High-reliability packaging quality evaluation method and system based on multi-source process data

Abstract

The application discloses a high-reliability packaging quality evaluation method and system based on multi-source process data, and relates to the technical field of quality evaluation.The application realizes online perception of the orientation consistency of the tape casting ceramic particles by adopting terahertz spectrum to nondestructively scan the unsintered green ceramic tape, avoids material waste and process interruption caused by destructive detection;secondly, by synchronously collecting the cutting edge morphology in the dicing process and quantifying the geometric characteristics, the identification of the edge collapse state is improved from subjective qualitative judgment to objective quantitative representation, which significantly improves the detection consistency and accuracy;and the spectrum coding data reflecting the internal quality and the morphology coding data reflecting the external defects are associated point by point to construct a two-dimensional mapping relationship of internal process parameters and external processing results, so that whether the cutting edge collapse is caused by potential abnormalities in the tape casting stage can be accurately traced back, and data closed loop is provided for the backtracking optimization of process parameters.

Description

Technical Field

[0001] This invention relates to the field of quality assessment technology, and in particular to a highly reliable packaging quality assessment method and system based on multi-source process data. Background Technology

[0002] Low-Temperature Co-fired Ceramic (LTCC) technology, as a key process for realizing microwave components, millimeter-wave circuits, and high-density system integration packaging, has been widely used in communications, radar, aerospace, and other fields in recent years. The LTCC process typically involves multiple steps, including green ceramic tape casting, slicing, drilling, via filling, conductor printing, stacking, lamination, sintering, and subsequent dicing. The green ceramic tape casting process determines the orientation of the ceramic particles, directly affecting the dielectric uniformity and mechanical strength of the substrate after sintering. The dicing process after sintering determines the geometric integrity and edge quality of the chip or circuit unit. Currently, the industry mostly uses offline sampling inspection for quality control of the LTCC substrate manufacturing process. For example, scanning electron microscopy is used to observe the cross-sectional morphology to assess the casting orientation, or optical microscopy is used to detect edge chipping during cutting. However, these methods have inherent limitations such as low detection efficiency, limited sample representativeness, and difficulty in achieving full-process online feedback.

[0003] In existing technologies, consistency assessment of the tape casting process mainly relies on dielectric property testing of the substrate after sintering or destructive cross-sectional microscopic analysis. The former can only reflect macroscopic average characteristics and cannot locate local tape casting anomalies; the latter is time-consuming and labor-intensive and cannot be applied to real-time monitoring of the production line. For quality control of the dicing process, conventional methods are manual visual inspection or traditional machine vision inspection based on two-dimensional images. These methods can usually only identify obvious edge damage, but lack the ability to quantitatively characterize micron-level edge chipping morphology and crack propagation trends, and the inspection standards are easily affected by subjective factors. More importantly, existing technical solutions often treat tape casting process control and dicing quality control as two independent links, failing to establish an intrinsic connection between them. This makes it impossible to trace whether the edge chipping problem in the dicing process originates from potential defects in the tape casting stage, thus limiting the closed-loop capability of process optimization. Summary of the Invention

[0004] In view of the problems existing in the above-mentioned background technology, the present invention is proposed.

[0005] Therefore, the problem to be solved by this invention is how to achieve efficient, non-destructive, online detection of the internal quality characteristic of casting consistency and the external geometric characteristic of cutting edge breakage during the manufacturing process of LTCC substrates, and to perform correlation analysis between the two to locate the abnormal areas of the process.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a highly reliable packaging quality assessment method based on multi-source process data, comprising: performing terahertz spectral scanning on unsintered green ceramic tape and spatially segmenting it according to preset chip positions; extracting terahertz frequency domain features within each segment to generate spectral coding data characterizing the consistency of ceramic particle casting orientation; synchronously acquiring the cutting edge morphology corresponding to each chip position during the dicing process of the sintered ceramic substrate; spatially segmenting and extracting geometric features of the scratch morphology at each position to generate morphology coding data characterizing the cutting edge breakage state; using chip position as an index, linking the spectral coding data and the morphology coding data point by point to construct a multi-source process data pair for each chip position; calculating a casting consistency index for the spectral coding data at each chip position and a breakage severity index for the morphology coding data; obtaining a comprehensive quality score for the corresponding chip through weighted fusion; and marking chips with a comprehensive quality score lower than a preset score threshold as abnormal process parameter areas.

[0008] As a preferred embodiment of the high-reliability packaging quality assessment method based on multi-source process data described in this invention, the method includes: performing terahertz spectral scanning on the unsintered green ceramic tape to obtain a three-dimensional terahertz data cube covering the entire surface of the green ceramic tape; the three dimensions of the three-dimensional terahertz data cube are the horizontal spatial coordinate X, the vertical spatial coordinate Y, and the terahertz time-domain sampling point, respectively, with each spatial coordinate point (X, Y) corresponding to a complete time-domain spectral curve; and dividing the three-dimensional terahertz data cube into several spatial slices according to the preset chip layout position. Each space is partitioned It corresponds to a chip location and contains time-domain spectral data of all sampling points within the region.

