Control equipment and control method of stepper

This application claims the benefit of People's Republic of China application Serial No. 201910999042.3, filed Oct. 21, 2019, the subject matter of which is incorporated herein by reference.

The invention relates in general to a control equipment and a control method, and more particularly to a control equipment and a control method of a stepper.

Along with the development in the semi-conductor technology, various electronic devices are provided one after another. The manufacturing process of chip includes an exposure process, a development process, and an etching process. After a photoresist layer is exposed in an exposure process, a development pattern is shown in the development process. In an etching process, a metal layer is etched, by using the development pattern as a mask, to form a circuit layer.

Therefore, the precision of controlling the exposure process greatly affects the circuit layer. Conventionally, an operator can adjust the exposure energy of the stepper to precisely develop the development pattern. However, as the trace width is getting smaller and smaller, the stepper needs more accurate calibration to form high precision circuits.

The present invention relates to a control equipment and a control method of a stepper, which calibrates the beam focal length and/or the exposure energy of the stepper to increase the precision of the stepper by using the artificial intelligence technology.

According to an embodiment of the present invention, a control equipment of a stepper is provided. The control equipment of the stepper includes an input device, a generating device and a processing device. The input device is configured to input a plurality of sample development patterns. The sample development patterns are obtained according to a plurality of sample focal length values. The generating device is configured to generate a plurality of generative categories corresponding to a plurality of generative focal length values by using a depth learning algorithm. The processing device is configured to analyze an estimated focal length value of an online development pattern according to the generative categories.

According to another embodiment of the present invention, a control method of a stepper is provided. The control method of the stepper includes the followings steps. A plurality of sample development patterns are obtained according to a plurality of sample focal length values. A plurality of generative categories corresponding to a plurality of generative focal length values are generated by using a depth learning algorithm. An online development pattern is obtained. An estimated focal length value of the online development pattern is analyzed according to the generative categories.

The above and other aspects of the invention will become better understood with regards to the following detailed description of the preferred but non-limiting embodiment(s). The following description is made with reference to the accompanying drawings.

FIG. 1 is a schematic diagram of a control equipment of a stepper according to an embodiment.

FIG. 2 is a flowchart of a control method of the stepper according to an embodiment.

FIG. 3 is a schematic diagram of each step of FIG. 2 .

FIG. 4 is a flowchart of a control method of the stepper according to another embodiment.

FIG. 5 is a schematic diagram of each step of FIG. 4 .

FIG. 6 is a flowchart of a control method of the stepper according to another embodiment.

FIG. 7 is a schematic diagram of each step of FIG. 6 .

Referring to FIG. 1 , a schematic diagram of a control equipment 100 of a stepper 700 according to an embodiment is shown. The control equipment 100 includes an input device 110 , a generating device 120 and a processing device 130 . The input device 110 is configured to input various types of data, and can be realized by such as a wired network connection, a wireless network transceiver module, a transmission line, or a USB port. The generating device 120 is configured to perform various training/generating procedures. The processing device 130 is configured to perform various processing/analyzing procedures. The generating device 120 and the processing device 130 can be realized by such as a circuit, a circuit board, a chip, a plurality of programming codes or a recording device for storing programming codes.

In the present embodiment, after the stepper 700 exposes and develops the photoresist layer on the production line, the scanning electron microscope (SEM) 800 captures and inputs an online development pattern P 70 to the control equipment 100 . The control equipment 100 can analyze the online development pattern P 70 to obtain an estimated focal length value F 7 by using the artificial intelligence technology. Thus, the beam focal length of the stepper 700 can be calibrated according to the estimated focal length value F 7 , such that the beam of the stepper 700 can be precisely focused on the photoresist layer and the precision of the stepper 700 can be increased. The operations of each element disclosed above are described below with accompanying flowcharts.

Refer to FIG. 2 and FIG. 3 . FIG. 2 is a flowchart of a control method of the stepper 700 according to an embodiment. FIG. 3 is a schematic diagram of each step of FIG. 2 . First, the method begins at step S 110 , a plurality of sample development patterns P 10 are obtained according to a plurality of sample focal length values F 1 (illustrated in FIG. 3 ) by the input device 110 . The sample focal length values F 1 are set by the stepper 700 , and different sample development patterns P 10 can be obtained according to different sample focal length values F 1 . The sample development patterns P 10 are obtained by the scanning electron microscope 800 by capturing the images of the photoresist layer which has been exposed and developed by the stepper 700 . Since the stepper 700 has limited number of calibration scales with respect to the sample focal length values F 1 , the number of the sample development patterns P 10 corresponding to the sample focal length values F 1 is also limited.

Then, the method proceeds to step S 120 , a plurality of generative categories CF 1 corresponding to a plurality of generative focal length values F 1 ′ are generated by the generating device 120 using a depth learning algorithm. The depth learning algorithm includes an auto-encoder algorithm and a generative adversarial network algorithm (GAN algorithm). As indicated in FIG. 3 , in the present step, the sample development patterns P 10 are converted into a plurality of encoded patterns P 11 by using the auto-encoder algorithm and the generative adversarial network algorithm. During the process of generating the encoded patterns P 11 , the generative categories CF 1 corresponding to different generative focal length values F 1 ′ are categorized and stored in a database 900 .

