WO2019219669A1

WO2019219669A1 - Method for evaluating measurement data

Info

Publication number: WO2019219669A1
Application number: PCT/EP2019/062321
Authority: WO
Inventors: Christoph Eichler
Original assignee: Osram Opto Semiconductors Gmbh
Priority date: 2018-05-14
Filing date: 2019-05-14
Publication date: 2019-11-21
Also published as: DE102018111378A1

Abstract

A method for evaluating a multiplicity of measurement datasets is proposed, which method determines a first statistical measurement variable vector for the measurement data of at least one measurement dataset. The multiplicity of measurement data are combined to form a multiplicity of partial measurement datasets, wherein an amount of the measurement data in the multiplicity of partial measurement datasets is substantially equal and the measurement data in any partial measurement dataset lie in a contiguous area. A second statistical measurement variable vector is then generated for each partial measurement dataset. Measurement data of the at least one measurement dataset are determined, which measurement data lie inside or outside an area spanned by the connecting lines of two adjacent second statistical measurement variable vectors.

Description

METHOD FOR EVALUATING MEASUREMENT DATA

The present disclosure claims the priority of German application DE 10 201 8 111 378.1 of May 15, 2018.

A method for evaluating a plurality of measurement records is given. Furthermore, an arrangement and a computer-readable storage medium as well as a use of the drive Ver disclosed.

In the manufacture and development of semiconductor chips, larger series of tests are indispensable for determining the exact parameter space. At the same time, process vari- ants require extensive testing in a few production steps to determine the yield or to divide the components into different batches.

For the evaluation of (measurement) data, the data must be prepared in such a way that the relationships are recognizable as quickly as possible and the correct conclusions can be drawn from the data. For example, in the component and in particular the LED development, large amounts of measured data are generated daily, which must be evaluated as efficiently as possible for the development and improvement of products and compared with the data of reference components. There are often measurement data that are considered as a function of other measured parameters. This includes, for example, the dependence on brightness vs. Wavelength or color temperature vs. Tension. Other examples are from component scattering of e.g. Transistors, amplifiers or other components known.

One aspect of the proposed principle is to specify a new, intuitive representation for the measurement data and the associated calculation algorithm. This can be useful in particular if a simple representation of larger amounts of measured data is no longer possible or if a larger number of components are no longer available. must be characterized and / or assigned to the measured values assigned to them. In the following, the term measuring date or measured data is understood to mean a set of interlinked measured variables. A measurement date comprises at least two measurands, of which at least one depends on the other. Each measurement datum can thus be represented as a form of a vector of individual measurands: M = (nl, n2, n3, ...), where nl, n2 ... are the measurands in the individual dimensions. Accordingly, the term "statistical measured variable vector" SM is understood to mean a vector of statistical measured variables in the individual dimensions SM = (sml, sm2, sm3...), Where sml, sm2 are the individual statistical measured variables in the dimensions.

One aspect of the proposed principle relates to a computer-implemented method for evaluating a plurality of measurement data records. In this case, at least one measurement data record is made available, wherein the at least one measurement data set comprises a plurality of measurement data, and each measurement data of the plurality of measurement data relates a first measurement variable to a second measurement variable. A first statistical measurement vector for the at least one measurement data set is generated. The plurality of measurement data of the at least one measurement data set are combined to form a plurality of partial measurement data records, so that each measurement data is assigned to a partial measurement data record. In this case, a number of the measurement data in the plurality of partial measurement data sets is essentially the same. Furthermore, the measurement data of each partial measurement data set can lie in a coherent area. In particular, no single measurement data is only surrounded by measurement data of another partial measurement data set.

Likewise, a second statistical measurement vector is generated for each partial measurement data set. Finally, measurement data of the at least one measurement data set are determined which lie within or outside a surface spanned by the connection lines of two neighboring second statistical measurement variable vectors. The proposed method makes it possible to combine a larger amount of measurement data into a closed polygon whose vertices are formed by statistical quantities of essentially the same number of measurement data, so-called "pie pieces." For a simpler representation, the statistical measure vector vectors of the at least one measurement data set or partial measurement datasets should be highlighted separately.

