A semiconductor device testing system under extreme conditions

Abstract

This application relates to the field of semiconductor device electrical performance testing technology, and discloses a semiconductor device testing system under extreme environments. The invention includes a test observation construction module, a contact window parsing module, a spatial order correlation module, and an anomaly interpretation output module. The test observation construction module acquires test observation data from wafer-level electrical performance testing under extreme environments, establishes a joint index of test points, and forms original observation records for the test points. The contact window parsing module generates contact state description quantities and effective stress window identifiers while retaining the original dimensions. The spatial order correlation module constructs wafer MAP spatial adjacency relationships and test order windows, generating spatial order drift correlation quantities. The anomaly interpretation output module generates device failure curve consistency results based on changes in electrical performance degradation direction, contact state description quantities, and spatial order drift correlation quantities, and labels the source of the anomalies.

Description

Technical Field

[0001] This application relates to the field of semiconductor device electrical performance testing technology, and in particular to a semiconductor device testing system under extreme environments. Background Technology

[0002] In the wafer-level testing and reliability screening of semiconductor devices, it is typically necessary to perform electrical performance parameter tests on the chip using probe stations, probe cards, wafer mounts, and testing instruments. These tests include parameters such as leakage current, on-resistance, threshold voltage, breakdown voltage, and current-voltage curves. For power semiconductor devices, whose subsequent applications often involve industrial control, new energy vehicles, wind and solar energy storage, and high-power conversion equipment, the electrical performance stability of the device under extreme environments such as high temperature, low temperature, temperature cycling, damp heat, and bias stress is a crucial criterion for determining whether the device can enter reliability screening and batch release. In actual testing, after the wafer is placed on the wafer mount, it needs to be... After processes such as leveling the substrate stage, adjusting the probe position, confirming the dot pattern, confirming the MAP diagram, and confirming the test data, a stable contact is formed between the probe tip and the wafer pad. Then, the test instrument collects the electrical parameter results of the corresponding chip. However, during extreme environmental loading, the thermal response speeds of the substrate stage, probe card, probe tip, wafer substrate, and test fixture are not completely consistent. Temperature changes, damp heat loading, and long-term bias stress will amplify tip wear, contact pressure changes, slight platform tilt, and dot pattern repeatability deviations, causing the contact state that meets the requirements during the room temperature debugging stage to gradually drift during continuous testing.

[0003] Existing wafer-level electrical performance testing technologies typically improve test data stability through methods such as probe calibration, contact resistance monitoring, repeated testing, GRR testing, and manual verification of MAP maps. These methods can confirm the device status before testing or identify some abnormal data after testing. However, they often rely on single calibration results, fixed thresholds, or offline verification conclusions as the basis for judgment. It is difficult to synchronously correlate probe contact status, test sequence, wafer spatial position, and abnormal changes in electrical parameters during continuous testing in extreme environments. Since sudden increases in leakage current, increased on-resistance, threshold voltage drift, or decreased breakdown voltage may originate from device degradation, probe contact instability, changes in wafer stage level, or dot offset, existing testing procedures are prone to misjudging local contact drift as collective failure of adjacent chips, or masking actual device degradation within test link anomalies. This leads to insufficient reliability of electrical performance test results under extreme environments, affecting wafer yield statistics, failure area location, reliability screening, and batch release judgment. There is an urgent need for a test interpretation mechanism that can simultaneously assess the reliability of contact status, spatial dot consistency, and abnormal distribution relationships of MAP maps during electrical performance testing. Summary of the Invention

[0004] This application proposes a testing system for semiconductor devices under extreme environments, which solves the problems mentioned in the background art.

[0005] To achieve the above objectives, this application adopts the following technical solution: a semiconductor device testing system under extreme environments, comprising:

[0006] The test observation construction module acquires extreme environment test observation data generated during wafer-level electrical performance testing in extreme environments. It establishes a joint index of test points using wafer MAP chip coordinates and test execution time scales, and collects the contents of the electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain of the same test point to form the original observation record of the test point.

[0007] The contact window parsing module retains the original dimensions in the original observation records of the test points and converts the contact changes, positioning offsets, repetition differences and electrical measurement fluctuations that enter the fusion judgment into dimensionless fusion features, generating contact state description quantities and effective stress window identifiers.

[0008] The spatial order association module constructs the wafer MAP spatial adjacency relationship and test order window based on the test point joint index, and generates spatial order drift association quantity according to the distribution change of contact state descriptor in the two.

[0009] The anomaly interpretation output module generates device failure curve consistency results based on the changes in electrical performance degradation direction, contact state description, and spatial order drift correlation within the effective stress window. It also marks the anomaly source label for test points that meet the anomaly judgment conditions and outputs the electrical performance test interpretation results.

[0010] Furthermore, the test observation construction module divides the extreme environment test observation data into electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain.

[0011] The electrical performance observation domain is formed by at least one of the following: leakage current curve, on-resistance curve, threshold voltage curve, and breakdown voltage curve, which characterize the change in electrical response at the test point; the contact observation domain is formed by contact resistance data, repeated measurement records, and contact resistance change records formed by the contact resistance data according to the test execution time scale, which characterize the contact state between the probe and the wafer pad; the spatial positioning observation domain is formed by the wafer MAP core coordinates, probe marking position, and pad center position, which characterize the spatial assignment and marking offset of the test point; and the environmental stress observation domain is formed by the stage temperature, extreme environment set temperature, and stage temperature change rate, which characterize the extreme environmental loading state.

[0012] Furthermore, the test observation construction module establishes a test point joint index by generating an object index based on the wafer number, wafer MAP row and column coordinates, and chip number; generating a sampling index based on the test start time, sampling time, and test execution sequence number; and binding the object index and the sampling index into a test point joint index, so that the same test point joint index points to the observation content of the same chip in the same extreme environment wafer-level electrical performance test process.

[0013] Furthermore, the test observation construction module generates original observation records for test points, including: performing object alignment and time-scale alignment on the observation content in the electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain according to the test point joint index, and retaining the original dimensions, sampling source, and sampling time of each observation content after alignment;

[0014] Among them, the electrical performance observation domain retains the dimensions of current, resistance or voltage according to the electrical performance type; the contact resistance data in the contact observation domain retains the dimension of resistance; the electrical performance test values ​​in the repeated measurement records retain the dimension of the electrical performance type they are measuring; the spatial positioning observation domain retains the dimension of length; and the environmental stress observation domain retains the dimensions of temperature and time.

[0015] Furthermore, the contact window parsing module converts contact changes, positioning offsets, repetition differences, and electrical measurement fluctuations that enter the fusion judgment into dimensionless fusion features, including: generating dimensionless fusion features corresponding to contact changes based on the difference between the current contact resistance and the stable batch contact resistance benchmark value, and the allowable fluctuation range of the contact resistance; generating dimensionless fusion features corresponding to positioning offsets based on the offset distance of the probe dot position relative to the center position of the pad, and the allowable offset distance of the pad; generating dimensionless fusion features corresponding to repetition differences based on the repetition measurement difference of the same electrical performance type at the same test point, and the allowable repeatability error of that electrical performance type; and generating dimensionless fusion features corresponding to electrical measurement fluctuations based on the fluctuation amplitude of the electrical performance observation domain content within the short-time sampling interval, and the corresponding electrical performance benchmark value.

[0016] Among them, the short-time sampling interval is determined by the sampling cycle of the test instrument and the electrical performance test specifications. The stable batch contact resistance reference value, the allowable fluctuation range of contact resistance and the electrical performance reference value are determined based on the stable batch statistical results, the repeatability and reproducibility test results of the measurement system and the measurement error of the test instrument. The allowable offset distance of the pad is determined based on the pad size of the wafer layout and the probe station dotting specifications.

[0017] Furthermore, the contact window parsing module generates contact state description quantities by: configuring judgment weights for the dimensionless fusion features corresponding to contact changes, positioning offsets, repetition differences, and electrical measurement fluctuations, and weighting and aggregating each dimensionless fusion feature to obtain the contact state description quantities.

[0018] The sum of the judgment weights is 1. The judgment weights are determined based on the statistical results of stable batches, the repeatability and reproducibility test results of the measurement system, and the measurement error of the test instrument. The contact state descriptor is a dimensionless quantity and is used to characterize the degree of probe contact instability at the test point.

[0019] Furthermore, the contact window analysis module generates effective stress window identifiers by: determining the environmental stress state of the test point based on the deviation between the stage temperature and the extreme environment set temperature, and the stage temperature change rate; when the deviation does not exceed the allowable temperature deviation and the stage temperature change rate does not exceed the temperature stability change rate threshold, the test point is marked as a test point within the effective stress window; when the deviation exceeds the allowable temperature deviation, or the stage temperature change rate exceeds the temperature stability change rate threshold, the test point is marked as an environmental transition state test point; the allowable temperature deviation is determined based on the extreme environment test specifications and the temperature sensor calibration error, and the temperature stability change rate threshold is determined based on the stage thermal response time and thermocouple calibration results.

