Analysis device, analysis method, and analysis program

Abstract

To implement a technique which can calculate a degree of contribution of each of elements constituting each process, to quality of a product manufactured through a plurality of processes.SOLUTION: An inspection / analysis device (1) comprises at least one processor (12) which performs an element contribution calculation step of calculating a contribution of each of elements constituting each of pieces of process information by using a model (M) which receives process information about a plurality of processes to output a class indicative of quality of a product manufactured through the plurality of processes and is constructed by machine learning.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an analysis apparatus, an analysis method, and an analysis program for analyzing defect factors regarding products manufactured through a plurality of processes.

Background Art

[0002] Products are usually manufactured through a plurality of processes. For example, semiconductor devices such as semiconductor laser units are manufactured through a plurality of processes including a crystal growth process, a separation process, an assembly process, and the like. Therefore, in order to establish a manufacturing method with a low occurrence rate of defective products, it is important to identify the factors when defective products occur. For this purpose, it is necessary to identify the contribution degree of each element constituting each process to the quality of the product. Patent Document 1 discloses a technique for calculating the contribution degree of each device used in manufacturing to the occurrence of defective products.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In Patent Document 1, it is described that the contribution degree of each device to the occurrence of defective products is calculated using an "element statistical model" (undefined in Patent Document 1), and it is suggested that a specific example of this "element statistical model" is a gradient boosting decision tree (see paragraphs 0026 to 0034 of Patent Document 1). However, Patent Document 1 does not even clarify what the gradient boosting decision tree is as a decision tree with what as the target variable and what as the explanatory variables. Even those skilled in the art who have access to Patent Document 1 cannot grasp the essence of the "element statistical model". For this reason, it has not been possible to realize a technique for calculating the contribution degree of each element constituting each process to the quality of the product.

[0005] One aspect of the present invention has been made in view of the above problems, and an object thereof is to realize a technology capable of calculating the contribution degree to the quality of a product for each element constituting each process.

Means for Solving the Problems

[0006] An analyzer according to Aspect 1 of the present invention is a model that takes process information regarding each of a plurality of processes as input and outputs a class indicating the quality of a product manufactured through the plurality of processes, and uses a model constructed by machine learning to execute an element-by-element contribution degree calculation step of calculating the contribution degree of each element constituting each process information by at least one processor.

[0007] According to the above configuration, it is possible to calculate the contribution degree to the quality of the product for each element constituting each process information.

[0008] In an analyzer according to Aspect 2 of the present invention, in addition to the configuration of Aspect 1, the processor further executes a step-by-step contribution degree calculation step of calculating the contribution degree of each process information by adding the contribution degrees of the elements constituting the process information, and such a configuration is adopted.

[0009] According to the above configuration, it is possible to calculate the contribution degree to the quality of the product for each process information.

[0010] In an analyzer according to Aspect 3 of the present invention, in addition to the configuration of Aspect 2, the processor executes the element-by-element contribution degree calculation step and the step-by-step contribution degree calculation step for each of a plurality of products, and for each element constituting each process, an average element-by-element contribution degree calculation step of calculating the average value of the contribution degrees calculated for each product, and for each process, an average step-by-step contribution degree calculation step of calculating the average value of the contribution degrees calculated for each product, and such a configuration is adopted.

[0011] According to the above configuration, for each element constituting each process, it is possible to calculate the average of the contribution degrees calculated for each product, and for each process, it is possible to calculate the average value of the contribution degrees calculated for each product.

[0012] In the analyzer according to Aspect 4 of the present invention, in addition to the configuration of any one of Aspects 1 to 3, the process information includes at least any one of information regarding the material used in the corresponding process, information regarding the product obtained in the corresponding process, information regarding the manufacturing apparatus used in the corresponding process, and information regarding the environment in which the corresponding process was carried out.

[0013] According to the above configuration, for each element constituting each process information, it is possible to accurately calculate the contribution degree to the quality of the product.

[0014] In the analyzer according to Aspect 5 of the present invention, in addition to the configuration of any one of Aspects 1 to 4, the product is a semiconductor device manufactured through a pre-process until a semiconductor chip is obtained and a post-process after obtaining the semiconductor chip, and the input of the model is the process information regarding the pre-process.

[0015] According to the above configuration, for each element constituting each process information, it is possible to calculate the contribution degree to the quality of the semiconductor device.

[0016] In the analyzer according to Aspect 6 of the present invention, in addition to the configuration of Aspect 5, the process information regarding the pre-process includes first process information regarding the semiconductor wafer obtained in the crystal growth process and second process information regarding the semiconductor chip obtained in the separation process.

[0017] According to the above configuration, for each element constituting each process information, it is possible to accurately calculate the contribution degree to the quality of the semiconductor device.