In one aspect, the first and / or the second statistical parameter vector is formed, in which a median is determined from the measured variables of the measured data. Alternatively, an average of the measured quantities of the measured data can be determined. The statistical measured variable thus forms a threshold value for later determination of the measurement data, which are located inside or outside the area defined by the threshold values. The median, or the average, form special thresholds for a number of measurement data. In another embodiment can be seen easily to make the threshold more flexible. For this purpose, it can be provided to use quantiles of predefined size as the statistical measured variable. Other functions, in particular statistical functions, are also suitable as a threshold value or as a measured variable vector, for example a standard deviation.

In another aspect, the method and step of generating the first and / or second statistical statistical vector comprises the following steps:

Generating a first statistical measurement variable from the first measurement variables of the plurality of measurement data;

Generating a second statistical measured variable from the two measured variables of the plurality of measured data;

- combining the first and second statistical measures to the statistical measure vector.

This can take account of the fact that each measurement datum comprises a pair, triple or quadruple on mutually dependent measured quantities. Dependent on each other means in In this case, that one measured variable can be represented as a function or parameter of another measured variable. In the proposed case, the respective statistical measured variable is determined separately, ie the statistical measured variable is determined for each of the interdependent measured variables of the measured data record or of the partial printed data set.

Mutually dependent measured variables can be different electrical, optical or other physical measured variables. Such measured variables may include, for example: different currents and / or voltages, colors, color temperatures, brightness, half-value bandwidth of a measured variable and others.

In another aspect, the first and / or second statistical measures are determined by determining a median from the measured quantities. Alternatively, it is also possible to determine an average value of the measured variables, or another statistical measured variable, such as standard deviation, etc. This also includes predefined quantiles from the measured variables. Depending on the application, suitable statistical variables can also be used in the process. Thus, if the statistical measure vector is to be more robust to outliers, then the statistical measures by the median, the geometric mean (for positive numbers), the trimmed mean, the windsorized mean, the quartile mean, or the weighted mean, where as weight For example, the distance or the reciprocal distance to the "first statistical measure" could be used. If the extremes tend to be in focus, the quadratic mean (for positive numbers), the cubic mean (for positive numbers), Range means or the arithmetic mean. Another aspect relates to the generation of the partial measurement data sets, the measured variable is finally used for the determination of the measured data. It can be provided that the plurality of partial measurement data sets is at least three. Alternatively, in order to obtain a smoother area given by the values of statistical measures of the partial measurement data sets, a larger number may be determined, for example in the range 6 to 12. For the processing of a large number of measurement data, the number of partial data sets may also be be determined by a mathematical dependency. For example, the number may be the nearest integer value of the decimal logarithm of the number of measurement data. Other options would be, for example, the square root or third root or fourth root or natural logarithm or just a fraction of the number such as a tenth or a hundredth of the number.

In this connection it has been found that the natural logarithm, or more generally the logarithm with the base between 1.5 and 3.5, preferably between 2 and 3 values, gives a good compromise between number of data sets and the corners of the polygon.

For a particularly clear representation, the number may be smaller than the root of the number of data points of a data record. Preferably, it is also at large data sets at less than 100, and for example less than 20.

In one aspect of the method, the step of collating comprises determining the angles between a reference value based on the first statistical measurement vector vector and a line through the first statistical parameter vector and the measurement data of the at least one measurement data set. In other words, an angle between a reference line that passes through the statistical measure and that through the connection between the statistical measured quantity and the measured data. As a result, one contains a characteristic size of each measurement datum relative to a reference value. For a classification into the partial measurement data sets, the measurement data can be sorted with regard to your associated angle and then summarized. Alternatively, measurement data with associated angles can also be combined in a predefined area. In this case, this range is given, for example, by the number of measurement data in the at least one measurement data set and the desired number of partial measurement data sets.

In this context, it is proposed in a further aspect that summarizing the sorted measurement data includes determining the number of measurement data in each partial measurement data set from a predetermined number of partial measurement data sets. If no integer value should result, the nearest integer value can be taken. Partial data records differ only by one measurement date in number. Subsequently, successive and sorted measurement data are summarized to a partial measurement data set until the predetermined number is reached.