[0020] Furthermore, the spatial order association module constructs the wafer MAP spatial adjacency relationship and test order window by: based on the wafer MAP core coordinates in the test point joint index, identifying cores that are adjacent to the target test point in the row direction, column direction, or diagonal direction and have original observation records of the test point as spatially adjacent test points, thus forming the wafer MAP spatial adjacency relationship; and based on the test execution time stamp in the test point joint index, grouping test points with consecutive test execution sequence numbers and consistent effective stress window identifiers into the same test order window.

[0021] Furthermore, the spatial order association module generates spatial order drift association quantities by: generating spatial adjacency contact aggregation results based on the average value of the contact state descriptors corresponding to the spatially adjacent test points of the target test point; dividing the test order window containing the target test point into a first half window and a second half window according to the test execution sequence number; when the number of test points in the test order window is odd, merging the test point corresponding to the median test execution sequence number into the first half window, and generating order drift results based on the absolute value of the difference between the average value of the contact state descriptors in the first half window and the average value of the contact state descriptors in the second half window; and aggregating the spatial adjacency contact aggregation results and the order drift results according to the association weights to obtain the spatial order drift association quantity; wherein, the sum of the association weights is 1, and the association weights are determined based on the number of spatially adjacent test points in the wafer MAP and the number of valid test points in the test order window.

[0022] Furthermore, the anomaly interpretation output module generates device failure curve consistency results and marks anomaly source tags, including: within the effective stress window, generating device failure curve consistency results based on the degradation direction change of the electrical performance observation domain content relative to the corresponding electrical performance reference value. Among them, the increase of leakage current, the increase of on-resistance, and the failure direction drift of threshold voltage as defined by the device test specification are configured as positive degradation changes, and the decrease of breakdown voltage is configured as a negative degradation change. The degradation direction changes of each electrical performance type are converted into the same-direction degradation characterization before consistency judgment is performed.

[0023] When the contact state description reaches the contact anomaly threshold, the spatial sequence drift correlation reaches the drift correlation threshold, and the device failure curve consistency result does not reach the device degradation threshold, the test point is marked as a local contact drift anomaly; when the contact state description does not reach the contact anomaly threshold, the spatial sequence drift correlation does not reach the drift correlation threshold, and the device failure curve consistency result reaches the device degradation threshold, the test point is marked as a true device degradation anomaly; when the effective stress window characterizes the environmental transition state and at least half of the test points within the same test sequence window have the same electrical performance degradation direction change, the test point is marked as an environmental transition anomaly; when the contact state description reaches the contact anomaly threshold and the dimensionless fusion feature corresponding to the repeated difference reaches the retest threshold, the test point is marked as a test link anomaly, and the electrical performance test interpretation result containing the anomaly source label and the original electrical performance test value is output;

[0024] Among them, the contact anomaly threshold, drift correlation threshold, device degradation threshold, and retest threshold are determined based on the stable batch statistical results, the repeatability and reproducibility test results of the measurement system, the device test specifications, and the measurement error of the test instrument.

[0025] The beneficial effects of this invention are as follows:

[0026] This invention establishes a joint index for test points using wafer MAP core coordinates and test execution timescales. It aggregates the electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain from extreme environment test observation data into the original observation records of test points. This solves the problems of scattered records of data from different sources and difficulty in synchronizing and tracing electrical performance anomalies of the same core with contact status, marking position, and environmental stress status during existing testing processes. This process unifies the object relationships, timescale relationships, and sampling sources of test points, and also reduces the misjudgment of anomaly sources caused by data mismatch.

[0027] This invention converts contact changes, positioning offsets, repetitive differences, and electrical measurement fluctuations into dimensionless fusion features while retaining the original dimensions, generating contact state descriptive quantities. It also combines the substrate temperature, extreme environmental set temperature, and substrate temperature change rate to generate an effective stress window identifier. This solves the problem of probe contact instability, repetitive measurement differences, and environmental transient disturbances being mixed in electrical performance test results. This processing can distinguish contact state evidence from environmental stress window evidence and avoid directly identifying electrical performance fluctuations or probe contact drift in the unstable temperature stage as actual device degradation anomalies.

[0028] This invention generates spatial order drift correlation quantities by constructing wafer MAP spatial adjacency relationships and test sequence windows. It then combines the changes in electrical performance degradation direction, contact state description quantities, and spatial order drift correlation quantities within the effective stress window to generate consistent results for device failure curves. By labeling test points with abnormal source tags, this invention solves the problem of difficulty in distinguishing the source of leakage current, on-resistance, threshold voltage, or breakdown voltage anomalies under extreme environments. This processing expands single-point electrical performance anomalies into a joint interpretation result of contact state, spatial distribution, test sequence, and device degradation curves. It also distinguishes the same electrical performance anomaly into local contact drift anomalies, test link anomalies, environmental transition anomalies, actual device degradation anomalies, or source tags to be verified, thereby improving the reliability and verifiability of electrical performance test interpretation results. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort:

[0030] Figure 1 This is a system framework diagram of the present invention;

[0031] Figure 2 This is a flowchart of the contact window parsing module of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example

[0034] like Figure 1 and Figure 2 As shown, the present invention discloses a test system for semiconductor devices under extreme environments, including: a test observation construction module, a contact window parsing module, a spatial order association module, and an anomaly interpretation output module.

[0035] In one implementation, the test observation construction module is used to acquire extreme environment test observation data generated during wafer-level electrical performance testing of semiconductor devices in extreme environments. It establishes a joint index of test points using wafer MAP chip coordinates and test execution time scales, and aggregates the contents of the electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain of the same test point to form the original observation record of the test point. The purpose of this module is to ensure that each observation content belongs to the same chip, the same test process, and the same sampling period before the subsequent contact window parsing module performs dimensionless fusion feature conversion, so as to avoid mismatches in the contents of the electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain at the object level or time scale level.

[0036] Extreme environments refer to test conditions that cause environmental stress to affect the wafer-level electrical performance test results of semiconductor devices. These include at least one of the following: high temperature environment, low temperature environment, temperature cycling environment, humid heat environment, bias stress environment, and a combination of humid heat and bias stress environment. The specific values ​​of high temperature environment, low temperature environment, humidity conditions, temperature cycling conditions, and bias stress conditions are determined by the device test specifications, reliability screening specifications, device rated operating conditions, and test equipment calibration results. The extreme environment set temperature is used to characterize the target environmental conditions applied by the test equipment, the stage temperature is used to characterize the actual environmental response of the wafer bearing position when the test point is sampled, and the stage temperature change rate is used to characterize whether the environmental stress is in the loading transition process. The above-mentioned environmental stress observation domain content is used to generate effective stress window identifiers in the subsequent process and is not directly weighted with the electrical performance observation domain content, contact observation domain content, or spatial positioning observation domain content.

[0037] The extreme environment test observation data is divided into four observation domains according to the observation function: the electrical performance observation domain is formed by at least one of the leakage current curve, on-resistance curve, threshold voltage curve and breakdown voltage curve characterizing the change of electrical response at the test point; the contact observation domain is formed by contact resistance data characterizing the contact state between the probe and the wafer pad, repeated measurement records and contact resistance change records formed by the contact resistance data according to the test execution time scale; the spatial positioning observation domain is formed by the wafer MAP core coordinates, probe marking position and pad center position characterizing the spatial assignment and marking offset state of the test point; and the environmental stress observation domain is formed by the stage temperature, extreme environment set temperature and stage temperature change rate characterizing the extreme environment loading state.

[0038] The electrical performance observation domain, contact observation domain, and environmental stress observation domain are provided by the test instrument, probe station control system, environmental chamber control system, and wafer stage temperature acquisition unit, respectively; the spatial positioning observation domain is provided by the wafer MAP file, probe station coordinate record, and spot detection record; the above data acquisition belongs to the basic acquisition process in wafer-level testing. This implementation does not change the electrical measurement principle of the data acquisition equipment. The focus is on the test point-level object aggregation and time-stamp aggregation of the observation content from different sources, so that the subsequent contact state description, effective stress window identifier, and spatial sequence drift correlation have the same data basis.

[0039] When the test observation construction module establishes the test point joint index, it first generates an object index based on the wafer number, the row and column coordinates of the wafer MAP, and the chip number. The wafer number is used to distinguish different wafers, the row and column coordinates of the wafer MAP are used to determine the spatial position of the chip in the wafer MAP, and the chip number is used to form a unique chip identifier within the same wafer. In one implementation, the object index is generated in the order of the fields wafer number, row coordinates, column coordinates, and chip number. When the wafer MAP file does not have a separate chip number configured, the combination of the row and column coordinates is used as the basis for generating the chip number.

[0040] Subsequently, the test observation construction module generates a sampling index based on the test start time, sampling time, and test execution sequence number. The test start time is used to limit the wafer-level electrical performance test process in the same extreme environment. The sampling time is used to characterize the actual acquisition time of the electrical performance observation domain, contact observation domain, and environmental stress observation domain. The test execution sequence number is used to characterize the sequential position of the test point in the continuous test process. The sampling index is generated according to the field order of test start time, test execution sequence number, and sampling time. When there are repeated measurement records for the same test point, the number of repeated measurements is used as an auxiliary field of the sampling index, so that repeated measurement records are included in the same test point composite index and their test sequence relationship is preserved.

