TWI801921B - Yield correlation-driven in-line tool matching management method - Google Patents

Abstract

A yield correlation-driven in-line tool matching management method is provided. The method includes collecting process parameters of plural tools and yields corresponding to the plural tools; creating a process parameter probability plot based on the process parameters of plural tools to calculate an in-line QC index based on a matching criteria; creating a yield probability plot based on the yields of the plural tools to calculate an in-line yield index based on the matching criteria; creating a process parameter-to-yield correlation plot according to the in-line QC index and the in-line yield index; determining a risk tool affecting the yield based on the process parameter-to-yield correlation plot; and performing a tool adjustment or margin check based on a preset condition.

Description

Yield-oriented online machine matching management method

The present invention relates to a machine matching management method, and in particular to a yield-oriented online machine matching management method.

Generally speaking, a semiconductor process requires hundreds of processes from wafer to product, such as thin film, deposition, chemical polishing (CMP), implantation, yellow light (exposure, etching) and so on. In addition, each process procedure may include several machines. For example, in the process of lithography etching, the measured critical dimension (CD) of each machine is different after exposure. In addition, at the beginning, there is not much difference between each machine of each process, but after a long period of mass production, each machine will have variability, which will affect the overall yield.

Customers will emphasize the importance of online machine matching according to their respective needs. That is to say, it is necessary to adjust the parameters of each machine so that the allocation can match as much as possible, so that the overall yield rate can be maintained at the required level. In this matching process, a box plot (box plot) is a commonly used method, but the box plot can only use visual or statistical verification to compare the differences in the online average (or median) of each production machine. However, the distribution characteristics of other machine differences (machine differences) cannot be compared and judged correctly by visual inspection. Therefore, if the simple Observe the box diagram, because the box diagram is only based on the average or median, so the visual results from the box diagram will suggest that the differences between machines are not large.

FIG. 1A is a box diagram showing the distribution of thickness values of each machine in the aluminum deposition process. As shown in Figure 1A, the box-shaped areas where the aluminum thickness (process parameters) of machine 1 to machine 4 are distributed are all about 3600; that is to say, there is no difference in the median of each machine. Therefore, if one looks at Figure 1A, it is possible to obtain that the machine differences between machine 1 and machine 4 are not large. In addition, from Figure 1A, it can be seen at most that the maximum value of machine 2 exceeds the upper limit by a little bit, but it is still unclear whether this situation will cause machine error and affect the yield rate.

FIG. 1B shows a probability diagram of the example in FIG. 1A . As shown in FIG. 1B , the horizontal axis is the aluminum thickness, and the vertical axis is the cumulative probability value (percentage). From the probability diagram in Figure 1B, it can be further understood that in this thickness distribution, there is a large machine error at the end of the curve. Therefore, the distribution of machine error can be further understood from the probability diagram. However, even so, the operator still does not know how this machine difference affects the yield rate of the overall process, and it is difficult to know how to match (adjust) between machines.

Therefore, the current monitoring mechanism lacks the function of two-way confirmation of the correlation between the process data and product test data of the production machine differential line specification. In other words, although the variability of each machine in each process can be monitored online, the current monitoring mechanism does not include yield as a monitoring factor. Therefore, even though online monitoring can control the variability (machine error) of each machine in a process program, it still cannot improve the yield rate in the end, that is, there is a lack of correlation between the yield rate and process parameters for two-way confirmation.

According to an embodiment of the present invention, a yield-oriented online machine matching management method is provided, which is executed by a processor and adjusts multiple machines in a process program. The online machine matching management method includes: collecting the measured values of the process parameters of the multiple machines, and the yield rate corresponding to each of the multiple machines; based on the process parameters of the multiple machines , make a process parameter probability map, and calculate the online quality control index based on the preset judgment criterion; make a yield probability map based on the yield rate of the plurality of machines, and calculate the online Yield index; according to the online quality control index and the online yield index, make a yield-process parameter correlation diagram; according to the yield-process parameter correlation diagram, it is determined that all machines will be affected risk machines with the above-mentioned yield rate; and based on preset conditions, perform a machine adjustment program or a process boundary adjustment program on the risk machines. The preset judgment criterion is to set a plurality of probability values in the process parameter probability map or the yield probability map, and calculate a plurality of maximum values among the plurality of machines corresponding to each of the plurality of probability values. process parameter differences or multiple maximum yield rate differences, and calculate the weighted average of the multiple maximum process parameter differences or multiple maximum yield rate differences in a statistical manner.

According to another embodiment of the present invention, a yield-oriented online machine matching management method is provided, which is executed by a processor. The online machine matching management method is suitable for an architecture composed of multiple process programs, and each of the multiple A process procedure includes a plurality of machines, and adjustments are made to the plurality of machines in each of the plurality of process procedures. The online machine matching management method includes: collecting the measured values of the process parameters of the multiple machines in each of the multiple manufacturing processes, and the yield rate corresponding to each of the multiple machines; Based on the process parameters of the plurality of machines, making a process a parameter probability map, and calculate an online quality control index based on a preset judgment criterion; based on the yield rates of the plurality of machines, make a yield probability map, and calculate an online yield index based on the preset judgment criterion ; According to the online quality control index and the online yield index, make a yield-process parameter correlation diagram; according to the yield-process parameter correlation diagram, determine that the yield rate will be affected in the plurality of machines risk machines; based on preset conditions, perform a machine adjustment program or a process boundary adjustment program on the risk machines; An indicator, finding at least one yield-critical process in the plurality of process procedures; and determining an optimal yield-rate machine path based on a combination of the plurality of machines in the at least one yield-critical process. The preset judgment criterion is to set a plurality of probability values in the process parameter probability map or the yield probability map, and calculate a plurality of maximum values among the plurality of machines corresponding to each of the plurality of probability values. process parameter differences or multiple maximum yield rate differences, and calculate the weighted average of the multiple maximum process parameter differences or multiple maximum yield rate differences in a statistical manner.

