel positions; pixel coordinates : Vertical pixel coordinates of the image acquired by the acquisition module, with values ​​ranging from real numbers to integers, used to index pixel positions; Unified coordinates : A unified coordinate system with lateral coordinates, taking values ​​in the range of finite real numbers, used for cross-modal space alignment; unified coordinates : A unified coordinate system with vertical coordinates, taking values ​​in the range of finite real numbers, used for cross-modal space alignment; The affine parameters are obtained by solving the reference point set, which contains at least three non-collinear point pairs. The data processing unit takes the pixel coordinates of the reference point set and the coordinates of the unified coordinate system as input, and solves the above six affine parameters using least squares or weighted least squares with truncation loss.

[0032] After obtaining the set of reference points, the data processing unit drives the navigation imaging module, the scattering microscopy module, and the thermal imaging module to image the same set of alignment references: the navigation imaging module covers the entire set of reference points, the scattering microscopy module covers the film label area and the local area of ​​the fixture reference, and the thermal imaging module covers the corner area of ​​the fixture reference.

[0033] Since the pixel size of the thermal imaging module is usually larger than that of the scattering microscopy module, the corner neighborhood is first determined in the thermal imaging module image by local maximum suppression, and then the corner center is determined by sub-pixel interpolation; in the scattering microscopy module image, the intersection neighborhood of the thin lines is determined by the consistency of the edge gradient direction, and then the center of the circular center marker is determined by the centroid method.

[0034] The data processing unit then uses a solver with truncation loss to solve for the coordinate mapping relationship, prioritizing the Gauss-Newton iteration method and reducing the weight of outlier reference points. The obtained coordinate mapping relationship is used to resample the coordinate transformation image using bilinear interpolation, so that subsequent modal data can correspond pixel by pixel in the same coordinate system. The termination condition is that the change in the calibration residual index is less than the set value in two consecutive iterations or the number of iterations is within the upper limit. After termination, the calibration residual is recorded.

[0035] The calibration residual index is preferably calculated using the following formula: ; Where: Calibration residual index Robust root mean square residuals for cross-module coordinate mapping, with a value range of [value range missing]. The dimensions are consistent with the deviation from the reference point, and it is used as the benchmark scale for subsequent coordinate uncertainty; the number of reference points The number of reference points used in the calculation, with a range of values. This is used to ensure that affine solution and residual statistics are feasible; index variables Reference point number, with a value range of: This is used to index residual items one by one; Reference point deviation : The distance between the observed position and the mapped predicted position of the i-th reference point, with a value range of . It is obtained by calculating the Euclidean distance in a unified coordinate system; truncation function The bias cost function, defined as a piecewise function, outputs a quantity in the form of squared distance, used to suppress the dominance of outlier bias on the residuals; the bias independent variable... : Input to the cost function, with a range of values. ; Cutoff threshold Outlier suppression inflection threshold, with a value range of: The preferred grid spacing is 1.5 to 1.5 times the grid spacing of the unified coordinate system, used to switch outlier bias from quadratic penalty to linear penalty; the grid spacing of the unified coordinate system... Spatial sampling interval of the resampling grid within the unified coordinate system, with a value range of [value range missing]. The effective spatial resolution of the thermal imaging module or navigation imaging module is used to unify the scale and lower threshold. When the calibration residual index is greater than the threshold, the data processing unit sends a realignment command to the operator, the platform returns to the alignment position, and the alignment reference image is reacquired and the coordinate mapping relationship is solved; when the calibration residual index is equal to the threshold, the data processing unit saves the coordinate mapping relationship and calibration residual of this round.

[0036] When used, the calibration residual index transforms the implicit empirical judgment of mapping quality into an executable criterion. The truncation loss and outlier weighting reduce the impact of abnormal reference points, making the coordinate mapping relationship more stable. After the image resampling rules are clear, subsequent steps can directly reuse the same coordinate transformation for multimodal alignment.

[0037] High-power pumped LDs exhibit heat accumulation and changes in current-carrying state under excitation. If the excitation waveform is inconsistent across samples, the thermal distribution data and electro-optical parameter data are difficult to correlate. Simultaneously, the thermal imaging module is sensitive to the emissivity of the packaged surface, and metallic reflections can introduce pseudo-thermal textures. To address this, we analyze the excitation dose budget to uniformly constrain the excitation waveform, stabilize the thermal imaging boundary conditions through reversible invasive surface treatment, and then synchronously trigger the acquisition of three types of data and solidify the process metadata.

[0038] Therefore, the excitation waveform is first generated based on the test formula, and the excitation dose budget is calculated and analyzed. If the limit is exceeded, the waveform is changed and recalculated. When the limit is exceeded, reversible invasive surface treatment is performed and synchronous acquisition is initiated. When the thermal image is saturated or the electrical parameters are abnormal, the degradation and reacquisition branches are executed and the process metadata is written.

[0039] The excitation effect is described as an energy injection with thermal memory, avoiding the use of peak current to characterize the excitation intensity. The data processing unit is coupled to the power supply / pulse source and the electro-optic parameter acquisition module. It first sends out the initial waveform parameters, and then the electro-optic parameter acquisition module reads back the voltage and current waveforms and calculates and analyzes the excitation dose budget. If the limit is exceeded, the pulse width and duty cycle are adjusted and read back again until the budget is met. To match the thermal memory weight with the package structure, the thermal relaxation time constant is obtained from a short pulse test excitation: after the test excitation, the electro-optic parameter acquisition module continues to sample the voltage and current waveforms, and the data processing unit uses the time scale of the waveform falling back to near the steady state as the initial value of the thermal relaxation time constant, and keeps this value unchanged in subsequent batches.

[0040] The preferred excitation dose budget for analysis is defined by the following formula: ; In the formula: Analysis of excitation dose budget Considering the equivalent energy injection amount after thermal memory, the value is taken as follows: Used to compare with a preset dose upper limit and constrain the excitation waveform; excitation duration : Total duration of excitation within a single acquisition window, with a value of , used to determine the integration interval and reflect the cycle time constraint; Excitation current waveform The driving current that varies with time, taking values ​​of It is used to characterize the injection intensity and, together with the voltage, determines the instantaneous power; excitation voltage waveform The terminal voltage varies with time, and its value is... Used to reflect the conduction state and together with the current to determine the instantaneous power; thermal relaxation time constant. The equivalent thermal relaxation parameters of the package and the chip, with a value range of [value range missing]. Used to set the decay scale of hot memory weights; time variable : Independent variable of integration, values , used to index waveforms and weights; When used, the excitation dose budget is analyzed to incorporate the risk of thermal accumulation on the same scale, reduce excitation drift between samples, and complete the budget constraint before entering synchronous acquisition, so that the subsequent thermal distribution data and electro-optic parameter data can correspond under the same excitation framework.

[0041] Furthermore, without altering the internal structure of the device, a reversible invasive surface treatment is applied to the packaging surface of the thermal imaging field of view to stabilize the emissivity and boundaries.