With the method according to the proposed principle, in one aspect, piecemeal pieces with substantially the same number of measured data are formed by the step of combining, starting from the first statistical measured quantity. In a slightly alternative way, the median point is subtracted from each measurement point, with the median being placed on the coordinates (0; 0) .Then the measurement data are converted into polar coordinates and thus entered Angle determined (the radius of the polar coordinates is irrelevant here) The resulting array of angles is sorted. Then the total number of measurements is divided by the number of segments, which gives the number of points contained in a segment (eg, 123 points per segment). The array with the angles is then indexed with integer multiples of this number, eg the 0., 123., 246, 369. array entry. These are the angles that each form an edge of a pie slice. Based on these "limit angles", each measuring point can be assigned to a segment between two such limiting angles.

In another aspect, the step of combining takes place in such a way that, starting from the first statistical measured variables, vector circle segments are spanned, it being assumed that the measured data can be displayed in one plane. The circle segments are chosen such that the number of measurement data within each circle segment is essentially the same. Subsequently, the measurement data within a circle segment are combined to form a partial measurement data set.

In another aspect, the measurement data of each partial measurement data record is distinguished by a special property. Thus, a connecting line between the first statistical measured variable vector to a measurement date adjacent to two connec tion lines from the first statistical measurement vector to each measurement date of the same partial measurement data set; or it is adjacent to a connecting line between the first statistical measurement vector vector to a measurement date of the same partial measurement data set and a connecting line between the first statistical measurement vector to a measurement date of an adjacent partial measurement data set. This ensures that "edges" or "boundaries" of two adjacent partial measurement data sets proceed essentially straight and in one direction, starting from the first statistical measurement vector. The "boundaries" between the partial measurement data records can thus be in a star shape starting from the first statistical measurement vector run. In particular, in one aspect, the Teilmessda records sets no bulges or bulges. Likewise, individual measurement data are completely enclosed by measurement data from another set.

Another aspect relates to an arrangement with which the pre-proposed method can be performed. In one aspect, an arrangement includes at least one processor and a memory configured to store measurement data sets. The memory contains one or more programs with instructions that, when executed on the at least one processor, perform the proposed method. The arrangement can be, for example, part of a control and control unit of a production line for the production of electronic components, in particular semiconductor components. The method can thus be used for the preparation of measurement data in the production and / or development of electronic components, in particular for the preparation of measurement data of components of different wafers or wafer batches.

Another aspect relates to a computer-readable storage medium comprising one or more programs. These are designed to be executed by a processor and contain instructions for the proposed method.

In the following, the invention will be explained in detail with reference to drawings.

Show it:

1 shows a representation of a measurement data set in XY form for

Explanation of some steps of the proposed method;

2 shows a representation of a measurement data set in XY form for

Explanation of further steps of the proposed method; FIG. 3 is an illustration of a measurement data set in XY form for explaining further steps of the proposed method;

4 shows a representation of a measurement data set in XY form for

Explanation of further steps of the proposed method;

5 shows a representation of a measurement data set in XY form for

Representation of various statistical measures;

FIG. 6 shows a representation of a plurality of measurement data sets for explaining further aspects of the proposed method

Figure 7 shows an embodiment of the proposed method

FIG. 8 shows an embodiment of some method steps

Figure 9 is an illustration of an embodiment of an arrangement for

Implementation of the proposed method.

In the production and also in the development of semicon terbauelementen fall on a variety of measurement data, which is of great importance for the understanding of the individual process steps or for the quality of the components. For example, during the manufacturing process measurement data can be recorded in order to characterize the components with regard to various electrical or optical parameters and to divide them into individual groups. Likewise, in development, a single parameter of components can be selectively changed whose electrical or optical properties subsequently determined who the.

In addition to the categorization of components and the following division into different groups, there is also the possibility of displaying the measurement data in different ways for further evaluation. This often uses programs such as Microsoft Excel ™, Origin ™ or others. A widely used, easy-to-create and easy-to-read display of interdependent parameters is the XY plot. In this case, different measured quantities are plotted on the X-axis and the Y-axis. A measuring point on the X-axis forms For example, an "independent" measured variable, the Y-axis contains the corresponding dependent measured variable, resulting in a value pair or a measurement date, which is displayed at the corresponding coordinate of the plot.

