TWI762364B - Ingot evaluation method - Google Patents

Abstract

An ingot evaluation method is provided, which includes the following steps. A plurality of statistical parameters are obtained based on the defect information of the first wafer at the first end and the second wafer at the second end of each of a plurality of ingots. A first bow value of the first and a second bow value of the second end of each ingot are measured, thereby marking the ingot to one corresponding category. A classification model is established based on the statistical parameters of the ingot and its marked category.

Description

Ingot Evaluation Method

The present invention relates to an evaluation and inspection method, and in particular, to an ingot evaluation method.

The semiconductor manufacturing industry has very strict requirements on product quality. In order to improve product quality, wafers that do not meet the quality specifications will be picked out before packaging and will not be packaged. As the wafer process becomes more and more complex, traditional visual inspection has been difficult to meet the current inspection needs. Therefore, a machine vision inspection method has been developed. Using machine vision to inspect can reduce inspection costs, improve inspection speed and reduce wrong judgment. The quality of incoming ingots has a great impact on the quality of wafers obtained after the processing process. Therefore, how to filter out the ingots with poor quality before the ingot processing process will be one of the subjects in the art.

The present invention provides a method for evaluating a crystal ingot, which can predict the quality of a crystal ingot from a wafer obtained by processing the crystal ingot.

The ingot evaluation method of the present invention includes: based on each of the plurality of ingots Defect information of the first wafer at the first end and the second wafer at the second end to obtain a plurality of statistical parameters; measure the first curvature of the first end of each ingot and the second curvature of the second end, by This marks the ingot to one of a corresponding plurality of categories; and builds a classification model based on the statistical parameters of the ingot and the category to which it is marked.

In an embodiment of the present invention, the step of obtaining statistical parameters based on the defect information of the first wafer at the first end and the second wafer at the second end of each ingot respectively includes: in a specified direction, The first wafer and the second wafer are each divided into at least two regions; based on the defect information of the first wafer, a first defect ratio and a second defect ratio corresponding to the two regions divided by the first wafer are calculated; Based on the defect information of the second wafer, calculate the third defect ratio and the fourth defect ratio corresponding to the two regions divided by the second wafer; calculate the first difference based on the first defect ratio and the second defect ratio; Calculate a second difference based on the third defect ratio and the fourth defect ratio; calculate a third difference based on the first difference and the second difference; and combine the first difference, the second difference, and the third difference as a statistical parameter.

In an embodiment of the present invention, the first defect ratio and the second defect ratio are respectively the ratio of at least one defect type in the two regions divided by the first wafer, the third defect ratio and the fourth defect ratio are respectively. The defect ratios are respectively the ratios of at least one defect type in the two regions divided by the second wafer.

In an embodiment of the present invention, the defect information includes a plurality of defect coordinates, and each defect coordinate corresponds to a defect type, and the defect type includes threading screw dislocation (TSD) and basal plane dislocation ( basal plane dislocation, BPD).

In an embodiment of the present invention, the step of measuring the first curvature of the first end of each ingot and the second curvature of the second end includes: measuring the curvature of a plurality of wafers included in each ingot one by one; Each ingot is divided into a first end section, a second end section and a middle section, wherein the middle section is located between the first end section and the second end section; the wafers located in the first end section are calculated The average of the curvatures of the wafers is used as the first curvature; and the average of the curvatures of the wafers located in the second end section is calculated as the second curvature.

In an embodiment of the present invention, the categories include a first category, a second category, and a third category, and the step of marking the ingot to one of the corresponding categories includes: judging the first curvature and the third Whether the curvature is within the preset range; if both the first curvature and the second curvature are within the preset range, mark the corresponding ingot as the first category; if one of the first curvature and the second curvature is within the preset range, and the other one of the first curvature and the second curvature is not within the preset range, marking the corresponding ingot as the second category; and if neither the first curvature nor the second curvature is within the preset range Within the range, mark the corresponding ingot as the third category.

In an embodiment of the present invention, after establishing the classification model, the method further includes: measuring the defect information of the first wafer and the second wafer at the front and rear ends of the ingot to be tested to obtain statistical parameters; The parameters are input into the classification model to predict one of the categories corresponding to the ingot to be tested.

Based on the above, the wafers obtained through the processing of the ingot are subjected to the calculation of defect concentration to identify the distribution of defects on the wafer, and finally the corresponding classification model is established through the artificial intelligence classification algorithm. Accordingly, the classification model is used to Ingot quality identification.

110: Measuring instruments

120: Analysis device

200: Ingot

210: First end segment

220: Intermediate Section

230: Second end segment

A, B, C, D: area

H1~H5, T1~T5, W: Wafer

S305~S320: Steps of Ingot Evaluation Method

FIG. 1 is a block diagram of an analysis system according to an embodiment of the present invention.