[0042] Preferably, after completing the aforementioned steps, the operator applies a peelable blackened film: the blackened film consists of a polyimide substrate layer, a carbon black filler resin blackening layer, and a soluble pressure-sensitive adhesive layer. The thickness of the polyimide substrate layer is set to 20 to 80 micrometers, the carbon black filler mass fraction of the blackening layer is set to 15% to 25%, and the thickness of the pressure-sensitive adhesive layer is set to 5 to 20 micrometers. Before application, the film is wiped with deionized water and dried, then degreased with electronic-grade isopropanol. During application, a constant pressing pressure is applied using a pressure head with a limiting setting, the pressing pressure being set to 5 to 30 Newtons, and maintained for 30 to 90 seconds. To prevent the film edges from lifting and entering the thermal imaging field of view, after application, a ceramic scriber is used to gently press along the edges to form a continuous adhesive strip, and a guide angle is reserved at one corner for easy peeling after acquisition.

[0043] Synchronous acquisition is accomplished by the trigger distributor: the data processing unit outputs a pre-trigger to put the three modules into a waiting state, and then outputs a main trigger to start excitation and sampling; the electro-optic parameter acquisition module starts sampling and generates voltage and current waveform records at the main trigger edge; the scattering microscopy module completes one exposure after a preset exposure delay following the main trigger edge to obtain scattering microscopy imaging data; and the thermal imaging module acquires thermal distribution data sequences at a fixed frame rate after the main trigger edge and outputs the frame timestamp associated with the main trigger. The data processing unit writes the clamping posture, main trigger timestamp, exposure delay, frame rate, sampling rate, and dose record corresponding to the analysis excitation dose budget into the process metadata, and outputs the three types of data together with the coordinate mapping relationship locked in the aforementioned steps.

[0044] To facilitate direct reading in subsequent steps, scattering microscopy data is saved as image frames with a master trigger timestamp, thermal distribution data is saved as a frame sequence with a frame timestamp sequence, and electro-optic parameter data is saved as a synchronous sampling sequence with sampling start and end timestamps. The three types of data are associated with the same acquisition event number.

[0045] Abnormal branch: When the heat distribution becomes saturated, first check the coverage of the blackened film. If necessary, reattach it and collect the excitation waveform again. When the electro-optic parameter acquisition module detects a sudden change in the voltage and current waveform, it identifies this round of data as having a reduced credibility level and writes the abnormal type into the process metadata for fusion, resulting in a reduced weight or secondary verification.

[0046] In use, reversible invasive surface treatment stabilizes the boundary conditions of thermal imaging and reduces pseudo-thermal textures introduced by reflection. Hardware synchronous triggering enables the three types of data to establish a correspondence around the same excitation window.

[0047] Furthermore, the alignment reference is not limited to film labels, but can also use peelable ink dots or jig-blasted diffuse reflection markings; the peelable black film is not limited to the attachment method, but can also use a peelable black spray coating and stand at 60 degrees Celsius for 10 minutes to form a film layer; the synchronous trigger is not limited to hardware pulse trigger, but can also use a unified timestamp alignment method and record the time synchronization error in the process metadata for subsequent error budgeting.

[0048] Step 2: Map the coordinates from Step 1 to the calibration residual index. By adjusting the scattering microscopy data, thermal distribution data, and electro-optic parameter data to a unified coordinate system, a set of candidate failure points and initial coordinate uncertainty are obtained, providing a directly callable object for error budget constraint inference and trigger decision-making in step three.

[0049] Among them, thermal distribution data describes the spatial distribution of energy dissipation, scattering microscopy data describes the scattering anomaly morphology near the packaging surface and light-emitting end, and electro-optic parameter data describes the changes in conduction and radiation states under the constraint of the analysis excitation dose budget. The observation geometry and sampling scales of the three types of data are different. If their respective pixel coordinates are directly compared, the hotspot center and scattering anomaly points will be spatially misaligned, leading to the subsequent decomposition of the same failure into multiple locations. This step uses a unified coordinate system as the sole spatial semantic carrier, takes the coordinate mapping relationship from step one as the transformation operator, and uses the calibration residual index as the transformation confidence boundary. Then, a candidate failure point set is generated using multi-source triggering-spatial correlation-availability check, and the initial coordinate uncertainty is output, enabling step three to propagate errors and allocate verification actions within the same scale.

[0050] This step is performed by the data processing unit. The data processing unit reads the coordinate mapping relationship and calibrates the residual index from step one. Analyze the excitation dose budget The system collects timestamps and clamping postures, and reads scattering microscopy imaging data frames, thermal distribution data frame sequences, and electro-optic parameter data sequences. After resampling on a unified coordinate system grid, it outputs a unified coordinate data packet. The unified coordinate data packet contains coordinate version identifiers, a set of candidate failure points, and initial coordinate uncertainty. It also saves grid parameters and extraction condition fields for direct reading in step three.

[0051] Packaging devices exhibit attitude differences after repeated clamping, flipping, and cross-station handling. Simultaneously, the pixel density of scattering microscopy data and thermal distribution data differs; if spatial positions are not interpreted on the same grid, spatial correlation will mistakenly interpret resolution differences as positional differences. By simultaneously locking both attitude semantics and grid scale, coordinate mapping relationships are invoked within controllable boundaries, and the attitude state is written into the coordinate version identifier.

[0052] Therefore, the data processing unit first generates a coordinate version identifier based on the clamping posture, and then projects the multimodal raw data onto a unified coordinate system grid; subsequently, it uses the calibration residual index to determine the registration threshold. If the threshold is not met, it triggers re-clamping or recalibration of the branch; if the threshold is met, it outputs a unified coordinate data packet.

[0053] The attitude status is fixed as a physical identifier that can be read by the navigation imaging module, avoiding the loss of attitude data when switching workstations due to relying solely on text records. The version identifier uses a polyimide substrate with a thickness of 25 to 75 micrometers. The surface is printed with directional arrows and intersecting short lines. The directional arrows are defined as the positive axis indicators of a unified coordinate system, and the intersection of the intersecting short lines is defined as the attitude reading point. The back is coated with a water-soluble pressure-sensitive adhesive layer, which uses polyvinyl alcohol as a film-forming agent and adds glycerin. The mass fraction of polyvinyl alcohol is 8% to 12%, the mass fraction of glycerin is 1% to 3%, and the balance is deionized water. The coating thickness is 10 to 30 micrometers, and it is dried at 50 to 70 degrees Celsius for 5 to 15 minutes to form an adhesive film.

[0054] After clamping, attach the version identification sheet to the non-light-emitting side of the packaging shell, aligning the directional arrow with the clamp's limiting edge. Apply a pressure of 3 to 15 Newtons and hold for 20 to 60 seconds. The navigation imaging module acquires the image of the identification sheet, and the data processing unit analyzes the direction of the directional arrow and the position of the attitude reading point, generating an attitude status field and writing the coordinate version identification. If removal is required, wipe the edges with deionized water, peel off the adhesive film, and wipe the film clean.