[0041] After the object index and sampling index are formed, the test observation construction module binds the object index and the sampling index to form a test point joint index. The test point joint index is used to point to the observation content of the same chip in the same extreme environment wafer-level electrical performance test. Through the test point joint index, the electrical performance observation domain content, contact observation domain content, spatial positioning observation domain content and environmental stress observation domain content of the same test point are collected into the original observation record of the same test point. This process enables the contact state, spatial position and environmental stress state of the test point to be traced simultaneously when judging electrical performance anomalies, avoiding the judgment of device degradation based solely on the electrical performance curve.

[0042] During object alignment, the test observation construction module uses the wafer MAP core coordinates as the main index to map the probe dot positions and pad center positions in the spatial positioning observation domain to the same core coordinates. Then, the electrical performance curves in the electrical performance observation domain and the contact resistance data in the contact observation domain are mapped to the same object index according to the core number or test execution sequence number recorded by the test equipment. If there are multiple test records for the same wafer MAP core coordinates, different records are distinguished according to the test execution sequence number and the number of repeated measurements. If the same test execution sequence number corresponds to multiple sampling points, a sampling sequence is established based on the sampling time, and curve records containing the sampling time and sampling value are retained.

[0043] During the time-scale alignment process, the test observation construction module uses the sampling time of the electrical performance observation domain as the primary time-scale, and classifies the observation content falling within the same time-scale allowable window in the contact observation domain and the environmental stress observation domain into the original observation record of the same test point. The time-scale allowable window is determined by the maximum period among the sampling period of the test instrument, the recording period of the probe station, and the sampling period of the environmental stress observation domain. The basis for this determination method is that when there are differences in the sampling periods of different acquisition sources, using the maximum sampling period as the collection boundary of the same sampling period can avoid the environmental stress data or contact observation data being incorrectly excluded because the sampling frequency is lower than that of the electrical performance data, and at the same time, it will not introduce the state data of other test points across the next sampling period.

[0044] For cases where multiple environmental stress observation domain records exist within the time-scale allowable window, the test observation construction module selects the stage temperature record with the smallest time difference from the electrical performance sampling time as the stage temperature for that test point. For cases where no environmental stress observation domain record exists within the time-scale allowable window, the test point is marked as a missing environmental stress record, and this missing mark is retained. Subsequently, the contact window parsing module will not directly identify this test point as a test point within the effective stress window when generating the effective stress window identifier. For test points with missing contact observation domain records, the test observation construction module retains the contents of the electrical performance observation domain, spatial positioning observation domain, and environmental stress observation domain, and generates a missing contact observation mark. This missing mark will be handled according to the missing mark rules when calculating the contact state descriptor.

[0045] When generating the original observation records for test points, the test observation construction module retains the original dimensions of each observation content; the electrical performance observation domain retains the dimensions of current, resistance, or voltage according to the electrical performance type; the contact resistance data in the contact observation domain retains the resistance dimension, and the electrical performance test values ​​in the repeated measurement records retain the dimension of the electrical performance type they are measuring; the probe dot position, pad center position, and dot offset distance in the spatial positioning observation domain retain the length dimension; the substrate temperature and extreme environment setting temperature in the environmental stress observation domain retain the temperature dimension, and the test start time, sampling time, and test execution timescale retain the time dimension; the above original dimension retention strategy is used to ensure that the subsequent dimensionless fusion features are only formed when they need to enter the fusion judgment, without changing the physical interpretation meaning of the original electrical performance test values ​​and the value of the test report output.

[0046] The original observation record of the test point includes at least the test point joint index, object index, sampling index, electrical performance observation domain content, contact observation domain content, spatial positioning observation domain content, environmental stress observation domain content, original dimension identifier, sampling source identifier, sampling time identifier, and missing status identifier. The original dimension identifier is used to record the unit attribute of each observation content. The sampling source identifier is used to record the source of each observation content. The missing status identifier is used to record the situation where the contact observation domain, spatial positioning observation domain, or environmental stress observation domain failed to be collected within the time scale allowable window. The missing status identifier is only used as the basis for subsequent modules to select the processing path and is not directly used as the judgment result of the anomaly source label.

[0047] The test observation construction module transforms multi-source observations from wafer-level electrical performance testing in extreme environments into raw test point observation records centered on a test point joint index. During the aggregation process, it simultaneously preserves object relationships, time-scale relationships, original dimensions, and sampling sources. As a result, the subsequent contact window parsing module can generate dimensionless fusion features based on clear dimensions and sources, the spatial order association module can construct wafer MAP spatial adjacency relationships and test order windows based on a unified test point joint index, and the anomaly interpretation output module can distinguish between local contact drift anomalies, test link anomalies, environmental transition anomalies, and actual device degradation anomalies on the same test point data chain.

[0048] In one implementation, the contact window parsing module, based on the original observation records of the test points, retains the original dimensions of contact resistance data, probe dot positions, pad center positions, repeated measurement records, and electrical performance observation domain content. It then converts contact changes, positioning offsets, repeatability differences, and electrical measurement fluctuations that enter the fusion judgment into dimensionless fusion features, generating contact state description quantities and effective stress window identifiers. This module inherits the original observation records of the test points formed by the test observation construction module. Its processing purpose is to convert contact resistance, dot offsets, repeatability differences, and electrical measurement fluctuations of different dimensions into dimensionless fusion features under a unified judgment scale, while retaining the physical meaning of the original electrical performance test values. This provides a contact state basis for subsequent spatial order drift correlation quantities and anomaly source labels.

[0049] Contact resistance data, probe dot positions, pad center positions, repeat measurement records, and electrical performance observation domain content are initially retained as raw test data upon entering the contact window parsing module. Contact resistance data retains the resistance dimension, probe dot positions and pad center positions retain the length dimension, electrical performance test values ​​in repeat measurement records retain the dimension of the measured electrical performance type, and electrical performance observation domain content retains the current, resistance, or voltage dimensions for leakage current, on-resistance, threshold voltage, or breakdown voltage, respectively. The above raw test data is used for test report output and anomaly verification. Dimensionless conversion is only performed when contact changes, positioning offsets, repeatability differences, and electrical measurement fluctuations are all included in the calculation of contact state descriptive quantities.

[0050] A stable batch refers to a historically qualified test batch of semiconductor devices of the same model that has completed repeatability and reproducibility testing of the measurement system under the same probe station, the same probe card, and the same test instrument configuration, and for which there are no missing records of environmental stress or missing contact observations in the original observation records of the test points; stable batches are used to provide the basis for forming contact resistance benchmarks, electrical performance benchmarks, statistical fluctuation ranges, and judgment weights.

[0051] The contact window parsing module generates dimensionless fusion features corresponding to contact changes based on the difference between the current contact resistance and the stable batch contact resistance benchmark value, as well as the allowable fluctuation range of the contact resistance. The stable batch contact resistance benchmark value is determined by the statistical results of the stable batch contact resistance. When the absolute value of the skewness coefficient of the stable batch contact resistance distribution is not greater than 1, the mean value of the stable batch contact resistance is used as the stable batch contact resistance benchmark value; when the absolute value of the skewness coefficient of the stable batch contact resistance distribution is greater than 1, the median value of the stable batch contact resistance is used as the stable batch contact resistance benchmark value. The allowable fluctuation range of the contact resistance is determined by the stable batch contact resistance standard. The contact resistance fluctuation is determined jointly by the difference, the resistance measurement error of the test instrument, and the stable batch contact resistance reference value. When the device test specification does not give an upper limit for contact resistance fluctuation, the allowable fluctuation range of contact resistance is determined by the maximum value among three factors: three times the standard deviation of stable batch contact resistance, three times the resistance measurement error of the test instrument, and 5% of the stable batch contact resistance reference value. When this maximum value exceeds 30% of the stable batch contact resistance reference value, the stable batch contact resistance reference value is marked as unavailable and enters the test link verification path. This process is used to exclude random fluctuations within the measurement resolution and identify contact changes that may affect the reliability of the electrical performance observation domain.

[0052] The contact window parsing module generates a dimensionless fusion feature corresponding to the positioning offset based on the offset distance of the probe dot position relative to the center position of the pad and the allowable offset distance of the pad. The allowable offset distance of the pad is determined by the pad size of the wafer layout and the probe station dot specification. When the probe station dot specification does not give a specific offset upper limit, the allowable offset distance of the pad is taken as 10% to 35% of the short side size of the pad, and the distance from the center of the probe mark to the boundary of the pad is not less than 1 / 2 of the equivalent diameter of the probe mark. The above value range is used to avoid identifying the normal repeatability positioning error of the probe station as a contact abnormality, and at the same time to prevent the probe mark from approaching the boundary of the pad, resulting in insufficient contact area.