[0018] In the analyzer according to Aspect 7 of the present invention, in addition to the configuration of Aspect 6, the first process information is an image representing the appearance of the semiconductor wafer, and the second process information includes a numerical sequence representing the attributes of the semiconductor chip and an image representing the appearance of the semiconductor chip.

[0019] According to the above configuration, for each element constituting each process information, it is possible to calculate more accurately the contribution degree to the quality of the semiconductor device.

[0020] The analysis method according to Aspect 8 of the present invention is a model in which at least one processor takes process information regarding each of a plurality of processes as input and outputs a class indicating the quality of a product manufactured through the plurality of processes, and includes an element-by-element contribution degree calculation step of calculating the contribution degree of each element constituting each process information using a model constructed by machine learning.

[0021] According to the above configuration, for each element constituting each process information, it is possible to calculate the contribution degree to the quality of the product.

[0022] The analysis program according to Aspect 9 of the present invention is an analysis program for operating a computer equipped with the processor as any one of the analyzers of Aspects 1 to 7, and is an analysis program for causing the processor to execute an element-by-element contribution degree calculation step.

[0023] According to the above configuration, for each element constituting each process information, it is possible to calculate the contribution degree to the quality of the product using a computer.

Effect of the Invention

[0024] According to one aspect of the present invention, it is possible to realize a technique for determining the quality of a product before finishing all the processes regarding the product manufactured through a plurality of processes.

Brief Description of the Drawings

[0025]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Embodiments for Carrying Out the Invention

[0026] (Configuration of the inspection / analysis device) An inspection / analysis device 1 (an example of the "analysis device" in the claims) according to an embodiment of the present invention will be described with reference to FIG. 1. The inspection / analysis device 1 is a device having a determination function and an analysis function. Here, the determination function is a function of determining the quality of a product based on process information regarding a plurality of processes included in the manufacturing method of the product. The analysis function is a function of calculating the degree of contribution to the determination result for each element constituting each process information. In the present embodiment, as the product, a semiconductor device including a semiconductor chip, particularly, a semiconductor laser unit including a semiconductor laser chip is assumed.

[0027] As shown in FIG. 1, the inspection / analysis device 1 includes a memory 11, a processor 12, and a storage 13. The memory 11, the processor 12, and the storage 13 are connected to each other via a bus (not shown). Further, an input / output interface (not shown) and a communication interface (not shown) may be connected to this bus. This input / output interface is used, for example, to input process information from an external device (e.g., a camera, a test device, a manufacturing device, etc.) to the inspection / analysis device 1, or to output a determination result and a degree of contribution from the inspection / analysis device 1 to an external device (e.g., a display, etc.). Also, this communication interface is used, for example, for the inspection / analysis device 1 to receive an input image provided from an external device (e.g., a camera, a test device, another computer connected to a manufacturing device), or for the inspection / analysis device 1 to transmit a determination result and a degree of contribution provided to an external device (e.g., another computer connected to a display).

[0028] The memory 11 is configured to record a determination program P1 for realizing a determination function, a program P2 for realizing an analysis function, and a model M constructed by machine learning that is used to realize the determination function and the analysis function. Note that, as the memory 11, for example, a semiconductor RAM (Random Access Memory) or the like can be used. Also, as the model M, for example, a CNN (Convolutional Neural Network) can be used. Further, a logistic regression model, a support vector machine, a random forest, etc. may be used as the model M.

[0029] The processor 12 is configured to realize a determination function by executing the determination program P1 stored in the memory 11, and to realize an analysis function by executing the analysis program P2 stored in the memory 11. As the processor 12, for example, a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), a TPU (Tensor Processing Unit), a digital signal processor, a microprocessor, a microcontroller, or a combination thereof, etc. can be used.

[0030] The storage 13 is configured to store (non-volatile storage) the determination program P1, the analysis program P2, and the model M. When realizing the determination function, the processor 12 expands and refers to the determination program P1 and the model M stored in the storage 13 onto the memory 11. Also, when realizing the analysis function, the processor 12 expands and refers to the analysis program P2 and the model M stored in the storage 13 onto the memory 11. Note that, as the storage 13, for example, a flash memory, an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a combination thereof, etc. can be used.

[0031] Here, the determination function and the analysis function are each described as being realized by a single processor 12 provided in a single computer, but the present invention is not limited to this. That is, it is also possible to adopt a configuration in which the determination function and the analysis function are jointly realized by a plurality of processors provided in a single computer or distributed among a plurality of computers.

[0032] Also, here, the configuration in which the model M is stored in a single memory 11 provided in a single computer is described, but the present invention is not limited to this. That is, it is also possible to adopt a configuration in which the model M is distributed and stored in a plurality of memories provided in a single computer or distributed among a plurality of computers.