2 is a schematic diagram of an ingot according to an embodiment of the present invention.

FIG. 3 is a flowchart of an ingot evaluation method according to an embodiment of the present invention.

4A and 4B are schematic diagrams of wafer partitioning according to an embodiment of the present invention.

FIG. 1 is a block diagram of an analysis system according to an embodiment of the present invention. Referring to FIG. 1 , the analysis system includes a measuring instrument 110 and an analysis device 120 . Data transmission between the measuring instrument 110 and the analysis device 120 can be performed, for example, through wired or wireless communication.

The measuring instrument 110 is, for example, an automated optical inspection (Automated Optical Inspection, AOI for short) instrument, but the present invention is not limited to this, and the measuring instrument 110 may be any instrument. AOI instrument is a high-speed and high-precision optical image detection system, including measurement lens technology, optical lighting technology, positioning measurement technology, electronic circuit testing technology, image processing technology and automation technology application, etc. It uses machine vision as the detection standard technology. The measuring instrument 110 uses an optical instrument to obtain the surface state of the finished product, and then uses computer image processing technology to detect defects such as foreign objects or abnormal patterns.

The analysis device 120 is an electronic device with a computing function, which can use a It can be realized by a personal computer, a notebook computer, a tablet computer, a smart phone, etc. or any device with computing functions, and the present invention is not limited to this. The analyzing device 120 receives defect information of a plurality of known wafers (ie, one or more defect coordinates with defects and the defect type corresponding to each defect coordinate) from the measuring instrument 110, thereby performing training to obtain a classification model , for the subsequent use of the measurement data of the wafers obtained by the processing of the ingot to be tested, to predict the quality of the ingot to be tested after the ingot is processed into a wafer, and then to determine whether the ingot to be tested meets the expectations, so as to determine the quality of the ingot to be tested. Ingot quality for identification. Wherein, the processing process may include cutting, grinding, polishing, etc., and the wafer may be etched before measuring the defect information, but the present invention is not limited thereto.

2 is a schematic diagram of an ingot according to an embodiment of the present invention. Referring to FIG. 2 , a plurality of wafers W can be obtained after the ingot 200 is processed. In this embodiment, the crystal ingot 200 includes a first end and a second end. For example, the first end of the crystal ingot 200 is the head end, and the second end of the crystal ingot 200 is the tail end. The first end of the ingot is regarded as the tail end, and the second end is regarded as the head end, or the left end of the ingot 200 is regarded as the first end, and the right end is regarded as the second end, or the right end is regarded as the first end, and the left end is regarded as the first end. For the second end, the present invention is not limited to this. In addition, the ingot 200 is divided into a first end section 210 , an intermediate section 220 and a second end section 230 . The middle section 220 is located between the first end section 210 and the second end section 230 . Also, the first end section 210 includes five wafers H1 - H5 , and the second end section 230 includes five wafers T1 - T5 . Here, the number of wafers included in the first end section 210 and the second end section 230 is only an example, and is not limited thereto. In other embodiments, the first end section 210 , the middle section 220 and the second end section 230 can also be set by dividing the ingot 200 into three equal parts, or as required The ingot 200 is divided into three parts.

FIG. 3 is a flowchart of an ingot evaluation method according to an embodiment of the present invention. Referring to FIG. 3 , in step S305 , a plurality of statistical parameters are obtained respectively based on defect information of the first wafer at the first end and the second wafer at the second end of the plurality of ingots.

Taking the ingot 200 of FIG. 2 as an example, the measuring instrument 110 is used to measure the wafer H1 (the first wafer) and the wafer T1 (the second wafer) at the head and tail ends (the first end and the second end). Optical inspection is used to detect whether the multiple coordinate positions included in the wafer H1 and the wafer T1 are defective, and record the coordinate positions with defects and the types of defects. Generally speaking, the types of defects output by AOI instruments include threading edge dislocation (TED), threading screw dislocation (TSD) and basal plane dislocation (BPD). After analysis, it can be known that TSD and BPD have a great influence on quality. Therefore, in this embodiment, TSD and BPD are used for statistical analysis.

Specifically, in a specified direction (such as a vertical direction or a horizontal direction), the wafer H1 and the wafer T1 are each divided into at least two regions, so as to calculate the proportions of various defect types in the two regions respectively, and then Obtain multiple statistical parameters. For example, FIGS. 4A and 4B are schematic diagrams of wafer partitioning according to an embodiment of the present invention. As shown in FIG. 4A , the wafer W is vertically divided into two regions A and B on the left and right by the center of the circle, so as to calculate the number of TSD plus BPD in the regions A and B respectively. In addition, as shown in FIG. 4B , the wafer W can also be divided into two upper and lower regions C and D in the horizontal direction by passing the center of the circle to calculate the TSD plus BPD in the regions C and D respectively. quantity.

