A method and system for testing a semiconductor device

Abstract

The application provides a kind of semiconductor device test method and system, method includes: obtaining the performance benchmark data of each test device, performance benchmark data includes the benchmark value of multiple performances;And obtain the performance test data of each test device at each time point in the aging test process;According to performance benchmark data, the benchmark sample of each test device is constructed, and the measured sample of each test device is constructed according to the performance test data at each time point;The benchmark sample and measured sample are clustered, and the clustering processing result is obtained;According to clustering processing result, whether each test device exists anomaly is determined.The application can accurately distinguish normal device and abnormal device by clustering processing to benchmark sample and measured sample, and automatically identify abnormal type, which is beneficial to improve the analysis efficiency and reliability of aging test data.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device testing technology, and more specifically, to a method and system for testing semiconductor devices. Background Technology

[0002] Semiconductor device aging tests are an important means of evaluating the long-term reliability of devices. By applying stresses such as high temperature, voltage, and current to the device, the potential defects inside are exposed, thereby obtaining reliability data of the device under normal operating conditions in a short period of time.

[0003] However, traditional aging test data processing methods have significant shortcomings. On the one hand, conventional methods typically rely on a single performance parameter or a preset fixed threshold to determine whether a device has failed, making it difficult to fully utilize the degradation information contained in multi-dimensional performance test data. On the other hand, different devices exhibit normal process fluctuations in their initial performance, and the magnitude and direction of degradation during the aging process also vary. Traditional threshold comparison methods are prone to misjudging normal fluctuations as abnormalities or missing truly abnormal devices. Furthermore, most existing methods can only provide a binary conclusion of pass / fail, failing to provide fine-grained classification of anomaly types, thus making it difficult to meet the growing demands for failure tracing and process improvement. Summary of the Invention

[0004] The purpose of this invention is to provide a testing method and system for semiconductor devices to improve the above-mentioned problems.

[0005] To achieve the above objectives, the embodiments of this application provide the following technical solutions: On one hand, embodiments of this application provide a method for testing semiconductor devices, the method comprising: Acquire performance baseline data for each device under test, including baseline values ​​for various performance characteristics; and acquire performance test data for each device under test at each time point during the aging test. A baseline sample for each device under test is constructed based on performance baseline data, and a measured sample for each device under test is constructed based on performance test data at each time point. Clustering is performed on the baseline samples and the measured samples to obtain the clustering results; based on the clustering results, it is determined whether each device under test has any abnormalities.

[0006] Secondly, embodiments of this application provide a testing system for semiconductor devices, the system comprising: The acquisition module is used to acquire performance baseline data for each device under test, including baseline values ​​for various performance characteristics; and to acquire performance test data for each device under test at each time point during the aging test. The construction module is used to build a baseline sample for each device under test based on performance baseline data, and to build a test sample for each device under test based on performance test data at each time point. The clustering module is used to cluster the benchmark samples and the measured samples to obtain the clustering results; based on the clustering results, it is determined whether each device under test has any abnormalities.

[0007] Thirdly, embodiments of this application provide a testing apparatus for semiconductor devices, the apparatus including a memory and a processor. The memory is used to store a computer program; the processor is used to execute the computer program to implement the steps of the aforementioned testing method for semiconductor devices.

[0008] Fourthly, embodiments of this application provide a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described semiconductor device testing method.

[0009] The beneficial effects of this invention are as follows: 1. This invention constructs a benchmark sample set that reflects the normal process fluctuation range, and performs cluster analysis on the measured samples of each device together with the benchmark samples. This can effectively distinguish between normal performance distribution and true abnormal degradation, avoid misjudgment caused by measurement noise or individual differences, and improve the accuracy of anomaly detection.

[0010] 2. This invention, through two-level clustering, can not only identify abnormal devices, but also further classify different types of abnormal patterns in a more refined manner. By calculating the centroid of each abnormal cluster and comparing it with the global mean dimension by dimension, the performance dimension with the largest deviation is accurately located as the abnormal type of the cluster, providing a clear direction for failure tracing.

[0011] 3. The test results output by this invention can be directly used for device quality sorting. Normal devices enter the qualified product warehouse, abnormal devices are classified and screened or scrapped according to the type of abnormality, and extreme abnormal devices are directly removed. At the same time, the distribution information of various abnormal devices can be fed back to the process stage, which can help locate the source of potential defects in the manufacturing process.

[0012] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the testing method for semiconductor devices described in the embodiments of the present invention; Figure 2 This is a schematic diagram of the test system structure for the semiconductor device described in this embodiment of the invention; Figure 3 This is a schematic diagram of the test equipment structure for the semiconductor device described in this embodiment of the invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0016] It should be noted that similar reference numerals or letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0017] Example 1 like Figure 1 As shown, this embodiment provides a method for testing semiconductor devices, which includes steps S1, S2 and S3.