[0033] Note that the determination program P1 and the analysis program P2 can each be recorded on a non-transitory tangible computer-readable recording medium. This recording medium may be the memory 11, the storage 13, or other recording media. For example, tapes, disks, cards, semiconductor memories, and programmable logic circuits can be used as other recording media.

[0034] (Method for manufacturing a semiconductor device using an inspection / analysis apparatus) A method S100 for manufacturing a semiconductor device using the determination function of the inspection / analysis apparatus 1 will be described with reference to FIG. 2. FIG. 2 is a flowchart showing the flow of the method S100 for manufacturing a semiconductor device. As described above, here, a semiconductor laser unit including a semiconductor laser chip is assumed as the semiconductor device.

[0035] As shown in FIG. 2, the manufacturing method S100 includes a pre-process S110 until a semiconductor chip is obtained and a post-process S120 after the semiconductor chip is obtained.

[0036] The previous process S110 is composed of, for example, as shown in FIG. 2, a crystal growth process S111, an electrode formation process S112, a first separation process 113 (an example of a "separation process"), and a second separation process S114 (an example of a "separation process"). The crystal growth process S111 is a process of obtaining a semiconductor wafer by crystal growth. The electrode formation process S112 is a process of forming a front surface electrode and a back surface electrode on the semiconductor wafer. The first separation process S113 is a process of obtaining semiconductor bars by cleaving or dicing the semiconductor wafer. A process of coating the end face of the obtained semiconductor bar may be included in the first separation process S113. The second separation process S114 is a process of obtaining semiconductor chips by cleaving or dicing the semiconductor bars.

[0037] The subsequent process S120 is composed of, for example, as shown in FIG. 2, an assembly process S121 and an evaluation process S122. The assembly process S121 is a process of assembling a semiconductor device using the semiconductor chips. The assembly process S121 is composed of, for example, a process of die-bonding the semiconductor chips to a heat sink mount and a wire bonding process for obtaining an electrical connection. The evaluation process S122 is a process of evaluating the semiconductor device. The evaluation process S122 is composed of, for example, a process of burning in the semiconductor device by continuous energization over a certain period of time and a process of inspecting the characteristics of the semiconductor device after burning in.

[0038] The characteristic point of the manufacturing method S100 according to the present embodiment is that, as shown in FIG. 2, a wafer imaging process S101, an attribute inspection process S102, a chip imaging process S103, and a determination process S104 are included.

[0039] The wafer imaging process S101 is a process of obtaining an image representing the appearance of each semiconductor chip included in the semiconductor wafer obtained in the crystal growth process S111 as process information (an example of "first process information") regarding the crystal growth process S111. The wafer imaging process S101 is performed, for example, by a camera, and the image obtained in the wafer imaging process S101 is supplied from the camera to the inspection / analysis apparatus 1.

[0040] The attribute inspection step S102 is a step of obtaining a numerical sequence representing the attributes of the semiconductor chips obtained in the second separation step S114 as process information (an example of "second process information") regarding the second separation step S114. The attribute inspection step S102 is implemented, for example, by a test apparatus, and the numerical sequence obtained in the attribute inspection step S102 is supplied from the test apparatus to the inspection / analysis apparatus 1. Note that the attribute inspection step S102 may be a step of obtaining a numerical sequence representing the attributes of each semiconductor chip included in the semiconductor wafer obtained in the first separation step S113 as process information regarding the first separation step S113.

[0041] The chip imaging step S103 is a step of obtaining an image representing the appearance of the semiconductor chips obtained in the second separation step S114 as process information (an example of "second process information") regarding the second separation step S114. The chip imaging step S103 is implemented, for example, by a camera, and the image obtained in the chip imaging step S103 is supplied to the inspection / analysis apparatus 1.

[0042] The determination step S104 is a step in which the inspection / analysis apparatus 1 determines the quality of the semiconductor chips using a model M constructed by machine learning. The inputs to the model M are (1) an image representing the appearance of the semiconductor chips to be determined, obtained in the wafer imaging step S101, (2) a numerical sequence representing the attributes of the semiconductor chips to be determined, obtained in the attribute inspection step S102, and (3) an image representing the appearance of the semiconductor chips to be determined, obtained in the chip imaging step S103. The output of the model M is a class indicating the quality of the semiconductor devices manufactured by subjecting the semiconductor chips to be determined to the subsequent process S120.