[0053] The contact window parsing module generates dimensionless fusion features corresponding to the repeatability differences based on the repeated measurement differences of the same electrical performance type at the same test point, and the allowable repeatability error of that electrical performance type. The allowable repeatability error is determined by the device test specifications, the repeatability and reproducibility test results of the measurement system, and the measurement error of the test instrument. For voltage and resistance electrical performance tests, the allowable repeatability error is determined by the larger of three times the standard deviation of repeated measurements in a stable batch and the upper limit of the measurement error of the test instrument. For leakage current electrical performance tests, the allowable repeatability error is also determined in conjunction with the current resolution of the test instrument and the magnitude of the leakage current to avoid distortion conversion when the leakage current is close to the resolution of the instrument.

[0054] The contact window parsing module generates dimensionless fusion features corresponding to electrical measurement fluctuations based on the fluctuation amplitude of the electrical performance observation domain within a short-time sampling interval and the corresponding electrical performance reference value. The short-time sampling interval is determined by the sampling period of the test instrument and the electrical performance test specification. In one embodiment, the short-time sampling interval includes 3 to 10 consecutive sampling points within the same test point and does not cross the next test execution sequence number. When the electrical performance test specification limits the sampling holding time, the short-time sampling interval is located within that sampling holding time. The electrical performance reference value is determined by the stable batch statistical results, device test specifications, and test instrument measurement error. When the electrical performance reference value is 0 or lower than the resolution of the test instrument, the allowable variation range of this electrical performance type is used as the denominator for the dimensionless conversion of electrical measurement fluctuations. The selection of 3 to 10 consecutive sampling points is based on the fact that fewer than 3 sampling points cannot form a stable fluctuation description, while more than 10 sampling points are prone to crossing the probe contact state change or environmental stress change stage.

[0055] The dimensionless fusion characteristic corresponding to contact change is obtained by dividing the absolute value of the difference between the current contact resistance and the stable batch contact resistance reference value by the allowable fluctuation range of the contact resistance; the current contact resistance, the stable batch contact resistance reference value, and the allowable fluctuation range of the contact resistance are all in the dimension of resistance, therefore the conversion result is a dimensionless quantity; the dimensionless fusion characteristic corresponding to positioning offset is obtained by dividing the offset distance of the pin mark center relative to the pad center position by the allowable offset distance of the pad; the offset distance of the pin mark center relative to the pad center position and the allowable offset distance of the pad are both in the dimension of length, therefore the conversion result is a dimensionless quantity; repeatability difference corresponds to The dimensionless fusion characteristic is obtained by dividing the absolute value of the difference between two measurements of the same electrical performance type at the same test point by the allowable repeatability error of that electrical performance type. Since both the absolute value of the difference between the two measurements and the allowable repeatability error have the same dimension of electrical performance type, the conversion result is a dimensionless quantity. The dimensionless fusion characteristic corresponding to electrical measurement fluctuations is obtained by dividing the standard deviation of the electrical performance observation domain within the short-time sampling interval by the corresponding electrical performance benchmark value or the allowable variation range of the corresponding electrical performance type. Since the standard deviation and the corresponding electrical performance benchmark value or allowable variation range have the same dimension, the conversion result is a dimensionless quantity.

[0056] When the same test point has repeated measurement records for multiple electrical performance types, the contact window parsing module generates dimensionless fusion features corresponding to the repeatability differences of each electrical performance type, and uses the dimensionless fusion feature with the largest value as the dimensionless fusion feature corresponding to the repeatability differences of the test point. When the same test point has short-term fluctuation results for multiple electrical performance types, the contact window parsing module generates dimensionless fusion features corresponding to the electrical measurement fluctuations of each electrical performance type, and uses the dimensionless fusion feature with the largest value as the dimensionless fusion feature corresponding to the electrical measurement fluctuations of the test point. The basis for using the maximum value as the synthesis result is that probe contact instability will cause a decrease in repeatability or an increase in short-term fluctuations in at least one electrical performance observation domain. Taking the maximum value can avoid weakening the evidence of contact anomalies after averaging multiple types of electrical performance results.

[0057] After obtaining the dimensionless fusion features corresponding to contact changes, positioning offsets, repeatability differences, and electrical measurement fluctuations, the contact window parsing module generates contact state descriptors; let the contact state descriptor for the i-th test point be... The calculation method is as follows:

[0058] ;

[0059] in, Indicates the test point number; Indicates the first Contact state description quantity for each test point; They represent the first Dimensionless fusion characteristics corresponding to contact changes, positioning offsets, repeatability differences, and electrical measurement fluctuations at each test point; These represent the judgment weights for contact change, positioning offset, repeatability difference, and electrical measurement fluctuation, respectively. Each judgment weight is a value between 0 and 1, and satisfies the following conditions: ; It is a dimensionless quantity used to characterize the first... The degree of probe contact instability at each test point.

[0060] Each judgment weight is determined based on the statistical results of stable batches, the repeatability and reproducibility test results of the measurement system, and the measurement error of the test instruments. In one embodiment, stable batches and manually confirmed contact anomaly records are used as weighted samples. The ratio of the mean difference to the within-group standard deviation of the dimensionless fusion features corresponding to contact changes, positioning offsets, repeatability differences, and electrical measurement fluctuations between the two types of samples is calculated, and the obtained ratios are normalized and used as initial weights. When no manually confirmed contact anomaly records are obtained, contact changes and positioning offsets are used as direct contact evidence, and the sum of their judgment weights is 0.55 to 0.80. Repeatability differences and electrical measurement fluctuations are used as evidence of the reliability of electrical performance acquisition, and the sum of their judgment weights is 0.20 to 0.45. The sum of all judgment weights is 1. This configuration makes the contact state description primarily based on direct contact evidence, while retaining the auxiliary characterization role of the reliability evidence of electrical performance acquisition for contact instability.

[0061] For handling missing dimensionless fusion features, if only one type of dimensionless fusion feature is missing among the contact change, positioning offset, repetition difference, and electrical measurement fluctuation, the missing item is deleted, and the remaining judgment weights are normalized according to the original weight ratio to generate the contact state description. If two or more types of missing items are missing, the contact state description of the test point is marked as unavailable, and a contact observation missing identifier is generated. Test points whose contact state descriptions are marked as unavailable do not enter the separate judgment path for actual device degradation anomalies, but need to enter the test link verification path. This process is used to prevent missing data from being used as evidence of normal contact, while retaining the verifiable test link processing path.

[0062] The contact window parsing module also generates an effective stress window identifier based on the content of the environmental stress observation domain. The effective stress window identifier is used to characterize whether the environmental stress state at the test point sampling time has a basis for judging device degradation. The stage temperature is used to characterize the actual temperature response of the wafer bearing position, the extreme environment setting temperature is used to characterize the target environmental conditions applied by the test equipment, and the stage temperature change rate is used to characterize whether the environmental stress is in the loading transition process. The stage temperature, extreme environment setting temperature, and stage temperature change rate retain the original temperature and time dimensions and are not included in the weighted calculation of the contact state description quantities.

[0063] Let the first The effective stress window for each test point is marked as follows: The method for determining it is as follows:

[0064] ;

[0065] in, Indicates the test point number; Indicates the first Effective stress window identifier for each test point; Indicates the first The temperature of the tray stage at each test point is expressed in °C. This indicates the set temperature for extreme environments, expressed in °C. This indicates the allowable temperature deviation, in °C. This represents the rate of temperature change of the stage at the i-th test point, in °C / min. This represents the threshold for the steady-state rate of change of temperature, expressed in °C / min; when When, it indicates the first Each test point is within the effective stress window; when When, it indicates the first The test points are in an environmental transition state.

[0066] The allowable temperature deviation is determined based on the extreme environment test specifications and the temperature sensor calibration error. When the test specifications do not provide an allowable deviation, the allowable temperature deviation is 2 to 3 times the temperature sensor calibration error, and is within the range of ±1℃ to ±5℃. The temperature stability rate of change threshold is determined based on the thermal response time of the substrate stage and the thermocouple calibration results. When the test specifications do not provide a rate of change threshold, the temperature stability rate of change threshold is 0.1℃ / min to 2℃ / min. The basis for the above range is that the substrate stage temperature deviating from the target environmental conditions will change the content of the electrical performance observation domain, such as leakage current, threshold voltage, and on-resistance. When the substrate stage temperature is still changing, the content of the electrical performance observation domain includes environmental transient disturbances and cannot be used as a direct basis for judging the actual degradation anomaly of the device.

[0067] When there are missing status indicators in the environmental stress observation domain, the contact window parsing module will not mark the test point as a test point within the effective stress window; when the deviation between the temperature of the support stage and the extreme environment setting temperature does not exceed the allowable temperature deviation, and the temperature change rate of the support stage does not exceed the temperature stability change rate threshold, the test point will be marked as a test point within the effective stress window; when the deviation between the temperature of the support stage and the extreme environment setting temperature exceeds the allowable temperature deviation, or the temperature change rate of the support stage exceeds the temperature stability change rate threshold, the test point will be marked as an environmental transition state test point.