[0018] Step S1: Obtain the performance baseline data for each device under test, which includes baseline values ​​for various performance characteristics; and obtain the performance test data for each device under test at each time point during the aging test. In this embodiment, the device under test (DUT) can be any type of semiconductor device, such as a metal-oxide-semiconductor field-effect transistor (MOSFET), a bipolar transistor (BPT), or an insulated-gate bipolar transistor (IGBT). The performance baseline data is obtained by performing initial performance tests on each DUT at room temperature before the aging test begins, and includes baseline values ​​for various performance parameters. These parameters may include, for example, electrical performance parameters such as threshold voltage, on-resistance, transconductance, and leakage current.

[0019] Aging testing refers to the process of applying certain stresses (such as high temperature, high humidity, voltage stress, current stress, etc.) to semiconductor devices to accelerate their aging. High Temperature Operating Life (HTOL) testing can be used for aging. During the aging test, a performance test is performed on each device under test at each preset time point, and the performance test data at that time point is recorded. The types of performance included in the performance test data are the same as those in the performance baseline data.

[0020] Step S2: Construct a baseline sample for each device under test based on the performance baseline data, and construct a test sample for each device under test based on the performance test data at each time point. The specific implementation steps of this step include step S21 and step S22; Step S21: Take the performance baseline data of each device under test as a basic sample, and generate expanded samples according to the preset number of expanded samples: For each basic sample, generate a random fluctuation amount independently in each performance dimension each time. The random fluctuation amount in each performance dimension follows a normal distribution with a mean of zero and a standard deviation of the absolute value of the baseline value of that performance dimension multiplied by 2%. Add the baseline value of each performance dimension to the random fluctuation amount generated in the corresponding performance dimension to obtain the new value of the baseline sample in each performance dimension. Use the new value of each performance dimension as an expanded sample. In this step, the performance baseline data of the device under test is first used as a basic sample, and then the basic sample is expanded multiple times. Each time, a random fluctuation that follows a normal distribution is added independently to each performance dimension, thereby generating multiple expanded samples that reflect the normal process fluctuation range. The purpose of this is to enable the subsequent clustering algorithm to identify the performance distribution range that a normal device should have, and to avoid misjudging normal fluctuations caused by measurement noise or small individual differences as abnormal. Step S22: Record each basic sample and its corresponding extended sample as a benchmark sample; during the aging test, take the average value of the performance test data at all time points in each dimension to obtain the measured mean value of the device under test in each performance dimension, and combine the measured mean values ​​in each dimension to form a measured sample.

[0021] In this step, the construction of the measured sample involves averaging the performance test data of the device at all time points during the aging test for each performance dimension, and finally merging them into a single measured sample. The purpose of averaging is to compress information from multiple time points into a comprehensive characterization, reflecting the overall degradation trend of the device throughout the aging process. Ultimately, each device corresponds to a set of benchmark samples and one measured sample. Clustering can then be used to determine whether the measured point falls within the normal dispersion range, thereby achieving anomaly detection.

[0022] In addition to steps S21 and S22, step S2 can also be implemented through steps S23 and S24. Step S23: Take the performance benchmark data of each device under test as a basic sample, and use the K-means algorithm to cluster all basic samples to obtain multiple clusters; calculate the average value of all basic samples in each cluster in each performance dimension to obtain the cluster center of each cluster; count the number of basic samples contained in each cluster and calculate the average number of all basic samples to obtain the average sample size; determine the cluster with the largest number of basic samples as the benchmark cluster, and classify the remaining clusters as non-benchmark clusters. Step S24: Calculate the Euclidean distance between the cluster center of each non-reference cluster and the cluster center of the reference cluster, as the deviation distance; designate non-reference clusters with deviation distances greater than a preset deviation distance threshold as far-biased clusters, and non-reference clusters with deviation distances less than or equal to the preset deviation distance threshold as near-biased clusters; for each far-biased cluster, calculate the ratio of the number of samples in the far-biased cluster to the average sample size to obtain the size coefficient of the far-biased cluster; multiply the preset basic expansion quantity by the size coefficient, and then multiply by a magnification factor greater than 1 to obtain the final expansion quantity of the far-biased cluster; for each near-biased cluster, calculate the ratio of the number of samples in the near-biased cluster to the average sample size to obtain the size coefficient of the near-biased cluster; multiply the preset basic expansion quantity by the size coefficient, and then multiply by a compression factor between 0 and 1 to obtain the expansion quantity of the near-biased cluster. In this step, the deviation distance threshold can be the average deviation distance of all non-benchmark clusters; the magnification factor can be 1.5~3.0, and the compression factor can be 0.2~0.8. Step S25: For each far-partial cluster and each near-partial cluster, perform the following operations respectively: Expansion steps: Extract the cluster center of the cluster and denote it as the baseline vector; calculate the standard deviation of each performance dimension within the cluster and combine the standard deviations of each performance dimension to form a standard deviation vector; independently and randomly sample from the standard normal distribution to generate a random perturbation vector with the same number of dimensions as the performance dimensions; multiply each component of the random perturbation vector by the corresponding dimension component of the standard deviation vector, and add the multiplied sample to the baseline vector to generate a new expanded sample; Repeat the expansion step until the number of expanded samples generated reaches the expansion number allocated to the cluster. Step S26: Record both the basic sample and the extended sample as reference samples; during the aging test, take the average value of the performance test data at all time points in each dimension to obtain the measured mean value of the device under test in each performance dimension, and combine the measured mean values ​​in each dimension to form a measured sample.

