Method and apparatus for reconstructing the position of semiconductor components on a wafer

Abstract

This invention relates to a method for determining an assignment rule to combine test results from different tests from the same semiconductor device. The method includes the following steps: (S23) fitting a model such as a linear regression model to predict test data; (S24) calculating a cost matrix based on the predictions; (S25) applying the Hungarian method to the cost matrix to obtain a new assignment rule, and repeating these steps multiple times.

Description

[Technical Field]

[0001] The present invention relates to a method for reconstructing the position of semiconductor components on a wafer, and to an apparatus configured to perform the method, wherein the semiconductor components are mounted on a wafer and thereafter the semiconductor components are separated from the wafer. [Previous Technology]

[0002] During the packaging process of a semiconductor component (specifically PowerMOS), the traceability of the semiconductor component to its original wafer and its original position on the wafer are lost. Specifically, this means that once the wafer is diced or split (a method of separating semiconductor components from the wafer) and packaged, the position of each semiconductor component on the wafer can no longer be retrieved. The packaging process provider can provide at least a rough match between the loose semiconductor components in final testing (the testing process of semiconductor components after packaging) and the semiconductor components on the wafer in wafer-level testing (testing before packaging). However, this still results in thousands of semiconductor components that cannot be assigned to multiple wafers. Since this is essentially a combinatorial problem, the complexity of the solution to this problem is factorial, due to the existence of factorially different ways to configure the semiconductor components in the correct order, where is the number of semiconductor components.

[0003] For ASIC semiconductor devices, there is a solution to this combined problem. For this purpose, a unique identifier is stored in the memory of the ASIC semiconductor device during wafer-level testing, which allows the final test to be assigned to wafer-level testing after packaging. However, due to the absence of memory, this is not possible for semiconductor devices such as PowerMOS. [Summary of the Invention]

[0004] In the first embodiment, the present invention relates to a method for determining an assignment rule for assigning variables from a first set of first variables to variables from a second set of second variables, specifically a computer-implemented method. The assignment rule can assign first variables to second variables in a one-to-one manner, that is, each first variable is assigned to at most one second variable by the assignment rule, and preferably the reverse. A set can be understood as a combination of individual variables. Preferably, the first and second sets are different sets that do not share common variables. Preferably, indices are assigned to each of the variables in the first and second sets. All indices in the first and second sets can be interpreted as an index set, thus, as a set, its elements consecutively index variables in the first or second set. The assignment rule then assigns indices from the second index set to the first index set. Therefore, the assignment rule describes which first variable belongs to which second variable, and preferably the reverse. Assignment rules can be lists, tables, or similar.

[0005] This method begins by initializing the assignment rules and providing a first and a second set. The initial assignment rule can be selected randomly or in the form of an identifier mapping. Other initial assignment rules may be used as alternatives, such as predefined, partially correct assignments.

[0006] This is followed by repeated execution of steps a) to d), as explained below. Execution can be repeated for a specified maximum number of repetitions, or a termination criterion can be defined, whereby repeated execution is terminated if the termination criterion is met. For example, the termination criterion is assigning the minimum modification to the rule.

[0007] a) Create a dataset containing a first variable and individual second variables assigned according to the assignment rules. This dataset may also be called a training dataset, in which the second variables are assigned as "labels" of the first variables. Note that this step is optional, as subsequent steps using this dataset essentially only require information about the current assignment rules between the first and second variables, which can be provided by the dataset or by the current assignment rules. The current assignment rules are the assignment rules that exist for the current repetition of steps a) through d), i.e., the assignment rules used in the most recent version of the dataset creation.

[0008] b) The machine learning system is trained in such a way that it determines the data set as a function of the assigned second variable ("label"); the training procedure can be understood as adjusting the parameters of the machine learning system so that the predictions made by the machine learning system through its determination are as close as possible to the second variable ("label") of the data set. Optimization can be performed relative to a cost function. The cost function preferably represents the mathematical difference between the output of the machine learning system and the label. Gradient descent is preferably used to perform optimization. The machine learning system can be one or more decision trees, neural networks, support vector machines, or the like. Training can be performed until any further improvement of the machine learning system during training becomes negligible, i.e., the second termination criterion is satisfied.