According to an embodiment of the present invention, in the above online machine matching management method, the online quality control index and online yield index are calculated based on the following formula:

Wherein α _i is the weight, △ _i is the maximum process parameter difference or the maximum yield rate difference corresponding to each of the multiple probability values, k σ is the full distribution range of the process parameter difference or the yield rate, σ is the standard deviation, k is a constant, and is the number of the multiple probability values.

According to an embodiment of the present invention, in the above online machine matching management method, when the online quality control index is greater than the first preset value and the online yield index is greater than When the second preset value, or when the online quality control index is less than the first preset value and the online yield index is greater than the second preset value, execute the machine adjustment procedure. The first preset value is the same as or different from the second preset value.

According to an embodiment of the present invention, in the above online machine matching management method, when the online quality control index is smaller than a first preset value and the online yield index is smaller than a second preset value, the process boundary adjustment is performed program. The first preset value is the same as or different from the second preset value.

According to an embodiment of the present invention, in the above-mentioned online machine matching management method, when the online yield index is less than the second preset value, before executing the process boundary adjustment program, further includes: according to the preset Setting a judgment criterion, calculating a test item index; judging whether the test item index is greater than the second preset value; and executing the machine adjustment program when the test item index is greater than the second preset value , when the test item index is greater than the second preset value, execute a process boundary adjustment program.

According to an embodiment of the present invention, in the above online machine matching management method, the machine adjustment procedure further includes: classifying the yield rate by risk process machines; according to the process parameter probability map and the yield rate a probability map to determine the risk machine; generate the yield-process parameter correlation map; and adjust the range of the process parameter of the risk machine according to the yield-process parameter correlation map.

According to an embodiment of the present invention, in the above online machine matching management method, the process boundary adjustment program further includes: classifying the yield rate by each of the machines in the process program; generating each of the machines The yield rate-process parameter correlation diagram; with and adjusting the range of the process parameter of the machine according to the yield rate-process parameter correlation diagram.

According to an embodiment of the present invention, in the above online machine matching management method, the process boundary adjustment procedure is to tighten the range of the process parameters, or change the center line of the range of the process parameters.

According to an embodiment of the present invention, in the above online machine matching management method, the preset condition is determined based on the yield rate.

According to an embodiment of the present invention, in the above-mentioned online machine matching management method, determining the best yield machine path is based on the yield probability map among the machines in the at least one yield-critical process , or use association rules.

Based on the above, according to the embodiment of the present invention, the monitoring mechanism not only considers the distribution difference of process parameters, but also considers the yield difference between machines. Therefore, through the two-way confirmation mechanism of yield rate and process parameters, the matching between machines can be more perfect and accurate, and the overall yield rate can be improved.

10_1, 10_2, ..., 10_k: process parameter probability map

20_1, 20_2, ..., 20_k: Yield probability map

CTY: Yield critical process

Yt_1, Yt_2, ..., Yt_N: Yield rate

Wf_1, Wf_2, ..., Wf_N: Wafers

Tool1_1~Tool1_n1: machine

Tool2_1~Tool2_n2: machine

Toolk_1~Tool1_nk: machine

S10~S22: Process steps of the online machine matching management method

S100~S230: Process steps of online machine matching management method

S100a~S100c: each step of calculating △QC

S102a~S102c: each step of calculating ΔYt

S110a~S110c: each step of calculating △Bin

S300~S306: Machine adjustment procedure

S400~S406: process boundary adjustment procedure

S500~S504: Determine the best yield rate machine path program

FIG. 1A is a box diagram showing the distribution of thickness values of each machine in the aluminum deposition process.

FIG. 1B shows a probability diagram of the example in FIG. 1A .

FIG. 2 is a system schematic diagram of a yield-oriented online machine matching management method according to the present invention.

FIG. 3 is a schematic flowchart of an online machine matching management method according to the present invention.

FIG. 4 is an explanatory diagram of a machine error judgment criterion in an embodiment of the present invention.

FIG. 5A and FIG. 5B show the situation where the data distribution and the probability map distribution of machine differences have a minimum value.

FIG. 6A and FIG. 6B show the situation where the data distribution and the probability map distribution of machine differences have a maximum value.

FIG. 7A and FIG. 7B show the situation that the data distribution and the probability map distribution of machine differences have offset distributions.

FIG. 8A and FIG. 8B show the situation that the data distribution and the probability map distribution of machine differences have a bimodal distribution.

FIG. 9 is a schematic flowchart of a detailed flow chart of an online machine matching management method according to the present invention.

FIG. 10A is a schematic flow chart of calculating _ΔQC .

FIG. 10B is a schematic flow chart of calculating _ΔYt .

FIG. 10C is a schematic flow chart of calculating _ΔBin .

FIG. 11 shows a schematic flow chart of machine adjustment.

12A to 12C are explanatory diagrams showing examples of machine adjustment.

13A to 13D are explanatory diagrams illustrating another example of machine adjustment.

FIG. 14A is a schematic flow chart of process boundary adjustment.

14B and 14C illustrate two examples of process boundary adjustment.

FIG. 15 is a schematic flow chart of determining the best yield machine path.

FIG. 16 shows an example of using the probability map to determine the best yield machine path.

FIG. 2 is a system schematic diagram of a yield-oriented online machine matching management method according to the present invention. FIG. 2 basically exemplarily draws a schematic diagram from the wafer as the raw material to the final product after entering various manufacturing processes.

As shown in FIG. 2 , from the raw material wafer (input side) Wf1~Wf_N to the product, there will be multiple process procedures, such as process procedure 1, process procedure 2, . . . , process procedure k (k is an integer). In addition, in each process procedure 1~k, one or more machines may be configured. As shown in the example in Figure 2, process program 1 includes machines Tool 1_1, Tool 1_2, Tool 1_3, ..., Tool 1_n1; process program 2 includes machines Tool 2_1, Tool 2_2, ..., Tool 2_n1; process Program k includes machines Tool k_1, Tool k_2, . . . , Tool k_nk, etc.