[0023] In the above expansion steps, the cluster with the largest number of basic samples is used as the benchmark cluster to represent the mainstream performance distribution. Clusters with large deviations and near deviations are differentiated by applying enhancement and attenuation factors, respectively. This allows a smaller number of samples located at the edge of the performance distribution to receive more sufficient supplementation, ensuring that the expanded benchmark sample set retains the distribution ratio of the original data while effectively covering the complete boundary range of normal fluctuations. Furthermore, this scheme uses the center of each cluster as the benchmark vector and the standard deviation of each dimension of each cluster as the standard deviation vector. Samples are taken from the standard normal distribution, multiplied term by term, and then superimposed onto the benchmark vector. This ensures that the generated expanded samples are naturally distributed within the original distribution range of each cluster, avoiding distribution distortion caused by simply adding a fixed proportion of noise.

[0024] Step S3: Perform clustering on the benchmark sample and the measured sample to obtain the clustering results; determine whether each device under test has any abnormalities based on the clustering results.

[0025] In this step, the specific implementation steps for clustering the benchmark sample and the measured sample to obtain the clustering result include steps S31 and S32. Step S31: Record both the benchmark sample and the measured sample as data points, and form the first dataset by collecting all the data points; obtain the first density neighborhood radius and the first density point threshold, and then, based on the DBSCAN clustering algorithm, traverse each data point in the first dataset in turn: The number of data points within the radius of the first density neighborhood centered on the data point is counted. If the number of data points is greater than or equal to the threshold of the first density points, the data point is determined to be a core point. Then, a first cluster is created starting from the core point, and all data points within the first density neighborhood radius of the data point are included in the first cluster. Then, each data point in the first cluster is traversed. If the data point is also a core point, all data points within its first density neighborhood are also included in the first cluster. In this step, the algorithm iterates through every data point in the dataset. For the currently selected data point, it counts the total number of data points contained within a circle centered on that data point and with a radius equal to the first density neighborhood. If this number reaches or exceeds the first density threshold, the data point is determined to be a core point, meaning that the data point is surrounded by a high-density region. After being determined to be a core point, the algorithm uses that data point as the starting point and groups all data points within its radius into the same cluster. Then, it checks whether newly added data points within the cluster also meet the core point condition. If they do, the data points within its radius are added to the cluster. This recursive expansion allows the clusters to grow continuously along density-connected regions until no more new data points can be added. Subsequently, the algorithm selects a new starting point from the unvisited data points and repeats the above process until all data points in the dataset have been visited.

[0026] Step S32: After traversing every data point in the first dataset, multiple first clusters are obtained; the clustering is checked for compliance based on the multiple first clusters. If the check is qualified, the first cluster with the most data points is recorded as the normal cluster, and the clustering is completed based on the normal cluster to obtain the clustering result; if the check is unqualified, the first density neighborhood radius and the first density point threshold are obtained again and the clustering is performed again until the check is qualified.

[0027] After traversal, multiple first clusters of varying sizes are obtained. The first cluster containing the most data points is designated as the normal cluster. This is because all data points are composed of baseline samples and measured samples from all devices. For normal devices, their initial performance baseline samples are naturally dense in space; after aging, their degradation is small and consistent in direction. Although the measured samples undergo overall translation, they remain dense, and the translation distance is limited. The baseline sample group and the measured sample group remain close to each other in the feature space, forming a large-scale, high-density region. Since normal devices constitute the vast majority, this region naturally contains far more data points than other regions. Abnormal devices, due to significant deviations in degradation magnitude or direction from the mainstream group, have samples far from the main group, forming small clusters or isolated points. Therefore, the largest cluster is designated as the normal cluster. Subsequent determinations only require checking whether the measured samples of each device fall into this cluster to identify any abnormalities.