Such a representation as an XY plot works well for smaller amounts of data. For larger data sets or for measurement data from different measurement series, the diagram becomes unclear or unreadable. It is uniquely complicated if each measurement datum has more than two dimensions, i. includes more than a few metrics, such as a triple or even more. Even with different types of representation of the individual NEN measured data sets, for example, by different symbols or different colors overlap with large amounts of data, the symbols and merge with each other. This makes it difficult for a user in manufacturing or in development to separate data from different samples in the same diagram of an XY plot or to assess their distribution.

The present application now proposes a method in which measurement data from at least one measurement data set with a larger number of measurement data are combined in a suitable manner so as to reduce the amount of data to be processed or also to be displayed. At the same time, despite the reduction, the approximate shape of the distribution of the measurement data is maintained, so that a statement is made possible despite the reduction. For example, a measurement data set of several thousand individual measurement data can be summarized in such a way that a polygon shape results with only a few corners.

The shape, location and size of the polygon may reflect predefined and appropriate statistical quantities of the original measurement data distribution of the entire set. For example, the polygon can be formed from the median of individual subregions, so that a simple classification of the individual measuring points is possible, please include. The shape of the polygon makes it easier to identify asymmetries or centroids in the distribution of measurement points. With several measurement series, these can be directly compared with each other in a simple way. Since the measurement data of a series are often linked to components, it is easier to classify components into groups and thus also quickly characterize large component quantities. This makes it possible to react faster to process fluctuations. Depending on the overall selected statistical function to generate the polygon surface, the method robust against outliers as _V orkommen more common in larger amounts of data.

FIG. 1 shows a representation of a measurement data set with a multiplicity, eg several hundred or a thousand individual measurement data in a YX plot view. The measurement data set originates from a semiconductor wafer with laser components. Due to different process variations in the individual production steps, slight differences in the doping or the thickness of the individual layers occur across the wafer. This ver changes the wavelength of the emitted light and the threshold current of the components. This dependence is shown in FIG. In this case, each measuring point 101 (here two examples be distinguished), referred to in the following measurement date, a pair of two measured variables assigned. A first measured quantity is shown on the X-axis and, in this exemplary embodiment, denotes the wavelength of the light emitted by the component. On the Y axis, the threshold current Ith of the device at the wavelength is plotted. 1 shows that the "center of gravity" of this measurement data set should be located slightly in the lower half of the plot, slightly offset from the left center The shape and center of gravity distribution can now be represented more simply using the proposed method. For this purpose, as shown in FIG. 1, a first statistical measurement data vector 200 is determined in a first step. This is the median in the embodiment. The measured variable vector 200 is determined individually for each dimension of the measured data. In this 2-dimensional representation, the median is thus determined for all measured variables on the X-axis and separately for all measured variables on the Y-axis. In the same way, the statistical size can also be determined for multi-dimensional measured data by calculating the statistical size for each dimension individually. Optionally, as shown here, the result, ie the median in the X and Y directions, can be entered as a point 200 with the correspondingly determined coordinates in the XY plot for better visualization. Figure 2 shows the next step. For this purpose, starting from the median, the measurement data set is subdivided into equally sized pie pieces 300 to 307. The term "equal size" is understood to mean that the number of measurement data 101 (only three of which are designated here by way of example) is approximately the same in each subarea 300 to 307. The subregions shown by the dashed line may be in this Thus, as the number of points in each subarea is substantially equal, the "representation" of the image will be different, so a "small" area will usually mean that the data will be closer together, while a larger area will indicate a larger distribution.

Furthermore, it is also apparent that the sub-areas of the type are designed so that they do not overlap. Therefore, there is no measurement date that is assigned to one subarea, but "lies" in another subarea, neither has the borderline between two subareas "bulges", but runs as shown starting from the median 200 in straight lines. The number of subsections in which a measurement record is subdivided affects the subsequent area che. A larger number results in a finer gradation and Be humor and allows just for larger data sets easier identification of focal points in the measured data set.

To determine the number of subregions can proceed in different ways. On the one hand, the number can be pre-divided, but must be at least 3. As a useful bar, a number of subregions in the range of 3 to 20, preferably 5 to 12 has been found. Alternatively, an estimate or calculation can be used. For example, the number of subregions may depend on the number of measurement data in the measurement data set. One way to determine the number would be according to the formula A = round (log (N)), where A represents the number of partial areas, which is formed by rounding down to the next integer number of the decimal logarithm of the number of measurement data N of the measurement data set. This results in n = N / A for the number n of measured data per subarea. "Remaining" measurement data (u = N mod A) are distributed as evenly as possible to the sub-areas, so that the difference between any two sub-areas is 0 or 1.