[0055] When in use, the posture status is fixed by a physical identification piece and can be repeatedly read. After being transported across workstations, it remains consistent with the direction of the unified coordinate system. The water-soluble pressure-sensitive adhesive layer makes the identification piece removable, reducing residual impact on the outer surface of the package and maintaining the continuity of subsequent processes.

[0056] A unified coordinate system grid is used as the sole scale for subsequent spatial calculations, enabling direct comparison of thermal distribution data and scattering microscopy data under the same grid index. The grid spacing is preferably set according to the pixel size of the thermal imaging module under the current lens. When mapping scattering microscopy data to the unified coordinate system grid, a reflective mask is first formed for the saturated reflective areas, followed by bilateral filtering outside the mask. Then, bilinear interpolation is used to resample the data to the unified coordinate system grid while preserving the original grayscale dynamic range. Thermal distribution data is aligned with the excitation window of the electro-optical parameter data according to the acquisition timestamp before being transformed to the unified coordinate system grid, forming a thermal distribution data subsequence bound to the excitation window. The registration threshold is determined by the calibration residual index; if the threshold is exceeded, the unified coordinate data packet is not output, and a re-clamping or recalibration branch is triggered; if the threshold is met, the grid parameters and resampling strategy are locked, and a coordinate version identifier is written.

[0057] When in use, a unified coordinate system grid unifies the spatial scale of the original multimodal data, and subsequent spatial association is completed under the same grid index; the calibration residual index is used as the registration threshold to clarify the usable boundary of the coordinate mapping relationship, and to avoid the calibration deviation being propagated as candidate coordinate deviation.

[0058] Thermal distribution data is affected by the encapsulation thermal path, resulting in hotspot broadening; scattering microscopy data is affected by particle contamination and edge reflection, resulting in pseudo-scattering; and electro-optic parameter data can only provide anomaly windows without location information. Therefore, the spatial correlation between thermal and scattering anomalies is used as the main thread, and the anomaly windows of the electro-optic parameter data are used to limit frame selection, so that the candidate failure point set has both spatial and temporal semantics. Then, the initial coordinate uncertainty is output to support the error budget propagation in step three.

[0059] Therefore, the data processing unit can obtain thermal distribution data frames and electro-optic parameter data anomaly windows from the thermal distribution data subsequence, thermal anomaly contour objects, and scattering anomaly objects from scattering microscopic imaging data; calculate spatial correlation weights to obtain a set of candidate failure points; finally, give the initial coordinate uncertainty of each candidate failure point and make reversible invasive positioning marks when cross-station repositioning is required.

[0060] Both thermal anomalies and scattering anomalies are represented as contourizable region objects, enabling spatial comparisons to consider not only point positions but also boundaries and shapes. Thermal anomaly extraction begins with background subtraction, using thermal distribution data frames of the same device without excitation, bound to the attitude state field. After subtraction, a dual-threshold segmentation of the thermal anomaly region is employed: the first threshold excludes overall temperature rise, and the second threshold locks in local hotspot cores. The search interval for both thresholds is limited by the analysis excitation dose budget and package type fields, ensuring that differences in excitation intensity across different batches do not alter the segmentation semantics. After segmentation, the outer contour is extracted, and the hotspot perimeter and area are calculated. Scattering anomaly extraction first creates a reflective mask, then performs connected component analysis outside the mask. Connected component filtering conditions are given by area and aspect ratio, distinguishing point scattering from line scattering as different source fields. Both thermal anomaly contour objects and scattering anomaly objects have extraction condition fields within a unified coordinate system for subsequent consistency checks.

[0061] In practice, thermal anomalies and scattering anomalies are treated as contour objects. Spatial correlation utilizes both position and shape information to reduce the inclusion of single-pixel noise in the candidate set. The coupling of dual thresholds with the analysis excitation dose budget introduces excitation intensity differences into the segmentation boundary conditions, ensuring that the thermal anomaly extraction rules do not change with operating conditions.

[0062] Then, the hotspot center point is defined first using the geometric centroid of the thermal anomaly profile object, and the scattering center point is defined using the centroid or midpoint of the scattering anomaly object. The heat dissipation distance and initial coordinate uncertainty are calculated to obtain the spatial correlation weight. When the threshold is met, the weighted position of the hotspot center point and the scattering center point is set as candidate coordinates and added to the candidate failure point set. The spatial correlation weight is calculated using the following formula: ; Where: Spatial correlation weight The correlation strength coefficient between thermal anomalies and scattering anomalies, with values ​​ranging from [value missing]. Used to determine whether two abnormal objects belong to the same candidate failure point; heat dissipation distance : The distance between the center of the hotspot and the center of the scattering, with a value of Used to measure spatial deviation; initial coordinate uncertainty : The uncertainty scale for the initial location of the candidate failure point, with a value of This is used to normalize the distance and reflect the uncertainty introduced by registration and hotspot broadening; The initial coordinate uncertainty is calculated using the following formula, whereby the calibration residual index and the hotspot shape factor jointly determine the uncertainty scale: ; Where: pi constant Pi, a constant with a value of 3.14159, is used to construct the dimensionless shape factor; initial coordinate uncertainty. : The uncertainty scale for the initial location of the candidate failure point, with values ​​ranging from 1 to 2. This is used as input for spatial correlation normalization and subsequent triggering decisions; calibration residual index Cross-module coordinate mapping residual strength, with a value of Used to provide the initial order of magnitude of uncertainty; shape weighting coefficient : Shape factor gain coefficient, with a value range of This is used to adjust the intensity of the influence of hotspot shape on uncertainty; Hotspot Perimeter : Perimeter of the outer contour of the thermal anomaly, with a value range of . , used to characterize the complexity of hotspot boundaries; the perimeter of a hotspot is calculated from the length of the boundary chain code of the outer contour of the thermal anomaly on the unified coordinate system grid, and the boundary chain code is composed of the sequence of outer contour points extracted from the connected components; Hotspot area Area of ​​the thermal anomaly region, with a value range of [value missing]. This is used to characterize the coverage area of ​​hotspots; the hotspot area is calculated by multiplying the number of grid cells within the thermal anomaly region, or by accumulating the pixel areas. The shape weight coefficient is given by the test formula, preferably [value missing]. When the hot spot is close to a circle, the uncertainty is mainly determined by the shape factor. When the hot spot is significantly elongated, the uncertainty increases with the shape factor to avoid passing overly optimistic coordinate outputs to step three under the packaging thermal path with significant thermal diffusion.

[0063] When candidate coordinates require repositioning across workstations, the data processing unit is coupled to a positioning point printing device to perform reversible invasive positioning marking. The positioning point printing device uses a micro-dispensing valve and a fine needle with an inner diameter of 50 to 150 micrometers. The dispensing material is water-soluble positioning ink, which uses polyvinylpyrrolidone (PVP) as a film-forming agent, titanium dioxide microparticles as a scattering filler, and deionized water as a solvent. The mass fraction of PPVP is 3% to 6%, and the mass fraction of titanium dioxide microparticles is 5% to 12%. After mixing, the mixture is magnetically stirred for 10 minutes and allowed to stand for 5 minutes to remove bubbles. The dispensing valve dispenses positioning ink dots at the candidate coordinate projection position, with a dot diameter of 150 to 500 micrometers. After the navigation imaging module confirms that the positioning ink dots are aligned with the coordinate version identifier, it writes the re-image into a unified coordinate data package. The positioning ink dots are removed by wetting with deionized water and wiping them away.