For example, taking the ingot 200 of FIG. 2 as an example, the analysis device 120 calculates the first defect ratio of the two regions divided by the wafer H1 according to the defect information corresponding to the wafer H1 (including the coordinate position of the defect and the defect type) and the second defect ratio. The first defect ratio and the second defect ratio are respectively the ratios of the two defect types TSD plus BPD in the two regions divided by the wafer H1. And based on the first defect ratio and the second defect ratio, a first difference is calculated.

Assume that the number of coordinate positions determined to have TSD in the region A of the wafer H1 is HA _TSD , and the number of coordinate positions determined to have BPD is HA _BPD . Furthermore, it is assumed that the number of coordinate positions determined to have TSD in the region B of wafer H1 is HB _TSD , and the number of coordinate positions determined to have BPD is HB _BPD . Based on this, the first defect ratio R ₁ and the second defect ratio R ₂ are respectively: R ₁ =(HA _BPD +HA _TSD )/(HA _BPD +HA _TSD +HB _BPD +HB _TSD ), R ₂ =(HB _BPD ) +HB _TSD )/(HA _BPD +HA _TSD +HB _BPD +HB _TSD ), wherein the first difference H _ΛB is R ₁ -R ₂ .

Furthermore, the analysis device 120 calculates the third defect ratio and the fourth defect ratio corresponding to the two regions divided by the wafer T1 according to the defect information (including the coordinate position and the defect type) corresponding to the wafer T1 . The third defect ratio and the fourth defect ratio are respectively the ratios of the two defect types TSD plus BPD in the two regions divided by the wafer T1 . And, a second difference is calculated based on the third defect ratio and the fourth defect ratio.

For example, it is assumed that the number of coordinate positions determined to have TSD in the region A of wafer T1 is TA _TSD , and the number of coordinate positions determined to have BPD is TA _BPD . In addition, it is assumed that the number of coordinate positions determined to have TSD in the region B of wafer T1 is TB _TSD , and the number of coordinate positions determined to have BPD is TB _BPD . Then the third defect ratio R ₃ and the fourth defect ratio R ₄ are respectively R ₃ =(TA _BPD +TA _TSD )/(TA _BPD +TA _TSD +TB _BPD +TB _TSD ), R ₄ =(TB _BPD +TB _TSD ) )/(TA _BPD +TA _TSD +TB _BPD +TB _TSD ), wherein the second difference T _ΛB is R ₃ -R ₄ .

After that, a third difference is calculated based on the first difference H _ΛB and the second difference T _ΛB . For example, the third difference D _ΛB is |H _ΛB -T _ΛB |. The first difference value H _ΛB , the second difference value T _ΛB and the third difference value D _{ΛB are} used as statistical parameters.

If the partition as shown in FIG. 4B is selected, the calculation process of the statistical parameters can refer to the calculation process of the first difference H _ΛB , the second difference T _ΛB and the third difference D _ΛB , which will not be repeated here.

In addition, in other embodiments, statistical parameters may also be calculated for one, three or more defect types.

Next, in step 5310, the first curvature of the first end and the second curvature of the second end of each ingot are measured, thereby marking the ingot to one of the corresponding multiple types. Specifically, the curvature of one or more wafers W included in one or more ingots 200 is measured one by one using the measuring instrument 110 . For example, the curvature of the wafer H1 among the wafers H1 to H5 of the first end section 210 is taken as the first curvature, and the curvature of the wafer T1 of the wafers T1 to T5 of the second end section 230 is taken as the curvature as the second curvature. In addition, in other embodiments, the average value of the curvature of the wafers H1 ˜ H5 located in the first end section 210 is calculated as the first curvature, and the calculated The average value of the curvatures of the wafers T1 to T5 at the second end section 230 is used as the second curvature.

Next, it is determined whether the first degree of curvature and the second degree of curvature are within a preset range. If both the first degree of curvature and the second degree of curvature are within the predetermined range, the corresponding ingot is marked as the first category. If one of the first curvature and the second curvature is within the predetermined range, and the other of the first curvature and the second curvature is not within the predetermined range, the corresponding ingot is marked as the second category. If neither the first curvature nor the second curvature is within the preset range, the corresponding ingot is marked as the third category.

For example, the preset range is set to -35 μm~+10 μm. Ingots whose curvatures at both ends of the head and the tail (the first end and the second end) are within the predetermined range are marked as the first category. Ingots whose bending values at both the head and tail ends (the first end and the second end) are not within the predetermined range are marked as the third category. Ingots with only one of the bending values at the head and tail ends (the first end and the second end) within the preset range are marked as the second category.