That is, the generative focal length values F 1 ′ are not set by the stepper 700 but are generated by using the auto-encoder algorithm and the generative adversarial network algorithm. The scale of the generative focal length values F 1 ′ is finer than that of the sample focal length values F 1 . That is, the number of the generative focal length values F 1 ′ is larger than that of the sample focal length values F 1 .

For example, the setting of the stepper 700 is limited to 10 sets of sample focal length values F 1 , and only 10 sample development patterns P 10 are obtained in step S 110 . In step S 120 , 100 generative categories CF 1 corresponding to 100 generative focal length values F 1 ′ can be generated by the generating device 120 .

Then, the method proceeds to step S 130 , a to-be-tested online development pattern P 70 is obtained by the input device 110 operating on the production line. The online development pattern P 70 is obtained by the scanning electron microscope 800 by capturing the images of the photoresist layer which has been exposed and developed by the stepper 700 .

Then, the method proceeds to step S 140 , the estimated focal length value F 7 of the online development pattern P 70 is analyzed by the processing device 130 according to the generative category CF 1 . As indicated in FIG. 3 , in the present step, the processing device 130 performs analysis according to the approximation between a hidden layer of the online development pattern P 70 and a hidden layer of each generative category CF 1 . That is, the online development pattern P 70 is converted into an encoded pattern P 71 by using the encoder algorithm and the generative adversarial network algorithm, and during the conversion process, the characteristics of the hidden layer HL 71 is compared with the characteristics of the hidden layer of the generative category CF 1 to find a set of best approximated generative category CF 1 . After the set of best approximated generative category CF 1 is obtained, its corresponding generative focal length value F 1 ′ is regarded as an estimated focal length value F 7 .

Then, the method proceeds to step S 150 , a control command CS 1 is outputted to the stepper 700 by the processing device 130 according to the estimated focal length value F 7 to calibrate the beam focal length of the stepper 700 .

With the said artificial intelligence technology, the beam focal length of the stepper 700 can be calibrated to the most accurate value, such that the precision of the stepper 700 can be greatly increased.

Apart from calibrating the beam focal length of the stepper 700 , the artificial intelligence technology of the present disclosure can be used to calibrate the exposure energy of the stepper 700 as well. Refer to FIG. 4 and FIG. 5 . FIG. 4 is a flowchart of a control method of the stepper 700 according to another embodiment. FIG. 5 is a schematic diagram of each step of FIG. 4 . First, the method begins at step S 210 , a plurality of sample development patterns P 10 are obtained by the input device 110 according to a plurality of sample focal length values F 1 and a plurality of sample energy values E 1 . The sample focal length values F 1 and the sample energy values E 1 are set by the stepper 700 . Different sample development patterns P 10 can be obtained according to different sample focal length values F 1 and sample energy values E 1 . The sample focal length values F 1 and the sample energy values E 1 form a matrix. Since the stepper 700 has limited number of calibration scales with respect to the sample focal length values F 1 and the sample energy values E 1 , the number of the sample development patterns P 10 corresponding to the sample focal length values F 1 and the sample energy values E 1 are also limited.

Then, the method proceeds to step S 220 , a plurality of generative categories CF 2 corresponding to a plurality of generative focal length values F 1 ′ and a plurality of generative energy values E 1 ′ are generated by the generating device 120 using a depth learning algorithm. In the present step, the sample development patterns P 10 are converted into a plurality of encoded patterns P 12 by using the auto-encoder algorithm and the generative adversarial network algorithm. During the process of generating the encoded patterns P 12 , a plurality of generative categories CF 2 corresponding to the generative focal length values F 1 ′ and the generative energy values E 1 ′ are classified. That is, the generative focal length values F 1 ′ and the generative energy values E 1 ′ are not set by the stepper 700 but are generated by using the auto-encoder algorithm and the generative adversarial network algorithm. The scale of the generative focal length values F 1 ′ is finer than that of the sample focal length values F 1 . That is, the number of the generative focal length values F 1 ′ is larger than that of the sample focal length values F 1 . The scale of the generative energy values E 1 ′ is finer than that of the sample energy values E 1 . That is, the number of the generative energy values E 1 ′ is larger than that of the sample energy values E 1 .

For example, the setting of the stepper 700 is limited to 10 sets of sample focal length values F 1 and 10 sets of sample energy value E 1 (form a 10*10 matrix), and only 100 sample development patterns P 10 can be obtained in step S 210 . In step S 120 , 10000 generative categories CF 2 corresponding to 100 generative focal length values F 1 ′ and 100 sets of generative energy values E 1 ′ (form a 100*100 matrix) can be generated by the generating device 120 .

Then, the method proceeds to step S 230 , a to-be-tested online development pattern P 70 is obtained by the input device 110 operating on the production line. The online development pattern P 70 is obtained by the scanning electron microscope 800 by capturing the images of the photoresist layer which has been exposed and developed by the stepper 700 .