[0064] As a supplement: the hot spot center point is taken as the geometric centroid of the thermal anomaly contour object, and the scattering center point is taken as the geometric centroid of the scattering anomaly connected domain; the thermal dissipation distance is the Euclidean distance between the two centroids in a unified coordinate system.

[0065] In use, the spatial correlation weights jointly constrain the heat-dissipation relationship with the heat dissipation distance and the initial coordinate uncertainty. The correlation conditions of the candidate failure point set are clear and facilitate subsequent consistency judgment. The initial coordinate uncertainty couples the calibration residual index with the hot spot shape factor, so that the influence of hot spot widening on the candidate position enters the error budget propagation in step three.

[0066] After completing step one, the operator attaches the version identification sheet to the non-light-emitting side of the package and presses it firmly. The navigation imaging module acquires an image of the identification sheet, and the data processing unit generates a coordinate version identifier and locks the unified coordinate system grid parameters. The data processing unit resamples the scattering microscopy data and thermal distribution data to the unified coordinate system grid and uses the anomaly window of the electro-optic parameter data to limit the selection of thermal distribution data frames. Subsequently, the thermal anomaly contour object and scattering anomaly object are obtained. The spatial correlation weight and initial coordinate uncertainty are calculated in the unified coordinate system to form a set of candidate failure points and candidate coordinates. The data processing unit sends coordinate commands to the positioning point printing device, and the dispensing valve ejects positioning ink dots on the outer surface of the package. After the navigation imaging module re-examines and confirms the position of the positioning ink dots, the unified coordinate data packet carries the set of candidate failure points and spatial correlation weights. 1. Initial coordinate uncertainty and retaken images proceed to step three.

[0067] Furthermore, the version identifier is replaced with a peelable ink orientation marker. The ink is formed by water-soluble acrylic resin and carbon black is added to create high contrast. The unified coordinate system grid resampling is replaced with cubic convolution interpolation and gradient constraints are applied before interpolation. The dual threshold for thermal anomaly segmentation is replaced with histogram threshold selection and the analysis excitation dose budget is used as the boundary of the search interval. The spatial correlation weight mapping is replaced with a monotonically decreasing rational function. The positioning ink dots are replaced with peelable wax-based positioning points. The wax-based positioning points are mainly composed of microcrystalline wax and silica particles are added to form scattering. They are removed by wiping after softening with low-temperature hot air.

[0068] Step 3: Using the initial coordinate uncertainty and spatial correlation weight as input, and under the constraints of calibration residual index and analysis excitation dose budget, perform error budgeting, interpretable characteristics, credibility classification, and joint triggering continuous output to obtain candidate failure point risk, failure category, and secondary verification instructions.

[0069] Encapsulating the thermal path stretches the hotspot boundary towards the heat flow direction, causing the hotspot center to deviate from the defect location; furthermore, scattering microscopy data is affected by reflection and particle contamination. To avoid mistaking thermal diffusion broadening and reflection artifacts as defect evidence, this step first uses the calibration residual index and coordinate uncertainty as the same error budget entry point, then uses spatial correlation weights to constrain the geometric consistency of thermal and scattering anomalies, subsequently aligning the electro-optic parameter anomaly window with the thermal and scattering local regions under the same temporal semantics, and finally determining whether to trigger secondary verification through confidence grading. This collaborative relationship enables the output to organize local evidence with candidate coordinates as the entry point and binds the secondary verification action to the same coordinate version identifier.

[0070] This step is performed by the data processing unit. The data processing unit reads the set of candidate failure points and, centered on the candidate coordinates, crops a local region from the thermal distribution data and scattering microscopy data, while simultaneously cropping anomaly window segments from the electro-optic parameter data. The data processing unit encapsulates the local region and segments into feature records and outputs the risk score, failure category, secondary verification instruction, and reversible risk labeling parameters.

[0071] If the frame selection and electro-optic parameter anomaly window of the thermal local region are inconsistent, different thermal and electro-optic features will appear; if there are strongly reflective pixels in the scattering local region, texture statistics will be replaced by reflective statistics. By analyzing the excitation dose budget to constrain the frame selection boundary and using a reflective mask to constrain the scattering statistics boundary, the three types of inputs—thermal, scattering, and electro-optic—are aligned under a single candidate coordinate and an excitation window.

[0072] Therefore, the data processing unit clips the thermal local region and the scattering local region along the candidate coordinates in the unified coordinate system grid, and then constrains the frame set of the thermal local region by the electro-optic parameter anomaly window to form an anomaly window heatmap; then, it extracts the hot spot gradient principal direction and hot spot shape factor in the thermal local region, extracts the connected component morphology and texture direction in the scattering local region, and extracts voltage trend segments in the electro-optic parameter anomaly window; finally, it correlates the above results with spatial weights. Coordinate uncertainty Calibration residual index Analyze the excitation dose budget binding as a feature record and output it.

[0073] In this process, candidate coordinates serve as the sole entry point, allowing the thermal local region and the scattering local region to share the same square window. The data processing unit extracts a square window using a unified coordinate system grid index, with the window side length set to 4 to 12 times the diameter of the positioning ink dot. After extracting the scattering local region, the reflective mask from step two is applied, and a grid is extended around the boundary of the reflective mask as an exclusion zone. Pixels within the exclusion zone do not participate in texture calculation. The frame set of the thermal local region is limited by an electro-optic parameter anomaly window: the data processing unit filters the thermal distribution data frames based on the start and end timestamps of the anomaly window and averages the filtered frames to obtain the anomaly window heatmap. Subsequently, band-limited differential is used to calculate the gradient within the anomaly window heatmap, and the gradient direction consistency is statistically analyzed. The direction with the highest gradient direction consistency is recorded as the main direction of the hotspot gradient and written into the feature record.

[0074] In practice, clipping within the same window aligns thermal and scattering spatial evidence under the same candidate coordinates, reducing window mismatches. Reflection boundary locking locks scattering statistics to non-reflective locations, ensuring the texture orientation closely approximates the true scattering structure. Anomaly window heatmaps are bound to electro-optic parameter anomaly windows, and the principal directions of hotspot gradients have consistent temporal semantics.

[0075] Furthermore, the electro-optic parameter data is transformed from a continuous sequence into interpretable trend segments. The data processing unit segments the electro-optic parameter data according to the excitation cycle given in the test formulation. Each segment includes a rising edge, a steady-state segment, and a falling edge. Within the steady-state segment, the first sampling point where the voltage waveform abruptly changes is defined as the failure point, and fixed-length samples before and after the failure point are extracted to form a voltage trend segment. The voltage trend segment retains the original sampled values ​​and timestamps, and the corresponding cycle segment and the analysis excitation dose budget are written at the beginning of the segment. The data processing unit encapsulates voltage trend segments, hotspot gradient principal directions, hotspot shape factors, and scattering connected component morphological parameters into a single feature record, and writes spatial correlation weights into the record header. Coordinate uncertainty Calibration residual index and analysis of excitation dose budget .