[0028] In addition, the specific implementation steps for performing qualification testing on this clustering based on multiple first clusters in this step include steps S321 and S322; Step S321: For each first cluster, determine whether the total number of data points within the first cluster is less than a threshold. If so, directly use all data points within the first cluster as edge points of the first cluster; otherwise, take the average value of all data points within the first cluster on each performance dimension to obtain the center point of the first cluster. From all data points within the first cluster, select the data point farthest from the center point as the first edge point and add it to the edge point set. Repeat the following operation until the number of data points in the edge point set reaches the preset number: For each data point in the first cluster except for the edge points, calculate the distance from each data point to the center point to obtain the first distance, and calculate the nearest distance from each data point to all edge points in the current edge point set to obtain the second distance; sum the first distance and the second distance according to preset weights to obtain the comprehensive score of the data point; select the data point with the highest comprehensive score as the edge point and add it to the edge point set; In this step, calculating the shortest distance from the data point to all edge points in the current edge point set specifically means: for the data point to be calculated, calculate the Euclidean distance between it and each edge point in the current edge point set, and then take the minimum value among these distance values. This minimum value is the shortest distance from the data point to be calculated to all edge points in the current edge point set. The weights can be 0.4 and 0.6, or can be customized according to user needs. Step S322: Perform a qualification test based on the edge points of each first cluster.

[0029] In step S322, the specific implementation steps for performing qualification detection based on the edge points of each first cluster include steps S3221 and S3222. Step S3221: For each first cluster, calculate the mean edge radius and the nearest edge distance of the first cluster; wherein, the mean edge radius is the average distance from each edge point of the first cluster to its center point; the nearest edge distance is the minimum value among the nearest distances from each edge point of the first cluster to all other edge points except for the first cluster. Step S3222: Perform a pass / fail check: Sort each first cluster in descending order of mean edge radius, and select the top three first clusters to form the first candidate set; Sort each first cluster in ascending order of nearest edge distance, and select the top three first clusters to form the second candidate set; If the first candidate set contains the first cluster with the most data points, and the second candidate set contains the first cluster with the fewest data points, then the current cluster is deemed pass / fail; otherwise, it is deemed fail / fail.

[0030] In this step, the method for calculating the minimum nearest distance to each edge point of the first cluster from all other edge points excluding the first cluster is as follows: First, calculate the minimum distance from each edge point of the first cluster to all other edge points (that is, the set of edge points corresponding to other first clusters besides the first cluster itself); then take the minimum value among all the nearest distances corresponding to all edge points of the first cluster as the nearest edge distance of the first cluster.

[0031] In step S32, the specific implementation steps for completing clustering based on normal clusters and obtaining the clustering results include steps S323 and S324. Step S323: Collect the data points in the first dataset that do not belong to the normal clusters to obtain the second dataset. Set the second density neighborhood radius and the second density point number threshold. The second density neighborhood radius is smaller than the first density neighborhood radius, and the second density point number threshold is smaller than the first density point number threshold. Iterate through each data point in the second dataset in turn. The system counts the number of data points within the radius of the second density neighborhood centered on the data point. If the number of data points is greater than or equal to the threshold for the number of second density points, the data point is determined to be a core point. A second cluster is then created starting from this core point, and all data points within the radius of the second density neighborhood of the data point are included in this second cluster. Then, the system iterates through each data point in the second cluster. If the data point is also a core point, all data points within its second density neighborhood are also included in the second cluster. If the data point is not a core point, it is marked as a point of high interest.

[0032] Step S321 involves performing a second-level refined clustering on the remaining data points from the first-level clustering that were not assigned to normal clusters, in order to distinguish different types of anomalous patterns. Since the first-level clustering uses a large density neighborhood radius and a high density point threshold, its function is to coarsely screen out the largest normal group. However, the number of anomalous samples is small and their distribution is sparse. If the same parameters are used continuously, effective clustering will not be possible. Therefore, the second-level clustering adjusts both parameters accordingly: reducing the radius is to avoid mixing different types of anomalous patterns together, and lowering the threshold is to ensure that even with sparse samples, a few closely spaced anomalous samples can still be identified as clusters.

[0033] When traversing the second dataset, a second cluster is created starting with the data points that meet the core point criteria. All data points within the density neighborhood radius of this cluster are included in this cluster, and the cluster is continuously expanded by other core points within the cluster. For data points that have been included in the cluster but do not meet the core point criteria, a new cluster is not created starting with them. Instead, they are marked as high-interest points, indicating marginal individuals with a large degree of deviation, which need to be given special attention in the future.

[0034] In this embodiment, the first clustering uses larger parameters to determine normal clusters, and their quality directly affects subsequent judgments. Therefore, a pass / fail check is set to ensure the reliability of normal clusters. The second clustering only subdivides the remaining data points to identify different types of abnormal patterns, so no further checks are set.