[0043] In machine learning, teacher data is used, which associates (1) an image representing the appearance of a semiconductor chip to be determined, obtained in the wafer imaging step S101, (2) numerical values representing the attributes of the semiconductor chip to be determined, obtained in the attribute inspection step S102, and (3) an image representing the appearance of the semiconductor chip to be determined, obtained in the chip imaging step S103, with (4) the inspection result of the semiconductor device including the semiconductor chip to be determined, obtained in the evaluation step S122, as a label. The inspection result is, for example, either a good product or a defective product. Typically, if deterioration occurs in the semiconductor device in the evaluation step S122, the inspection result of that semiconductor device is a defective product, and if no deterioration occurs in the semiconductor device in the evaluation step S122, the inspection result of that semiconductor device is a good product. Note that the inspection / analysis apparatus 1 may have a function of constructing the model M by performing such machine learning in addition to the function of determining the quality of the semiconductor device using the model M.

[0044] When the output of the model M in the determination step S104 is the good product class, the subsequent process S120 for the semiconductor chip to be determined is performed. On the other hand, when the output of the model M in the determination step S104 is the defective product class, the subsequent process S120 for the semiconductor chip to be determined can be omitted. In this case, it is possible to avoid a decrease in production efficiency caused by performing the evaluation step S122, which is time-consuming and costly, on semiconductor chips that are highly likely to be defective semiconductor devices. Therefore, the production efficiency of the semiconductor device can be improved.

[0045] (Performance of the model) The performance of the model M used in the determination step S104 will be described with reference to FIGS. 3 and Tables 1 to 4.

[0046] FIG. 3 is a graph showing the number of non-defective products and defective products included in each lot arranged in chronological order. Here, the number of non-defective products is the number of semiconductor devices determined to be non-defective in the evaluation step S122, and the number of defective products is the number of semiconductor devices determined to be defective in the evaluation step S122. From the graph shown in FIG. 3, it was found that there are periods with a large number of defective products and periods with a small number of defective products.

[0047] Therefore, for each of the period with a large number of defective products and the period with a small number of defective products, learning (Training), validation (Validation), and evaluation (test) of the model M were performed. Table 1 is a table showing the number of semiconductor devices used for learning, validation, and evaluation of the model M for the period with a large number of defective products, divided into non-defective products and defective products. Table 2 is a table showing the number of semiconductor devices used for learning, validation, and evaluation of the model M for the period with a small number of defective products, divided into non-defective products and defective products.

[0048] [Table 1]

[0049] [Table 2]

[0050] Table 3 is a table showing the inference matrix of the model M for the period with a large number of defective products.

[0051] [Table 3]

[0052] As shown in Table 3, out of the 1957 semiconductor chips included in the semiconductor devices determined to be non-defective in the evaluation step S122, 1702 semiconductor chips were classified into the non-defective class in the determination step S104. That is, the non-defective reproduction rate was 86.97%. Also, out of the 153 semiconductor chips included in the semiconductor devices determined to be defective in the evaluation step S122, 108 semiconductor chips were classified into the defective class in the determination step S104. That is, the defective reproduction rate was 70.59%.

[0053] Furthermore, as shown in Table 3, out of the 1747 semiconductor devices including the semiconductor chips classified into the non-defective class in the determination step S104, 1702 semiconductor devices were determined to be non-defective in the evaluation step S122. That is, the non-defective conformity rate was 97.42%. Also, out of the 363 semiconductor devices including the semiconductor chips classified into the defective class in the determination step S104, 108 semiconductor devices were determined to be defective in the evaluation step S122. That is, the defective conformity rate was 29.75%.

[0054] From these results, it was found that for the period with many defective products, the F value of non-defective products of model M was 0.919, and the accuracy rate of model M was 85.78%.

[0055] Table 4 is a table showing the inference matrix of model M for the period with few non-defective products.

[0056]

Table 4

[0057] As shown in Table 4, out of the 4,660 semiconductor chips included in the semiconductor devices determined to be non-defective in the evaluation step S122, 3,369 semiconductor chips were classified into the non-defective class in the determination step S104. That is, the non-defective reproduction rate was 72.30%. Also, out of the 110 semiconductor chips included in the semiconductor devices determined to be defective in the evaluation step S122, 67 semiconductor chips were classified into the defective class in the determination step S104. That is, the defective reproduction rate was 60.91%.

[0058] Furthermore, as shown in Table 4, out of the 3,412 semiconductor devices including the semiconductor chips classified into the non-defective class in the determination step S104, 3,369 semiconductor devices were determined to be non-defective in the evaluation step S122. That is, the non-defective conformity rate was 98.74%. Also, out of the 1,358 semiconductor devices including the semiconductor chips classified into the defective class in the determination step S104, 67 semiconductor devices were determined to be defective in the evaluation step S122. That is, the defective conformity rate was 4.93%.

[0059] From these results, regarding the period with few defective products, it was concluded that the F value of non-defective products of model M was 0.8347 and the accuracy rate of model M was 72.03%.