If the number A of subregions is fixed, the measurement data set can be subdivided into the pie wedges as shown in the figure so that there are n readings in each subrange (n + 1 readings are obtained in some subranges, unless n is an integer divisor of N) ).

There are several possibilities for this, one of which is shown below, and for the sake of simplicity it is assumed that N = A * n. Thus, starting from a perpendicular through the median 200 (or any other virtual reference line), one determines the angle of each connecting line between the median 200 and the individual measurement data. In other words, each measurement datum is assigned an angle number that can lie between 0 ° and 360 °. The measured data are then displayed according to the size of the ordered angle number arranged. Then the first n measurement data are assigned to the first subarea 300, the measurement data n + 1 to 2n to the second subarea and so on to the last subarea. The dashed lines in FIG. 2 show the boundary between the partial regions 300 to 307 thus formed. These represent circle segments in the XY plane, the number of measurement data in each circular segment being approximately equal. In an alternative possibility of the summary, the subregions are formed in such a way that virtual lines are formed from the median or the first statistical measured variable vector to a measurement datum. Each of these lines is adjacent to other lines, ie, there are second and third lines which are "left" and "right," respectively, adjacent to a "first" line, as indicated by lines 306, 307, and 308 in FIG. In this case, each line is characterized by one of two properties: Either the first line has a left and right "neighbor line", which are jointly assigned to the same partial measuring data set, or the first line, which is assigned to a part measuring data set, has a neighboring line, which is assigned to the same partial measurement data record and a further after-line, which is assigned to the adjacent partial measurement data record. Each measurement data set therefore comprises exactly two such lines, which include differently assigned neighboring lines. FIG. 3 shows the next step of the method. For each partial measurement data set that was generated in the previous step, the second statistical measured variable is now formed. For example, in the embodiment, this is again the median, which thus represents the "center of gravity" of the distribution of the measured data in the corresponding part-measuring data set In the exemplary embodiment, the same size is used for the first and the second measured variable, but this is not necessary In a final step, the determined measuring points, the medians of the individual part-measuring data sets, are connected. This takes place along adjacent partial measuring data sets, so that an enclosed area results. The corner points of this surface are determined by the determined second statistical measured quantities. In this respect, FIG. 4 shows the area thus created without the individual measurement data in the XY plot representation. The corner points of the polygon are connected to lines 400, so that the illustrated 8 corner is formed. With the illustrated surface, it is thus easier to display larger measurement data records or to compare measurement data from different sentences

On the basis of the shape and distribution of the surface can be made in a simple way statements about the entire measurement data set. For example, the position of the median within the area can be used to make a statement about the "density distribution" of the dataset.Measurement points that lie within the area have a smaller deviation from the median.in particular, asymmetries are more easily recognized by the shape of the cloud Measurement data sets whose measurement data are presented together Darge, the proposed form of representation allows a simple comparison of the various measurement data sets.

FIG. 5 shows a variant of the proposed method in which the statistical measured quantity is varied. This also changes the shape of the surface. Such a method can be used, for example, in the manufacture, in order to estimate the number of components in a predefined range. In the example shown, instead of the median (= 50% quantile), any other quantile Q is used for the partial measurement data records. This can be used to control how "large" the area of the enclosed measurement data is, for which a further case distinction is appropriate.

In the example, first the same steps as described with reference to FIGS. 1 to 3 are carried out, ie the first statistical measured variable qx and qy are entered into the respective dimensions. In this specific case, the median is determined. Likewise, the respective second statistical measured variable qx 'and qy' in the respective dimensions is determined for each partial measuring data record. These steps are still done without consideration of the special quantiles.

Subsequently, the case distinction must be carried out for the presentation. This becomes necessary because now the measuring points in each dimension are not exactly divided in half, as is the case with the median. Therefore, if the second statistical measured variable in the first dimension is greater than the first statistical measured variable (qx '> qx), the desired quantile Q is calculated for the partial measuring data set. If the second measured variable is smaller than the first measured variable (qx '<qx), the quantile to be calculated applies: 100% -Q. The same applies to the other dimensions.