In addition, the process procedures 1~k are, for example, deposition, implantation, etching, exposure, etc., the steps required for each process procedure from the raw wafer to the product, and the machines referred to here are those corresponding to each process procedure. required machine. Each process procedure can include one or more machines, the number of which can be set according to requirements. After the wafers (input side) Wf1˜Wf_N go through a complete process procedure (k channels in this embodiment), the corresponding yields (yield) Yt1_1, Yt1_2, . . . , Yt1_n can be calculated on the product side. Yield rate can use any method that can be used in the semiconductor manufacturing process, such as WAT (Wafer Acceptance Test), WT (Wafer Test), CP (chip probing) and other quality-related monitoring items.

According to the embodiment of the present invention, after each process procedure 1~k is finished, the in-line QC index (In-line QC index) Δ _QC corresponding to the process is calculated. For example, in process procedure 1, online process parameters are measured for each machine and converted into a probability plot. Generally speaking, in the process procedure 1, there are equipped machines that can generate corresponding probability maps. When these probability maps are plotted on the same probability map, the online quality control index (QC index) △ _QC1 corresponding to this process parameter can be calculated. In addition, in the example shown in Figure 2, there are process procedures 1~k, so the corresponding online quality control indicators △ _QC1 ~△ _QCk can be calculated. And the online yield index △ _Yt1 ~ △ _Ytk . Finally, the critical to yield process can be obtained through the judging criteria of the embodiments of the present invention.

FIG. 3 is a schematic flowchart of an online machine matching management method according to the present invention. The method shown in FIG. 3 can be executed by a processor of a computer, a server, or a terminal device. For each process i (i=1~k) shown in Figure 2, measure the process parameters and corresponding yield of each machine of the process i Procedures 2~k are handled in the same way.

First, in step S10, the system will collect the measured values of the process parameters of multiple machines Tool 1_1, Tool 1_2, Tool 1_3, ..., Tool 1_n1 as shown in Figure 2, and the corresponding Tool Yields Yt1_1, Yt1_2, ..., Yt1_n1 of 1_1, Tool 1_2, ..., Tool 1_n1. This collection means that the process parameters of each process can be measured immediately after each machine in the process is finished. These measurement results can be sent to computers, servers or terminal devices for control through the network.

Next, in step S12, based on the process parameters of multiple machines Tool 1_1, ..., Tool 1_n1, a process parameter probability map (such as QC data 10_1, 10_2, ..., 10_k in Figure 2) is produced, and based on the following The default judgment criteria are used to calculate the online quality control index △ _QC1 . In addition, in step S14, based on the yield rates Yt1_1, ..., Yt1_n1 of multiple machines Tool 1_1, ..., Tool 1_n1, a yield rate probability map (as shown in Figure 2 Yt data 20_1, 20_2, ... , 20_k), and based on the preset judgment criteria described later, calculate the online yield index ΔYt ₁ . The calculation method and judgment criteria of the online quality control index △ _QC1 and the online yield index △Yt ₁ will be described in detail later.

Next, in step S16, according to the online quality control index _ΔQC1 and the online yield index _ΔYt1 , a yield rate-process parameter correlation diagram is created. In step S18 , according to the yield rate-process parameter correlation diagram, the risk tool that will affect the yield rate among the plurality of tools Tool 1_1 , . . . , Tool 1_n1 is determined. In step S20 , based on a preset condition, such as a required yield value, a tool adjustment procedure or a tool margin adjustment procedure is performed on the risky machine. The machine adjustment procedure or process boundary adjustment procedure will be described in detail later.

In addition, according to the above procedure, in step S22 , a critical to yield (CYT) process can be determined from the yield-process parameter correlation graph, that is, a process that has a critical impact on the overall yield. Therefore, according to the embodiment of the present invention, the yield-critical process can be determined from hundreds of process procedures, so as to determine the optimal yield machine path (golden path). Similarly, the method of determining the path of the machine with the best yield rate will be described in detail later.

The management flow described above is basically the basic processing flow of the embodiment of the present invention, and then the detailed flow of the embodiment of the present invention will be further described in detail. The judgment criteria used to calculate the online quality control index _ΔQC and the online yield index ΔYt in the embodiment of the present invention will be described in detail below.

FIG. 4 is an explanatory diagram of a machine error judgment criterion in an embodiment of the present invention. The judging criterion in this embodiment is the multi-slices matching criterion (MSMC) on the probability map, which will be described in detail below using FIG. 4 . In addition, two curves are used for illustration in FIG. 4 , and each curve represents a graph of data value (process parameter) and cumulative probability (percentage) of a machine. Therefore, Fig. 4 shows the probability diagrams of two machines in one process, that is, in the same process, the difference in the probability distribution of different machines for the same process parameter.

According to the embodiment of the present invention, as shown in FIG. 4 , the positions of multiple cut segments can be obtained between two curves, and the difference _Δi is calculated at the positions of each cut segment. Here, in FIG. 4 , the X-coordinate axis represents the data value of the process parameter and the Y-coordinate represents the cumulative probability value. The acquisition position of multiple cut segments is basically to select the probability value (Y value) that needs to be considered, and calculate the difference (X value) of the data value between machines under this probability value. In this example, the difference △ ₁ ~ △ ₇ of 7 data values is calculated by taking 7 segments as an illustration example. In this example, the probability values (percentages) of the seven cut segments are about 0.135, 2.28, 15.87, 5, 84.13, 97.72, and 99.87, and the differences of the corresponding data values are △ ₁ ~ △ ₇ . Next, a weight α _i is assigned to the difference of each data value. After that, use the following formula to calculate the online quality control index △ _QC .

，其中

,in

Among them, σ _p is the standard deviation, and α _i is the weight of the difference of each data value.

The standard deviation σ _p can be calculated in any statistical manner, and the present invention is not particularly limited. In addition, kσ in Figure 4 represents the full distribution range, and generally plus or minus 6σ can be used as the entire distribution range, that is, the above-mentioned constant k can be set to 12. Of course, k can also adopt other values, which are set according to actual operation requirements.

In addition, regarding the weight α _i , it can be set according to actual operation requirements. For example, if the importance of each selected probability value segment is the same, then α ₁ to α ₇ in the above example can be set to 1/7 respectively. In addition, if it is necessary to specially consider the machine differences between the probability values 99.87 and 0.135, the weight α ₁ and weight α ₇ of the corresponding difference can be increased. Therefore, the setting of each weight can be set according to actual needs.