[0009] c) Calculate the cost matrix, where each entry in the cost matrix represents the distance between the prediction of the machine learning system and the second variable according to the assignment rule, specifically the distance between the prediction of the machine learning system and all variables in the second set. This distance can be determined by a norm. Other distance measures can also be conceived. The cost matrix can be structured such that columns and rows are each assigned to the first variable or, based on the first and second variables, to the prediction of the machine learning system, where each entry represents the distance between the respective specified variables in the column and row. Entries not located on the diagonal of the cost matrix can be considered as transportation costs, which, contrary to the assignment rule, must be consumed to assign the first variable to the respective second variable in the corresponding column / row.

[0010] d) Optimize the assignment rule based on the cost matrix such that the assignment rule produces the minimum total cost based on the cost matrix entries. The total cost is the sum of the cost matrix entries necessary to assign variables of the first set to variables of the second set according to the current assignment rule from the cost matrix. In other words, optimize (specifically minimize) this sum for the entries selected from the cost matrix according to the assignment rule. Note that the entries are selected according to the assignment rule in such a way that the entries of each row and column of the cost matrix selected according to the assignment rule are assigned to each other by the first and second variables according to the assignment rule.

[0011] The assignment rule determined in the last repetition of step d) is the final assignment rule, which is output in the optional step.

[0012] The variables can be scalars or vectors, such as time series, especially sensor data acquired or indirectly determined by sensors. Preferably, the first and second variables are one or more measurement results from a single measurement or from a plurality of different measurements, and each of these measurements is performed on one of the plurality of objects. In other words, each variable is assigned to one of the objects. In the step of creating a dataset, only a predetermined number of measurement results from the plurality of measurement results can be used for the second variable. The assignment rule can specify which first and second variables are measurement results of the same object. Particularly preferably, at least one measurement of the object is performed for the first variable at a first time point, and a measurement is performed for the second variable at a second time point, the second time point being after the first time point. The second time point can be defined after the object has been modified or changed.

[0013] It is proposed to use a cost minimization algorithm to optimize the assignment rules for a given cost matrix. For example, the optimization system can be optimized using the Hungarian method applied to the cost matrix. The Hungarian method (also known as the Kuhn-Munkres algorithm) is an algorithm used to solve the weighted mapping problem. Alternatively, a greedy implementation of the cost minimization algorithm can be used.

[0014] It is also proposed that the machine learning system be a regression model, which determines the second variable as a function of the first variable and the parameters of the regression model, wherein the parameters of the regression model are adjusted during training.

[0015] Regression is used to model the relationship between a dependent variable (often also called the response variable) and one or more independent variables (often also called explanatory variables). Regression can parameterize more complex functions so that the data is best represented according to specific mathematical criteria. For example, the generalized least squares method calculates a unique straight line (or hyperplane) that minimizes the sum of squared deviations (i.e., the sum of squared residuals) between the true data and this line (or hyperplane).

[0016] It is also proposed that the first and second variables characterize the product according to different production process steps during product manufacturing. For example, the second time point here could be the time when a manufacturing process step is completed. The product can be any product produced in the manufacturing facility. Preferably, when manufacturing a product, for example, if it is no longer possible to directly assign the product from a block material (e.g., a screw) to a production batch, the traceability of the aforementioned process steps is lost (the so-called "block material"). It is conceivable that the first variable characterizes a component (specifically, a part), and the second variable characterizes the final product, wherein the assignment rule describes which component is processed to produce which product, or which component is installed in which product. An example of this operation is whether the components in the product can be read out without non-destructive removal. With this invention, it is then possible to assign the production batch of the component by measuring the product.