[0009] As a preferred embodiment of the high-reliability packaging quality assessment method based on multi-source process data described in this invention, the generation of the spectral coding data includes: processing each spatial segment... All time-domain spectral curves within the space are subjected to Fourier transform to obtain the frequency-domain spectrum of each sampling point; for the frequency-domain spectrum of each sampling point, the average amplitude in the preset characteristic frequency band is calculated as the frequency-domain characteristic amplitude of the corresponding sampling point; the standard deviation of the frequency-domain characteristic amplitude of all sampling points within the spatial segment is calculated as the casting consistency characteristic value of the spatial segment; and each spatial segment is then... The casting consistency characteristic values ​​are arranged in order of chip position to form a spectral encoded data sequence characterizing the casting orientation consistency of ceramic particles.

[0010] As a preferred embodiment of the high-reliability packaging quality assessment method based on multi-source process data described in this invention, the step of spatially segmenting and extracting geometric features of the scratch morphology at each location includes: during the ceramic substrate dicing process, simultaneously acquiring the cutting edge morphology at each chip location using a line-scanning white light interferometer to obtain three-dimensional surface contour data covering all dicing paths; and dividing the three-dimensional surface contour data into multiple spatial segments. Each space is partitioned Corresponding to the complete cut boundary region of a chip; for each spatial slice The contour data within the chip is used to extract the maximum chipping depth and the area of ​​the chipping region at the cut edge, which are then used as chipping feature parameters of the chip.

[0011] As a preferred embodiment of the high-reliability packaging quality assessment method based on multi-source process data described in this invention, the following steps are taken: the maximum value of the chipping depth and the area of ​​the chipping region of each chip are combined to form a two-dimensional feature vector, which is then arranged in order of chip position to generate a morphology-encoded data sequence characterizing the chipping state; each element in the morphology-encoded data sequence corresponds to a two-dimensional feature vector of a chip position.

[0012] As a preferred embodiment of the high-reliability packaging quality assessment method based on multi-source process data described in this invention, the point-by-point association includes: using the coordinates of each chip location as an index, pairing the corresponding position's tapering consistency feature value in the spectrum-encoded data sequence with the two-dimensional feature vector at the same position in the topography-encoded data sequence to generate multi-source process data pairs corresponding to the chip locations.

[0013] As a preferred embodiment of the high-reliability packaging quality assessment method based on multi-source process data described in this invention, the following steps are taken: For each chip location, a casting consistency feature value is extracted from the spectral encoded data sequence; the Euclidean distance between the casting consistency feature value and a preset standard casting feature value is calculated; the normalized reciprocal of the Euclidean distance is used as the casting consistency index value for the corresponding chip; For each chip location, the maximum chipping depth and the chipping area are extracted from the topography encoded data sequence; the two are weighted, summed, and normalized to obtain the chipping severity index value for the corresponding chip; the 1-chipping severity index value is used as the chipping quality index, and weighted and fused with the casting consistency index to obtain a comprehensive quality score, with the weights of the two being 1.

[0014] Secondly, the present invention provides a highly reliable packaging quality assessment system based on multi-source process data, which includes: a terahertz spectrum encoding module, which performs terahertz spectrum scanning on unsintered green ceramic tape, and spatially segments it according to a preset chip position, extracts the terahertz frequency domain features in each segment, and generates spectrum encoding data characterizing the consistency of ceramic particle casting orientation.

[0015] The cutting morphology encoding module synchronously collects the cutting edge morphology corresponding to each chip position during the dicing process of the sintered ceramic substrate, performs spatial segmentation and geometric feature extraction on the scratch morphology of each position, and generates morphology encoding data that characterizes the cutting edge breakage state.

[0016] The multi-source data association module uses chip location as an index to associate the spectrum encoding data with the topography encoding data point by point, thereby constructing a multi-source process data pair for each chip location;

[0017] The comprehensive quality scoring module calculates the flow-spinning consistency index for the spectral encoded data at each chip location and the edge-break severity index for the morphology encoded data; it obtains the comprehensive quality score of the corresponding chip through weighted fusion; and it marks chips with comprehensive quality scores below a preset scoring threshold as areas with abnormal process parameters.

[0018] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein the computer program instructions, when executed by the processor, implement the steps of the high-reliability packaging quality evaluation method based on multi-source process data as described in the first aspect of the present invention.

[0019] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the high-reliability packaging quality assessment method based on multi-source process data as described in the first aspect of the present invention.