In addition, the embodiments of the present invention do not specifically limit the selection of the segment of the probability value. In addition, the number of cuts can be set according to actual needs. How many cuts should be taken to calculate the online quality control index △ _QC .

In addition, in the example in Figure 4, two curves (two machines) are used as an illustration, but if there are multiple machines in the process procedure, the corresponding number of curves will appear on the probability map drawn at this time. curve. In this case, the above data value difference is taken as _Δi which has the largest difference (ie the largest process parameter difference) under the selected probability value.

Through the calculation of the online quality control index △ _QC in the above-mentioned embodiment of the present invention, it is possible to take a plurality of segments in the entire probability distribution to perform the difference of the data values of each segment, so that the differences between machines can be considered more effectively. variation between. For example, in the known method, usually only the part of the difference _Δ4 in Fig. 4 is considered, which is equivalent to the median part of the box diagram shown in Fig. 1A above. However, the present invention further takes into account the factors of larger variation (machine error) with larger differences in the curves, such as the tail end part in FIG. 4 . Therefore, it can be managed more effectively when performing online machine matching management (adjustment of process parameters, etc.).

In addition, in the above description example, the online quality control index △ _QC of the machine in each process procedure is taken as an example. However, the embodiment of the present invention further considers the yield rate. Therefore, after the yield measurement shown in Figure 2, the related yield probability map will also be drawn. At this time, for this yield probability map, the same calculation of the online yield index △ _Yt will also be performed. The calculation method of the online yield index △ _Yt is basically the same as that of the online quality control index △ _QC , so I won’t go into details here.

According to the embodiment of the present invention, the key yield correlation can be calculated according to the online yield index _ΔYt and the online quality control index _ΔQC , and then the process parameters of the machines most affecting the yield can be adjusted. This section will be described in detail next.

Next, use Figures 5A, 5B ~ Figures 8A, 8B to illustrate the distribution differences (machine differences) of several machine data values (process parameter values) and the examples of the corresponding probability diagrams, that is, to illustrate some data values (process parameter values) An example of a distribution with outliers. In the following examples, two machines are used as an example to illustrate, but it is not limited thereto. FIG. 5A and FIG. 5B show the situation where the data distribution and the probability map distribution of machine differences have a minimum value. FIG. 5A is a diagram showing the data distribution of machine 1 and machine 2 . From Figure 5A, the data distributions of machine 1 and machine 2 between data values (process parameters) of about 42~60 are almost overlapping, but machine 2 (difference machine) is between data values 18~24 There is a distribution in between. It can be seen from this that among machines 1 and 2, machine 2 still has a part of the minimum value distribution. This minimum value distribution is the difference between machine 1 and machine 2.

In addition, FIG. 5B shows the distribution of machine 1 and machine 2 after the data values (process parameters) are converted into a probability map, where the horizontal axis is the data value (process parameter) and the vertical axis is the probability value. It can also be clearly seen from Figure 5B that machine 1 and machine 2 overlap in a section of the distribution of the probability map (the data value is between 42 and 60), but in the minimum value part (the data value is 18~ 24) and the above-mentioned overlapping area are offset, that is, there is a machine difference Δ ₁ .

6A and 6B show the situation where the data distribution and probability map distribution of machine differences have a maximum value. FIG. 6A is a diagram showing the data distribution of machine 1 and machine 2 . From Figure 6A, the data distribution of machine 1 and machine 2 between the data value (process parameter) of about 40~60 is almost overlapping, but the data value of machine 2 (difference machine) is between before and after the data value of 70 There is a distribution. It can be seen from this that among machine 1 and machine 2, there is still a part of the maximum value distribution in machine 2. This maximum value distribution is the difference between machine 1 and machine 2.

In addition, FIG. 6B shows the distribution of machine 1 and machine 2 after the data values (process parameters) are converted into a probability map, where the horizontal axis is the data value (process parameter) and the vertical axis is the probability value. It can also be clearly seen from Figure 6B that machine 1 and machine 2 have a section of overlap in the distribution of the probability map (the data value is between 40 and 60), but in the part of the maximum value (the data value is around 70) ) is offset from the above overlapping area, that is, there is a machine difference △ ₁ .

FIG. 7A and FIG. 7B show the situation that the data distribution and the probability map distribution of machine differences have offset distributions. FIG. 7A is a diagram showing the data distribution of machine 1 and machine 2 . From FIG. 7A , the distribution of data values (process parameters) of machine 1 is approximately between 42 and 58, while the distribution of data values of machine 2 is approximately between 49 and 61. Machine 1 and machine 2 There is a shift between the measured value distributions, that is, the data distribution pattern is shifted to the right as a whole.

In addition, FIG. 7B shows the distribution of machine 1 and machine 2 after the data values (process parameters) are converted into a probability map, where the horizontal axis is the data value (process parameter) and the vertical axis is the cumulative probability value. It can also be clearly seen from Fig. 7B that machine 1 and machine 2 are two almost parallel lines on the probability diagram, which reflects the general situation shown in Fig. 7A, where the data distribution is the entire translation. Therefore, in this case, there is an obvious machine difference between machine 1 and machine 2 between the data values of about 42~60. At this point, the MSMC judgment criterion illustrated in Figure 4 above can be used to calculate the data value difference (such as △ ₁ , △ ₂ , △ ₃ , etc.) Quality control index △ _QC .

FIG. 8A and FIG. 8B show the situation that the data distribution and the probability map distribution of machine differences have a bimodal distribution. . FIG. 8A shows the obtained data distribution diagram of Machine 1 and Machine 2 . From Fig. 8A, compared with the data distribution of machine 1, the data value of machine 2 almost presents two peak distribution states, and the data values are about 12-50 and 50-84. In addition, FIG. 8B shows the distribution of machine 1 and machine 2 after the data values (process parameters) are converted into a probability map, where the horizontal axis is the data value (process parameter) and the vertical axis is the cumulative probability value. It can also be seen from Figure 8B that machine 1 and machine 2 will intersect together in the probability diagram, and then there is an offset between the front and back of the data data, that is, the machine difference (such as △ ₁ , △ _2, etc.) is large Partially occurs in the header and trailer.

