CN1279599C - Defect detection parameter analysis method - Google Patents

Abstract

A defect inspection parameter analysis method is used for analyzing a plurality of batches of products with a batch number respectively, each batch of products is manufactured by a plurality of machines, one or more wafers in each batch of products are at least detected by a defect inspection item to generate a defect inspection parameter value, the defect inspection item and the parameter value thereof as well as a process station related to the defect inspection item are respectively stored in a database, the method comprises the following steps: searching a database to obtain defect detection parameter values of a plurality of batches of products; dividing a plurality of batches of products into a qualified product group and an unqualified product group according to the defect detection parameter values; searching the process station related to the defect detection item from the database; searching the machines through which the qualified product group passes in the process station; searching the machines passed by the unqualified product group in the process station; and judging the machine station with the unqualified product group passing probability higher than the qualified product group passing probability.

Description

Defect detection parameter analysis method

technical field

The invention relates to a process parameter analysis method, in particular to a defect detection parameter analysis method.

Background technique

In semiconductor manufacturing technology, to complete a semiconductor product, it usually has to go through many processes, such as lithography process, etching process, ion implantation process, etc.; that is to say, a large number of machines must be applied in the semiconductor manufacturing process, and Many cumbersome procedures. Therefore, those who are familiar with this technology are committed to ensuring the normal operation of the machine, maintaining or improving product yield, detecting and confirming problems, and machine maintenance, so as to make the production speed and quality of semiconductor products meet customer needs.

Generally speaking, to discuss the problems of semiconductor manufacturing process, the following data can be analyzed, including process parameter data, online quality test (In-line QC) data, defect inspection (defect inspection) data, sample test (sample test) data, wafer test (wafer test) data, and post-package test (final test) data. Among them, the defect detection data is to detect defects for each layer of the wafer, such as the total number of defects (total count), the number of defects added (adder count), or the number of defect categories (class count), so The resulting test values, which are usually represented by a defect distribution map.

In the prior art, as shown in FIG. 1, step 101 is first performed. At this time, those skilled in the art will test various defect detection items for each wafer, such as the inter-metal dielectric layer (inter-metal dielectric layer) Defect detection, etc.

Next, in step 102 , those skilled in the art observe the results of various defect detection items of each wafer, so as to find products with deviations in defect detection results. As shown in FIG. 2, a plurality of lattices (die) 21 will be cut in a wafer, which includes a plurality of black spots, representing the positions of defects 22 in a certain layer of the wafer, as shown in FIG. 2 A map showing the distribution of defects is displayed.

In step 103, a skilled person judges the process stations that may have problems based on experience and the defect distribution map of the abnormal product selected in step 102, such as the polysilicon layer formation process, the metal layer formation process, the inner metal dielectric layer formation process, etc.

Finally, in step 104 , those skilled in the art check each machine in the process station determined in step 103 to find out the abnormal machine. For example, a person skilled in the art can detect out-of-standards based on the total number of defects in the inner metal dielectric layer, determine that the problematic process station is the deposition process station of the inner metal dielectric layer, and check out abnormal machines, such as Deposition machines, etching machines, etc.

However, since the existing technology uses human experience judgment to determine the analysis results (step 103), the accuracy and credibility of the final analysis results will be questionable; The experience among engineers is not easy to pass on, and each engineer has limited ability and cannot take into account the operating status of all machines in the factory. Therefore, when the defect detection results of semiconductor products are abnormal, the engineers may not have enough experience to quickly and correctly It may take a lot of time to conduct relevant research to determine which link has a problem, and it may even make a wrong judgment. In this way, not only will the efficiency of the manufacturing process be reduced, the production cost will be increased, and the production situation will not be improved in time. Improve yield.

Therefore, how to provide an analysis method that can quickly and correctly determine which link is faulty when abnormality occurs in the defect detection data of semiconductor products is one of the important issues in the current semiconductor manufacturing technology.

Contents of the invention

In order to overcome the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a defect detection parameter analysis method that can quickly and accurately determine which link is faulty when abnormalities occur in the defect detection data of semiconductor products.

Another object of the present invention is to provide a defect detection parameter analysis method capable of correcting the kill ratio of defect detection according to the results of defect detection and wafer testing.

The feature of the present invention is to cooperate with a database that records various defect detection items and related process machines, and use the commonality analysis method to analyze the defect detection parameters.

