Control method of multi-stage etching process and processing device using the same

Abstract

A method for controlling a multi-stage etching process and an apparatus for using the same are disclosed. The method for controlling a multi-stage etching process includes the following steps: setting stack composition information for one of several hard masking layers; setting an etching target condition; and generating parameter setting information for one of these hard masking layers using a machine learning model, provided the etching target condition is met. The machine learning model is trained based on the stack composition information of these hard masking layers, several process parameters, and a process result.

Description

Technical Field

[0001] The present disclosure relates to a control method for a semiconductor process and an electronic device using the same, and more particularly to a control method for a multi-stage etching process and a processing device using the same. Prior Art

[0002] As semiconductor devices become increasingly finer in size, multi-stage etching technology has been developed to reduce the etching width. In multi-stage etching technology, multiple hard mask layers are stacked and etched step by step.

[0003] Traditionally, in a multi-stage etching process, one or two hard mask layers are primarily monitored in the hope of achieving the final etching target. However, with the rapid advancement of semiconductor products, the hard mask layers used in multi-stage etching techniques have a wide variety of material variations, and the stacking relationships of the layers are also different. Therefore, it is no longer possible to monitor these two hard mask layers.

[0004] In a multi-stage etching process, the materials, etching methods, and process conditions of each hard mask layer contribute to variations, all of which can potentially impact the final etching target. Engineers struggle to manage these numerous variations, creating a bottleneck in the development of multi-stage etching technology. Summary of the Invention

[0005] This disclosure relates to a control method for a multi-stage etching process and a processing device employing the same. Using information about the stacked composition of hard mask layers and the set etching target conditions, a machine learning model generates parameter setting information for each hard mask layer while satisfying the etching target conditions. This machine learning technology accurately controls numerous factors of variation, breaking through the technological bottleneck of multi-stage etching technology.

[0006] According to one aspect of the present disclosure, a method for controlling a multi-stage etching process is provided. The method includes the following steps: Setting stack composition information for one of a plurality of hard mask layers; Setting an etching target condition; and, using a machine learning model, generating parameter setting information (parameter setting recipe) for the hard mask layers while satisfying the etching target condition. The machine learning model is trained based on the stack composition information for the hard mask layers, a plurality of process parameters, and a process result.

[0007] According to another aspect of the present disclosure, a processing device is provided. The processing device is used to control a multi-stage etching process. The processing device includes a multi-layer setting unit, a target setting unit, and a machine learning model. The multi-layer setting unit is used to set stacking composition information of a plurality of hard mask layers. The target setting unit is used to set an etching target condition. The machine learning model is used to generate parameter setting information (parameter setting recipe) for the hard mask layers when the etching target condition is satisfied. The machine learning model is trained based on the stacking composition information of the hard mask layers, a plurality of process parameters, and a process result.

[0008] In order to better understand the above and other aspects of the present disclosure, the following embodiments are specifically described in detail with reference to the accompanying drawings: Simple diagram description

[0009] FIG. 1 is a schematic diagram illustrating a multi-step etching process according to one embodiment. Figure 2 illustrates an example of how to set the electrostatic chuck temperature. FIG3 is a schematic diagram of a processing device according to an embodiment. Figure 4 illustrates an example of online operation of a machine learning model. FIG5 is a block diagram of a processing device according to an embodiment. FIG. 6 illustrates a method for controlling a multi-stage etching process according to one embodiment. Implementation Method

[0010] Please refer to Figure 1, which illustrates a multi-stage etching process according to one embodiment. To achieve the target etching condition of a narrow etching width CD1 in the target layer TG, a multi-stage etching of the hard mask layer HMi is required. During the etching of the hard mask layer HMi, various variation factors must be controlled, such as material, thickness, gas flow rate, bias RF power, and electrostatic chuck temperature.

[0011] Refer to Figure 2, which illustrates the electrostatic chuck temperature setting. Taking the electrostatic chuck temperature as an example, the electrostatic chuck EC is divided into zones Z1 through Z4. Each zone Z1 through Z4 requires a specific temperature setting. When etching different hard mask layers HMi, the temperature of each zone Z1 through Z4 must be adjusted accordingly. Consequently, the electrostatic chuck temperature setting alone creates multiple variables that affect the target etching conditions. In this situation, achieving the target etching conditions by simply monitoring specific steps becomes difficult.

[0012] In addition to the electrostatic chuck temperature, the material and thickness of each hard mask layer HMi, the gas flow rate used to etch each hard mask layer HMi, and the RF power bias used to etch each hard mask layer HMi all contribute to several variables that affect the target etching conditions. In this situation, achieving the target etching conditions by monitoring specific steps becomes difficult.

[0013] In this embodiment, a machine learning technique is proposed to accurately set these numerous factors of variation. Through machine learning, the numerous factors of variation can be accurately set.