It can be seen from the above several examples in Fig. 5A, Fig. 5B to Fig. 8A, Fig. 8B that according to the data obtained from the quality control audit of the online process parameters of each machine in each process, it can be drawn in the The distribution status of the process parameters (that is, the data values in the above example) of each machine in the process. Afterwards, the data distribution obtained online can be converted into a percentage line diagram (probability diagram) of the process parameters of each machine. Through the converted probability map, the online quality control index △ _QC and the online yield index △ _Yt between machines in each process can be obtained through the MSMC judgment criterion illustrated in Figure 4 above.

In this embodiment, each machine of each process is a probability map of performing the above-mentioned process parameters when operating online. In addition, after all the manufacturing processes, the yield rate of the product will also have a corresponding probability map.

FIG. 9 is a schematic flowchart of the detailed flow of the online machine matching management method according to the present invention, which is a further description of the flow shown in FIG. 3 . This online machine matching management method can be implemented in a system constructed by various available devices such as servers, computers, tablets, and terminal devices. In addition, these devices can realize online process parameter measurement control, data acquisition, machine parameter adjustment and other operations in real time through the internal network connected to each machine of each process procedure.

As shown in Fig. 9 and Fig. 2, firstly, in step S100, in each process procedure 1~k, the online quality control indicators _ΔQC1 ~△ _QCk between each machine are calculated. Here, the calculation method of the online quality control index △ _QC can refer to the schematic flowchart shown in FIG. 10A . As shown in FIG. 10A, in step S100a, the data of the process parameters are classified by machine. That is, as shown in FIG. 5A , FIG. 6A , FIG. 7A , and FIG. 8A , the distribution of process parameters is calculated according to the machines of each process procedure. Next, in step S100b, a process parameter probability map is generated, that is, the distribution percentage of the above-mentioned obtained process parameters in each machine is drawn, as shown in Figure 5B, Figure 6B, Figure 7B, Figure 8B, etc., to generate the distribution of each machine Probability plot of process parameters versus probability values. Afterwards, in step S100c, the online quality control index Δ _QC is calculated by using the MSMC judgment criterion.

In step S102, in each process procedure 1-k, the online yield index Δ _Yt1 -Δ _Ytk of each process procedure 1-k is calculated. Here, the calculation method of the online quality control index _ΔYt can refer to the schematic flowchart shown in FIG. 10B . As shown in FIG. 10B , in step S102 a , the yield data is classified by machine. Next, in step S102b, a yield probability map is generated. Afterwards, in step S102c, the online yield index Δ _Yt is calculated by using the MSMC judgment criterion.

In step S104, it is judged whether the online quality control index _ΔQC is greater than a first preset value p (for example, it may be 12% in the example shown in FIG. 4 ). Through the first preset value p, it can be judged whether there is a machine difference in the process parameters among the machines in the process procedure. If the online quality control index _ΔQC is smaller than the first preset value p, it means that there is no or very little difference in the process parameters among the machines in the process procedure. At this time, step S106 will be further executed to determine whether there is a difference in the yield rate.

Conversely, if the online quality control index _ΔQC is greater than the first preset value p, it indicates that there are differences in process parameters (machine differences) among the machines in the process procedure. In this case, step S108 is further executed to determine whether there is a difference in the yield (basically equivalent to step S106).

In step S106, it is judged whether the online yield index △ _Yt is greater than the first preset value q, and through the second preset value q, it can be judged whether there is machine difference in the yield rate among the machines in the process procedure . Here, the first preset value p and the first preset value q can be set according to needs, and both can be set to the same or different values. In addition, according to whether the online quality control index △ _QC is greater than the first preset value p and whether the online yield rate index △ _Yt is greater than the first preset value q, the quality control measurement (△ _QC ) of the process parameters can be compared with the yield rate The measurement (△ _Yt ) is correlated.

In addition, when the online quality control index △ _QC is greater than the first preset value p and the online yield index △ _Yt is greater than the second preset value q, it means that there are obvious differences in the process parameters and yield between machines. Or when the online quality control index △ _QC is less than the first preset value p and the online yield rate index △ _Yt is greater than the second preset value q, it means that there is no or no obvious difference in the process parameters between machines, but there are chances for the yield rate Difference. In these two cases, the tool adjustment program (sub-process) of S200 will be further executed, which will perform tool adjustment on risk tools with large machine differences according to the yield rate conditions, that is, the process parameters Adjustment (eg, CD value, film thickness, etc.). The machine adjustment procedure of step S200 will be further described below.

In addition, when it is determined in step S106 (or S108 ) that the online yield index Δ _Yt is smaller than the second preset value q, step S110 may be further performed to calculate the test item index Δ _Bin . The test item Bin is, for example, the electrical test of the product, which is also used to test the yield rate of the product. The test item Bin can be, for example, various feasible items such as capacitance test, leakage current test, etc., and the implementation of the present invention does not specifically limit what kind of items to be used. The yield rate is to evaluate the quality of the overall product. If the yield rate meets the conditions, further finer tests can be carried out on the details. Here, the test item Bin can select an item to be tested according to actual requirements. The calculation of the test item index △ _Bin is basically the same as the online quality control index △ _QC and the online yield index △ _Yt ; that is, the MSMC judgment criterion shown in Figure 4 is used for calculation. Here, the calculation method of the online quality control index _ΔBin can refer to the schematic flowchart shown in FIG. 10C . As shown in FIG. 10C , in step S110a, the data of the test item Bin is classified by machine. Next, in step S110b, a test item probability map is generated. Afterwards, in step S110c, the test item index Δ _Bin is calculated by using the MSMC judgment criterion.

Next, in step S112, according to the calculation result of step S110, it is judged whether the test item index _ΔBin is greater than the second preset value q. If the test item index △ _Bin is greater than the second preset value q, it also means that there is a difference in the yield rate (machine error). At this time, the machine adjustment procedure of step S200 will be further executed. The machine adjustment procedure of step S200 will be further described below.