[0035] In step S3, the specific implementation steps for determining whether each device under test has an anomaly based on the clustering results include steps S33 and S34. Step S33: Take the average of the performance baseline data of all tested devices on each performance dimension to obtain the performance mean data; take the average of all data points in the second cluster on each performance dimension to obtain the centroid of the second cluster; compare the centroid with the performance mean data dimension by dimension, and the performance dimension with the largest deviation is the anomaly type of the second cluster. If the deviation of multiple performance dimensions is the same and all are the maximum value, then the anomaly type of the second cluster is marked as a comprehensive anomaly. In this step, for each second cluster, the centroid and the performance mean data are compared on the same performance dimension. The absolute value of the difference between the two is divided by the performance mean data on that performance dimension, and then multiplied by 100%. The resulting percentage is the deviation on that performance dimension.

[0036] Step S34: After clustering is completed, check the classification of the test sample of each device under test in the clustering results. If the test sample falls into the normal cluster, the device under test is marked as a normal device. If the test sample falls into any second cluster, the abnormality type corresponding to the second cluster is used as the label of the device under test. If the test sample is a point of high interest, the device under test is marked as an extreme abnormal device.

[0037] Steps S33 and S34 together complete the final anomaly determination for each device under test. First, for each second cluster formed in the second-level clustering, its centroid is calculated, and the global mean of all device performance benchmark data is calculated simultaneously. The two are compared dimension by dimension. If only one dimension has the largest deviation, then that dimension is determined as the anomaly type of that cluster (for example, the largest deviation in the threshold voltage dimension corresponds to "threshold voltage anomaly"). If the deviations of multiple performance dimensions are the same and all are the maximum values, then the anomaly type of that second cluster is marked as "comprehensive anomaly" to characterize that the anomaly involves the coupling degradation of multiple performance parameters. Subsequently, the belonging of the measured sample of each device under test in the clustering results is checked: if it falls into a normal cluster, it is marked as a normal device; if it falls into a second cluster, the anomaly type corresponding to that cluster is used as the label for that device; if it is marked as a point of high concern, it is marked as an extreme anomaly device, indicating that it has the largest deviation and the highest reliability risk in that anomaly type. After these two steps, each device under test obtains a clear test result.

[0038] After obtaining the test results, the entire batch of components can be sorted for quality: normal components enter the qualified product warehouse, abnormal components are classified and screened or scrapped according to the type of abnormality, and extremely abnormal components are directly removed. Simultaneously, the distribution information of various abnormal components can be fed back to the process stage, helping to locate potential sources of defects in the manufacturing process. Furthermore, the abnormality rate and degradation degree data of this batch of components can also be used to assess the overall reliability level of the product, providing a basis for life prediction and quality assurance strategies. In short, the conclusions output by this method can guide production decisions, failure tracing, and reliability assessment.

[0039] Example 2 like Figure 2 As shown, this embodiment provides a testing system for semiconductor devices, the system including an acquisition module 1, a construction module 2, and a clustering module 3.

[0040] Module 1 is used to acquire performance baseline data for each device under test, including baseline values ​​for various performance characteristics; and to acquire performance test data for each device under test at each time point during the aging test. Module 2 is used to construct a baseline sample for each device under test based on performance baseline data, and to construct a test sample for each device under test based on performance test data at each time point. Clustering module 3 is used to perform clustering processing on the benchmark sample and the measured sample to obtain the clustering processing results; and to determine whether each tested device has any abnormalities based on the clustering processing results.

[0041] In one specific embodiment of this disclosure, the building module 2 further includes an expansion unit 21 and a combination unit 22.

[0042] The expansion unit 21 is used to take the performance benchmark data of each device under test as a basic sample and generate expansion samples according to a preset number of expansion samples: for each basic sample, a random fluctuation is generated independently in each performance dimension each time. The random fluctuation in each performance dimension follows a normal distribution with a mean of zero and a standard deviation of the absolute value of the benchmark value of that performance dimension multiplied by two percent; the benchmark value of each performance dimension is added to the random fluctuation generated in the corresponding performance dimension to obtain the new value of the benchmark sample in each performance dimension, and the new value of each performance dimension is used as an expansion sample; Combination unit 22 is used to record each basic sample and its corresponding extended sample as a reference sample; during the aging test, the performance test data at all time points are averaged in each dimension to obtain the measured mean of the device under test in each performance dimension, and the measured mean of each dimension is combined to form a measured sample.

[0043] In one specific embodiment of this disclosure, the clustering module 3 further includes a traversal unit 31 and a labeling unit 32.