[0060] (Method for Analyzing Defect Causes Using an Inspection / Analysis Apparatus) A method S200 for analyzing defect causes using the analysis function of the inspection / analysis apparatus 1 will be described with reference to FIG. 4. FIG. 4 is a flowchart showing the flow of the method S200 for analyzing defect causes.

[0061] As shown in FIG. 4, the analysis method S200 includes a contribution degree calculation step S201 for each element, a contribution degree calculation step S202 for each process, an average contribution degree calculation step S203 for each element, and an average contribution degree calculation step S204 for each process. In the present embodiment, the execution subject of the contribution degree calculation step S201 for each element, the contribution degree calculation step S202 for each process, the average contribution degree calculation step S203 for each element, and the average contribution degree calculation step S204 for each process is the processor 12 of the inspection / analysis device 1. The contribution degree calculation step S201 for each element and the contribution degree calculation step S202 for each process are executed for each semiconductor device in which the determination result in the determination step S104 and the determination result in the evaluation step S122 match.

[0062] The contribution degree calculation step S201 for each element is a step of calculating the contribution degree of each element constituting each process information to the determination result of the model M. In the present embodiment, (1) each pixel value constituting an image representing the appearance of a semiconductor chip included in the semiconductor wafer obtained in the crystal growth step S111 (process information related to the crystal growth step S111), (2) each numerical value constituting a numerical sequence representing the attributes of the semiconductor chip obtained in the second separation step S114 (first process information related to the second separation step S114), (3) each pixel value constituting an image representing the appearance of the semiconductor chip obtained in the second separation step S114 (second process information related to the second separation step S114), the contribution degree to the determination result of the model M is calculated. Note that the method for calculating the contribution degree of each element is not particularly limited, but in the present embodiment, the marginal contribution degree (marginal contribution) of each element is calculated using a known library called SHAP (SHapley Additive exPlanations).

[0063] Hereinafter, regarding the semiconductor device i, the contribution degree of each pixel value constituting an image representing the appearance of the semiconductor chip included in the semiconductor wafer obtained in the crystal growth step S111 is denoted as αij. Also, regarding the semiconductor device i, the contribution degree of each numerical value constituting a numerical sequence representing the attributes of the semiconductor chip obtained in the second separation step S114 is denoted as βik. Further, regarding the semiconductor device i, the contribution degree of each pixel value constituting an image representing the appearance of the semiconductor chip obtained in the second separation step S114 is denoted as γil.

[0064] The contribution degree calculation step S202 for each step is a step of calculating the contribution degree of each process information with respect to the determination result of the model M. The contribution degree of each process information is, for example, the sum of the contribution degrees of each element constituting the process information. In the present embodiment, the contribution degree αi of the process information regarding the crystal growth step S111 is calculated by adding the contribution degrees αij of each pixel value constituting an image representing the appearance of the semiconductor chip included in the semiconductor wafer obtained in the crystal growth step S111. Also, the contribution degree βi of the first process information regarding the second separation step S114 is calculated by adding the contribution degrees βik of each numerical value constituting a numerical sequence representing the attributes of the semiconductor chip obtained in the second separation step S114. Further, the contribution degree γi of the second process information regarding the second separation step S114 is calculated by adding the contribution degrees γil of each pixel value constituting an image representing the appearance of the semiconductor chip obtained in the second separation step S114.

[0065] The average contribution degree calculation step S203 for each element is a step of calculating the average value of the contribution degrees calculated for each product for each element constituting the process information. In the present embodiment, for each pixel value constituting an image representing the appearance of semiconductor chips included in the semiconductor wafer obtained in the crystal growth step S111 (process information related to the crystal growth step S111), the average value αj of the contribution degrees αij calculated for each semiconductor device is calculated. Also, for each numerical value constituting a numerical sequence representing the attributes of the semiconductor chips obtained in the second separation step S114 (the first process information related to the second separation step S114), the average value βk of the contribution degrees βik calculated for each semiconductor device is calculated. Also, for each pixel value constituting an image representing the appearance of the semiconductor chips obtained in the second separation step S114 (the second process information related to the second separation step S114), the average value γl of the contribution degrees γil calculated for each semiconductor device is calculated. Hereinafter, the average values αj, βk, γl are referred to as the average contribution degrees αj, βk, γl for each element.

[0066] The average contribution degree calculation step S204 for each process is a step of calculating the average value of the contribution degrees calculated for each product for each process information. In the present embodiment, for the process information related to the crystal growth step S111, the average value α of the contribution degrees αi calculated for each semiconductor device is calculated. Also, for the first process information related to the second separation step S114, the average value β of the contribution degrees βi calculated for each semiconductor device is calculated. Also, for the second process information related to the second separation step S114, the average value γ of the contribution degrees γi calculated for each product is calculated. Hereinafter, the average values α, β, γ are referred to as the average contribution degrees α, β, γ for each process.