In addition, in step S112, if it is determined that the test item index _ΔBin is smaller than the second preset value q, this also means that there is no difference in the yield rate (machine error) or the difference is small. In other words, according to the embodiment of the present invention, when the online yield index _ΔYt and the test item index _ΔBin are mostly smaller than the second preset value, the operator may consider executing the subroutine of step S220 , ie process boundary adjustment. The adjustment of the process boundary usually means that there is no machine difference or is not obvious between the machines. At this time, the operator can decide whether to make better adjustments to the range of the process parameters according to the correlation diagram between the process parameters and the yield rate in step S220.

After step S200 is executed, step S210 is performed to find out the yield-critical process. In this step, based on the above-mentioned steps, it is possible to find out the process that has a critical impact on the yield rate in the entire hundreds of process procedures, and find out the yield-critical process (CTY) from it.

Afterwards, the subroutine in step S230 is executed, and according to the key yield rate process obtained in step S210, various combinations of machines in each key yield rate process can be performed, so as to find out the machine combination with the highest yield rate, that is, the best Yield machine path (golden path).

Next, the tool adjustment procedure (tool adjustment) of step S200 in FIG. 9 will be described. FIG. 11 is a schematic flowchart of the machine adjustment procedure, and FIGS. 12A to 12C are explanatory diagrams illustrating examples of the machine adjustment procedure.

As shown in FIG. 11 , in step S300 , the yield rate is classified by risk process tool. Here, the risk process refers to the key process that will affect the yield. Next, in step S302, the risk machines are determined according to the process parameter probability map and the yield probability map (and/or the test item probability map). Here, a risk machine refers to a machine with a machine error in a key process that will affect the yield rate, that is, the machine will produce a poor yield rate. Using the process parameter probability map and the yield probability map, together with the analysis of the aforementioned online quality control index △ _QC and online yield index △ _Yt (and/or test item index △ _Bin ), it is possible to determine the The risk machine in .

Next, in step S304, a yield-process parameter correlation map is generated. From the correlation diagram of yield rate-process parameters, we can see the influence of the distribution of process parameters among machines on the yield rate. Next, in step S306, according to the yield rate-process parameter correlation diagram, the process parameter range of the risk machine is adjusted. Through this procedure, in the yield-critical process, the risk machine can be adjusted according to the required reservation conditions (determined by the yield rate, if it needs to be set above 95%), the process parameter range of the risk machine can be adjusted. This enables the yield rate of all machines in the yield-critical process to reach the preset target.

Next, a specific example of the machine adjustment process will be described with the example shown in FIG. 12A to FIG. 12C . This example corresponds to the situation where the online quality control indicator _ΔQC is greater than the first preset value p, and the online yield indicator _ΔYt is also greater than the second preset value q, that is, the situation in step S108 of FIG. 9 is "Yes". In addition, in this example, the critical dimension (CD) of two machines in a process program is used as the online monitoring object of process parameters. Figure 12A presents the distribution of the CD value measurement data of machine 1 and machine 2 obtained online, and then converted into a probability map, in which the online CD value (process parameter, data value) is the horizontal axis, and the probability value is the vertical axis axis. FIG. 12B presents the yield probability diagram corresponding to machine 1 and machine 2 in this manufacturing process, wherein the yield of each machine is on the horizontal axis, and the cumulative probability value is on the vertical axis.

FIG. 12C is a schematic diagram of machine tool adjusting process parameters. FIG. 12C is a distribution diagram of each machine's process parameter versus yield rate obtained by plotting the process parameter probability map of each machine in FIG. 12A and the yield rate probability map of each machine in FIG. 12B . As shown in Figure 12C, if it is set to meet the yield rate greater than 90%, it can be seen that the yield rate of machine 1 in the entire CD value (process parameter) range can meet more than 90%, but machine 2 is in the CD value. When it is greater than 58, the yield rate will be lower than 90%, and the requirement of yield rate cannot be met.

Therefore, in order to make the overall yield rate reach more than 90%, it is necessary to adjust the process parameters of the machine. It can be seen from Figure 12C that for Machine 2, as long as the CD value is lowered to below 58, the yield rate of Machine 2 can be maintained above 90%. Therefore, according to the method of the embodiment of the present invention, it can be found that the machine difference between machine 1 and machine 2 has a greater impact on the yield rate, so that the process parameters of machine 2 can be adjusted online in real time range, so that the yield rate of machine 2 can reach more than 90%.

Next, another machine adjustment example using the embodiment of the present invention will be described. FIG. 13A to FIG. 13D show another example to illustrate the machine adjustment process. In this example, it is equivalent to the situation where the online quality control index △ _QC is smaller than the first preset value p, while the online yield index △ _Yt is larger than the second preset value q, that is, step S108 in Figure 9 is "Yes" situation. In this case, it can be seen from Figure 13A that there is almost no difference in the online quality control index △ _QC between Machine 1 and Machine 2, but it can be seen from Figure 13B that the yield rate between Machine 1 and Machine 2 is If there is a difference, that is, the online yield index Δ _Yt exceeds a predetermined value.

As shown in FIG. 13C , it is a distribution diagram of process parameters (CD values) versus yield between machine 1 and machine 2 . As mentioned above, because the result presented by the online quality control index △ _QC is no difference between the two machines, the distribution ranges on the horizontal axis in Fig. 13C are almost the same. However, there are differences in the yield rates between the two machines, that is, the distribution ranges on the vertical axis in FIG. 13C show separate states. Therefore, as shown in Figure 13D, according to the results of the correlation diagram in Figure 13C, the operator can adjust the process parameter CD value of machine 1 with a relatively low yield rate to above 53, so that the yield rate of machine 1 can be improved Satisfy more than 95% of the conditions, and machine 2 can also achieve a yield of more than 95%.

Next, step S220 in FIG. 9 will be described. Step S220 is a process boundary adjustment procedure. FIG. 14A is a schematic flowchart of a process boundary adjustment procedure. 14B and 14C illustrate two examples of process boundary checking.