Then, the method proceeds to step S 240 , the estimated focal length value F 7 and an estimated energy value E 7 of the online development pattern P 70 are analyzed by the processing device 130 according to the generative category CF 2 . As indicated in FIG. 5 , in the present step, the processing device 130 performs analysis according to the approximation between a hidden layer of the online development pattern P 70 and a hidden layer of each generative category CF 2 . That is, the online development pattern P 70 is converted into an encoded pattern P 72 by using the encoder algorithm and the generative adversarial network algorithm. During the process, the characteristics of the hidden layer HL 72 is compared with of the characteristics of the hidden layer of the generative category CF 2 to find a set of best approximated generative category CF 2 . After the set of best approximated generative category CF 2 is obtained, the generative focal length value F 1 ′ and the generative energy value E 1 ′ corresponding to the set of best approximated generative category CF 2 are regarded as the estimated focal length value F 7 and the estimated energy value E 7 .

Then, the method proceeds to step S 250 , the control command CS 2 is outputted to the stepper 700 by the processing device 130 according to the estimated focal length value F 7 and the estimated energy value E 7 to calibrate the beam focal length and the exposure energy of the stepper 700 .

With the artificial intelligence technology, the beam focal length and the exposure energy of the stepper 700 can be calibrated to the most accurate value, such that the precision of the stepper 700 can be greatly increased.

Besides, the artificial intelligence technology of the present disclosure can be used to calibrate the exposure energy of the stepper 700 only. Refer to FIG. 6 and FIG. 7 . FIG. 6 is a flowchart of a control method of the stepper 700 according to another embodiment. FIG. 7 is a schematic diagram of each step of FIG. 6 . Firstly, the method begins at step S 310 , a plurality of sample development patterns P 10 are obtained by the input device 110 according to a plurality of sample energy values E 1 . The sample energy values E 1 are pre-set by the stepper 700 , and different sample development patterns P 10 can be obtained according to different sample energy values E 1 . Since the stepper 700 has limited number of calibration scales with respect to the sample energy value E 1 , the number of the sample development patterns P 10 corresponding to the sample energy value E 1 is also limited

Then, the method proceeds to step S 320 , a plurality of generative categories CF 3 corresponding to a plurality of generative energy values E 1 ′ are generated by the generating device 120 using a depth learning algorithm. In the present step, the sample development patterns P 10 are converted into a plurality of encoded patterns P 13 by using an auto-encoder algorithm and a generative adversarial network algorithm. During the process of generating encoded patterns P 13 , a plurality of generative categories CF 3 corresponding to the generative energy values E 1 ′ are classified. That is, the generative energy values E 1 ′ are not set by the stepper 700 but are generated by using the auto-encoder algorithm and the generative adversarial network algorithm. The scale of the generative energy values E 1 ′ is finer than that of the sample energy values E 1 . That is, the number of the generative energy values E 1 ′ is larger than that of the sample energy values E 1 .

For example, the setting of the stepper 700 is limited to 10 sets of sample energy values E 1 , and only 10 sample development patterns P 10 can be obtained in step S 310 . In step S 320 , 100 generative categories CF 3 corresponding to 100 sets of generative energy values E 1 ′ can be generated by the generating device 120 .

In step S 330 , a to-be-tested online development pattern P 70 is obtained by the input device 110 operating on the production line. The online development pattern P 70 is obtained by the scanning electron microscope 800 by capturing the images of the photoresist layer which has been exposed and developed by the stepper 700 .

Then, the method proceeds to step S 340 , the estimated energy value E 7 of the online development pattern P 70 is analyzed by the processing device 130 according to the generative category CF 3 . As indicated in FIG. 5 , in the present step, the processing device 130 performs analysis according to the approximation between a hidden layer of the online development pattern P 70 and a hidden layer of each generative category CF 3 . That is, the online development pattern P 70 is converted into an encoded pattern P 73 by using the encoder algorithm and the generative adversarial network algorithm. During the process, the characteristics of the hidden layer HL 73 is compared with the characteristics of the hidden layer of the generative category CF 3 to find a set of best approximated generative category CF 3 . After the set of best approximated generative category CF 3 is obtained, the generative energy value E 1 ′ corresponding to the set of best approximated generative category CF 3 is regarded as an estimated energy value E 7 .

Then, the method proceeds to step S 350 , the control command CS 3 is outputted to the stepper 700 by the processing device 130 according to the estimated energy value E 7 to calibrate the beam focal length and the exposure energy of the stepper 700 .

With the artificial intelligence technology, the exposure energy of the stepper 700 can be calibrated to the most accurate value, such that the precision of the stepper 700 can be greatly increased.

While the invention has been described by way of example and in terms of the preferred embodiment(s), it is to be understood that the invention is not limited thereto. On the contrary, it is intended to cover various modifications and similar arrangements and procedures, and the scope of the appended claims therefore should be accorded the broadest interpretation so as to encompass all such modifications and similar arrangements and procedures.