[0067] Referring to the average contribution degrees α, β, γ for each process calculated as described above, it is possible to know which of the process information regarding the crystal growth process S111, the first process information regarding the second separation process S114, and the second process information regarding the second separation process S114 strongly affects the quality of the semiconductor device. Also, referring to the average contribution degree αj for each element calculated as described above, it is possible to know which of the pixel values constituting the image representing the appearance of the semiconductor chips included in the semiconductor wafer obtained in the crystal growth process S111 strongly affects the quality of the semiconductor device. Further, referring to the average contribution degree βk for each element calculated as described above, it is possible to know which of the numerical values constituting the numerical sequence representing the attributes of the semiconductor chips obtained in the second separation process S114 strongly affects the quality of the semiconductor device. Moreover, according to the average contribution degree γl for each element calculated as described above, it is possible to know which of the pixel values constituting the image representing the appearance of the semiconductor chips obtained in the second separation process S114 strongly affects the quality of the semiconductor device.

[0068] (Example) First, for the period with a large number of defective products (see Fig. 3), the average contribution degrees α, β, γ for each process and the average contribution degrees αj, βk, γl for each element were calculated using the analysis method S200. As samples, out of 2,110 semiconductor devices manufactured during the period with a large number of defective products, (1) 1,702 semiconductor devices determined to be non-defective in both the determination process S104 and the evaluation process S122, and (2) 108 semiconductor devices determined to be defective in both the determination process S104 and the evaluation process S122 were used (see Table 3).

[0069] FIG. 5 is a graph showing the average contribution per process α, β, γ and the average contribution per element βk calculated using the analysis method S200 for the period with a large number of defective products. The average contribution per process α of the process information regarding the crystal growth process S111 was 20%, the average contribution per process β of the first process information regarding the second separation process S114 was 50%, and the average contribution per process γ of the second process information regarding the second separation process S114 was 30%. From this, it was confirmed that the first process information regarding the second separation process S114 has the strongest influence on the quality of the semiconductor device. Also, among the attributes constituting the first process information regarding the second separation process S114, the attribute with the highest average contribution per element βk was attribute X. From this, it was confirmed that attribute X has the strongest influence on the quality of the semiconductor device.

[0070] Note that the analysis of the cause of defects by the inspector (human) was performed independently of the analysis of the cause of defects using the model M. As a result, the inspector determined that the element having the strongest influence on the quality of the semiconductor device during the period with a large number of defective products is attribute Y that constitutes the first process information regarding the second separation process S114. This attribute Y has a strong correlation with the above-described attribute line X.

[0071] Note that in the analysis method S200, not only the average contribution per process α, β, γ and the average contribution per element αj, βk, γl, but also the contribution per process αi, βi, γi and the contribution per element αij, βik, γil for each semiconductor device are calculated. Therefore, for each individual semiconductor device, it is possible to know which process information has the strongest influence on the quality of the semiconductor device, and which element of each process information has the strongest influence on the quality of the semiconductor device.

[0072] FIG. 6 is a graph showing the per-process contribution degrees αi, βi, γi and the per-element contribution degrees βik of a certain semiconductor device determined to be a defective product in both the determination step S104 and the evaluation step S122. For this semiconductor device, the per-process contribution degree αi of the process information regarding the crystal growth step S111 was 9%, the per-process contribution degree βi of the first piece of process information regarding the second separation step S114 was 73%, and the per-process contribution degree γi of the second piece of process information regarding the second separation step S114 was 18%. From this, it was confirmed that for this semiconductor device, the first piece of process information regarding the second separation step S114 has the strongest influence on the quality of the semiconductor device. Also, for this semiconductor device, among the attributes constituting the first piece of process information regarding the second separation step S114, the attribute with the highest average per-element contribution degree βk was the M attribute. From this, it was confirmed that for this semiconductor device, the M attribute has the strongest influence on the quality of the semiconductor device.

[0073] Next, for the period with few defective products (see FIG. 3), the average per-process contribution degrees α, β, γ and the average per-element contribution degrees αj, βk, γl were calculated using the analysis method S200. As samples, out of 4770 semiconductor devices manufactured during the period with few defective products, (1) 3369 semiconductor devices determined to be non-defective products in both the determination step S104 and the evaluation step S122, and (2) 67 semiconductor devices determined to be defective products in both the determination step S104 and the evaluation step S122 were used (see Table 4).