[0020] The beneficial effects of this invention are as follows: Compared with traditional methods that rely on offline cross-sectional microscopic analysis or manual visual inspection, this invention uses terahertz spectroscopy to perform non-destructive scanning of unsintered green ceramic strips, achieving online perception of the orientation consistency of cast ceramic particles, thus avoiding material waste and process interruption caused by destructive testing; secondly, by simultaneously acquiring the morphology of the cutting edge during the dicing process and quantifying its geometric features, the identification of edge breakage is elevated from subjective qualitative judgment to objective quantitative characterization, significantly improving detection consistency and accuracy; furthermore, by linking the spectral coding data reflecting internal quality with the morphological coding data reflecting external defects point by point, a two-dimensional mapping relationship between internal process parameters and external processing results is constructed, thereby enabling precise tracing of whether the cutting edge breakage originates from potential anomalies in the casting stage, providing a data closed loop for backtracking and optimizing process parameters.

[0021] By using comprehensive quality scoring, automated grading and screening of each chip area on the substrate was achieved, transforming the marking of abnormal areas from post-event sampling to online full inspection, which greatly improved the overall yield and reliability of LTCC products and provided a new technical path for quality control of high-density microwave integrated packaging. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a high-reliability packaging quality assessment method based on multi-source process data.

[0024] Figure 2 This is a structural diagram of a high-reliability packaging quality assessment system based on multi-source process data. Detailed Implementation

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0027] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0028] Figure 1 This is a flowchart illustrating a high-reliability packaging quality assessment method based on multi-source process data according to an embodiment of the present invention. Figure 1 As shown, the high-reliability packaging quality assessment method based on multi-source process data includes:

[0029] S1: Perform terahertz spectral scanning on the unsintered green ceramic strip and spatially segment it according to the preset chip position. Extract the terahertz frequency domain features in each segment to generate spectral coding data that characterizes the consistency of the ceramic particle casting orientation.

[0030] It should be noted that the present invention aims to monitor quality from the source of the manufacturing of the packaging carrier (ceramic substrate).

[0031] First, terahertz spectroscopy is performed on the unsintered green ceramic tape to obtain a three-dimensional terahertz data cube covering the entire surface of the tape. Specifically, the unsintered green ceramic tape, after being cast, is laid flat and fixed on a two-dimensional translation stage. A transmission or reflection terahertz time-domain spectroscopy system is used to scan the surface of the green ceramic tape point by point. The system controls the synchronous movement of the terahertz transmitting and receiving probes, covering the effective area of ​​the entire green ceramic tape according to a preset spatial sampling interval (e.g., the values ​​of the X-axis and Y-axis spatial sampling steps need to be determined comprehensively based on the minimum chip size and the spatial variation rate of the casting orientation characteristics to ensure that each chip area contains at least 9 (3×3) effective sampling points).

[0032] At each sampling point, a picosecond-level terahertz pulse is emitted, and the time-domain electric field intensity variation curve after passing through the green ceramic tape is recorded. This curve contains information about the absorption and dispersion response of the microstructure inside the green ceramic tape to the terahertz wave. By synchronously acquiring the time-domain curves of all sampling points, a three-dimensional data structure is constructed, denoted as . .in, The dimension corresponds to the horizontal spatial coordinates of the raw ceramic zone, with a total of M sampling points; The dimension corresponds to the vertical spatial coordinates, with a total of N sampling points; The dimension corresponds to the terahertz time domain sampling time point, with a total of L points (determined by the time window and sampling rate).

[0033] therefore, In essence, it is a three-dimensional array of M×N×L, and a complete time-domain spectral curve can be indexed for any spatial coordinate.

[0034] Secondly, based on the preset chip layout positions, the three-dimensional terahertz data cube is divided into several spatial segments. Each space is partitioned It corresponds to a chip location and contains time-domain spectral data of all sampling points within the region.

[0035] In practice, to correlate the quality characteristics of the green ceramic tape with the performance of subsequent individual chips, spatial partitioning is required based on a predefined chip layout scheme in the photolithography pattern or design file. Specifically, this involves obtaining a pre-defined set of chip layout coordinates. ,in This refers to the total number of chips on a single ceramic tape. Usually the first The geometric center coordinates or lower left corner coordinates of each chip. Based on this layout, the three-dimensional data structure is... Perform spatial partitioning, define the first Spatial partitioning corresponding to each chip The delineation method is as follows:

[0036] coordinates Using this as a baseline, expand outwards to include all scan points covered by the complete outline of the chip (e.g., a rectangle). ,in , The chips in their unsintered state are respectively , The range of scan points covered by the design dimensions in the direction after coordinate transformation.

[0037] Therefore, each Each data cube contains all the original time-domain spectral data within the chip region, forming a sub-data cube. It should be noted that this invention spatially segments the data based on the chip layout, establishing a precise mapping from material regions to chip units. This allows subsequently extracted features to directly correspond to individual chips in the final product, providing a spatial indexing basis for single-chip-level quality traceability.

[0038] Furthermore, for each spatial partition All time-domain spectral curves within the range are subjected to Fourier transform to obtain the frequency-domain spectrum of each sampling point. For each sampling point's frequency-domain spectrum, the average amplitude over a preset characteristic frequency band is calculated as the corresponding frequency-domain characteristic amplitude.