[0017] The first and second variables may be measurement / test results or other attributes of the product, component, etc. The first and second variables are usually slightly different from each other, for example, due to manufacturing tolerances, but describe the same measurement / attribute of the product, component, etc.

[0018] It is also proposed that the first variable be a first test result or measurement result of a semiconductor component on the wafer, and the second variable be a second test result or measurement result of the semiconductor component after it has separated from the wafer. The semiconductor component can be a portion of an electronic component already grown on the wafer, such as a group of transistors in an integrated circuit. The test result can also pertain to the entire semiconductor component. Linear regression has proven particularly effective in finding optimal assignment rules for machine learning systems. This is based on a linear relationship, which in this case is a reasonable assumption about the assignment test results. Linear regression is a special case of regression. In linear regression, a linear function is assumed. It only uses a linear combination of the regression coefficients (but not necessarily the independent variables) where the dependent variable is a linear combination of the regression coefficients.

[0019] It is also proposed that the first test result be a wafer-level test result, and the second test result be a final test result. Preferably, the final test result is fewer than the wafer-level test result. The tests are, for example, voltage tests and / or contact tests.

[0020] It is also proposed to produce semiconductor components on multiple different wafers. This is because it has been demonstrated that the method can even find the correct assignment rule across multiple wafers within a reasonable computation time.

[0021] It is also proposed to assign rules to determine which second test result belongs to which first test result, and then determine the location of the semiconductor component within the wafer based on the associated first test result. This allows for the reconstruction of the location, which for the first time makes it possible to uniquely trace the semiconductor component from the final manufacturing process step to the previous processing step.

[0022] It is also proposed that, in addition to these locations, other variables characterizing the wafer and / or semiconductor components on the wafer and separately assigned second test results be determined, wherein this data is combined into other training data sets, wherein other machine learning systems are trained based on the training data sets to predict the second test results.

[0023] The advantage of this is that the assignment can be used to create other training datasets to train other machine learning systems, thereby predicting the characteristics of packaged semiconductor devices in the early stages of the manufacturing process. This significantly reduces the time spent detecting deviations in process parameters, specifically for parameters that can only be correctly evaluated during final testing (e.g., RDSon).

[0024] Another advantage obtained here is that the assignment can also be used to train other machine learning systems that actively identify defective semiconductor wafers. This saves process resources and reduces waste.

[0025] In other embodiments, the present invention relates to a device and a computer program, wherein each is configured to perform the above methods; and a machine-readable storage medium on which the computer program is stored. Advantages of the present invention

[0026] The present invention, which has the features of the independent claim 1, has the following advantages: it enables the determination of potential assignment between semiconductor components that depend on the results of wafer-level testing and packaged semiconductor components that depend on the results of final testing, without the need to retrospectively add metadata, such as unique identifiers or the like.

[0027] The present invention also has the following advantages: it realizes a one-to-one assignment between semiconductor components and their original positions on the wafer, thus achieving better process control (e.g., root cause analysis of defective parts).

[0028] Other aspects of the present invention are the subject matter of the independent patent application. It is advantageous to develop them into the subject matter of the subsidiary patent application.

Implementation Method

[0030] During the packaging process of a semiconductor component or device, the traceability of the component to its original wafer and its original position on the wafer are lost. After the semiconductor component elements have been separated, individual semiconductor components may sometimes be mixed together, meaning that their position on the wafer is lost without being uniquely marked. This is schematically illustrated in Figure 1. Each wafer 10 has a plurality of semiconductor components or semiconductor devices 11. At this stage, each semiconductor device 11 has a known position on the wafer 10. Typically, the semiconductor device 11 undergoes a plurality of tests at this stage, also known as wafer-level testing. The wafer 10 is then diced into parts, separating the semiconductor devices 11 from each other. Dicing can be performed using a saw 12 or by laser. Finally, the semiconductor devices diced to a certain size are packaged, for example, mounted in a microcontroller 13. The next stage is when information about the wafer 10 on which the semiconductor device was originally located and its position within the wafer 10 is lost. Typically, the microcontroller 13, which has semiconductor device 11, undergoes several tests, also known as final tests. However, due to the mixing that occurs as wafer 10 is diced into parts, it is not easy to definitively determine where a given semiconductor device 11 of the microcontroller 13 is located on wafer 10 and which wafer-level test corresponds to which final test; that is, which are test results of the same semiconductor device. The semiconductor device can be a microelectronic module, such as an integrated circuit (hereinafter also referred to as a chip), a sensor, etc.