In order to achieve the above object, the defect detection parameter analysis method according to the present invention is used to analyze multiple batches of products with a batch number respectively, each batch of products is made through multiple machines, and each wafer in each batch of products passes through at least one The defect detection item is detected to generate a defect detection parameter value, and the defect detection item and its parameter value and a process station related to the defect detection item are stored in a database. The method includes the following steps: searching the database To obtain the defect detection parameter values of multiple batches of products; divide multiple batches of products into a qualified product group and an unqualified product group according to the defect detection parameter values; search for process stations related to defect detection items from the database; search for qualified products The machines that the group passes through at the process station; search for the machines that the unqualified product group passes through at the process station; and the machines that judge that the passing probability of the unqualified product group is higher than the passing probability of the qualified product group.

In addition, each wafer in each batch of products is also inspected by a wafer test item related to the defect inspection item to generate a wafer test parameter value, and the wafer test item and the wafer test item are stored in the database. Test parameter values, and according to the defect detection parameter analysis method of the present invention, also use the overlay method to compare the wafer test parameter value distribution map and the defect distribution map, so as to find a better kill ratio for defect detection.

As mentioned above, because the defect detection parameter analysis method according to the present invention cooperates with the database that records various defect detection items and the process machines related to it and uses the commonality analysis method to analyze the defect detection parameters, it can be used in semiconductor products When the defect detection data of the system is abnormal, it can quickly and accurately determine which part of the problem is caused, and find out the abnormal machine. In addition, it can also correct the kill ratio of defect detection according to the results of defect detection and wafer testing (kill ratio), so it can effectively reduce the error of human judgment to improve the efficiency of the process, reduce production costs, and improve the online production situation in time to increase the yield rate.

Description of drawings

Fig. 1 shows the flowchart of the existing defect detection parameter analysis method;

Fig. 2 shows the defect detection parameter value distribution map of wafer;

Fig. 3 shows the flowchart of the defect detection parameter analysis method according to a preferred embodiment of the present invention;

FIG. 4 shows a flowchart of a defect detection parameter analysis method according to another preferred embodiment of the present invention; and

Figure 5 shows a distribution of wafer test parameter values for a wafer.

Explanation of symbols in the figure

101～104 The flow of the existing defect detection parameter analysis method

21 lattice

22 Defects

23 Defect lattice

301~311 The flow of the defect detection parameter analysis method in the preferred embodiment of the present invention

401~411 The flow of the defect detection parameter analysis method in another preferred embodiment of the present invention

51 Disqualification lattice

52 qualified lattice

Detailed ways

The defect detection parameter analysis method according to the preferred embodiment of the present invention will be described below with reference to the accompanying drawings, wherein the same components are represented by the same symbols.

As shown in Figure 3, the flow chart of the defect detection parameter analysis method of the preferred embodiment of the present invention is shown in the figure, which is used to quickly and correctly determine which machine is out of order when the defect detection data of semiconductor products is abnormal. problem.

First, step 301 searches a database to obtain defect detection parameter values of multiple batches of products. Among them, each batch (lot) of products has a lot number (lot number), and each batch of products includes 25 wafers, and each batch of products has gone through multiple machines with multiple processes, and one or more wafers in each batch of products of the wafers are inspected by at least one defect inspection item to generate a defect inspection parameter value. In this embodiment, the defect detection results can be divided into total count of defects, adder count of defects or class count of defects; and defect detection parameter values can be represented by a defect distribution diagram , taking the defect distribution diagram of the increased number of defects as an example, as shown in FIG. 2 , wherein a plurality of black dots distributed in a plurality of lattices 21 of the wafer represent a defect 22 respectively. It should be noted that a wafer may have defects in different layers, and at this time, a wafer may have more than one defect distribution map.

Next, step 302 displays the defect detection results of each batch of products in a graph. In this embodiment, this step uses a histogram to represent the defect detection parameter value of each batch of products, such as the total number of defects, the number of defect increases or the number of defect categories, so engineers can observe the histogram and understand the defect detection parameters distribution of values.

In step 303, the multiple batches of products obtained in step 301 are divided into a qualified product group and a non-conforming product group, and the criterion for the distinction is whether they meet the preset specifications of the defect detection parameter values. In this embodiment, this step sets several batches of products whose defect detection parameter values are within the preset specification range as Group A (qualified product group) products, for example including batch numbers 1, 2, 3, 4, and 5 (such as Shown in step 304); and set the batches of products whose defect detection parameter values are not within the preset specification range as Group B (unqualified product group) products, for example including batch numbers 6, 7, 8, 9, and 10 (such as shown in step 305).