[0039] In this embodiment of the invention, each spatial fragment obtained above It contains multiple spatial sampling points, each corresponding to a time-domain spectral curve. First, for Each time-domain curve within the range undergoes a Fast Fourier Transform (FFT) to transform it from the time domain to the frequency domain: specifically, for The Middle Time-domain signal of each sampling point After Fourier transform, the frequency domain complex signal is obtained. ,in For amplitude spectrum, For phase spectrum, ω is the angular frequency.

[0040] Considering that the orientation differences of ceramic particles primarily affect the absorption and scattering of terahertz waves in specific frequency bands, one or more pre-defined characteristic frequency bands sensitive to orientation changes need to be selected. These frequency bands are typically calibrated through preliminary experiments: standard samples with different casting orientations (e.g., samples taken parallel or perpendicular to the casting direction) are prepared, terahertz measurements are performed, and the spectral differences are compared and analyzed to identify the frequency band range with significant and stable differences, denoted as . .

[0041] After determining the characteristic frequency band, Frequency domain amplitude spectrum of the r-th sampling point Calculation in the characteristic frequency band The integral average or arithmetic average within the range is used as the frequency domain characteristic amplitude of that sampling point.

[0042] This process iterates through The set of frequency domain characteristic amplitudes of all points within the segment is obtained by sampling all points within that segment. It should be noted that since the uniformity of the orientation of the flow reflects the uniformity of the particle orientation distribution in the region, if the orientation uniformity is good, the frequency domain characteristic amplitudes of each sampling point should be close to each other; conversely, if there is local turbulence or orientation disorder, the characteristic amplitudes will show large dispersion.

[0043] Finally, the standard deviation of the frequency domain characteristic amplitude of all sampling points within the spatial slice is calculated as the tape-out consistency characteristic value of the spatial slice. Then, according to the physical position order of the chip on the green ceramic strip (e.g., row-major, from left to right, from top to bottom), each spatial slice is... The casting consistency characteristic values ​​are arranged in order of chip position to form a spectral encoded data sequence H that characterizes the casting orientation consistency of ceramic particles.

[0044] S2: During the dicing process, the morphology of the cutting edge corresponding to each chip position is collected simultaneously on the sintered ceramic substrate. The scratch morphology of each position is spatially segmented and geometric features are extracted to generate morphology coding data that characterizes the cutting edge breakage state.

[0045] It should be noted that after the ceramic substrate is sintered at high temperature, it needs to be divided into individual chip units through a dicing process. During the dicing process, due to tool wear, fluctuations in cutting parameters, or the release of internal stress in the material, chipping defects are easily generated at the cutting edges. The presence of chipping not only reduces the mechanical strength of the chip, but may also cause crack propagation in subsequent packaging processes, ultimately affecting the long-term reliability of the device. Traditional manual microscope sampling inspection methods are inefficient, have incomplete coverage, and are difficult to quantify. To solve the above problems, the specific operation of this invention is as follows:

[0046] S2.1: During the dicing process of the ceramic substrate, the cutting edge morphology of each chip position is simultaneously acquired by a line scanning white light interferometer to obtain three-dimensional surface contour data covering all cutting paths.

[0047] In this embodiment of the invention, during the dicing process of the ceramic substrate, the substrate is fixed on the worktable of the dicing machine, and the dicing blade separates the substrate along a preset cutting path. To simultaneously acquire the morphology of the cutting edge, a line-scanning white light interferometric probe is integrated behind or to the side of the dicing machine's blade holder. This probe uses the principle of white light interferometry, employing a beam splitter to divide broadband light into two beams: one beam is directed towards a reference mirror, and the other is focused onto the surface being measured. As soon as the dicing blade passes over the substrate, the probe immediately performs a line scan on the newly formed cutting edge.

[0048] Specifically, the interference probe is positioned perpendicular to the cutting path direction. Using a built-in linear CCD or CMOS camera, it acquires interference signals along an entire line perpendicular to the cutting path direction in a single exposure. As the stage moves continuously along the cutting path direction, the probe triggers acquisition at a fixed frequency, thereby constructing three-dimensional surface contour data covering the entire cutting path and the areas on both sides.

[0049] The three-dimensional surface contour data is stored in the form of a three-dimensional point cloud, denoted as . .in, The axial direction is parallel to the cutting path (dicing feed direction). The axial direction is perpendicular to the cutting path (the direction that crosses the edge). The axis represents the height value obtained through interferometric calculations, with a resolution down to the nanometer level. For the longitudinal and transverse dicing network on a rectangular substrate, the system requires two scans: the first scan covers all longitudinal dicings, and the second scan covers all transverse dicings, or a multi-probe layout can achieve full coverage in one go. The final data volume covers the four dicing boundaries of all chips on the substrate, meaning that all four edges of each chip are completely recorded.