FIG. 5 shows a measurement data set in which various quantiles were determined with the proposed method and plotted as an area. The point 200 in this embodiment represents the median of the entire measurement data set. The outer polygon 402 corresponds to the 75% quantile of each part measurement data set, the inner polygon 401 shows the 25% quantile. The polygon 400 between the other two polygons encloses an area corresponding to the 50% quantile of each partial measurement data set. In this example, various evaluations are made for a Messda tensatz that flow into the Dar position. This makes it possible to make further statements about the distribution of the measurement data in the measurement data record. It is also possible that polygon corners overlap.

FIGS. 6A and 6B show an exemplary embodiment in which a plurality of measurement data sets are combined into groups. Such different measurement data records can result, for example, in production if electrical parameters of semiconductor components are different Wafer batches are compared. In the embodiment shown, a total of six different semiconductor wafers have been processed in pairs. The resul tierenden semiconductor components of each wafer are measured in their electrical parameters and summarized the measurement data for the illustrated areas including their jewei time priorities. The illustration shown immediately shows that the processed wafers concentrate in different areas of the XY plot diagram. The areas 500a and 500b belong to the first wafer pair, the areas 510a and 510b to the second wafer pair and the areas 520a and 520b to the third wafer pair. By combining the wafer pairs into a group, shown in FIG. 6B, further properties become clear. Thus, the second group with the area 510 is characterized by a significantly larger distribution or spread. This results primarily from the fact that the individual areas 510a and 510b of the group are clearly separated so that the measured data in this group have a greater distribution. It can also be seen that the center 510c of the second group has barely shifted, but essentially corresponds to the center of the surface closed by the polygon 510a. Through the summary representation can be found with different knowledge about the distribution of the measured data and thus the electrical parameters of the semiconductor components for the different wafer lots ge so with different. In addition, since the individual measurement results can be assigned to the semiconductor components, process fluctuations in production or during development can also be shot back in a simple manner. FIG. 7 shows an embodiment of a flow chart for the individual method steps for evaluating a multiplicity of measured data. According to the proposed principle, at least one measurement data set is provided in the first step S1, this measurement data set comprising a plurality of measurement data. In this case, the term measuring data is understood to mean a summary of individual measured variables, of which at least one measured variable is dependent on the others. Each measurement data thus contains a vector (nl, n2...) Of at least 2 measurands n1 and n2. After provision of the at least one measurement data set, a first statistical measurement vector is generated for the measurement data set in the next step S2. In this case, the first statistical measured variable vector has the same dimension as the measured data, and therefore also comprises at least 2 individual statistical values. The first statistical value is formed from the first measured quantities of the measured data of at least one measured data set, the second statistical measured variable from the respectively second measured variables of the measured data of the at least one measured data set, etc.

In step S3, the plurality of measurement data are combined into a plurality of partial measurement data. In this case, a number of the measured data of the partial measuring data sets is essentially the same. In other words, each partial measurement data set comprises substantially the same number of measurement data. In addition, the measurement data of each partial data set are in a contiguous area.

Then, in step S4, a second statistical measurement vector is generated for each partial measurement data set. Here, too, the second statistical measured variable vector of each partial measuring data set comprises at least 2 statistical values Quantities, each of which has been formed from the corresponding first and second measured variables of the measurement data in the part measuring data set IN ANY.

Finally, in step S5, the measurement data of at least one measurement data set are determined, which are located in or outside of a tensioned area through the connecting lines of two adjacent two statistical measurement variables.

For this purpose, it can be provided to connect respectively adjacent second statistical measured quantities by a line, so that a surface enclosed by a polygon is formed. Similarly, measurement data with a dimension greater than 2 (which results in an area) can also be represented. If the measurement data or parameter vectors have two or more elements, corresponding objects of higher order, e.g. in three dimensions, the connecting lines would enclose a volume. Thus one could e.g. also form three-dimensional (or multi-dimensional) clouds / objects using the same procedure and use them to assess "lies inside / outside".

In a further optional step, the measurement data of the at least one measurement data set can be displayed, wherein each measurement data is linked to an axis of the representation. For example, the representation in the measurement data can take place in an XY plot, so that the first measured variable of each measurement datum is shown on the X-axis and the second measured variable of each measurement datum is shown on the Y-axis. In addition, any statistical measurement vector can also be entered in such a plot. By Furthermore optional representation of the connecting lines of two adjacent statistical measured variable vectors can be represented by the area spanned thereby.