[0044] Traversal unit 31 is used to record both the benchmark sample and the measured sample as data points, and to form the first dataset by collecting all the data points; it obtains the first density neighborhood radius and the first density point threshold, and then, based on the DBSCAN clustering algorithm, it traverses each data point in the first dataset in turn: The number of data points within the radius of the first density neighborhood centered on the data point is counted. If the number of data points is greater than or equal to the threshold of the first density points, the data point is determined to be a core point. Then, a first cluster is created starting from the core point, and all data points within the first density neighborhood radius of the data point are included in the first cluster. Then, each data point in the first cluster is traversed. If the data point is also a core point, all data points within its first density neighborhood are also included in the first cluster. The labeling unit 32 is used to traverse every data point in the first dataset to obtain multiple first clusters; the clustering is tested for compliance based on the multiple first clusters. If the test is successful, the first cluster with the most data points is recorded as the normal cluster, and the clustering is completed based on the normal cluster to obtain the clustering result; if the test is unsuccessful, the first density neighborhood radius and the first density point number threshold are obtained again and the clustering is performed again until the test is successful.

[0045] In one specific embodiment of this disclosure, the marking unit 32 further includes a judgment unit 321 and a detection unit 322.

[0046] The judgment unit 321 is used to determine whether the total number of data points in each first cluster is less than a threshold. If so, all data points in the first cluster are directly used as edge points of the first cluster; otherwise, the average value of all data points in the first cluster is taken on each performance dimension to obtain the center point of the first cluster. From all data points in the first cluster, the data point farthest from the center point is selected as the first edge point and added to the edge point set. The following operation is repeated until the number of data points in the edge point set reaches a preset number: For each data point in the first cluster except for the edge points, calculate the distance from each data point to the center point to obtain the first distance, and calculate the nearest distance from each data point to all edge points in the current edge point set to obtain the second distance; sum the first distance and the second distance according to preset weights to obtain the comprehensive score of the data point; select the data point with the highest comprehensive score as the edge point and add it to the edge point set; The detection unit 322 is used to perform qualification detection based on the edge points of each first cluster.

[0047] In one specific embodiment of this disclosure, the detection unit 322 further includes a calculation unit 3221 and a determination unit 3222.

[0048] The calculation unit 3221 is used to calculate the mean edge radius and the nearest edge distance for each first cluster; wherein, the mean edge radius is the average distance from each edge point of the first cluster to its center point; and the nearest edge distance is the minimum value among the nearest distances from each edge point of the first cluster to all other edge points except for the first cluster. The judgment unit 3222 is used to perform a pass / fail test: sort each first cluster in descending order of the mean edge radius, and select the top three first clusters to form a first candidate set; sort each first cluster in ascending order of the nearest edge distance, and select the top three first clusters to form a second candidate set; if the first candidate set contains the first cluster with the most data points, and the second candidate set contains the first cluster with the fewest data points, then the current cluster is judged to be qualified, otherwise it is unqualified.

[0049] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0050] Example 3 Corresponding to the above method embodiments, this disclosure also provides a test device for semiconductor devices. The test device for semiconductor devices described below can be referred to in correspondence with the test method for semiconductor devices described above.

[0051] Figure 3 This is a block diagram illustrating a test apparatus 300 for a semiconductor device according to an exemplary embodiment. Figure 3 As shown, the test equipment 300 for the semiconductor device may include a processor 301 and a memory 302. The test equipment 300 for the semiconductor device may also include one or more of a multimedia component 303, an I / O interface 304, and a communication component 305.

[0052] The processor 301 controls the overall operation of the semiconductor device testing equipment 300 to complete all or part of the steps in the aforementioned semiconductor device testing method. The memory 302 stores various types of data to support the operation of the semiconductor device testing equipment 300. This data may include, for example, instructions for any application or method operating on the semiconductor device testing equipment 300, and application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 302 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 303 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 302 or transmitted via the communication component 305. The audio component also includes at least one speaker for outputting audio signals. I / O interface 304 provides an interface between processor 301 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 305 is used for wired or wireless communication between the semiconductor device test equipment 300 and other devices. Wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 305 may include a Wi-Fi module, a Bluetooth module, or an NFC module.

[0053] In an exemplary embodiment, the test equipment 300 for the semiconductor device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the aforementioned test method for the semiconductor device.

[0054] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the semiconductor device testing method described above. For example, the computer-readable storage medium may be the memory 302 including program instructions described above, which may be executed by the processor 301 of the semiconductor device testing equipment 300 to complete the semiconductor device testing method described above.

[0055] Example 4 Corresponding to the above method embodiments, this disclosure also provides a readable storage medium. The readable storage medium described below can be referred to in conjunction with the test method for the semiconductor device described above.

[0056] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the testing method for the semiconductor device described in the above method embodiments.