[0050] It should be noted that by integrating online white light interferometric scanning during the dicing process, 100% inspection of cutting edge defects was achieved, avoiding secondary positioning errors and efficiency losses caused by offline detection. Furthermore, the acquisition of three-dimensional surface contour data not only includes the depth information of the edge breakage, but also fully records the spatial morphology of the edge breakage (such as length, width, and volume), providing high-precision raw data for subsequent multi-dimensional feature extraction.

[0051] S2.2: Divide the 3D surface contour data into multiple spatial pieces. Each space is partitioned This corresponds to the complete cutting boundary region of a chip. Specifically, the design size of the chip on each sintered substrate is approximately... For the first The chip has four cutting boundaries located at: the upper boundary lower boundary left boundary right boundary . Therefore, the first The dicing boundary region corresponding to each chip consists of four strip-shaped regions: each strip extends a certain width (e.g., ±50μm) to both sides of the dicing track to ensure complete coverage of the area where edge chipping may occur. It should be noted that the coordinates used here are the green ceramic tape design coordinates; the actual chip position will change due to sintering shrinkage and will be corrected later using coordinate registration in S3.

[0052] In actual processing, the strip-shaped areas around the edges of the chip are usually merged into a single outline data block to facilitate programming.

[0053] S2.3: Partition each space The contour data within the chip is used to extract the maximum chipping depth and the area of ​​the chipping region at the cut edge, which are then used as chipping feature parameters for the corresponding chip.

[0054] For each cutting boundary region, feature parameters are needed to quantify the severity of edge chipping. Edge chipping is mainly manifested as material peeling at the cutting edge, forming pits. First, edge detection algorithms (such as the Canny operator or threshold segmentation) are used to locate the theoretical edge line of the cutting path (usually the location of abrupt height changes) from the 3D contour, and the equation of the straight line is obtained by fitting using the least squares method. Then, regions unaffected by edge chipping (e.g., flat areas more than 50 μm away from the edge line) are selected on both sides of the edge line for plane fitting to obtain the reference plane. Next, all data points within the segment are traversed, and the signed distance from each point to the reference plane is calculated. Points with a distance greater than a preset depth threshold (e.g., 2 μm) and a connected region area greater than the minimum edge chipping area (e.g., 10 μm²) are marked as edge chipping regions.

[0055] Based on this, the maximum chipping depth in all chipping regions of the chip is calculated. That is, the value with the largest absolute depth among all marked points; at the same time, the total area of ​​all collapsed areas is calculated. That is, the sum of the projected areas of all marked points (which can be obtained by multiplying the pixel count by the area corresponding to a single pixel).

[0056] Therefore, for the first Each chip yields a two-dimensional feature vector. .

[0057] To facilitate subsequent association with terahertz spectrum encoded data, these feature vectors need to be arranged in the same chip position order as in S1, forming a shape encoded data sequence of length K. In this sequence, each element is a vector containing two values, representing the first, second, and third elements respectively. The sequence contains the maximum chipping depth and the area of ​​the chipped region for each chip. For chip locations where features could not be extracted due to damage or other reasons, the corresponding elements in the sequence are filled with invalid values ​​(such as NaN) to ensure that the sequence length is consistent with the total number of chips in the design. This sequence can be directly stored as a matrix, such as a two-dimensional array, where the first column is the depth value and the second column is the area value.

[0058] Optionally, to ensure data consistency and traceability, the physical coordinate index of each chip and the corresponding dicing process parameters (such as the cutter number and cutting speed of the current dicing channel) should be recorded while constructing the sequence, so that the specific process step can be quickly located when anomalies are found in subsequent analysis.

[0059] As can be seen, the morphology-encoded data sequence of the present invention simplifies complex two-dimensional / three-dimensional morphology into numerical vectors with clear physical meaning, greatly reducing the data dimension, while retaining the core information for assessing the severity of edge collapse. Furthermore, the ordered arrangement of the sequence provides a standardized interface for point-by-point association with spectral-encoded data, enabling subsequent multi-source data fusion to be performed directly based on the chip index, thus avoiding complex spatial registration calculations.

[0060] S3: Using chip location as an index, the spectrum coding data and the topography coding data are associated point by point to construct a multi-source process data pair for each chip location.

[0061] Preferably, using the coordinates of each chip location as an index, the corresponding position's tapering consistency feature value in the spectrum-encoded data sequence and the corresponding two-dimensional feature vector in the topography-encoded data sequence are paired to generate multi-source process data pairs for the corresponding chip locations. For example, the spectrum-encoded data sequence output by S1 is obtained. Each element Corresponding to the The film casting consistency characteristic values ​​of each chip region are obtained, and the order of this sequence is based on the chip layout coordinate set in the green ceramic tape design layout; simultaneously, the morphology encoded data sequence output by S2 is acquired. The arrangement order is based on the actual layout coordinate set of the chips on the substrate after sintering.

[0062] It should be noted that, due to the isotropic or anisotropic linear shrinkage (the shrinkage rate is usually 10%~20%) that occurs during the sintering process of the green ceramic tape, there is an overall scaling and translation deviation between the chip layout coordinates and the actual chip layout coordinates. If the correlation is directly performed by the sequence index, a systematic misalignment will be introduced.