[0068] The processing results of the contact window parsing module include contact state descriptors and effective stress window identifiers. The contact state descriptors characterize the probe contact instability at the test point, while the effective stress window identifiers characterize the environmental stress state at the time of test point sampling. The contact state descriptors are used by the subsequent spatial sequence association module to identify the distribution changes of the contact state in the wafer MAP spatial adjacency relationship and test sequence window. The effective stress window identifiers are used by the subsequent anomaly interpretation output module to limit the calculation range of the device failure curve consistency results. Through the above processing, electrical performance anomalies form a contact credibility basis and an environmental stress window basis before entering the anomaly source label determination, thereby avoiding directly identifying data disturbances or fluctuations caused by probe contact instability under environmental transition states as actual device degradation anomalies.

[0069] In one implementation, the spatial order association module is used to construct the wafer MAP spatial adjacency relationship and test order window based on the test point joint index, and generate spatial order drift association quantity according to the distribution change of the contact state description quantity in the wafer MAP spatial adjacency relationship and test order window. This module receives the test point joint index, wafer MAP core coordinates and test execution time scale formed by the test observation construction module, and receives the contact state description quantity and effective stress window identifier generated by the contact window parsing module. The purpose of this module is to extend the contact state description quantity of a single test point to the wafer spatial relationship and test process relationship, so as to identify contact anomaly clusters in the local spatial range and contact state drift that occurs as the test progresses.

[0070] The wafer MAP die coordinates and test execution timestamps are fundamental test records in the wafer-level electrical performance testing process. This implementation does not change the wafer MAP coordinate generation method or the test execution sequence number recording method. The key processing point of the spatial sequence association module is to use the same test point joint index to bind the wafer MAP spatial adjacency relationship, test sequence window, and contact state description quantity to the same test point data chain, so that the subsequent anomaly interpretation output module can combine the spatial sequence drift association quantity to distinguish between contact anomalies caused by test process drift and electrical performance anomalies caused by device degradation.

[0071] When constructing the spatial adjacency relationship of the wafer MAP, the spatial order association module determines the spatially adjacent test points around the target test point that have original observation records of the test point, based on the wafer MAP core coordinates in the test point joint index. The spatially adjacent test points are composed of cores that are adjacent to the target test point in the row direction, column direction, or diagonal direction, and the core has the original observation record of the test point and available contact state descriptions. For test points in the internal region of the wafer MAP, the number of spatially adjacent test points does not exceed 8. For test points at the wafer edge or wafer notch region, the wafer MAP spatial adjacency relationship is determined according to the actual existing spatially adjacent test points that have original observation records of the test point. If the contact state description of a spatially adjacent test point is marked as unavailable, the spatially adjacent test point does not participate in the calculation of the spatially adjacent contact aggregation result.

[0072] The spatial sequence association module prioritizes using adjacent cores in the row and column directions to form a basic adjacency set, and then incorporates adjacent cores in the diagonal direction into the extended adjacency set. Adjacent cores in the row and column directions are used to reflect the local continuity of directly adjacent regions in the wafer MAP, while adjacent cores in the diagonal direction are used to supplement oblique anomalous clusters in local regions of the wafer. Through the above spatial adjacency relationship of the wafer MAP, the spatial sequence association module can determine whether there is a clustering phenomenon of increased contact state descriptor in the area around the target test point, providing spatial evidence for identifying local contact drift anomalies.

[0073] When constructing the test sequence window, the spatial sequence association module, based on the test execution time stamp in the test point joint index, groups test points with consecutive test execution sequence numbers and consistent effective stress window identifiers into the same test sequence window. The test execution sequence number is used to characterize the order in which the probe station performs wafer-level electrical performance tests on each test point. The effective stress window identifier is used to limit the test points within the same test sequence window to be in the same environmental stress state. This process allows the test sequence window to be used to observe the changes in contact state as the test progresses, avoiding the mixing of environmental transition states and states within the effective stress window into the same test sequence window.

[0074] A valid test point is a test point with original observation records, consistent effective stress window identifiers, and contact state descriptions that are not marked as unavailable. The number of valid test points within a test sequence window should be no less than four, ensuring that the first and second halves of the window each have at least two usable contact state descriptions to reduce the impact of fluctuations in a single test point on the sequence drift results. When the number of valid test points within the same continuous test segment exceeds the window capacity defined by the test equipment recording specifications or device test specifications, it is split into multiple test sequence windows according to the test execution sequence number. If no window capacity is given, the split boundary is determined by changes in the effective stress window identifier, switching of the test path segment, or switching of the probe cleaning record.

[0075] When the number of valid test points in the test sequence window is even, the spatial sequence association module divides the test sequence window into a first half and a second half according to the test execution sequence number. When the number of valid test points in the test sequence window is odd, the valid test points corresponding to the median test execution sequence number are merged into the first half of the window. This division method gives the test sequence window a definite sequential relationship, avoiding the lack of a clear calculation object for the sequence drift result due to the odd number of valid test points in the window. Both the first half and the second half of the window use only valid test points available for contact state description quantities for calculation.

[0076] When generating the spatial order drift correlation quantity, the spatial order correlation module first generates a spatial adjacency contact aggregation result based on the average value of the contact state descriptors corresponding to the spatially adjacent test points of the target test point. The spatial adjacency contact aggregation result is used to characterize the degree of contact anomaly aggregation in the area surrounding the target test point. Subsequently, the spatial order correlation module generates a sequence drift result based on the absolute value of the difference between the average value of the contact state descriptors in the first half of the test sequence window and the average value of the contact state descriptors in the second half of the test sequence window where the target test point is located. The sequence drift result is used to characterize whether the contact state changes as the test progresses within the same test sequence window. Finally, the spatial order correlation module aggregates the spatial adjacency contact aggregation result and the sequence drift result according to the correlation weight to obtain the spatial order drift correlation quantity.

[0077] Given that the number of spatially adjacent test points is not less than 1, and the test sequence window can form a first half window and a second half window, let the first half... The spatial order drift correlation of each test point is The calculation method is as follows:

[0078] ;

[0079] in, Indicates the target test point number; Indicates the first Spatial order drift correlation of each test point; Indicates the first The set of spatially adjacent test points for each test point; Indicates the first The number of spatially adjacent test points for each test point; Indicates the spatial adjacency test point number; Indicates the first A quantity describing the contact state of adjacent test points in space; Indicates the first The test sequence window containing each test point; This represents the average value of the contact state description of the test points in the first half of the test sequence window. This represents the average value of the contact state description of the test points in the second half of the test sequence window. Indicates spatial adjacency weight; Indicates the test order weight; and All are values ​​between 0 and 1, and satisfy the following conditions: Since the contact state descriptor, the average value of the first half window, and the average value of the second half window are all dimensionless, the spatial order drift correlation is also dimensionless.

[0080] The correlation weights are determined based on the number of spatially adjacent test points in the wafer MAP and the number of valid test points within the test order window. When the number of spatially adjacent test points is not less than 4 and the number of valid test points within the test order window is not less than 4, both the spatial adjacency weight and the test order weight are set to 0.5. When the number of spatially adjacent test points is between 1 and 3 and the number of valid test points within the test order window is not less than 4, the spatial adjacency weight is set to 0.25 and the test order weight is set to 0.75. When the number of spatially adjacent test points is 0 and the number of valid test points within the test order window is not less than 4, spatial adjacency contact aggregation results are not calculated, the spatial adjacency weight is set to 0, and the test order weight is set to 1. When the number of valid test points within the test order window is less than 4 and the number of spatially adjacent test points is not less than 1, order drift results are not calculated, the spatial adjacency weight is set to 1, and the test order weight is set to 0. The above weight settings are used to ensure that the side with sufficient sample quantity to meet the calculation conditions makes an effective contribution to the spatial order drift correlation, avoiding unstable results introduced by spatial relationships or test order relationships with insufficient sample quantity.

[0081] When the number of spatially adjacent test points is 0 and the number of valid test points in the test order window is less than 4, the spatial order association module marks the spatial order drift association as unavailable and generates a spatial order association missing identifier. The spatial order association missing identifier is not used as the result of the anomaly source label, but is only used to indicate that the anomaly interpretation output module does not use the spatial order drift association to perform anomaly source labeling.

[0082] For wafer edge test points, if the number of spatially adjacent test points is less than the number of test points inside the wafer MAP, the spatial order association module calculates the spatial adjacent contact aggregation result based on the actual existing spatially adjacent test points with original observation records, and adjusts the spatial adjacent weight and test order weight according to the association weight rules. For test points in the test order window where the contact state description is unavailable, the spatial order association module deletes the test point from the average calculation of the first or second half of the window. If the number of available contact state descriptions in either half of the window after deletion is less than 2, the order drift result is not calculated, and the test order weight is set to 0.