[0057] Specifically, the readable storage medium can be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for testing semiconductor devices, characterized in that, include: Acquire performance baseline data for each device under test, which includes baseline values ​​for various performance characteristics; And to obtain performance test data for each device under test at each time point during the aging test; A baseline sample for each device under test is constructed based on performance baseline data, and a measured sample for each device under test is constructed based on performance test data at each time point. Clustering is performed on the baseline sample and the measured sample to obtain the clustering results; The clustering results are used to determine whether each device under test has any abnormalities. The clustering process involves performing clustering on the baseline samples and the measured samples to obtain the clustering results, including: Both the baseline sample and the measured sample are recorded as data points, and the set of all data points is used to form the first dataset. The radius of the first density neighborhood and the threshold of the number of first density points are obtained, and then each data point in the first dataset is traversed sequentially based on the DBSCAN clustering algorithm. The number of data points within the radius of the first density neighborhood centered on the data point is counted. If the number of data points is greater than or equal to the threshold of the first density points, the data point is determined to be a core point. Then, a first cluster is created starting from the core point, and all data points within the first density neighborhood radius of the data point are included in the first cluster. Then, each data point in the first cluster is traversed. If the data point is also a core point, all data points within its first density neighborhood are also included in the first cluster. After traversing every data point in the first dataset, multiple first clusters are obtained. The clustering is then checked for compliance based on these first clusters. If the check is successful, the first cluster with the most data points is designated as the normal cluster, and clustering is completed based on the normal cluster to obtain the clustering result. If the check fails, the first density neighborhood radius and the first density point threshold are re-acquired, and clustering is performed again until the check is successful. Among them, the clustering was qualified based on multiple first clusters, including: For each first cluster, determine whether the total number of data points within the first cluster is less than a threshold. If so, directly use all data points within the first cluster as edge points. Otherwise, take the average of all data points within the first cluster across each performance dimension to obtain the center point of the first cluster. From all data points within the first cluster, select the data point farthest from the center point as the first edge point and add it to the edge point set. Repeat the following operation until the number of data points in the edge point set reaches a preset number: For each data point in the first cluster except for the edge points, calculate the distance from each data point to the center point to obtain the first distance, and calculate the nearest distance from each data point to all edge points in the current edge point set to obtain the second distance; sum the first distance and the second distance according to preset weights to obtain the comprehensive score of the data point; select the data point with the highest comprehensive score as the edge point and add it to the edge point set; Perform a pass / fail check based on the edge points of each first cluster; The qualification check is performed based on the final edge points of each first cluster, including: For each first cluster, calculate the mean edge radius and the nearest edge distance for that first cluster; where the mean edge radius is the average distance from each edge point of the first cluster to its center point; and the nearest edge distance is the minimum of the nearest distances from each edge point of the first cluster to all other edge points except for the first cluster. The following steps are performed to determine the suitability of the clusters: First clusters are sorted from largest to smallest based on their mean edge radius, and the top three first clusters are selected to form the first candidate set. First clusters are then sorted from smallest to largest based on their nearest edge distance, and the top three first clusters are selected to form the second candidate set. If the first candidate set contains the first cluster with the most data points, and the second candidate set contains the first cluster with the fewest data points, then the current cluster is considered acceptable; otherwise, it is considered unacceptable. Among them, clustering is completed based on normal clusters to obtain clustering results, including: Data points in the first dataset that do not belong to normal clusters are grouped together to obtain the second dataset. A second density neighborhood radius and a second density point count threshold are set, where the second density neighborhood radius is smaller than the first density neighborhood radius, and the second density point count threshold is smaller than the first density point count threshold. Each data point in the second dataset is then iterated through sequentially. The system counts the number of data points within the radius of the second density neighborhood centered on the data point. If the number of data points is greater than or equal to the threshold for the number of second density points, the data point is determined to be a core point. A second cluster is then created starting from this core point, and all data points within the radius of the second density neighborhood of the data point are included in this second cluster. Then, the system iterates through each data point in the second cluster. If the data point is also a core point, all data points within its second density neighborhood are also included in the second cluster. If the data point is not a core point, it is marked as a point of high interest.

2. The testing method for semiconductor devices according to claim 1, characterized in that, A baseline sample is constructed for each device under test based on performance baseline data, and a measured sample is constructed for each device under test based on performance test data at each time point, including: The performance baseline data of each device under test is used as a basic sample. Expanded samples are generated according to the preset number of expanded samples: For each basic sample, a random fluctuation is generated independently for each performance dimension. The random fluctuation for each performance dimension follows a normal distribution with a mean of zero and a standard deviation equal to the absolute value of the baseline value for that performance dimension multiplied by 2%. The baseline value for each performance dimension is added to the random fluctuation generated for the corresponding performance dimension to obtain the new value of the baseline sample for each performance dimension. The new value for each performance dimension is used as an expanded sample. Each basic sample and its corresponding extended sample are recorded as a baseline sample. During the aging test, the performance test data at all time points are averaged in each dimension to obtain the measured mean of the device under test in each performance dimension. The measured mean of each dimension is then combined to form a measured sample.