[0031] One objective of this invention is to restore traceability in semiconductor manufacturing after the packaging process. This assignment achieves other benefits, such as better process control or earlier prediction of final wafer properties. Furthermore, root cause analysis of deviations measured at the wafer level in final testing can be extended to the wafer manufacturing process. This, in turn, enables a deeper understanding of the process and results in better process control and thus improved quality.

[0032] The proposed assignment algorithm consists of an alternating sequence of optimized regression parameters (when regressing from chip-level test to final test data), followed by optimized test matching assignments. The current assignment of the final test chip is used as a 'regression marker' in each iteration.

[0033] This invention also uses a cost minimization algorithm, which determines the optimal one-to-one assignment under a specified cost matrix. To construct a suitable cost matrix, regression error is applied by calculating an appropriate distance measure (e.g., norm) between the final test prediction of the trained regressor and the regression label. Based on this cost matrix, the algorithm rearranges the chips in the final test to minimize regression loss. The regressor or regression model can be freely chosen depending on the characteristics of the data (e.g., linear regression with linear dependence).

[0034] Figure 2 shows a schematic flowchart 20 of a method for determining an assignment rule that maps the test results of the final test to the corresponding wafer-level test results. Upon completion of this method, an assignment rule should be obtained that assigns the associated test results of the wafer-level test to the final test. Therefore, this rule description originates from associated test results of the same semiconductor component.

[0035] The method begins at step S21. This step initializes the assignment rules. The test results of the wafer-level test (WLT) and the final test (FT) are also provided in this step.

[0036] This is followed by step S22. In this step, a training dataset is created containing the WLT test results and the individual FT test results assigned according to the assignment rules.

[0037] After step S22 is completed, step S23 follows. In this step, depending on wafer-level testing (WLT): a regressor is trained such that it determines the assigned final test based on the training dataset. The regressor can be a linear regression model. The regressor is trained in a known manner, for example, by adjusting the parameters of the regressor to minimize the regression error on the training dataset.

[0038] Once the regressor has been trained, proceed to step S24. In this step, a cost matrix is ​​created. Columns and rows are assigned to wafer-level tests and final tests, respectively. Based on the corresponding WFT test results for each column and the corresponding FT test results for each row, entries in the cost matrix are determined from the training data using the norm between regression predictions, and stored in the cost matrix.

[0039] After step S24 is completed, the assignment rules are optimized in step S25. Optimization is performed by applying the Hungarian method to the cost matrix to obtain improved assignment rules based on the cost matrix.

[0040] If the termination criteria are not met, then steps S22 to S25 are repeated. The termination criteria can be a specified maximum number of repetitions.

[0041] If the termination criteria are met, the method terminates and the assignment rule can be output.

[0042] In an optional step following step S25, an assignment rule is used to reconstruct the position of the semiconductor component 11 on the wafer 10. The assignment rule can be used to reverse-engineer the WLT test results starting from the FT test results. Since the storage of WLT test results typically includes the locations where individual tests were performed within the wafer, it is possible to accurately reconstruct the location on the wafer where the corresponding semiconductor device is generated.

[0043] It is conceivable that, depending on the position reconstruction after step S25, control signals can be activated to control physical systems, such as computer-controlled machines, such as manufacturing machines, specifically processing machines for wafers. For example, if the FT test results are not optimal, the control signals can adjust previous production steps accordingly to obtain better FT test results later.