[0083] The spatial order drift correlation is only used to characterize the degree of contact drift correlation of test points in the wafer MAP spatial adjacency relationship and test sequence window, and does not directly output anomaly source labels; when the spatial order drift correlation reaches the drift correlation threshold, it indicates that there is evidence of contact drift correlation in the spatial relationship and test process relationship of the test point; whether to mark the test point as a local contact drift anomaly, test link anomaly, environmental transition anomaly, or device real degradation anomaly is completed by the anomaly interpretation output module in combination with the contact state description quantity, effective stress window identifier, and device failure curve consistency results; the drift correlation threshold is determined by the anomaly interpretation output module based on stable batch statistical results, measurement system repeatability and reproducibility test results, device test specifications, and test instrument measurement errors.

[0084] The processing result of the spatial order correlation module is the spatial order drift correlation quantity. This result jointly expresses the aggregation of contact state descriptors in the wafer MAP spatial adjacency relationship and the changes before and after in the test sequence window. Compared with threshold judgment only for the contact state descriptors of a single test point, this implementation can identify contact drift accumulated along the test sequence and contact anomaly aggregation in a local spatial range, thereby providing process evidence and spatial evidence for the subsequent anomaly interpretation output module to distinguish between local contact drift anomalies, test link anomalies, environmental transition anomalies and device real degradation anomalies.

[0085] In one implementation, the anomaly interpretation output module is used to generate device failure curve consistency results based on the electrical performance observation domain content, contact state description quantity, effective stress window identifier, and spatial order drift correlation quantity, and to mark the test points that meet the anomaly judgment conditions with anomaly source labels, and output electrical performance test interpretation results; this module accepts the electrical performance observation domain content formed by the test observation construction module, the contact state description quantity and effective stress window identifier generated by the contact window parsing module, and the spatial order drift correlation quantity generated by the spatial order correlation module.

[0086] In this embodiment, the leakage current curve, on-resistance curve, threshold voltage curve, and breakdown voltage curve in the electrical performance observation domain are the original test results in wafer-level electrical performance testing. The original electrical performance test values ​​retain the dimensions of current, resistance, or voltage and are output as test report content in the electrical performance test interpretation results. The anomaly interpretation output module does not directly output the actual device degradation anomaly based on whether a single electrical performance test value exceeds the device test specification limit. Instead, within the effective stress window, it converts the degradation direction of different electrical performance types into a unidirectional degradation characterization and then combines the contact state description quantity and spatial order drift correlation quantity to determine the anomaly source label. This processing is used to distinguish the similar manifestations of probe contact instability, test link repeatability anomaly, environmental transition state, and device body degradation in the electrical performance observation domain.

[0087] The anomaly interpretation output module first checks whether the test point has an available effective stress window identifier, contact state descriptor, and spatial order drift correlation. When the effective stress window identifier indicates that the test point is within the effective stress window, it proceeds to calculate the consistency result of the device failure curve. When the effective stress window identifier indicates that the test point is in an environmental transition state, the electrical performance change of the test point is not directly used as the basis for determining the actual degradation anomaly of the device, and it proceeds to the environmental transition anomaly determination path. When the contact state descriptor is marked as unavailable, or the spatial order drift correlation is marked as unavailable, the anomaly interpretation output module retains the corresponding unavailable identifier and transfers the test point to the test link verification path, without replacing the normal determination result with the missing result.

[0088] Before generating the consistency results of device failure curves, the anomaly interpretation output module determines the electrical performance benchmark value, allowable variation range, and degradation direction for each electrical performance type. The electrical performance benchmark value is determined by the statistical results of stable batches, device test specifications, and measurement errors of test instruments. When the device test specifications provide electrical performance limits or allowable variation ranges, the device test specifications shall be used as the priority. When the device test specifications do not provide corresponding benchmarks, the electrical performance statistical results of the same model of device in the stable batch within the effective stress window shall be used as the benchmark. The allowable variation range is determined by the device test specifications, the statistical fluctuation range of stable batches, and the measurement errors of test instruments.

[0089] The normalized degradation variation is obtained by dividing the difference between the electrical performance test value at the test point within the effective stress window and the corresponding electrical performance reference value by the corresponding electrical performance reference value or the allowable variation range. When the corresponding electrical performance reference value is 0 or lower than the resolution of the test instrument, the allowable variation range of the electrical performance type is used as the normalization denominator. This process converts different electrical performance types into dimensionless quantities before entering the consistency result calculation of the device failure curve, while retaining the physical dimensions of the original electrical performance test values ​​and the meaning of the test report.

[0090] The degradation direction coefficient is used to convert the change direction of each electrical performance type into a unidirectional degradation characterization. When the failure direction of an electrical performance type is that the test value is higher than the electrical performance reference value, the degradation direction coefficient is 1; when the failure direction of an electrical performance type is that the test value is lower than the electrical performance reference value, the degradation direction coefficient is -1. The degradation direction coefficient corresponding to the increase in leakage current and the increase in on-resistance is 1, and the degradation direction coefficient corresponding to the decrease in breakdown voltage is -1. The degradation direction coefficient of the threshold voltage is determined according to the failure direction defined in the device test specification. Through this processing, the increase in leakage current, the increase in on-resistance, the drift of the threshold voltage towards the failure direction, and the decrease in breakdown voltage can participate in the calculation of the consistency results of the failure curve of the same device with a unidirectional degradation characterization.

[0091] Let the first The consistency results of the device failure curves at each test point are as follows: The calculation method is as follows:

[0092] ;

[0093] in, Indicates the test point number; Indicates the first Consistency results of device failure curves at each test point; Indicates the electrical performance type number; This indicates the number of electrical performance types involved in the degradation consistency assessment, excluding electrical performance types with missing samples, range overflow, or unavailable references. Indicates the first The test point The normalized degradation change of a type of electrical performance relative to the corresponding electrical performance benchmark value within the effective stress window; Indicates the first Degradation direction coefficient for electrical performance type; Indicates the first The degradation judgment threshold for electrical performance types is determined by the device test specifications, the statistical fluctuation range of stable batches of devices of the same model, and the measurement error of the test instrument. Since the normalized degradation change and the degradation judgment threshold are both dimensionless, the consistency result of the device failure curve is also dimensionless, with a value range of 0 to 1. The larger the value, the more the electrical performance change at the corresponding test point conforms to the device degradation direction.

[0094] When the number of electrical performance types participating in the degradation consistency judgment is 1, the device failure curve consistency result is determined by the same-direction degradation characterization of that single electrical performance type. When the number of electrical performance types participating in the degradation consistency judgment is greater than 1, the anomaly interpretation output module averages the same-direction degradation characterization of each electrical performance type, so that the device failure curve consistency result reflects whether multiple electrical performance changes jointly point to device degradation. When the original test value of any electrical performance type has a sampling missing, range overflow, or reference unavailable flag, that electrical performance type does not participate in the calculation of the device failure curve consistency result. If the number of electrical performance types participating in the degradation consistency judgment is 0, the device failure curve consistency result is marked as unavailable, and the test point is written into the test link review record. The test link review record includes at least the test point joint index, the reason for unavailability, the electrical performance type involved, the original electrical performance test value, the sampling missing flag, the range overflow flag, and the reference unavailable flag, and the test point does not enter the direct judgment path of the device's true degradation anomaly.

[0095] The test link verification includes sampling integrity verification, range validity verification, and reference availability verification. Sampling integrity verification is used to determine if there are any missing samples in the electrical performance observation domain, if the sampling time does not fall within the effective stress window, or if the test point joint index does not correspond to the original electrical performance curve. When there is a retrievable curve record for the same test point in the original test record, the original curve record is used to supplement the electrical performance observation domain content for the corresponding electrical performance type, and the normalized degradation change calculation is re-executed. When there is no retrievable curve record in the original test record, the sampling missing identifier is retained, and the test point is marked as the source test point to be verified.

[0096] Range validity verification is used to determine whether the original electrical performance test value has range overflow, test instrument limitation, or exceeds the recording range allowed by the test specification. When there is an alternative range retest value for the same test point in the test instrument record, the original electrical performance test value of the corresponding electrical performance type is updated with the alternative range retest value, and the device failure curve consistency result calculation is re-executed. When there is no alternative range retest value, the range overflow flag is retained, and the electrical performance type is excluded from the device failure curve consistency result calculation.

[0097] The baseline availability verification is used to determine whether the corresponding electrical performance baseline value, allowable variation range, or degradation judgment threshold can be determined by the device test specification, stable batch statistics, and test instrument measurement error. When the device test specification provides a baseline value or limit for the corresponding electrical performance type, the device test specification is used to regenerate the electrical performance baseline value and degradation judgment threshold. When the device test specification does not provide a corresponding baseline value or limit, stable batch samples with the same device model, the same probe station, the same probe card, and the same test instrument configuration are re-screened. Test points with missing environmental stress records, missing contact observation indicators, unavailable contact state description quantities, range overflow indicators, and environmental transition state indicators are removed. The electrical performance baseline value, allowable variation range, and degradation judgment threshold are regenerated based on the screened stable batch samples.

[0098] After review, if the corresponding electrical performance type can form a valid electrical performance benchmark value, allowable variation range, and degradation judgment threshold, then the electrical performance type is rewritten into the set of electrical performance types participating in the degradation consistency judgment, and the consistency result of the device failure curve is recalculated; if a valid benchmark still cannot be formed after review, then the benchmark unusable mark is retained, and the test point is marked as the source test point to be reviewed, and the device true degradation anomaly label is not directly output.