[0074] FIG. 7 is a graph showing the average contribution per process α, β, γ and the average contribution per element βk calculated using the analysis method S200 for the period with few defective products. The average contribution per process α of the process information regarding the crystal growth process S111 was 14%, the average contribution per process β of the first process information regarding the second separation process S114 was 45%, and the average contribution per process of the second process information regarding the second separation process S114 was 41%. From this, it was confirmed that the first process information regarding the second separation process S114 has the strongest influence on the quality of the semiconductor device. Also, among the attributes constituting the first process information regarding the second separation process S114, the attribute with the highest average contribution per element βk was the Z attribute. From this, it was confirmed that the Z attribute has the strongest influence on the quality of the semiconductor device.

[0075] Note that the analysis of the cause of defects by the inspector (human) was performed independently of the analysis of the cause of defects using the model M. However, the inspector could not identify which element has the strongest influence on the quality of the semiconductor device during the period with few defective products. This means that there are defect causes that cannot be identified by the analysis by the inspector but can be identified by the analysis using the model M.

[0076] (a) of FIG. 8 is a histogram of the Z attribute regarding the semiconductor device determined to be a non-defective product in both the determination process S104 and the evaluation process S122. (b) of FIG. 8 is a histogram of the Z attribute regarding the semiconductor device determined to be a defective product in both the determination process S104 and the evaluation process S122. According to FIG. 8, it can be seen that semiconductor devices with particularly low values of the Z attribute specifically appear only in defective products. This suggests the possibility of estimating the quality of the semiconductor device by paying attention to the Z attribute.

[0077] (c) of FIG. 8 is a graph showing the per-process contribution degrees αi, βi, γi and the per-element contribution degrees βik of a certain semiconductor device with particularly low Z-attribute values. For this semiconductor device, the per-process contribution degree αi of the process information regarding the crystal growth process S111 was 8%, the per-process contribution degree βi of the first piece of process information regarding the second separation process S114 was 69%, and the per-process contribution degree γi of the second piece of process information regarding the second separation process S114 was 23%. From this, it was confirmed that for this semiconductor device, the first piece of process information regarding the second separation process S114 has the strongest influence on the quality of the semiconductor device. Also, for this semiconductor device, among the attributes constituting the first piece of process information regarding the second separation process S114, the attribute with the highest average per-element contribution degree βk was the Z-attribute. From this, it was confirmed that for this semiconductor device, the Z-attribute has the strongest influence on the quality of the semiconductor device.

[0078] Note that in the analysis method S200, not only the average per-process contribution degrees α, β, γ and the average per-element contribution degrees αj, βk, γl, but also the per-process contribution degrees αi, βi, γi and the per-element contribution degrees αij, βik, γil for each semiconductor device are calculated. Therefore, for individual semiconductor devices, it is possible to find out which process information has the strongest influence on the quality of the semiconductor device, and which element of each process information has the strongest influence on the quality of the semiconductor device.

[0079] FIG. 9 is a graph showing the per-process contribution degrees αi, βi, γi and the per-element contribution degree βik of a certain semiconductor device determined to be a defective product in both the determination step S104 and the evaluation step S122. For this semiconductor device, the per-process contribution degree αi of the process information regarding the crystal growth step S111 was 11%, the per-process contribution degree βi of the first piece of process information regarding the second separation step S114 was 52%, and the per-process contribution degree γi of the second piece of process information regarding the second separation step S114 was 37%. From this, it was confirmed that for this semiconductor device, the first piece of process information regarding the second separation step S114 has the strongest influence on the quality of the semiconductor device. Also, for this semiconductor device, among the attributes constituting the first piece of process information regarding the second separation step S114, the attribute with the highest average per-element contribution degree βk was the N attribute. From this, it was confirmed that for this semiconductor device, the N attribute has the strongest influence on the quality of the semiconductor device.

[0080] (Modification example of inspection / analysis) In the present embodiment, a semiconductor device, particularly a semiconductor laser unit, was taken as the inspection target, but the present invention is not limited thereto. That is, any product manufactured through a plurality of processes can be taken as the inspection target.

[0081] Also, in the present embodiment, an image representing the appearance of the semiconductor wafer obtained in the crystal growth step S111, a numerical value representing the attributes of the semiconductor chip obtained in the second separation step S114, and an image representing the appearance of the semiconductor chip obtained in the second separation step S114 were used as the input to the model M, but the present invention is not limited thereto. That is, process information regarding any process can be used as the input to the model M. As the process information regarding each process, in addition to information regarding the product obtained in that process, for example, information regarding the material used in that process, information regarding the manufacturing apparatus used in that process, and information regarding the environment in which that process is carried out can be mentioned. There is a certain correlation between these pieces of information and the quality of the product. Therefore, if these pieces of information are used as the input to the model M, the quality of the product can be effectively estimated.