In addition, the process boundary adjustment procedure is usually aimed at the situation that the online yield index _ΔYt or the test item index _ΔBin is smaller than the second preset value, that is, the partial difference of the yield rate is small. In this case, not only the online yield index △ _Yt or the test item index △ _Bin is smaller than the second preset value, but also the online quality control index △ _QC is also smaller than the first preset value. At this time, in the relationship diagram between process parameters and yield rate drawn based on this, the distribution diagrams of each machine are basically almost overlapping, that is, the machine difference is small or even non-existent. However, even so, if the process parameter range of the machine is not set properly, the yield rate in some areas will be lower than the target value. Therefore, in this case, even if the machines are matched, it is still necessary to reset the range.

As shown in FIG. 14A , in step S400 , the yields are classified by process equipment. Next, in step S402, the matching of each machine is confirmed. Next, in step S404, a yield-process parameter correlation map is generated. From the yield-process parameter correlation diagram, we can see the relationship between the distribution of process parameters among machines and the yield rate. In this case, even if the difference between the machines is not obvious or not, but the process parameter range is not properly adjusted, it will cause the yield rate of some areas to be lower in the process parameter range. Worse things happen. Next, in step S406 , according to the yield rate-process parameter correlation diagram, the process parameter range of the machine is adjusted (boundary adjustment, etc.). Here, the adjustment methods include, for example, tightening the parameter range, or adjusting the centerline to adjust the parameter range, and the like. Two specific examples will be described below.

The process boundary check shown in FIG. 14B is to tighten the range of process parameters according to yield. In the example of FIG. 14B , the horizontal axis represents the value of the process parameter CD, and the vertical axis represents the yield Yt. As mentioned above, in this case, the distributed process parameter CD values of machine 1 and machine 2 are 40~60 and almost overlap. However, even if they overlap, when the cd value of machine 1 and machine 2 is greater than 58, the corresponding yield value is below 93. Therefore, in order to make the yield rate performance of each machine better, even if the machine difference between machine 1 and machine 2 is not obvious, the operator can control the process parameter CD value of machine 1 and machine 2, and tighten it to the CD value Between 42~58. Therefore, through such process boundary adjustment (in this case, the boundary range is tightened), the yield rate of each machine can be maintained above 93%.

The process boundary check shown in FIG. 14C adjusts the center line of the parameter range according to the yield. Similarly, in the example of FIG. 14C, the horizontal axis is the process parameter CD value, The vertical axis is the yield rate Yt. As mentioned above, in this case, if the distribution of the process parameter CD values of machine 1 and machine 2 is 35~65, its center line falls at 50. However, in this case, there will be a problem of uneven distribution, that is, the CD values of the process parameters with a yield rate above 94 are almost all on the right side of the center line 50. That is to say, the best yield is produced when the CD value is about 53. So in this case, the operator can reset the center line to improve the uneven distribution. In other words, the operator can adjust the entire CD value distribution range from the original 35~65 to 43~63, and then reset the centerline to 53. As shown in FIG. 15B , after adjusting the centerline, the yield rate of the entire process parameter CD value compared with the distribution of the centerline 53 can be further improved.

Next, a method for determining the best-yield machine path by using the results of the embodiments of the present invention will be described. FIG. 15 is a schematic flow chart of determining the best yield machine path. As shown in FIG. 15 , in step S500 , the yield is classified according to the yield-critical process. Next, in step S502, a corresponding probability map is generated based on the above classification. Afterwards, in step S504, according to the probability map, find out the machine combination with the best yield rate as the machine path with the best yield rate.

Figure 16 is an example of determining the best yield machine path. Figure 16 shows the probability diagram of the yield rate of each machine in the yield-critical process. As shown in FIG. 16 , the yield-critical process 2 includes machines C and D, the yield-critical process 5 includes machines P and Q, and the yield-critical process 6 includes machines W and Z. It can be seen from Figure 16 that the combination of machines C, P, and W has the highest yield rate of 95.32, so including machines C, P, and W is the path with the best yield rate.

In addition, the best yield rate machine path can also use big data to determine defined, for example using association rules. Association rules are a way to find the correlation between variables in the big data database, and use this to find the best yield machine path. Table 1 is an example of using association rules.

It can be seen from Table 1 that in the entire process, if a combination of machine C including yield-critical process (CYT) 2, machine P with yield-critical process 5, and machine W with yield-critical process 6 is used, The overall yield can reach more than 95%. Therefore, the combination including machines C, P, and W marked in gray is the machine path with the best yield rate.

To sum up, according to the embodiment of the present invention, not only the distribution difference of process parameters is considered in the monitoring mechanism, but also the yield difference between machines is considered. Therefore, through the two-way confirmation mechanism of yield rate and process parameters, the matching between machines can be more perfect and accurate, and the overall yield rate can be improved.

Claims (20)

A yield-oriented online machine matching management method is executed by a processor and adjusts multiple machines in a process program. The online machine matching management method includes: Collecting the measured values of the process parameters of the plurality of machines, and the yield rate corresponding to each of the plurality of machines; making a process parameter probability map based on the process parameters of the plurality of machines, and calculating online quality control indicators based on preset judgment criteria; Making a yield probability map based on the yield rates of the plurality of machines, and calculating an online yield index based on the preset judgment criteria; Making a yield-process parameter correlation diagram according to the online quality control index and the online yield index; According to the yield rate-process parameter correlation diagram, determine the risk machine that will affect the yield rate among the plurality of machines; and Based on preset conditions, a machine adjustment program or a process boundary adjustment program is performed on the risk machine, Wherein the preset judgment criterion is to set a plurality of probability values in the process parameter probability map or the yield probability map, and calculate a plurality of probability values among the plurality of machines corresponding to each of the plurality of probability values. The maximum process parameter difference or the multiple maximum yield rate differences, and calculate the weighted average of the multiple maximum process parameter differences or the multiple maximum yield rate differences in a statistical manner. According to the yield-oriented online machine matching management method described in claim 1, the online quality control index and the online yield index are calculated based on the following formula:

Wherein α _i is the weight, Δ _i is the maximum process parameter difference or the maximum yield rate difference corresponding to each of the multiple probability values, k σ is the full distribution range of the process parameter difference or the yield rate, σ is the standard deviation, k is a constant, and is the number of the multiple probability values. The yield-oriented online machine matching management method as described in claim 1, wherein when the online quality control index is greater than the first preset value and the online yield index is greater than the second preset value, or the online When the quality control index is less than the first preset value and the online yield index is greater than the second preset value, execute the machine adjustment procedure, Wherein the first preset value is the same as or different from the second preset value. The yield-oriented online machine matching management method as described in claim 1, wherein when the online quality control index is smaller than a first preset value and the online yield index is smaller than a second preset value, the process is executed boundary adjustment procedure, Wherein the first preset value is the same as or different from the second preset value. According to the yield-oriented online machine matching management method described in claim 4, when the online yield index is less than the second preset value, before executing the process boundary adjustment procedure, it further includes: Calculating test item indicators according to the preset judgment criteria; judging whether the test item index is greater than the second preset value; and When the test item index is greater than the second preset value, the machine adjustment program is executed; when the test item index is greater than the second preset value, the process boundary adjustment program is executed. The yield-oriented online machine matching management method as described in claim 1, wherein the machine adjustment procedure further includes: Classifying the yield rate by risk process equipment; determining the risk machine according to the process parameter probability map and the yield probability map; generating said yield-process parameter correlation map; and According to the yield rate-process parameter correlation diagram, the range of the process parameter of the risk machine is adjusted. The yield-oriented online machine matching management method as described in claim 1, wherein the process boundary adjustment procedure further includes: Classifying the yield rate by each of the machines of the process procedure; generating the yield rate-process parameter correlation diagram for each of the machines; and The range of the process parameter of the machine is adjusted according to the yield rate-process parameter correlation diagram. The yield-oriented online machine matching management method according to claim 7, wherein the process boundary adjustment procedure is to tighten the range of the process parameters. The yield-oriented online machine matching management method according to claim 7, wherein the process boundary adjustment program changes the centerline of the range of the process parameters. The yield-oriented online machine matching management method according to claim 1, wherein the preset condition is determined based on the yield. A yield-oriented online machine matching management method is executed by a processor. The online machine matching management method is suitable for an architecture composed of multiple process programs, each of which includes multiple machines. And adjust the multiple machines in each of the multiple manufacturing processes, the online machine matching management method includes: Collecting the measured values of the process parameters of the plurality of machines in each of the plurality of process steps, and the yield rate corresponding to each of the plurality of machines; making a process parameter probability map based on the process parameters of the plurality of machines, and calculating online quality control indicators based on preset judgment criteria; Making a yield probability map based on the yield rates of the plurality of machines, and calculating an online yield index based on the preset judgment criteria; Making a yield-process parameter correlation diagram according to the online quality control index and the online yield index; According to the yield rate-process parameter correlation diagram, determine the risk machine that will affect the yield rate among the multiple machines; Based on preset conditions, perform a machine adjustment procedure or a process boundary adjustment procedure on the risky machine; According to the online quality control index and the online yield index associated with each of the plurality of manufacturing procedures, find out at least one yield-critical process among the plurality of manufacturing procedures; and Based on the combination of the plurality of machines of the at least one yield-critical process, determining the best yield machine path, Wherein the preset judgment criterion is to set a plurality of probability values in the process parameter probability map or the yield probability map, and calculate a plurality of probability values among the plurality of machines corresponding to each of the plurality of probability values. The maximum process parameter difference or the multiple maximum yield rate differences, and calculate the weighted average of the multiple maximum process parameter differences or the multiple maximum yield rate differences in a statistical manner. According to the yield-oriented online machine matching management method described in claim item 11, the online quality control index and the online yield index are calculated based on the following formula:

Wherein α _i is the weight, Δ _i is the maximum process parameter difference or the maximum yield rate difference corresponding to each of the multiple probability values, k σ is the full distribution range of the process parameter difference or the yield rate, σ is the standard deviation, k is a constant, and is the number of the multiple probability values. The yield-oriented online machine matching management method according to claim 11, wherein when the online quality control index is greater than a first preset value and the online yield index is greater than a second preset value, or the online When the quality control index is less than the first preset value and the online yield index is greater than the second preset value, execute the machine adjustment procedure, Wherein the first preset value is the same as or different from the second preset value. The yield-oriented online machine matching management method according to claim 11, wherein when the online quality control index is smaller than a first preset value and the online yield index is smaller than a second preset value, the process is executed boundary adjustment procedure, Wherein the first preset value is the same as or different from the second preset value. According to the yield-oriented online machine matching management method described in claim 14, when the online yield index is less than the second preset value, before executing the process boundary adjustment procedure, it further includes: Calculating test item indicators according to the preset judgment criteria; judging whether the test item index is greater than the second preset value; and When the test item index is greater than the second preset value, the machine adjustment program is executed; when the test item index is greater than the second preset value, the process boundary adjustment program is executed. The yield-oriented online machine matching management method as described in claim 11, wherein the machine adjustment procedure further includes: Classifying the yield rate by risk process equipment; determining the risk machine according to the process parameter probability map and the yield probability map; generating said yield-process parameter correlation map; and According to the yield rate-process parameter correlation diagram, the range of the process parameter of the risk machine is adjusted. The yield-oriented online machine matching management method as described in claim 11, wherein the process boundary adjustment procedure further includes: Classifying the yield rate by each of the machines of the process procedure; generating the yield rate-process parameter correlation diagram for each of the machines; and The range of the process parameter of the machine is adjusted according to the yield rate-process parameter correlation diagram. The yield-oriented online machine matching management method according to claim 17, wherein the process boundary adjustment procedure is to tighten the range of the process parameter or change the centerline of the range of the process parameter. The yield-oriented online machine matching management method according to claim 11, wherein the preset condition is determined based on the yield. The yield-oriented online machine matching management method as described in claim 11, wherein determining the best yield machine path is based on the yield rate among the machines in the at least one yield-critical process Probabilistic graphs, or exploit association rules.

Images