[0099] The degradation judgment threshold is determined based on the device test specification, stable batch statistical results, and test instrument measurement error. When the device test specification provides a degradation judgment standard for the corresponding electrical performance type, the degradation judgment threshold is calculated by converting the corresponding limit value in the device test specification. When the device test specification does not provide a degradation judgment standard, the degradation judgment threshold is determined by the larger of the 99th percentile of the normalized degradation change of the electrical performance type in the stable batch and the converted value of the test instrument measurement error. This determination method ensures that the degradation judgment threshold is higher than the normal fluctuation range formed by the stable batch statistical fluctuation and the test instrument measurement error, thus avoiding the identification of normal measurement fluctuations as actual device degradation anomalies.

[0100] The anomaly interpretation output module also determines the contact anomaly threshold, drift correlation threshold, device degradation threshold, and retest threshold. The contact anomaly threshold is determined based on the statistical results of the contact state descriptor of the stable batch, the repeatability and reproducibility test results of the measurement system, and the measurement error of the test instrument. When the device test specification does not provide a contact anomaly threshold, the contact anomaly threshold is taken as the larger of the 99th percentile of the contact state descriptor of the stable batch and 0.75. When the larger value exceeds 1.25, the corresponding stable batch is marked as having unavailable contact reference and enters the stable batch review path.

[0101] The drift correlation threshold is determined based on the statistical results of the spatial order drift correlation of stable batches. When the device test specification does not provide a drift correlation threshold, the drift correlation threshold is the larger of the 99th percentile of the spatial order drift correlation of stable batches and 0.25. When this larger value exceeds 1.00, the corresponding stable batch is marked as having an unavailable spatial order correlation benchmark and enters the stable batch review path. The device degradation threshold is used to determine whether the consistency result of the device failure curve meets the requirements for determining the actual degradation of the device. When the device test specification does not provide a device degradation threshold, the device degradation threshold is the larger of the 99th percentile of the consistency result of the device failure curve of stable batches and 0.60. When this larger value exceeds 0.85, the corresponding stable batch is marked as having an unavailable device degradation benchmark and enters the stable batch review path. The retest threshold is used to determine whether the dimensionless fusion feature corresponding to the repeated differences meets the requirements for the test link review. When the device test specification does not provide a retest threshold, the retest threshold is 1.00.

[0102] The batch verification path is used to reconfirm whether stable batch samples can be used as the basis for threshold formation when contact references, spatial sequence correlation references, or device degradation references are unavailable. During stable batch verification, candidate stable batch samples formed by the same type of semiconductor device under the same probe station, the same probe card, the same test instrument configuration, and the same extreme environmental test conditions are first read; then, test points with missing environmental stress records, missing contact observation indicators, range overflow indicators, unavailable contact state descriptive quantities, unavailable spatial sequence drift correlation quantities, and manually confirmed abnormal records are removed from the candidate stable batch samples to obtain the verified stable batch samples.

[0103] When the verified stable batch samples meet the minimum sample size specified in the device test specification or the measurement system repeatability and reproducibility test rules, the statistical distribution of contact state descriptive quantities, the statistical distribution of spatial order drift correlation quantities, and the statistical distribution of device failure curve consistency results are regenerated based on the verified stable batch samples. When neither the device test specification nor the measurement system repeatability and reproducibility test rules specify a minimum sample size, the verified stable batch samples shall be no less than 3 historical qualified test batches, and each historical qualified test batch shall have no less than 30 test points with original observation records of available test points.

[0104] When regenerating the contact anomaly threshold, the larger of the 99th percentile of the contact state descriptor in the verified stable batch samples and 0.75 is used; when regenerating the drift correlation threshold, the larger of the 99th percentile of the spatial order drift correlation in the verified stable batch samples and 0.25 is used; when regenerating the device degradation threshold, the larger of the 99th percentile of the consistency result of the device failure curve in the verified stable batch samples and 0.60 is used. If the regenerated contact anomaly threshold still exceeds 1.25, the drift correlation threshold still exceeds 1.00, or the device degradation threshold still exceeds 0.85, the corresponding benchmark continues to be marked as unusable, and the benchmark unusable verification mark is output in the electrical performance test interpretation results; the corresponding test point does not directly enter the device actual degradation anomaly judgment path, but enters the source pending verification output path.

[0105] The above threshold values ​​are based on the fact that both the contact state description quantity and the spatial order drift correlation quantity are evidence of contact drift, and their thresholds should be higher than the statistical upper limit formed by normal contact fluctuations in the stable batch; the consistency result of the device failure curve is a unidirectional degradation characterization between 0 and 1, and the device degradation threshold is not lower than 0.60 to require multiple electrical performance types to form unidirectional degradation evidence. When the statistical upper limit of the stable batch exceeds 0.85, it indicates that the stable batch itself is not suitable as a benchmark for device degradation judgment; the retest threshold of 1.00 indicates that the repeatability difference reaches the upper limit of the allowable repeatability error for the corresponding electrical performance type.

[0106] After completing the consistency results of the device failure curve and determining each threshold, the anomaly interpretation output module performs anomaly source label determination. When the same test point meets multiple anomaly source label conditions at the same time, the anomaly interpretation output module determines the final anomaly source label according to the anomaly source label determination priority. The determination priority is as follows: environmental transition anomaly, test link anomaly, local contact drift anomaly, and device real degradation anomaly. This priority is only used for conflict resolution when the same test point meets multiple anomaly source label conditions at the same time, and does not change the determination conditions of each anomaly source label itself.

[0107] When the effective stress window indicates the environmental transition state, and at least half of the test points within the same test sequence window have the same direction of electrical performance degradation, the anomaly interpretation output module will mark the corresponding test point as an environmental transition anomaly. The same direction of electrical performance degradation means that at least half of the test points within the same test sequence window show the same direction of change in the same electrical performance type, and this same direction of change occurs synchronously with the environmental stress loading process. This determination is used to identify batch directional fluctuations formed when the environmental temperature has not yet stabilized or the environmental stress is in the loading transition stage.

[0108] When the contact state description quantity reaches the contact anomaly threshold and the dimensionless fusion feature corresponding to the repeated difference reaches the retest threshold, the anomaly interpretation output module marks the corresponding test point as a test link anomaly. This judgment is used to identify the decrease in test link reliability caused by the combination of contact resistance change, repeated measurement difference and unstable electrical measurement acquisition. For test points marked as test link anomalies, retest suggestions are generated in the electrical performance test interpretation results. The retest suggestions include re-performing probe contact confirmation, repeated measurement and contact resistance verification.

[0109] When the effective stress window identifier characterizes the test point as being within the effective stress window, the contact state description quantity reaches the contact anomaly threshold, the spatial sequence drift correlation quantity reaches the drift correlation threshold, and the device failure curve consistency result does not reach the device degradation threshold, the anomaly interpretation output module marks the test point as a local contact drift anomaly. This judgment indicates that the electrical performance anomaly is related to the probe contact state and spatial sequence drift evidence, but the device failure curve consistency result does not form true device degradation evidence. Therefore, the test point is not directly judged as a true device degradation anomaly.

[0110] When the effective stress window identifier characterizes the test point as being within the effective stress window, the contact state description quantity does not reach the contact anomaly threshold, the spatial sequence drift correlation quantity does not reach the drift correlation threshold, and the device failure curve consistency result reaches the device degradation threshold, the anomaly interpretation output module marks the test point as a true device degradation anomaly. This judgment indicates that the test point is under effective environmental stress conditions, the contact state and spatial sequence drift evidence do not point to contact anomalies, and the change in the direction of electrical performance degradation meets the device degradation judgment requirements.

[0111] If the same test point meets multiple anomaly source label conditions simultaneously, the anomaly interpretation output module outputs an anomaly source label according to the aforementioned judgment priority, and writes the judgment conditions that are not output as the final label into the auxiliary judgment record in the electrical performance test interpretation result; if the same test point does not meet the judgment conditions of environmental transition anomaly, test link anomaly, local contact drift anomaly, and device real degradation anomaly, but its original electrical performance test value exceeds the device test specification limit, it is marked as a source to be verified label, and the electrical performance type exceeding the limit, the original electrical performance test value, and the missing evidence that needs to be verified are recorded in the electrical performance test interpretation result.

[0112] The final anomaly interpretation output module outputs the electrical performance test interpretation results. These results include a test point joint index, anomaly test point location, anomaly source label, original electrical performance test values, contact state description, effective stress window identifier, spatial sequence drift correlation, device failure curve consistency results, auxiliary judgment records, and retest suggestions. The original electrical performance test values ​​retain current, resistance, or voltage dimensions. The contact state description, spatial sequence drift correlation, and device failure curve consistency results are output as dimensionless interpretation results. Retest suggestions are generated based on the anomaly source label. Through this processing, the anomaly interpretation output module can distinguish the same electrical performance anomaly into environmental transition anomalies, test link anomalies, local contact drift anomalies, actual device degradation anomalies, or source-to-review labels, thereby reducing the risk of misjudging probe contact drift or environmental transition disturbances as actual device failure.