In optional embodiments of step S2, the first statistical measured variable vector is generated by forming a first statistical measured variable from the first measured variables of the plurality of measured data (S21). In the same way, a second statistical measured quantity is generated from the two measured variables of the plurality of measured data (S22). If the measurement data have a higher dimension, the person who repeats the aforementioned steps accordingly, so that statistical measures are formed separately from the different measures of the plurality of measurement data. Subsequently, the statistical measured variables are combined to form a first statistical measured variable vector (S23).

The second statistical measured variable vectors of the individual partial measuring data sets can be generated in the same way. This is expressed by the optional steps of S41, S42 and S43 in Fig. 7 at step S4. As described above, the statistical measurement variable vectors are formed for each partial data set in which statistical measured variables are generated from the individual measured quantities of the plurality of measured data.

The formation or generation of the statistical measured quantities can take place in various ways. For example, a median can be determined from the measured variables of the measured data. Alternatively, the determination of an average, the determination of a predefined quantile or the determination of a standard deviation is also possible. In general, the method can be applied to any statistical functions in which measured variables of measured data of a measurement data set form the basis.

FIG. 8 shows an optional embodiment of step S3 of FIG. 7. For this, an angle is determined for the step of combining, which is determined by a reference line from the first statistical measured variable vector and a line by the first statistical measured variables vector and a measuring datum of the at least one Measuring data set is spanned, step S31. Thus, a Win kelzahl each measurement date of at least one Messdatensat zes be assigned. Subsequently, the measurement data are sorted with regard to their assigned specific angle numbers, step S32, for example in ascending or descending form. The thus sorted measurement data can then be summarized in a suitable manner to the partial measurement data sets in step S33, so that the number of measurement data of each partial measurement data set remains substantially the same.

Figure 9 shows another optional step for summarizing sorted measurement data. For this purpose, in step S331 the number of measurement data of each part of the measurement data set is determined from a predetermined number of part measurement data sets. For example, this can be done by a simple quotient of the total number of measurement data and the desired number of partial measurement data sets, wherein the quotient is rounded to the next larger or next smaller integer value. Subsequently, the sorted successive measurement data are assigned to a partial measurement data set until the previously determined number of measurement data in their partial measurement data set is reached.

An arrangement in which the proposed use fin det, Figure 10 shows this is a module or plant 10 for semiconductor manufacturing and includes, inter alia, a test table 10a, ter with a shelf 40 for semicon ter and a test device 12 for testing various measurement variables of semiconductor components of the semiconductor wafer. For example, the test apparatus can be used to test electrical and optical parameters of the various semiconductor components before the singulation. The system 10 is connected to a computer 20 for transmitting the measured data. This comprises a memory 21 in which the measurement data of the individual measurement data records are stored and one or more processors 22. The computer 20 is designed to perform the proposed method and to display the result on a display 30 in a predefined representation. As a result, the method can be used to characterize semiconductor components on wafers and classify them into different classes. By assigning the components to a position on the semiconductor wafer, fluctuations in a production process can also be investigated. In another application, components that are located "on the edge" of the polygon (or more dimensional body) according to the mea- surement data assigned to them, could potentially pose a quality risk and thus be marked as bad (without human intervention) and / or sorted out. LIST OF REFERENCE NUMBERS

10 plant

10a test table

12 test device

20 computers

21 memory

22 processors

30 screen

40 storage area

10 0 Measurement data set

10 1 measured data

20 0 first statistical measurement vector

30 0 - 307 Partial measuring data set

30 6 - 308 lines

31 0 - 317 second statistical measurement vector 40 0 line

40 1 - 402 areas of different quantiles 50 0 - 520 grouped areas

50 0a - 520b areas of different measurement data sets

Sl, S2, S3, S4, S5 process steps

S21, S22, S23 optional process steps

S31, S32, S33 optional process steps

S331, S332 optional process steps

S41, S42, S43 optional process steps

Claims

A computer-implemented method for evaluating a plurality of measurement data records, comprising the steps:

Providing (Sl) at least one measurement data set (100), wherein the at least one measurement data set (100) comprises a multiplicity of measurement data (101), and each measurement data of the plurality of measurement data relates a first measurement variable to a second measurement variable;

Generating (2) a first statistical measured value vector for the at least one measured data record;

Combining (S3) the plurality of measurement data into a plurality of partial measurement data sets (300-307), wherein a number of the measurement data in the plurality of partial measurement data sets (300-307) is substantially equal and the measurement data of each partial measurement data set in a coherent area lie.