[0063] To address the above issues, a mapping relationship needs to be established beforehand from the green ceramic tape coordinate system to the sintered substrate coordinate system. In practice, multiple optical alignment marks (such as crosshairs or circular holes) can be pre-fabricated at the edge or specific locations during the green ceramic tape casting process, and the coordinates of these marks can be synchronously recorded using S1 scanning data. Before or during the dicing process of the sintered substrate, the coordinates of the same set of marks are measured again using a white light interferometer or an additional high-resolution vision system. Based on the two sets of corresponding point coordinates, rigid transformation parameters (including scaling factor g, rotation angle θ, and translation vector Δ) are calculated to obtain the transformation matrix G, which ensures that for any chip center coordinate on the green ceramic tape... The corresponding coordinates on the substrate after sintering This transformation can be expressed in homogeneous coordinate form:

[0064] ;

[0065] in, This represents the translation amount.

[0066] After obtaining the optimal transformation parameters through least squares fitting, the chip layout coordinates in the green ceramic tape design layout will be used. All chip coordinates are transformed to the post-sintering space to obtain the registered coordinate set. Meanwhile, the actual chip coordinates measured in S2 can be obtained directly from the positioning system of the dicing machine or image recognition.

[0067] Finally, using the registered coordinates Based on this, the morphology encoding sequence is obtained through nearest neighbor search or bilinear interpolation. Each element in the sequence is remapped to the chip index that best matches its spatial location, forming an aligned morphology sequence. .like It has already been with If the data is highly consistent (e.g., sintering shrinkage has been compensated for by the process), this registration step can be skipped, and the original index can be used for direct association. During nearest neighbor matching, a maximum allowable distance (e.g., 100 μm) is set. If the minimum distance exceeds the set maximum allowable distance, it is determined that the corresponding design chip does not have corresponding topology data. Null values ​​(e.g., NaN) are filled into the corresponding row in the matrix, and the chip is skipped or processed separately in subsequent S4 steps.

[0068] After completing coordinate registration and sequence alignment, two sequences of length are obtained. And the data sequences with strictly corresponding order: spectral feature sequences (Reordered according to the spatial arrangement after sintering) and morphological feature sequence At this point, using the unique index of each chip... Using the chip ID as the primary key, the corresponding casting consistency feature value is combined with the two-dimensional topography vector to generate the first... A pair of multi-source process data for a single chip. This data pair can be viewed as a triplet or a three-dimensional feature vector.

[0069] In practice, to facilitate subsequent calculation and storage, a K×3 matrix can be constructed, with the row index corresponding to the chip number and the three columns being the consistent feature value of the tapering, the maximum depth of the chipping edge, and the area of ​​the chipping edge region, respectively.

[0070] Meanwhile, to preserve spatial information, the actual sintered coordinates of each chip can be stored as an additional column in the data table to form a multi-source process dataset containing spatial location. Each record includes: chip ID, X coordinate, Y coordinate, tape casting feature value, chipping depth, and chipping area.

[0071] S4: Calculate the flow consistency index for the spectral encoded data of each chip location, and calculate the edge collapse severity index for the topography encoded data; obtain the comprehensive quality score of the corresponding chip through weighted fusion; mark the chips with comprehensive quality scores below the preset score threshold as abnormal process parameter areas.

[0072] First, for each chip location, the delay consistency feature value is extracted from the spectrum-coded data sequence, and the Euclidean distance between the delay consistency feature value and the preset standard delay feature value is calculated. (The larger the Euclidean distance, the greater the deviation of the casting orientation from the standard state in that region, i.e., the worse the consistency). The normalized reciprocal of the Euclidean distance is used as the casting consistency index value of the corresponding chip.

[0073] In this embodiment of the invention, the specific transformation form of the normalized reciprocal can be either a linear mapping or a nonlinear mapping. For example, the following formula can be used to calculate the flow delay consistency index. :

[0074] ;

[0075] in, A section factor, used to control Sensitivity to indicator values. The value can be determined experimentally: select a batch of samples with known casting quality (good, medium, poor), and calculate... and adjust Make It can effectively distinguish different levels, usually making Distributed between 0 and 1. When hour, , indicates complete agreement; when As it approaches infinity, It approaches 0. This transformation ensures consistency metrics. The dimensionless and normalized values ​​reflect the degree of deviation in a monotonically decreasing manner.

[0076] The preset standard casting characteristic value can be set to 0, representing an ideal uniform state; or it can be calculated by measuring multiple batches of high-quality green ceramic tape samples and using the average value of the casting consistency characteristic value of all chip regions as a benchmark. Once the preset standard casting characteristic value is determined, it will serve as the benchmark for subsequent batch comparisons.