[0082] Note that the process information may be represented by an image, a numerical value or a numerical sequence, or a character or a character string. Further, the process information can be obtained by various sensors such as a camera, a microphone, a gas sensor, an acceleration sensor, a gyro sensor, a current sensor, a voltage sensor, a temperature sensor, and a humidity sensor. Information about the manufacturing apparatus (for example, set values) may be obtained from a controller that controls the manufacturing apparatus. Also, information about a material or a product (for example, characteristic values) may be obtained from a test apparatus that tests the material or the product.

[0083] In the present embodiment, the inspection / analysis apparatus 1 having the determination function and the analysis function has been described, but the determination function can be omitted. Regardless of whether or not it has a determination function, an apparatus having an analysis function (analysis apparatus) in general is included in the scope of the present invention.

[0084] (Supplementary Notes) The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope shown in the claims. Embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiments are also included in the technical scope of the present invention. Further, the application target of the present invention is not limited to semiconductor devices, and can be generally applied to products manufactured through a plurality of processes.

Description of Reference Numerals

[0085] 1 Inspection / Analysis Apparatus 11 Memory 12 Processor 13 Storage P1 Determination Program P2 Analysis Program M Model S100 Manufacturing Method S101 Wafer Imaging Step S102 Attribute Inspection Step S103 Chip Imaging Step S104 Determination Step

Claims

1. A model that takes process information for each of a plurality of processes as input and outputs a class indicating the quality of a product manufactured through the plurality of processes. The model is constructed by machine learning, and at least one processor is provided that executes an element-by-element contribution calculation step of calculating the contribution degree of each element constituting each process information. Among the process information for each of the plurality of processes, the process information for at least one process is an image. In the element-by-element contribution calculation step, the at least one processor calculates the contribution degree of each pixel value constituting the image. An analyzer characterized by the above.

2. The processor further executes a step-by-step contribution calculation step of calculating the contribution degree of each process information by adding the contribution degrees of the elements constituting the process information. The analyzer according to claim 1, characterized by the above.

3. The processor executes the element-by-element contribution calculation step and the step-by-step contribution calculation step for each of a plurality of products, and for each element constituting each process, calculates an average value of the contribution degrees calculated for each product, which is an average element-by-element contribution calculation step, and for each process, calculates an average value of the contribution degrees calculated for each product, which is an average step-by-step contribution calculation step. The analyzer according to claim 2, characterized by the above.

4. The process information includes at least any one of information on the material used in the corresponding process, information on the product obtained in the corresponding process, information on the manufacturing apparatus used in the corresponding process, and information on the environment in which the corresponding process was carried out. The analyzer according to any one of claims 1 to 3, characterized by the above.

5. The product is a semiconductor device manufactured through a pre-process until a semiconductor chip is obtained and a post-process after the semiconductor chip is obtained. The input of the model is the process information related to the pre-process. The analyzer according to any one of claims 1 to 4, characterized by the above.

6. The process information related to the pre-process includes first process information on a semiconductor wafer obtained in a crystal growth process and second process information on a semiconductor chip obtained in a separation process. The analyzer according to claim 5, characterized by the above.

7. The first process information is an image representing the appearance of the semiconductor wafer. The second process information includes a numerical sequence representing the attributes of the semiconductor chip and an image representing the appearance of the semiconductor chip. The analyzer according to claim 6, characterized in that.

8. An analyzer comprising at least one processor, The at least one processor is A model that takes process information regarding each of a plurality of processes as input and outputs a class indicating the quality of a product manufactured through the plurality of processes, and uses a model constructed by machine learning to calculate the contribution degree of each element constituting each process information. An element-by-element contribution degree calculation step, A step-by-step contribution degree calculation step of calculating the contribution degree of each process information by adding the contribution degrees of the elements constituting the process information is executed for each of a plurality of products, An average element-by-element contribution degree calculation step of calculating the average value of the contribution degrees calculated for each product for each element constituting each process, For each process, an average step-by-step contribution degree calculation step of calculating the average value of the contribution degrees calculated for each product is further executed. The analyzer according to claim 8, characterized in that.

9. At least one processor is a model that takes process information regarding each of a plurality of processes as input and outputs a class indicating the quality of a product manufactured through the plurality of processes, and uses a model constructed by machine learning to calculate the contribution degree of each element constituting each process information. An element-by-element contribution degree calculation step is included, Among the process information regarding each of the plurality of processes, the process information regarding at least one process is an image, In the element-by-element contribution degree calculation step, the at least one processor calculates the contribution degree of each pixel value constituting the image. The analysis method according to claim 13, characterized in that.

10. An analysis program for operating a computer equipped with the processor as the analyzer according to any one of claims 1 to 8, the analysis program for causing the processor to execute an element-by-element contribution degree calculation step.

Images