Then, step 306 searches for the process station related to the layer of the analyzed defect detection item from an experience accumulation database; for example, if the layer of the analyzed defect detection item is the inner metal dielectric layer, then the relevant The process station may be a deposition process station, a lithography process station or an etching process station for the dielectric layer after the first metal layer. In this embodiment, the accumulated experience database includes experience accumulated by senior engineers in tracking problems in the past; in addition, the data deduced by the computer system according to the method of the present invention will also be stored in the database.

After step 306 searches the database for the process station related to the level of the analyzed defect detection item, step 307 shows that after the search in step 306, the item to be tracked is a certain process station.

Next, in step 308 , it is searched which machines are included in the tracked process stations, such as E1, E2, E3 . . . . Next, step 309 calculates the probabilities of the group B products passing through the other machines of the process station. In addition, step 310 calculates the probabilities of the group A products passing through the other machines of the process station. Finally, in step 311 , the commonality analysis method is used to find out the machines whose passing probability of the products of the group B is higher than that of the products of the group A. The machines with high passing probability of the Group B products obtained in step 311 are the machines that may have problems analyzed according to the defect detection parameter analysis method of the preferred embodiment of the present invention.

In addition, as shown in FIG. 4 , according to the flow chart of another preferred embodiment of the present invention, this implementation provides a method for correcting the defect quantity control standard by using defect distribution and wafer test results. In this embodiment, each wafer in each batch of products is also tested by a wafer test item to generate a wafer test parameter value, and the database also stores wafer test items and their parameter values, as well as defects. Correlation between detection items and wafer test items.

First, step 401 searches a database to obtain defect detection parameter values of multiple batches of products. As mentioned earlier, each batch of products has a batch number, and each batch of products includes 25 wafers, and one or more wafers in each batch of products have passed the defect inspection project, and each wafer will pass through the wafer The test items are detected to generate defect detection parameter values and wafer test parameter values. In this embodiment, the defect detection parameter value may be represented by a defect distribution map (as shown in FIG. 2 ), wherein a plurality of black dots distributed in a plurality of lattices 21 of the wafer represent a defect 22 respectively. , and the lattice with black spots is the defective lattice 23. It should be noted that a wafer may have defects in different layers, and at this time, a wafer may have more than one defect distribution map.

Next, in step 402, it is judged whether the defect detection parameter value of each batch of products acquired through step 401 exceeds a preset specification. Generally speaking, the preset specification of the defect detection parameter value can be within a certain range. This step judges whether the defect detection parameter value of each batch of products obtained exceeds the upper limit (UCL) of the preset specification. In addition, the analysis and judgment of this step The defect detection item may be the total number of defects, the number of increased defects, or the number of defect categories. In this embodiment, step 402 searches for each wafer of each batch of products, and if a batch of products contains more than one wafer with a defect exceeding a predetermined specification, then proceed to step 403 to pick out defective products Batch number, if not, stop analysis. Next, in step 404, the defect distribution map of the wafer with defects is found.

Then, step 405 judges whether the wafer test parameter value of the batch of products with the batch number obtained in step 403 is stored in the database. In this implementation, the wafer test parameter value can be represented by a wafer test parameter value distribution map, as shown in Figure 5, a wafer will be cut into multiple lattices, including multiple disqualified lattices 51 (shown in black) and a plurality of qualified lattices 52 (shown in white). At this time, if step 405 judges that there are wafer test parameter values stored in the database, proceed to step 406 to obtain the distribution map of wafer test parameter values of each wafer of the batch of products; if not, stop the analysis. It should be noted that the wafer test parameter values analyzed and searched in steps 405 and 406 are wafer test items related to defect detection items, such as function test items or power supply current test (IDDQ test) items.

Next, in step 407, the defect distribution map found in step 404 is compared with the wafer test parameter value distribution map obtained in step 406 by means of overlapping maps to obtain the overlapping lattice numbers of the two distribution maps, In order to calculate the ratio of the number of overlapping lattices to the number of disqualified lattices; in this step, the number of overlapping lattices is the overlapping quantity of these defect lattices and these disqualified lattices. Then, in step 408, it is judged whether the ratio is greater than or equal to a default value, such as 50%, if not, then skip this layer, and stop the analysis when all layers are skipped; if yes, go to step 409.

In step 409, the data such as product batch number, layer data and number of defects after the analysis of the above steps are sorted out. In this embodiment, in this step, the analyzed layer is first marked as a defective layer, and then the batch of products and their batch numbers including wafers having at least this defective layer are searched, so as to pick out the product batch number, layer classification Data and number of defects, etc.

In addition, step 410 will perform statistical analysis to obtain a representative value as the kill ratio of the number of defects in this level. At the same time, in step 411, according to the kill ratio of the number of defects, the defect detection parameter analysis method according to the preferred embodiment of the present invention can predict the yield rate of this product in the subsequent production of this layer of products .