After the statistical parameters of each crystal ingot and its corresponding category are obtained, in step S315, a classification model is established based on the statistical parameters of the crystal ingot and its marked category. For example, the statistical parameters included in the ingot are used as the input of the classification model, and the type of the ingot is used as the output, thereby training the classification model to adjust the parameters of the classification model.

After the classification model is established, in step S320, the classification model may be used to identify the quality of the ingot. That is, the statistical parameters obtained by measuring the defect information of both the first wafer and the second wafer at the front and rear ends of the ingot to be tested are input. Enter the classification model, and then predict the corresponding category of the ingot to be tested.

For example, Table 1 shows the training data. The class of the ingot is marked with the curvature of the head and tail ends (the first end and the second end), and the statistical parameters H _ΛB , T _ΛB and D _{ΛB are used} .

Here, a multi-class logistic regression (Multinomial Logistic Regression) algorithm can be used to establish a classification model. This embodiment uses three types, therefore, three types of extended logistic regression are used to reconstruct the classifier. The first classifier selects the first category as positive, and makes the second and third categories negative. The second classifier selects the second class as the positive class, and the first class and the third class as the negative class. The third classifier selects the third class as the positive class, and the first class and the second class as the negative class. Using these three classifiers, in the prediction stage, each classifier can obtain the probability of the current positive class according to the test sample. After that, select the highest calculated result The classifier of , its positive class can be used as the prediction result.

In addition, the collected data can also be divided into a training data set and a validation data set. Use the validation dataset to verify the accuracy of the classification model.

To sum up, the wafers obtained through the ingot processing process are used to calculate the concentration of defects to identify the distribution of defects on the wafers. Finally, the artificial intelligence classification algorithm is used to establish the corresponding classification model. Accordingly, the classification model is used to identify the quality of the ingot.

S305~S320: Steps of Ingot Evaluation Method

Claims (7)

An ingot evaluation method, comprising: obtaining a plurality of statistical parameters based on defect information of a first wafer at a first end and a second wafer at a second end respectively of a plurality of ingots; measuring each of the ingots a first curvature of the first end of the ingots and a second curvature of the second end thereof, thereby marking the ingot to one of a corresponding plurality of categories; and based on the ingots These statistical parameters and their labeled classes build a classification model. The ingot evaluation method according to claim 1, wherein the statistical parameters are obtained based on defect information of the first wafer at the first end and the second wafer at the second end respectively of the ingots The step includes: dividing the first wafer and the second wafer into at least two areas in a specified direction; a first defect ratio and a second defect ratio of the two divided regions; based on the defect information of the second wafer, calculating a third defect ratio corresponding to the two divided regions of the second wafer; and a fourth defect ratio; based on the first defect ratio and the second defect ratio, a first difference is calculated; based on the third defect ratio and the fourth defect ratio, a second difference is calculated; based on the first The difference and the second difference, a third difference is calculated; and the first difference, the second difference and the third difference are used as the statistical parameters number. The ingot evaluation method according to claim 2, wherein the first defect ratio and the second defect ratio are respectively the proportions of at least one defect type in the two regions divided by the first wafer, and the first defect ratio The three defect ratio and the fourth defect ratio are respectively the proportions of the at least one defect type in the two regions divided by the second wafer. The ingot evaluation method according to claim 1, wherein the defect information includes a plurality of defect coordinates, each of the defect coordinates corresponds to one of a plurality of defect types, and the defect types include threading screw dislocations , TSD) and basal plane dislocation (BPD). The ingot evaluation method of claim 1, wherein the step of measuring the first curvature of the first end and the second curvature of the second end of each of the ingots comprises: measuring each of the ingots one by one The curvature of a plurality of wafers included in the ingots; dividing each of the ingots into a first end section, a second end section and a middle section, wherein the middle section is located in the first end section between the end section and the second end section; calculating the average value of the curvature of the wafers located at the first end section as the first curvature; and calculating the curvature at the second end section The average of the curvatures of the wafers is used as the second curvature. The ingot evaluation method of claim 1, wherein the categories include a first category, a second category, and a third category, and the step of marking the ingot to one of the corresponding categories includes: : determine whether the first curvature and the second curvature are within a preset range; if both the first curvature and the second curvature are within the preset range, mark the corresponding ingot as the The first category; if one of the first curvature and the second curvature is within the preset range, and the other of the first curvature and the second curvature is not within the preset range, the corresponding The ingot is marked as the second category; and if neither the first curvature nor the second curvature is within the predetermined range, the corresponding ingot is marked as the third category. The ingot evaluation method according to claim 1, wherein after establishing the classification model, the method further comprises: measuring defect information of both the first wafer and the second wafer at the front and rear ends of an ingot to be tested. to obtain the statistical parameters; and input the statistical parameters into the classification model, so as to predict one of the categories corresponding to the ingot to be tested.

Images