[0077] Secondly, for each chip location, the maximum value of the edge collapse depth is extracted from the topography encoding data sequence. and the area of ​​the collapsed edge The weighted sum of the two values ​​is then normalized to obtain the corresponding chip edge damage severity index value.

[0078] Before weighted summation, both parameters are first rendered dimensionless. An upper limit threshold for depth can be set based on historical data or process specifications. and the upper limit threshold of area (e.g., the upper limit threshold of depth) Take 1 / 3 of the dicing blade thickness as the upper limit threshold of the area. (Take 5% of the chip area). Then calculate the relative depth value. :

[0079] ;

[0080] relative area :

[0081] ;

[0082] This compresses both of them into the [0,1] interval.

[0083] When performing a weighted summation, the weights of the two factors equal 1. The weights can be set based on reliability test data: for example, by comparing the failure probabilities of chips with different chipping patterns after temperature cycling or mechanical shock, the contribution ratio of depth and area to failure can be determined. If there is no prior knowledge, they can be temporarily set to equal weights. The larger the chipping severity index value, the more severe the chipping (i.e., the worse the quality).

[0084] Furthermore, to comprehensively evaluate the overall process quality of the chip, the tape casting consistency index and the 1-edge chipping quality index need to be integrated into a comprehensive score. The 1-edge chipping severity index value is used as the edge chipping quality index, and it is weighted and fused with the tape casting consistency index to obtain the comprehensive quality score. The sum of their weights is 1. The weight allocation depends on the product's sensitivity to the two types of defects: if the device has extremely high requirements for substrate dielectric uniformity (such as microwave devices), the weight of the tape casting consistency index should be increased; if higher mechanical strength requirements are required (such as high-stress packaging), the weight of the edge chipping quality index should be increased. The weights can be determined through the analytic hierarchy process (AHP) or regression analysis based on historical failure data.

[0085] The overall quality score of each chip is compared with a preset scoring threshold. The preset scoring threshold needs to be set considering process capability and product reliability requirements: it can be based on the statistical distribution of overall scores from a large number of previously qualified products, taking the mean minus a certain number of standard deviations as the lower limit; or it can be directly set according to the customer's minimum acceptable quality requirements. If the overall quality score is less than the preset scoring threshold, then the chip is judged to be... The chip has abnormal process parameters and is marked as an abnormal area.

[0086] Furthermore, such as Figure 2 As shown, this embodiment also provides a high-reliability packaging quality assessment system based on multi-source process data, including:

[0087] The terahertz spectrum coding module performs terahertz spectrum scanning on the unsintered green ceramic tape, and spatially segments it according to the preset chip position. It extracts the terahertz frequency domain features in each segment and generates spectrum coding data that characterizes the consistency of the ceramic particle casting orientation.

[0088] The cutting morphology encoding module synchronously collects the cutting edge morphology corresponding to each chip position during the dicing process of the sintered ceramic substrate, performs spatial segmentation and geometric feature extraction on the scratch morphology of each position, and generates morphology encoding data that characterizes the cutting edge breakage state.

[0089] The multi-source data association module uses chip location as an index to associate spectrum-encoded data with topography-encoded data point by point, constructing multi-source process data pairs for each chip location;

[0090] The comprehensive quality scoring module calculates the tapering consistency index for the spectral encoded data of each chip location and the edge collapse severity index for the topography encoded data; it obtains the comprehensive quality score of the corresponding chip through weighted fusion; and it marks chips with comprehensive quality scores below the preset scoring threshold as areas with abnormal process parameters.

[0091] This embodiment also provides a computer device applicable to a high-reliability packaging quality assessment method based on multi-source process data, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the high-reliability packaging quality assessment method based on multi-source process data as proposed in the above embodiment.

[0092] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0093] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the high-reliability packaging quality assessment method based on multi-source process data as proposed in the above embodiments.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A high-reliability packaging quality assessment method based on multi-source process data, characterized in that: include: Terahertz spectral scanning was performed on the unsintered green ceramic tape, and spatial segmentation was performed according to the preset chip position. The terahertz frequency domain features in each segment were extracted to generate spectral coding data characterizing the uniformity of the ceramic particle casting orientation. During the dicing process, the morphology of the cutting edge corresponding to each chip position is collected simultaneously on the sintered ceramic substrate. The scratch morphology at each position is spatially segmented and geometric features are extracted to generate morphology coding data that characterizes the cutting edge breakage state. Using chip location as an index, the spectrum encoding data and the topography encoding data are associated point by point to construct a multi-source process data pair for each chip location; For each chip location, a flow delay consistency index is calculated based on the spectral encoded data, and a chipping severity index is calculated based on the topography encoded data. A comprehensive quality score for the corresponding chip is obtained through weighted fusion. Chips with a comprehensive quality score lower than a preset score threshold are marked as abnormal process parameter regions.