[0076] When in use, voltage trend segments are entered into feature records in a fixed format, so that subsequent consistency judgments are based on segments rather than abstract conclusions. Feature records are bound to spatial and temporal elements within the same data structure, and grading and triggering can be completed without relying on external queries.

[0077] Relying solely on coordinate uncertainty would push all hotspot broadening samples into the trigger queue, causing congestion of secondary verification resources; relying solely on spatial correlation weights would ignore the mutually exclusive relationship between voltage trend segments and thermal and scattering features. Therefore, a confidence-based weighting method is used to incorporate spatial correlation weights. Coordinate uncertainty The calibration residual index is associated with the triggering rule, and a reversible risk marker is printed after the secondary verification instruction is generated, so that step four performs the verification action at the local entry point.

[0078] Therefore, the data processing unit first calculates the confidence weight based on the feature records and completes the consistency conflict determination, then writes the conflict samples and low confidence samples into the trigger queue. Before writing into the trigger queue, the data processing unit generates a risk score and failure category based on interpretable features. The risk score is jointly determined by the hotspot shape factor, the density of scattering connected domains, and the abrupt change direction of the voltage trend segment, and its effective range is modulated by the confidence weight. The failure category is determined by a category dictionary, which at least includes end-face catastrophic optical damage (COD), mirror catastrophic optical damage (COMD), and volume catastrophic optical damage (COBD). Each category is given by a combination of the positional conditions of the scattering anomaly relative to the end face, the change direction conditions of the voltage trend segment, and the main direction conditions of the hotspot gradient. To meet the minute-level cycle time, the calculation of the risk score and failure category adopts a rule tree structure. The rule tree node conditions directly reference the feature record fields and complete the branch selection in constant time. Subsequently, the data processing unit selects the priority action according to the ratio of expected information gain to action cost and generates a secondary verification instruction. Finally, it is coupled to the positioning point printing device to print reversible risk marks and re-confirm the position deviation, with the position deviation using coordinate uncertainty as the judgment boundary.

[0079] Supplementing action library fields and scoring functions: Setting convergence factors and action costs for each action entry in the action library. (Equipment capabilities and engineering calibrations are written into the test formula or action library) Define action scoring: ; Action Scoring The range of values ​​is Used to select priority actions; convergence factor The range of values ​​is This characterizes the expected strength of the convergence of the coordinate uncertainty due to the action; action cost The range of values ​​is This represents the combined cost of time and resources consumed by the action.

[0080] Similarly, geometric consistency and uncertainty level are used as confidence weights, and the lower limits of coordinate uncertainty and calibration residual index are merged. The larger product of these two lower limits and the lower limit of grid spacing in the same coordinate system is then used as the coordinate uncertainty and calibration residual index. The denominator is greater than zero. The credibility weight is calculated using the following formula for the response feature record: ; Where: Credibility weight : Dimensionless weight of the credibility of evidence for candidate failure points, with a value range of Used to determine the effectiveness of trigger queue generation and risk scoring; spatial correlation weight The correlation strength coefficient between thermal anomalies and scattering anomalies, with a value range of [value missing]. , used to provide the underlying weights for geometric consistency; Coordinate uncertainty : Uncertainty scale for candidate failure point location, with a value range of Used to reflect the impact of hotspot widening and registration residuals on position; calibration residual index : Cross-module coordinate mapping residual intensity, with a value range of It is used to provide a reference scale for registration errors and normalize coordinate uncertainties. ; The data processing unit then begins to determine the conflict rule base: the conflict rule base is published in the text format of the appendix to the instruction manual, and the order of the rules is feature condition - mutual exclusion relationship - trigger action; if the direction of change of the voltage trend segment and the broadening direction shown by the main direction of the hot spot gradient are mutually exclusive in the rule base, then the failure point is identified as a consistency conflict and enters the trigger queue; if there is no consistency conflict and the confidence weight is less than the confidence threshold given by the test recipe, then it enters the trigger queue.

[0081] When used, the credibility weight will be spatially correlated with the weight. The coupling of coordinate uncertainty and calibration residual index makes the trigger queue subject to both geometric consistency constraints and uncertainty constraints. The conflict rule base converts mutually exclusive relationships into reproducible entries, so that voltage trend segments can stably enter the trigger queue when they conflict with heat and scattering.

[0082] Furthermore, the triggering results are mapped to executable instructions and package surface markings. The data processing unit generates a set of secondary verification candidate actions for each candidate failure point in the trigger queue, and selects the priority action based on the ratio of expected information gain to action cost; the expected information gain is expressed as the degree to which the number of observation steps required to converge the coordinate uncertainty, and the action cost is expressed as the number of clamping operations and the equipment occupancy time. The data processing unit writes the priority action, candidate coordinates, coordinate version identifier, and confidence weight into the secondary verification instruction. The secondary verification instruction is a structured text record, with fields separated by fixed delimiters, and includes a checksum generated by cyclic redundancy check to determine the record integrity.

[0083] The data processing unit is then coupled to a positioning point printing device to print reversible risk markers. These reversible risk markers are double-layered concentric rings: the inner ring is made of water-soluble positioning ink, and the outer ring is made of water-soluble risk ink. The water-soluble risk ink uses polyvinyl alcohol as a film-forming agent, phthalocyanine blue as a pigment, and deionized water as a solvent. The polyvinyl alcohol mass fraction is 6% to 10%, and the phthalocyanine blue mass fraction is 1% to 3%. The ink is stirred for 15 minutes and allowed to stand for 10 minutes to remove bubbles. The inner ring line width is 100 to 300 micrometers, and the outer ring line width is 150 to 500 micrometers. The number of broken segments in the outer ring is given by the test formula and used to encode the risk level. After printing, the navigation imaging module re-photographs the markers. If the deviation between the center of the concentric ring obtained from the re-photograph and the candidate coordinates exceeds the coordinate uncertainty, a re-printing is performed. The markers are removed by soaking and wiping with deionized water, followed by degreasing with electronic-grade isopropanol and drying.

[0084] In use, the secondary verification instruction includes the action sequence and check code, ensuring that step four is executed according to the instruction and that record integrity can be determined. The double-layered concentric rings fix the candidate coordinates and risk level on the outer surface of the package, allowing secondary verification to enter the local field of view from the concentric ring entrance. The water-soluble formula makes the markings removable and compatible with rework processes, and the outer ring break segment coding reduces additional labeling steps.