[0113] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A semiconductor device testing system under extreme environments, characterized in that, include: The test observation construction module acquires extreme environment test observation data generated during wafer-level electrical performance testing in extreme environments. It establishes a joint index of test points using wafer MAP chip coordinates and test execution time scales, and collects the contents of the electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain of the same test point to form the original observation record of the test point. The contact window parsing module retains the original dimensions in the original observation records of the test points and converts the contact changes, positioning offsets, repetition differences and electrical measurement fluctuations that enter the fusion judgment into dimensionless fusion features, generating contact state description quantities and effective stress window identifiers. The spatial order association module constructs the wafer MAP spatial adjacency relationship and test order window based on the test point joint index, and generates spatial order drift association quantity according to the distribution change of contact state descriptor in the two. The anomaly interpretation output module generates device failure curve consistency results based on the changes in electrical performance degradation direction, contact state description, and spatial order drift correlation within the effective stress window. It also marks the anomaly source label for test points that meet the anomaly judgment conditions and outputs the electrical performance test interpretation results.

2. The semiconductor device extreme environment testing system according to claim 1, characterized in that, The test observation construction module divides the extreme environment test observation data into electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain; The electrical performance observation domain is formed by at least one of the following: leakage current curve, on-resistance curve, threshold voltage curve, and breakdown voltage curve, which characterize the change in electrical response at the test point; the contact observation domain is formed by contact resistance data, repeated measurement records, and contact resistance change records formed by the contact resistance data according to the test execution time scale, which characterize the contact state between the probe and the wafer pad; the spatial positioning observation domain is formed by the wafer MAP core coordinates, probe marking position, and pad center position, which characterize the spatial assignment and marking offset of the test point; and the environmental stress observation domain is formed by the stage temperature, extreme environment set temperature, and stage temperature change rate, which characterize the extreme environmental loading state.

3. The semiconductor device extreme environment testing system according to claim 2, characterized in that, The test observation construction module establishes a test point joint index by generating an object index based on the wafer number, wafer MAP row and column coordinates, and chip number; generating a sampling index based on the test start time, sampling time, and test execution sequence number; and binding the object index and the sampling index into a test point joint index, so that the same test point joint index points to the observation content of the same chip in the same extreme environment wafer-level electrical performance test process.

4. The semiconductor device extreme environment testing system according to claim 3, characterized in that, The test observation construction module generates original observation records for test points, including: performing object alignment and time-scale alignment on the observation content in the electrical performance observation domain, contact observation domain, spatial positioning observation domain, and environmental stress observation domain according to the test point joint index, and retaining the original dimensions, sampling source, and sampling time of each observation content after alignment; Among them, the electrical performance observation domain retains the dimensions of current, resistance or voltage according to the electrical performance type; the contact resistance data in the contact observation domain retains the dimension of resistance; the electrical performance test values ​​in the repeated measurement records retain the dimension of the electrical performance type they are measuring; the spatial positioning observation domain retains the dimension of length; and the environmental stress observation domain retains the dimensions of temperature and time.

5. The semiconductor device extreme environment testing system according to claim 4, characterized in that, The contact window parsing module converts contact changes, positioning offsets, repeatability differences, and electrical measurement fluctuations that enter the fusion judgment into dimensionless fusion features, including: generating dimensionless fusion features corresponding to contact changes based on the difference between the current contact resistance and the stable batch contact resistance benchmark value, and the allowable fluctuation range of the contact resistance; generating dimensionless fusion features corresponding to positioning offsets based on the offset distance of the probe dot position relative to the center position of the pad, and the allowable offset distance of the pad; generating dimensionless fusion features corresponding to repeatability differences based on the repeatability difference of the same electrical performance type at the same test point, and the allowable repeatability error of that electrical performance type; and generating dimensionless fusion features corresponding to electrical measurement fluctuations based on the fluctuation amplitude of the electrical performance observation domain content within the short-time sampling interval, and the corresponding electrical performance benchmark value. Among them, the short-time sampling interval is determined by the sampling cycle of the test instrument and the electrical performance test specifications. The stable batch contact resistance reference value, the allowable fluctuation range of contact resistance and the electrical performance reference value are determined based on the stable batch statistical results, the repeatability and reproducibility test results of the measurement system and the measurement error of the test instrument. The allowable offset distance of the pad is determined based on the pad size of the wafer layout and the probe station dotting specifications.

6. The semiconductor device extreme environment testing system according to claim 5, characterized in that, The contact window parsing module generates contact state description quantities by: configuring judgment weights for the dimensionless fusion features corresponding to contact changes, positioning offsets, repetition differences, and electrical measurement fluctuations, and weighting and aggregating each dimensionless fusion feature to obtain the contact state description quantities. The sum of the judgment weights is 1. The judgment weights are determined based on the statistical results of stable batches, the repeatability and reproducibility test results of the measurement system, and the measurement error of the test instrument. The contact state descriptor is a dimensionless quantity and is used to characterize the degree of probe contact instability at the test point.

7. The semiconductor device extreme environment testing system according to claim 6, characterized in that, The contact window parsing module generates effective stress window identifiers by: determining the environmental stress state of the test point based on the deviation between the stage temperature and the extreme environment set temperature, and the stage temperature change rate; when the deviation does not exceed the allowable temperature deviation and the stage temperature change rate does not exceed the temperature stability change rate threshold, the test point is marked as a test point within the effective stress window; when the deviation exceeds the allowable temperature deviation, or the stage temperature change rate exceeds the temperature stability change rate threshold, the test point is marked as an environmental transition state test point; the allowable temperature deviation is determined based on the extreme environment test specifications and the temperature sensor calibration error, and the temperature stability change rate threshold is determined based on the stage thermal response time and thermocouple calibration results.

8. The semiconductor device extreme environment testing system according to claim 7, characterized in that, The spatial order association module constructs the wafer MAP spatial adjacency relationship and test order window by: based on the wafer MAP core coordinates in the test point joint index, identifying cores that are adjacent to the target test point in the row direction, column direction, or diagonal direction and have original observation records of the test point as spatially adjacent test points, thus forming the wafer MAP spatial adjacency relationship; and based on the test execution time stamp in the test point joint index, grouping test points with consecutive test execution sequence numbers and consistent effective stress window identifiers into the same test order window.

9. The semiconductor device extreme environment testing system according to claim 8, characterized in that, The spatial order association module generates spatial order drift association quantities by: generating spatial adjacency contact aggregation results based on the average value of the contact state descriptors corresponding to the spatially adjacent test points of the target test point; dividing the test sequence window containing the target test point into a first half window and a second half window according to the test execution sequence number; when the number of test points in the test sequence window is odd, merging the test point corresponding to the median test execution sequence number into the first half window, and generating order drift results based on the absolute value of the difference between the average value of the contact state descriptors in the first half window and the average value of the contact state descriptors in the second half window; and aggregating the spatial adjacency contact aggregation results and the order drift results according to the association weights to obtain the spatial order drift association quantity; wherein, the sum of the association weights is 1, and the association weights are determined based on the number of spatially adjacent test points in the wafer MAP and the number of valid test points in the test sequence window.

10. A semiconductor device testing system under extreme environments according to claim 9, characterized in that, The anomaly interpretation output module generates device failure curve consistency results and marks anomaly source tags, including: within the effective stress window, generating device failure curve consistency results based on the degradation direction change of the electrical performance observation domain content relative to the corresponding electrical performance reference value. Among them, the increase of leakage current, the increase of on-resistance, and the failure direction drift of threshold voltage as defined by the device test specification are configured as positive degradation changes, and the decrease of breakdown voltage is configured as a negative degradation change. The degradation direction change of each electrical performance type is converted into the same-direction degradation characterization before consistency judgment is performed. When the contact state description reaches the contact anomaly threshold, the spatial sequence drift correlation reaches the drift correlation threshold, and the device failure curve consistency result does not reach the device degradation threshold, the test point is marked as a local contact drift anomaly; when the contact state description does not reach the contact anomaly threshold, the spatial sequence drift correlation does not reach the drift correlation threshold, and the device failure curve consistency result reaches the device degradation threshold, the test point is marked as a true device degradation anomaly; when the effective stress window characterizes the environmental transition state and at least half of the test points within the same test sequence window have the same electrical performance degradation direction change, the test point is marked as an environmental transition anomaly; when the contact state description reaches the contact anomaly threshold and the dimensionless fusion feature corresponding to the repeated difference reaches the retest threshold, the test point is marked as a test link anomaly, and the electrical performance test interpretation result containing the anomaly source label and the original electrical performance test value is output; Among them, the contact anomaly threshold, drift correlation threshold, device degradation threshold, and retest threshold are determined based on the stable batch statistical results, the repeatability and reproducibility test results of the measurement system, the device test specifications, and the measurement error of the test instrument.

Images