3. A testing system for semiconductor devices, characterized in that, include: The acquisition module is used to acquire performance benchmark data for each device under test. The performance benchmark data includes benchmark values ​​for various performance characteristics. And to obtain performance test data for each device under test at each time point during the aging test; The construction module is used to build a baseline sample for each device under test based on performance baseline data, and to build a test sample for each device under test based on performance test data at each time point. The clustering module is used to perform clustering processing on the benchmark samples and the measured samples to obtain the clustering results; The clustering results are used to determine whether each device under test has any abnormalities. The clustering module includes: The traversal unit is used to record both the benchmark sample and the measured sample as data points, and to form the first dataset by collecting all the data points; the radius of the first density neighborhood and the threshold of the first density point count are obtained, and then each data point in the first dataset is traversed sequentially based on the DBSCAN clustering algorithm: The number of data points within the radius of the first density neighborhood centered on the data point is counted. If the number of data points is greater than or equal to the threshold of the first density points, the data point is determined to be a core point. Then, a first cluster is created starting from the core point, and all data points within the first density neighborhood radius of the data point are included in the first cluster. Then, each data point in the first cluster is traversed. If the data point is also a core point, all data points within its first density neighborhood are also included in the first cluster. The labeling unit is used to traverse every data point in the first dataset to obtain multiple first clusters. Based on these multiple first clusters, the current clustering is checked for compliance. If the check is successful, the first cluster with the most data points is recorded as the normal cluster, and the clustering is completed based on the normal cluster to obtain the clustering result. If the check fails, the first density neighborhood radius and the first density point threshold are re-acquired and the clustering is performed again until the check is successful. The detection unit includes: The judgment unit is used to determine, for each first cluster, whether the total number of data points in the first cluster is less than a threshold. If so, all data points in the first cluster are directly used as edge points of the first cluster; otherwise, the average value of all data points in the first cluster is taken on each performance dimension to obtain the center point of the first cluster. From all data points in the first cluster, the data point farthest from the center point is selected as the first edge point and added to the edge point set. The following operation is repeated until the number of data points in the edge point set reaches a preset number: For each data point in the first cluster except for the edge points, calculate the distance from each data point to the center point to obtain the first distance, and calculate the nearest distance from each data point to all edge points in the current edge point set to obtain the second distance; sum the first distance and the second distance according to preset weights to obtain the comprehensive score of the data point; select the data point with the highest comprehensive score as the edge point and add it to the edge point set; A detection unit is used to perform pass / fail checks based on the edge points of each first cluster; The detection unit includes: The calculation unit is used to calculate the mean edge radius and the nearest edge distance for each first cluster; wherein, the mean edge radius is the average distance from each edge point of the first cluster to its center point; and the nearest edge distance is the minimum value among the nearest distances from each edge point of the first cluster to all other edge points except for the first cluster. The judgment unit is used to perform qualification detection: sort the first clusters in descending order of their mean edge radius, and select the top three first clusters to form the first candidate set; sort the first clusters in ascending order of their nearest edge distance, and select the top three first clusters to form the second candidate set; if the first candidate set contains the first cluster with the most data points, and the second candidate set contains the first cluster with the fewest data points, then the current cluster is judged to be qualified; otherwise, it is deemed unqualified. Among them, clustering is completed based on normal clusters to obtain clustering results, including: Data points in the first dataset that do not belong to normal clusters are grouped together to obtain the second dataset. A second density neighborhood radius and a second density point count threshold are set, where the second density neighborhood radius is smaller than the first density neighborhood radius, and the second density point count threshold is smaller than the first density point count threshold. Each data point in the second dataset is then iterated through sequentially. The system counts the number of data points within the radius of the second density neighborhood centered on the data point. If the number of data points is greater than or equal to the threshold for the number of second density points, the data point is determined to be a core point. A second cluster is then created starting from this core point, and all data points within the radius of the second density neighborhood of the data point are included in this second cluster. Then, the system iterates through each data point in the second cluster. If the data point is also a core point, all data points within its second density neighborhood are also included in the second cluster. If the data point is not a core point, it is marked as a point of high interest.

4. The semiconductor device testing system according to claim 3, characterized in that, Build modules, including: The expansion unit is used to take the performance baseline data of each device under test as a base sample and generate expansion samples according to a preset number of expansion samples: For each base sample, a random fluctuation is generated independently for each performance dimension. The random fluctuation for each performance dimension follows a normal distribution with a mean of zero and a standard deviation equal to the absolute value of the baseline value for that performance dimension multiplied by 2%. The baseline value for each performance dimension is added to the random fluctuation generated for the corresponding performance dimension to obtain the new value of the baseline sample for each performance dimension. The new value for each performance dimension is used as an expansion sample. The combination unit is used to record each basic sample and its corresponding extended sample as a reference sample; during the aging test, the performance test data at all time points are averaged in each dimension to obtain the measured mean of the device under test in each performance dimension, and the measured mean of each dimension is combined to form a measured sample.

Images