To sum up, since the defect detection parameter analysis method according to the present invention cooperates to record various defect detection items and the database of the process machine related thereto, and uses the commonality analysis method to analyze the defect detection parameters, it can be used in semiconductor products When the defect detection data of the system is abnormal, it can quickly and accurately determine which part of the problem is caused, and find out the abnormal machine. In addition, it can also correct the kill ratio of defect detection according to the results of defect detection and wafer testing (kill ratio), so it can effectively reduce human judgment errors to improve process efficiency, reduce production costs, and improve online production in time to increase yield.

The foregoing are examples only, not limitations. Any equivalent modification or change without departing from the spirit and scope of the present invention shall be included in the claims of the present invention.

Claims (10)

1, a kind of defects detection parameters analysis method, in order to analyze the many batches of products that have a lot number respectively, these many batches of products are obtained through a plurality of boards, and a slice in the every batch of product or above wafer at least through the detection of a defects detection project to produce a defects detection parameter value, this defects detection project and a processing procedure station relevant with this defects detection project are not stored in the database, this database also stores this defects detection parameter value, it is characterized in that this method comprises:

Search this database to obtain the defects detection parameter value of these many batches of products;

According to this defects detection parameter value should many batches of product differentiations be a qualified products group and a substandard product group;

It is other to search this processing procedure station relevant with this defects detection project from this database;

Search the board of this qualified products group other process at this processing procedure station;

Search the board of this substandard product group other process at this processing procedure station; And

Judge that this substandard product group is higher than the board of this qualified products group through probability through probability.

2, defects detection parameters analysis method as claimed in claim 1 is characterized in that, utilizes intercommunity to analyze gimmick and judges that this substandard product group is higher than the board of this qualified products group through probability through probability.

3, defects detection parameters analysis method as claimed in claim 1 is characterized in that, this defects detection parameter value be a defective total quantity, defective accelerate and defective categorical measure one of them.

4, defects detection parameters analysis method as claimed in claim 3 is characterized in that, more comprises:

Utilize this defects detection parameter value of these many batches of products of histogram graph representation.

5, defects detection parameters analysis method as claimed in claim 1, it is characterized in that, each wafer in the every batch of product also through the detection of a wafer sort project relevant with this defects detection project to produce a wafer test parameters value, this database also stores this wafer sort project and this wafer test parameters value, and this defects detection parameters analysis method more comprises:

Whether the defects detection parameter value of judging these many batches of products is greater than one first standard value;

When the defects detection parameter value of judging these many batches of products greater than this first standard value, obtain the lot number of product with defective;

In this batch product, obtain on each wafer at defective place a plurality of layers not, wherein each layer do not have a defect map respectively, this defect map has a plurality of defect lattices;

Search has the wafer test parameters value distribution map of each wafer of this batch product of defective, and this wafer test parameters value distribution map has a plurality of lattices that overstep the bounds;

Other defect map of each layer and this wafer test parameters value distribution map are folded the figure action, and to obtain an overlapping lattice number, this overlapping lattice number is these defect lattices and the overlapping quantity of these lattices that overstep the bounds;

Calculate the ratio of the quantity of this overlapping lattice number and this lattice that oversteps the bounds;

Judge that whether this ratio is more than or equal to one second standard value;

When judging this ratio less than this second standard value, it is other to skip over this layer;

When judging this ratio, should not be denoted as a defect layer by layer more than or equal to this second standard value; And

Search includes this batch product and the lot number thereof of the wafer that has this defect layer at least.

6, defects detection parameters analysis method as claimed in claim 5 is characterized in that, more comprises:

Defective number according to this defect layer produces the 3rd standard value that a defective as this defect layer is closed down ratio in the statistical analysis mode.

7, defects detection parameters analysis method as claimed in claim 6 is characterized in that, more comprises:

According to the 3rd standard value, prediction proceeds to the yield of this batch product of this defect layer in successive process.

8, defects detection parameters analysis method as claimed in claim 5 is characterized in that, more comprises:

Judge and whether store this wafer test parameters value in this database;

When in judging this database, not storing this wafer test parameters value, stop seek actions; And

When in judging this database, storing this wafer test parameters value, obtain the wafer test parameters value distribution map of each wafer of this batch product with defective.

9, defects detection parameters analysis method as claimed in claim 5 is characterized in that, this wafer sort project is the functional test project.

10, defects detection parameters analysis method as claimed in claim 5 is characterized in that, this wafer sort project is a power supply supply of current test event.

Images