Generating (S4) a second statistical measured variable vector for each partial measurement data set;

Determining (S5) measurement data of the at least one measurement data set which lie within or outside a surface spanned by the connecting lines of two adjacent second statistical measured variable vectors.

The computer-implemented method of claim 1, further comprising at least one of the steps (S6):

Representing (S61) the measurement data of the at least one measurement data set, wherein each measurement data is linked to an axis of the representation;

Representing (S62) the first and / or second statistical measure

Representing (S63) the area spanned by the connecting lines of two adjacent second statistical measured variable vectors.

3. The computer-implemented method of claim 1, wherein the step of generating the first and / or second statistical statistical vector comprises:

Generating (S21, S41) a first statistical measured variable from the first measured variables of the plurality of measured data; Generating (S22, S42) a second statistical measured quantity from the second measured variables of the plurality of measured data;

Combining (S23, S43) the first and second statistical measured variable to the statistical measured variable vector.

4. The computer-implemented method according to claim 3, wherein generating (S21, S41, S22, S42) of the first and / or second statistical measured quantity comprises at least one of the following steps:

Determining (S211) a median from the measurands;

Determining (S222) an average value or a quantity derived therefrom from the measured quantities;

Determining a predefined quantile from the measured quantities;

Determining a geometric mean or a weighted average, or range means

Determining a standard deviation from the measured quantities.

5. Computer-implemented method according to one of the previous ones

Claims in which the plurality of partial data sets (300-307) is at least three, and preferably in the range of 6 to 12, or preferably approximately proportional to the natural logarithm of the number of measurement data, or preferably smaller than the root of the number of measurement data is.

The computer-implemented method of any one of the preceding claims, wherein the step of summarizing comprises:

Determining (S31) the angles between a reference line on the basis of the first statistical measured variable vector and a line through the first statistical measurement vector and the measurement data of the at least one measurement data set;

- sorting (S32) the measurement data with respect to the determined angle;

- summarizing (S33) the sorted measurement data to the

Variety of partial measurement data sets.

The computer-implemented method of claim 6, wherein the step of summarizing (S33) the sorted measurement data comprises:

Determining (S331) the number of measurement data in each part of the measurement data set from a predetermined number of partial measurement data sets;

Associating (S332) successive sorted measurement data with a partial measurement data set until the determined number has been reached.

8. The computer-implemented method according to claim 1, wherein the step of combining takes place in such a way that, starting from the first statistical measurement, pie slices having essentially the same number of measurement data are formed.

9. The computer-implemented method of claim 1, wherein the step of combining comprises: starting from the first statistical measured variable vector, clamping of circle segments in such a way that the number of measurement data within each circle segment is substantially the same;

Combining measurement data within a circle segment to a partial measurement data set.

10. The computer-implemented method according to claim 1, wherein the step of combining follows in such a way that a line from the first statistical measured variable vector is adjacent to a measurement datum to two lines, which lead from the first statistical measurement vector to a measurement date of the same part of the measurement data set; or

to a line leading from the first statistical measurement vector to a measurement date of the same sub-measurement data set and a line leading from the first statistical measurement vector to a measurement date of a neighboring sub-measurement data set.

11. A computer-implemented method according to one of the preceding claims, in which the specific measured data and in particular the associated measured variables are assigned to components which thereby contain a predefined characteristic.

12. Arrangement comprising:

at least one processor and a memory adapted to store measurement records;

one or more programs with instructions that when executed on the at least one processor, the United drive according to one of claims 1 to 10 perform.

A computer-readable storage medium comprising one or more programs configured to be executed by a processor and including instructions for the method of any one of the preceding claims.

14. Use of the method for processing measurement data in the production and / or development of electronic components, in particular for the preparation of measurement data of components of different wafers or wafer batches.