[0085] After reading the unified coordinate data packet, the data processing unit displays a set of candidate failure points and presents an overlay of the thermal local area and the scattering local area at the candidate coordinates. The operator keeps the stage locked, and the data processing unit filters the thermal distribution data frames according to the electro-optic parameter anomaly window and forms an anomaly window heatmap. Then, it removes reflective areas from the scattering microscopy data and extracts the texture direction, subsequently generating a feature record containing voltage trend segments. The data processing unit calculates the confidence weight and forms a trigger queue based on the conflict rule base, then generates a secondary verification instruction and prints a reversible risk mark for the double-layer concentric ring. After the navigation imaging module re-examines and confirms that the center of the concentric ring falls within the judgment boundary of coordinate uncertainty, the operator transfers the device to the next station. The personnel at the next station align the microscope field of view with the concentric ring and execute step four according to the secondary verification instruction.

[0086] Furthermore, the reversible risk marker has been changed from a concentric ring to a combination of arrows and short lines, and the risk level has been encoded at the location of the broken segment; the water-soluble risk ink has been changed from phthalocyanine blue to iron oxide red; the secondary verification instruction check code has been replaced from cyclic redundancy check to Hamming code, while the field structure remains unchanged; and the conflict rule base has been replaced from text rules to decision tree rules.

[0087] Step 4: Execute triggered secondary verification under the constraints of the secondary verification instruction, and if convergence is still not achieved, perform destructive mechanism confirmation on a small number of samples, write the results back to the unified coordinate system and test formula, and output the final conclusion package.

[0088] Step 3 has assigned spatial correlation weights to each candidate coordinate. Coordinate uncertainty and confidence weight The thermal path of the encapsulation causes hotspot widening and shifting, and reflections from the encapsulation surface can create false scattering in the scattering microscopy data. If destructive sample preparation is performed directly, the cross-sectional window is forced to enlarge, amplifying the sample preparation defects. Step four involves first obtaining supplementary observations coaxial with the candidate coordinates at the secondary verification station using partial opening and exposure combined with lock-in thermal imaging, photoluminescence (PL), or photon emission microscopy (PEM). This is used to prioritize convergence of coordinate uncertainties and resolve consistency conflicts. Only a small number of samples that still conflict are subjected to destructive mechanism confirmation. The secondary verification and destructive mechanism confirmation use reversible risk markers as the field of view entry point, and write back the calibration residual index and coordinate uncertainty under the same coordinate version identifier. This allows subsequent batches to reuse the updated thresholds and action sequences.

[0089] Step four is executed by the data processing unit. The data processing unit is communicatively connected to the navigation imaging module, microscopic imaging module, thermal imaging module, and electro-optic parameter acquisition module of the secondary verification station, and also communicatively connected to the cutting device, grinding and polishing device, focused ion beam (FIB) device, and electron beam induced current (EBIC) acquisition device of the destructive mechanism confirmation station. The secondary verification station and the destructive mechanism confirmation station share fixtures and alignment references, ensuring that the reversible risk markers and coordinate version identifiers maintain their geometric relationship after transfer between stations.

[0090] Among the issues, the candidate failure point set suffers from excessively high coordinate uncertainty due to low confidence weights, as well as consistency conflicts in the conflict rule base. If the cross-section or FIB is entered before the coordinate uncertainty converges, the cross-section window expands, introducing redundant cutting, and the stress from opening and grinding can induce cracks near the solder layer. Reversible risk markers limit the verification action to the vicinity of the candidate coordinates, and secondary evidence is used to prioritize the convergence of coordinate uncertainty. .

[0091] Therefore, the data processing unit reads the secondary verification instructions and determines the entry sequence and verification actions. It then transports the device under test (DUT) with reversible risk markers to the secondary verification station. Using the navigation imaging module, it determines the outer ring break segment code, aligns the microscopic field of view with the inner ring center, and collects the locally exposed data in the secondary observation. The data processing unit aligns the secondary data to a unified coordinate system and updates the calibration residual index and coordinate uncertainty. Using the center of the inner ring of the reversible risk marker as the entry point, the opening is exposed near the candidate coordinates to reduce the global search. The device under test is fixed on the secondary verification fixture while keeping the attitude state field unchanged. The navigation imaging module acquires the image of the reversible risk marker. By analyzing the position of the outer ring break segment and verifying that the risk level code is consistent with the secondary verification command, the microscopic imaging module is driven to align the center of the field of view with the center of the inner ring.

[0092] Following this, partial opening and exposure is performed, employing a continuous process of cover removal and cleaning and protection of the exposed surface. When the encapsulation cover is a transparent sealant, the operator places the device in a solvent-resistant tray and adds electronic-grade N-methylpyrrolidone. The tray is maintained at 60°C to 90°C for 20 to 60 minutes to soften the sealant. The sealant is then peeled off layer by layer using a ceramic scraper, with the scraping depth limited under microscopic observation to prevent it from penetrating the light-emitting end face. When the encapsulation cover is a metal cap, the operator uses a micro-cutting blade to cut along the weld seam and lift the cap. The blade thickness is 0.1 mm to 0.3 mm, and the cutting depth is limited to the weld seam thickness. After cover removal, the operator rinses off residual solvent with deionized water and degreases with electronic-grade isopropanol. A peelable protective adhesive based on a polyvinyl alcohol-glycerin system is then applied around the exposed area to form a dike, with a dike thickness of 0.2 mm to 0.6 mm. This dike is allowed to stand at room temperature for 10 to 25 minutes to form a continuous boundary. If the reversible risk marker is eroded by solvent, causing the inner ring boundary to become blurred, the data processing unit instructs the operator to reprint the reversible risk marker and use the navigation imaging module to re-photograph to confirm that the center of the new marker falls into the coordinate uncertainty determination boundary of the original candidate coordinates.

[0093] In use, the reversible risk marker locates the local open exposure near the candidate coordinates and reduces the global search path, while the dike limits the exposure window to a local area and inhibits solvent penetration into the solder layer, thereby reducing contamination of the exposed surface.

[0094] Furthermore, after local exposure, secondary observations are acquired, and the secondary evidence fragments are written back to the unified coordinate system grid to converge coordinate uncertainty. The secondary observation and acquisition are executed according to the secondary verification instruction field, with the preferred action sequence being phase-locked thermal imaging lightweight acquisition, PL observation and acquisition, and PEM observation and acquisition. Phase-locked thermal imaging lightweight acquisition applies periodic power perturbations under the analysis excitation dose budget constraint and acquires a thermal distribution data sequence. The data processing unit performs phase-sensitive detection on the sequence to output a local heat source map. The phase-sensitive detection uses an orthogonal reference sequence with the same frequency as the power perturbation to multiply and accumulate the thermal distribution data sequence frame by frame, obtaining the in-phase component and the orthogonal component respectively. The amplitude of the in-phase component and the orthogonal component is then synthesized to obtain the local heat source map.