2. The high-reliability packaging quality assessment method based on multi-source process data as described in claim 1, characterized in that: Terahertz spectral scanning was performed on the unsintered green ceramic strip to obtain a three-dimensional terahertz data cube covering the entire surface of the green ceramic strip. The three dimensions of the three-dimensional terahertz data cube are the horizontal spatial coordinate X, the vertical spatial coordinate Y, and the terahertz time-domain sampling point, respectively. Each spatial coordinate point (X, Y) corresponds to a complete time-domain spectral curve. Based on the preset chip layout, the three-dimensional terahertz data cube is divided into several spatial segments. Each space is partitioned It corresponds to a chip location and contains time-domain spectral data of all sampling points within the region.

3. The high-reliability packaging quality assessment method based on multi-source process data as described in claim 2, characterized in that: The generation of the spectrum-coded data includes: For each spatial partition Fourier transforms were performed on all time-domain spectral curves within the sampled area to obtain the frequency-domain spectrum of each sampling point. For the frequency domain spectrum of each sampling point, the average amplitude in the preset characteristic frequency band is calculated as the frequency domain characteristic amplitude of the corresponding sampling point; Calculate the standard deviation of the frequency domain characteristic amplitude of all sampling points within the spatial segment, and use it as the flow-spread consistency characteristic value of the spatial segment; Divide each space into segments The standard deviations of the frequency domain characteristic amplitudes are arranged in order of chip position to form a spectral encoded data sequence that characterizes the consistency of the ceramic particle casting orientation.

4. The high-reliability packaging quality assessment method based on multi-source process data as described in claim 3, characterized in that: The spatial segmentation and geometric feature extraction of the scratch morphology at each location includes: During the dicing process of ceramic substrates, the morphology of the cutting edges at each chip position is simultaneously acquired by a line-scanning white light interferometer to obtain three-dimensional surface contour data covering all cutting paths. The three-dimensional surface contour data is segmented into multiple spatial pieces. Each space is partitioned This corresponds to the complete cutting boundary region of a chip; For each spatial partition The contour data within the chip is used to extract the maximum chipping depth and the area of ​​the chipping region at the cut edge, which are then used as chipping feature parameters of the chip.

5. The high-reliability packaging quality assessment method based on multi-source process data as described in claim 4, characterized in that: The maximum chipping depth and the area of ​​the chipping region of each chip are combined to form a two-dimensional feature vector, which is then arranged in order of chip position to generate a morphology-encoded data sequence that characterizes the chipping state. Each element in the morphology-encoded data sequence corresponds to a two-dimensional feature vector of a chip position.

6. The high-reliability packaging quality assessment method based on multi-source process data as described in claim 5, characterized in that: The point-to-point association includes: Using the coordinates of each chip location as an index, the standard deviation of the frequency domain feature amplitude at the corresponding position in the spectrum-encoded data sequence is paired with the two-dimensional feature vector at the same position in the topography-encoded data sequence to generate multi-source process data pairs for the corresponding chip locations.

7. The high-reliability packaging quality assessment method based on multi-source process data as described in claim 6, characterized in that: For each chip location, the standard deviation of the frequency domain feature amplitude is extracted from the spectrum-encoded data sequence. The Euclidean distance between the standard deviation of the frequency domain feature amplitude and the preset standard drag-out feature value is calculated. The normalized reciprocal of the Euclidean distance is used as the drag-out consistency index value of the corresponding chip. For each chip location, the maximum chipping depth and chipping area are extracted from the topography encoding data sequence. The two are then weighted, summed, and normalized to obtain the chipping severity index value for the corresponding chip. The 1-edge chipping severity index is used as the edge chipping quality index, and it is weighted and fused with the casting consistency index to obtain a comprehensive quality score. The weights of the two are summed to 1.

8. A high-reliability packaging quality assessment system based on multi-source process data, based on the high-reliability packaging quality assessment method based on multi-source process data as described in any one of claims 1 to 7, characterized in that: Also includes: The terahertz spectrum coding module performs terahertz spectrum scanning on the unsintered green ceramic tape, and spatially segments it according to the preset chip position. It extracts the terahertz frequency domain features in each segment and generates spectrum coding data that characterizes the consistency of the ceramic particle casting orientation. The cutting morphology encoding module synchronously collects the cutting edge morphology corresponding to each chip position during the dicing process of the sintered ceramic substrate, performs spatial segmentation and geometric feature extraction on the scratch morphology of each position, and generates morphology encoding data that characterizes the cutting edge breakage state. The multi-source data association module uses chip location as an index to associate the spectrum encoding data with the topography encoding data point by point, thereby constructing a multi-source process data pair for each chip location; The comprehensive quality scoring module calculates the flow-spinning consistency index for the spectral encoded data at each chip location and the edge-break severity index for the morphology encoded data; it obtains the comprehensive quality score of the corresponding chip through weighted fusion; and it marks chips with comprehensive quality scores below a preset scoring threshold as areas with abnormal process parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the high-reliability packaging quality assessment method based on multi-source process data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the high-reliability packaging quality assessment method based on multi-source process data as described in any one of claims 1 to 7.

Images