[0044] Figure 3 shows a schematic diagram of the device 30 used to perform the method according to Figure 2.

[0045] The device includes a supplier 51 that provides the training dataset as described in step S22. The training dataset is then fed to a regressor 52, which uses this dataset to determine the output variable. The output variable and the training dataset are fed to an evaluator 53, which uses the output variable and the training dataset to determine the updated parameters of the regressor 52, and transmits these parameters to a parameter memory P, whereby these parameters replace the current parameters. The evaluator 53 is configured to perform step S23.

[0046] The steps performed by the device 30 can be implemented as a computer program on the machine-readable storage medium 54 and executed by the processor 55.

[0047] The term "computer" covers any device used for processing predefined computational rules. Such computational rules may be provided in software form, hardware form, or a hybrid of software and hardware. [Simplified Explanation of the Diagram]

[0029] In the following, exemplary embodiments are described in more detail with reference to the accompanying drawings. In the drawings: [Figure 1] shows a schematic diagram of the packaging process; [Figure 2] shows an exemplary embodiment of the flowchart of the present invention; [Figure 3] shows a schematic diagram of the training device.

Claims

1. A method for determining an assignment rule for assigning a first variable from a first set of first variables to a second variable from a second set of second variables, comprising the steps of: initializing (S21) the assignment rule and providing (S21) the first set and the second set; repeating steps a) to c): a) training (S23) the machine learning system in such a way that the machine learning system determines the second variable assigned according to the assignment rule as a function of the first variable; b) calculating (S24) a cost matrix, wherein entries in the cost matrix represent the distance between the prediction of the machine learning system as a function of the first variable and the second variable; and c) optimizing (S25) the assignment rule based on the cost matrix such that assigning the first variable to the second variable according to the assignment rule results in the minimum total cost based on the entry in the cost matrix. The first variable is the first test result of the semiconductor component on the wafer, and the second variable is the second test result of the semiconductor component after it has been separated from the wafer. The assignment rule indicates which test results of the first test result and the second test result originate from the same semiconductor component.

2. The method of request item 1, wherein the optimization (S25) of the assignment rule is performed by means of the Hungarian algorithm or a greedy implementation.

3. The method of request item 2, wherein the cost matrix is ​​square, wherein if the number of the first variable and the number of the second variable do not match, the maximum value in that entry of the cost matrix is ​​used to fill the blank entry of the cost matrix.

4. The method of any one of requests 1 to 3, wherein the machine learning system is a regression model (53) that determines the second variable as a function of the first variable and the parameters of the regression model.

5. The method of any one of claims 1 to 3, wherein the first variable and the second variable characterize the product according to different production process steps during the production of the product, wherein the assignment rule characterizes which variables in the first variable and the second variable of the first set and the second set characterize the same product.

6. The method of request item 1, wherein the first test result is a wafer-level test result and the second test result is a final test result.

7. The method of any one of claims 1 to 3, wherein the semiconductor component is produced on a plurality of different wafers.

8. The method of claim 1, wherein the assignment rule is used to determine which second test result belongs to which first test result, and then, based on the associated first test result, to determine the location of the semiconductor device within the wafer.

9. The method of claim 8, wherein, in addition to the location, other variables characterizing the wafer and / or the semiconductor component elements on the wafer and the assigned second test results are determined, wherein this data is combined into a training dataset, and other machine learning systems are trained based on the training dataset to predict the second test results.

10. The method of claim 1, wherein the semiconductor component element is a power MOSFET.

11. An apparatus (30) configured to perform any of the methods requested in items 1 to 10.

12. A computer program product comprising instructions that cause the computer to perform the methods described in claims 1 to 10 during execution of the computer program product.

13. A machine-readable storage medium having a computer program product as described in claim 12 stored thereon.

Images