[0095] PL observations acquire images of the emitted light by illuminating the exposed area with an excitation source and using a microscopic imaging module. The excitation source wavelength is given by the test formula field. PEM observations acquire images of emitted light events by a PEM detector under dark conditions and map the event coordinates to a unified coordinate system grid. Secondary data registration reuses the coordinate mapping relationship from step one, and at the secondary verification station, the coordinate mapping relationship is re-solved to eliminate minor attitude differences introduced by station transport. The re-solving process involves the data processing unit extracting the corner point set of the alignment reference from the navigation imaging module image, using Gauss-Newton iteration to solve for the affine transformation parameters, and using the corner point reprojection error as the objective function. Subsequently, the calibration residual index is recalculated at the secondary verification station and overwritten with the original record. Finally, the data processing unit updates the coordinate uncertainty based on the confidence weight. ; Where: coordinate uncertainty The position uncertainty scale for candidate coordinates, with a range of values ​​of [value range missing]. This is used to define the coordinate window after secondary verification and as the basis for confirming the destructive mechanism; calibration residual index The robust root mean square residual recalculated at the secondary verification station has a range of values. As a lower bound for uncertainty updates, it prevents the uncertainty from falling below the upper limit of registration capability; confidence weight : Weight of evidence credibility, with a value range of This is used to determine the shrinkage strength; the higher the reliability, the more pronounced the shrinkage. Updated coordinate uncertainty and local heat map, PL luminescence image, PEM luminescence events Figure 1 Write the unified coordinate system data packet and attach a secondary verification action sequence field to the conclusion packet.

[0096] In use, the secondary observation acquisition solidifies the local anomaly center into evidence fragments on a unified coordinate system grid, enabling coordinate updates to be performed using candidate coordinates as the entry point. The calibration residual index is recalculated at the secondary verification station and overwrites the original record, ensuring that the error budget entry point is consistent with the current station's geometric state.

[0097] Furthermore, some devices under test still exhibit consistency conflicts after secondary verification, and root cause closure cannot be achieved solely through secondary observations. If all samples enter the FiB or cross-section, the production line cycle time cannot be maintained, and destructive sample preparation stress will increase new cracks. Only a small number of samples that still meet the high-risk criteria and whose coordinate uncertainty has not converged are subjected to destructive mechanism confirmation, and the conflict rule base and test formula are written back. Therefore, the data processing unit reads the updated coordinate uncertainty and secondary evidence fragments, and the test formula selects samples for destructive mechanism confirmation based on the sampling ratio field; the operator uses reversible risk markers and positioning ink dots as repositioning entry points at the destructive mechanism confirmation station to perform embedding, cutting, and cross-sectional positioning, and performs scanning electron microscopy imaging and EBIC acquisition on the cross-section; the data processing unit generates truth labels, writes them back to the conflict rule base and test formula, outputs the final conclusion package, and performs exit condition determination.

[0098] As a supplement, the judgment order is first the conflict rule base and then the category dictionary. Example 1 of the conflict rule base: When the scattering anomaly object is located on the end face side in the normal direction of the end face boundary under the unified coordinate system, and the voltage trend segment is an upward jump at the failure point, and the main direction of the hot spot gradient in the anomaly window heatmap points to the inside of the cavity, a consistency conflict occurs and it enters the trigger queue. The trigger action is to prioritize the acquisition of PeM observations. Another example 2: When the scattering anomaly object is near the end face, and the hot spot core of the anomaly window heatmap coincides with the end face boundary, and the voltage trend segment is a downward jump at the failure point, the failure category is COD.

[0099] The destructive sample preparation process is divided into embedding and fixation, coarse positioning cutting, and fine positioning sectioning to ensure that the candidate coordinates remain repositionable during sample preparation. The operator pours low-shrinkage two-component epoxy resin to form an embedding block. The mass ratio of epoxy resin to curing agent is set to 10:3 to 10:5. After mixing, vacuum degassing is performed for 3 to 8 minutes before pouring. After pouring, curing is carried out at 50 to 80 degrees Celsius for 30 to 120 minutes. After curing, the operator cuts along the direction indicated by the inner ring center, leaving a 0.2 mm to 0.8 mm margin near the candidate coordinates as a fine positioning section window.

[0100] The precise positioning section is achieved through mechanical grinding and polishing or FIB section path. Mechanical grinding and polishing involves grinding with 600-4000 grit sandpaper under deionized water lubrication and polishing with alumina polishing slurry for 3-10 minutes to achieve a mirror finish. For FIB section, a platinum protective layer with a thickness of 0.5-2 micrometers is first deposited, followed by rough milling with a 30-10 kV ion beam and progressively decreasing current for fine milling, allowing the section to penetrate the interface between the epitaxial layer and the solder layer. After section completion, the operator performs EBIC acquisition and records the current response as the scanning position changes. The data processing unit maps the EBIC response map back to a unified coordinate system grid to form a truth evidence fragment. The mapping process uses bilinear interpolation or cubic convolution interpolation, and the interpolation strategy field is recorded in the unified coordinate system data packet to maintain consistency in subsequent playback.

[0101] In use, the embedding and allowance retention fix the surrounding structure of the candidate coordinates, allowing the fine positioning section window to expand around the candidate coordinates and reducing the risk of overcutting. Mechanical grinding and polishing, along with the FIB section, provide two feasible paths and maintain consistent repositioning entry points, enabling the destructive mechanism confirmation to adapt to different sampling scales.

[0102] Furthermore, the cross-sectional evidence is converted into truth labels and written back to the conflict rule base and test recipe. After receiving the cross-sectional image and EBIC response map, the data processing unit first confirms that the cross-sectional direction is consistent with the candidate coordinate direction using the attitude state field in the coordinate version identifier. Then, based on the preset category dictionary, it classifies the truth evidence fragments into end-face catastrophic optical damage (COD), mirror catastrophic optical damage (COMD), or volume catastrophic optical damage (COBD), and writes the truth labels into the truth field of the unified coordinate system data packet. Subsequently, the data processing unit performs conflict rule base write-back and test recipe write-back. The conflict rule base write-back writes rule entries into the update section and includes version numbers and cyclic redundancy check codes. The test recipe write-back updates the confidence threshold, secondary verification action sequence threshold, and sampling ratio fields, and binds the update to the packaging type field. After completing the write-back, the data processing unit outputs the final conclusion packet, which includes candidate coordinates and updated coordinate uncertainties. The system includes risk level, failure category, secondary verification action sequence, evidence fragment index and truth value field, and performs exit condition judgment and diversion marking accordingly.

[0103] The repair personnel clamped the device, carrying a double-layered concentric ring reversible risk marker, to the secondary verification station and aligned it with the center of the inner ring. They then softened and peeled off the sealant, followed by rinsing, degreasing, and applying a protective coating to create an exposure window. The data processing unit executed phase-locked thermal imaging lightweight acquisition, PL observation acquisition, and PEM observation acquisition according to the secondary verification instructions, and wrote back the evidence fragments, while simultaneously updating the coordinate uncertainty. For a still conflicting device, the rework personnel proceed to the destructive mechanism verification station to complete the epoxy resin embedding and curing and FIB cross-section, and perform EBIC data acquisition on the cross-section; the data processing unit generates truth labels and writes back the conflict rule base and test formula, and outputs a final conclusion package carrying the truth field.

[0104] In use, truth label write-back binds the conflict rule base and test recipe to the package type, reducing duplicate conflicts in subsequent batches and shortening the sequence of secondary verification actions. The final conclusion package solidifies the candidate coordinates and evidence fragment indexes, ensuring that rework and process handling are performed based on the same conclusion package.

[0105] Furthermore, the localized opening and exposure was replaced by swelling and peeling with oxygen and argon plasma etching to remove the resist, with RF power ranging from 50 watts to 200 watts and processing for 1 to 5 minutes; the PEM in the secondary observation acquisition was replaced by electroluminescence microscopy acquisition and end face images were acquired under the analysis excitation dose budget constraints; the destructive mechanism confirmation was replaced by FIB cross section with ion beam oblique cutting while keeping EBIC acquisition unchanged.

[0106] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0107] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0109] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0110] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A failure analysis method for optically pumped laser chips, targeting packaged optically pumped high-power semiconductor laser devices under test; characterized in that: include, The device is clamped in a fixture with an alignment reference, the alignment reference is identified, the coordinate mapping relationship between the navigation imaging module, the scattering microscopy module and the thermal imaging module to a unified coordinate system is calculated and the calibration residual index is recorded, and scattering microscopy imaging data, thermal distribution data and electro-optic parameter data are acquired under controlled excitation. The data is registered to a unified coordinate system, and candidate failure points are generated based on thermal anomalies, scattering anomalies, and electro-optic parameter anomalies. The coordinate uncertainty of the candidate failure points is calculated based on the calibration residual index and the thermal diffusion effect. Thermal features, scattering features, and electro-optic features are extracted to generate risk scores and failure categories. Secondary verification instructions are triggered based on risk scores and coordinate uncertainties. In response to the Level 2 verification command, Level 2 data is collected and registered to a unified coordinate system, and the suspected coordinates, coordinate uncertainty, risk score and failure category are output. The controlled excitation is set according to the preset test formula to analyze the excitation dose budget, and the temperature rise and power are monitored during the acquisition of thermal distribution data; When the analysis stimulus dose budget is exceeded, the stimulus conditions are adjusted, including reducing the stimulus intensity and shortening the stimulus time, and the data is reacquired. The over-budget event, the adjusted stimulus conditions, and the confidence degradation flag are written into the process metadata. The process of generating candidate failure points includes: extracting hotspot regions from thermal distribution data and determining hotspot centers and hotspot boundaries; extracting scattering anomaly connected domains from scattering microscopy data and determining scattering centers; calculating the thermal dissipation distance between hotspot centers and scattering centers in a unified coordinate system and generating spatial correlation weights; and filtering candidate failure points based on spatial correlation weights. In response to the Level 2 verification command, the device is partially exposed while maintaining the identifiable alignment reference. Subsequently, lock-in thermal imaging lightweight acquisition data, photoluminescence observation acquisition data, and photon emission microscopy acquisition data are collected as Level 2 data. The Level 2 data are then registered to a unified coordinate system to update the risk score, failure category, and coordinate uncertainty. The data processing unit reads the secondary verification instructions and determines the entry sequence and verification actions. It transports the device under test with reversible risk markers to the secondary verification station, determines the outer ring break segment code through the navigation imaging module, aligns the microscopic field of view with the center of the inner ring, and collects the local open exposure in the secondary observation. The data processing unit aligns the secondary data with the unified coordinate system and updates the calibration residual index and coordinate uncertainty. The data processing unit is coupled to the positioning point printing device to print reversible risk marks. The reversible risk marks are double-layer concentric rings: the inner ring material is water-soluble positioning ink, and the outer ring material is water-soluble risk ink. The water-soluble risk ink uses polyvinyl alcohol as a film-forming agent, phthalocyanine blue as a pigment, and deionized water as a solvent. The mass fraction of polyvinyl alcohol is 6% to 10%, and the mass fraction of phthalocyanine blue is 1% to 3%. The ink is stirred for 15 minutes and allowed to stand for 10 minutes to remove bubbles. The inner ring line width is 100 micrometers to 300 micrometers, and the outer ring line width is 150 micrometers to 500 micrometers. The number of broken segments in the outer ring is given by the test formula and used to encode the risk level. After printing, the navigation imaging module is re-photographed. If the deviation between the center of the concentric ring obtained by the re-photograph and the candidate coordinates exceeds the coordinate uncertainty, the overlay is re-printed. When removing the mark, it is soaked in deionized water and wiped off, and then degreased with electronic grade isopropanol and blown dry.

2. The method for failure analysis of an optically pumped laser chip according to claim 1, characterized in that: When acquiring scattering microscopy imaging data, thermal distribution data, and electro-optic parameter data, the device clamping posture, acquisition timestamp, excitation conditions, and module status are recorded simultaneously. The records are associated with the coordinate mapping relationship and calibration residual index and written into the process metadata. The process metadata is output along with the registration results.

3. The method for failure analysis of an optically pumped laser chip according to claim 2, characterized in that: When scattering microscopy data, thermal distribution data and electro-optic parameter data are registered to a unified coordinate system, a coordinate version identifier is generated and associated with the clamping posture. When the clamping posture changes or the sample preparation stage changes, a new coordinate version record is established and the historical coordinate version record is retained.

4. The method for failure analysis of an optically pumped laser chip according to claim 3, characterized in that: The coordinate uncertainty is calculated by the error budget, which includes the calibration residual index, the pixel quantization error corresponding to the grid spacing of the unified coordinate system, and the hot spot widening term characterized by the area and perimeter of the hot spot region. The coordinate uncertainty is output along with the candidate failure point.

5. The method for failure analysis of an optically pumped laser chip according to claim 4, characterized in that: After extracting thermal features, scattering features, and electro-optic features, credibility grading is performed based on the credibility downgrade flag in the process metadata, and credibility weights are generated. The thermal features, scattering features, and electro-optic features are then written into the feature record. when When a feature record meets the conflict conditions of the conflict rule base, the corresponding candidate failure point is marked as requiring secondary verification, and a secondary verification instruction is generated based on the risk score and coordinate uncertainty, with a verification code attached.

6. The method for failure analysis of an optically pumped laser chip according to claim 5, characterized in that: When the risk score is high and the coordinate uncertainty does not converge after secondary verification, the device is selected for destructive mechanism confirmation according to the preset sampling strategy. The destructive mechanism confirmation process includes cross-sectional sample preparation, electron beam induced current observation and acquisition, and focused ion beam cutting. The obtained truth labels are then written back to update the trigger threshold, confidence grading rules, and thermal diffusion uncertainty parameters.

7. The method for failure analysis of an optically pumped laser chip according to claim 6, characterized in that: The thermal anomaly extraction uses dual-threshold segmentation to identify hotspot regions in the thermal distribution data and output the hotspot center and hotspot boundary. The second threshold of the dual-threshold segmentation is higher than the first threshold. The secondary verification instruction is generated based on the trigger thresholds set by the preset test recipe. The trigger thresholds include the threshold for risk score and the threshold for coordinate uncertainty, and the risk score and coordinate uncertainty are jointly determined.

Images