[0014] Please refer to FIG. 3 , which illustrates a schematic diagram of a processing apparatus 100 according to one embodiment. A wafer 900 undergoes an etching process for each hard mask layer HMi in a semiconductor machine 200. Stack composition information ST of the hard mask layer HMi, process parameters PRi of each etching step, and process results RS form a set of training data. By performing multiple etching processes on the same product or different products, multiple sets of stack composition information ST, process parameters PRi, and process results RS can be collected, forming a large dataset.

[0015] The processing device 100 of this embodiment performs training by stacking a large data set of composition information ST, process parameters PRi, and process results RS to obtain a machine learning model 130.

[0016] Please refer to Figure 4, which illustrates an example of the online operation of the machine learning model 130. After the machine learning model 130 is trained, specific stack composition information ST* and etching target conditions RS* can be input into the machine learning model 130 to obtain an appropriate set of parameter setting information PRi*. The etching target condition RS* can be, for example, an etching rate, an edge profile, or a critical dimension. An edge profile can be, for example, the angle of the etched sidewall. A critical dimension can be, for example, the etching width CD1 described above. The parameter setting information PRi* can be, for example, a gas flow rate, a bias RF power, or an electrostatic chuck temperature. In this way, machine learning technology can accurately set a variety of factors of variation.

[0017] Please refer to FIG. 5 , which illustrates a block diagram of a processing device 100 according to one embodiment. The processing device 100 includes a multi-layer setting unit 110, a target setting unit 120, and a machine learning model 130. The multi-layer setting unit 110, the target setting unit 120, and the machine learning model 130 may be, for example, a circuit, a chip, a circuit board, a program code, or a storage device storing program code. The processing device 100 of this embodiment can use the multi-layer setting unit 110 to set the stacking composition information ST* of the hard mask layer HMi, and use the target setting unit 120 to set the etching target condition RS*. The machine learning model 130 can then generate parameter setting information PRi* for each hard mask layer HMi when the etching target condition RS* is satisfied. The operation of each component is described in detail below with the aid of a flowchart.

[0018] Referring to FIG. 6 , a method for controlling a multi-stage etching process according to one embodiment is illustrated. In step S110 , the multi-layer configuration unit 110 configures stacking composition information ST* for a plurality of hard mask layers HMi. The stacking composition information ST* includes, for example, the material of each hard mask layer HMi, the thickness of each hard mask layer HMi, and the stacking order of the hard mask layers HMi.

[0019] Next, in step S120, the target setting unit 120 sets an etching target condition RS*. The etching target condition RS* may be, for example, an etching rate, an edge profile, or a critical dimension. In this step, the etching target condition RS* may be for a single hard mask layer HMi. Alternatively, the etching target condition RS* may be for several adjacent and stacked layers of the hard mask layers HMi. Alternatively, the etching target condition RS* may be for all hard mask layers HMi.

[0020] Then, in step S130, the machine learning model 130 generates parameter setting information PRi* for the hard mask layers HMi while satisfying the etching target conditions RS*. The parameter setting information PRi* may include, for example, gas flow rate, RF bias power, electrostatic chuck temperature, or a combination thereof. In this step, the parameter setting information PRi* may be for a single hard mask layer HMi. Alternatively, the parameter setting information PRi* may be for several adjacent and stacked layers of the hard mask layers HMi. Alternatively, the parameter setting information PRi* may be for all hard mask layers HMi.

[0021] According to the above embodiment, using the stack composition information ST* of the hard mask layer HMi and the etching target condition RS*, the machine learning model 130 can generate parameter setting information PRi* for these hard mask layers HMi while satisfying the etching target condition RS*. Through machine learning technology, numerous factors of variation can be accurately determined, breaking through the technical bottleneck of multi-stage etching technology.

[0022] In previous multi-stage etching techniques, parameters were controlled only for each hard mask layer HMi in order to achieve the desired etching result for that hard mask layer HMi. However, this embodiment utilizes big data to string together stacked hard mask layers HMi into stacking composition information ST. Deep learning technology also incorporates the stacking relationship of the hard mask layers HMi into analysis as a reference.

[0023] In particular, when there is no historically identical stacking composition information ST, conventional techniques are unable to successfully analyze the parameter setting information PRi*. This embodiment utilizes big data fusion, allowing any similar pieces of stacking composition information ST in historical data to be incorporated into the training of the machine learning model 130. Even when there is no historically identical stacking composition information ST, the machine learning model 130 of this embodiment can still successfully analyze the parameter setting information PRi*.

[0024] Furthermore, in previous multi-stage etching techniques, only a single hard mask layer HMi could be monitored for etching results. In this embodiment, the machine learning model 130, trained using big data, can arbitrarily define which hard mask layers HMi (or even all) are targeted by the etching target condition RS*, resulting in more stable control of the critical hard mask layer HMi.

[0025] Furthermore, in previous multi-stage etching techniques, only the process parameters of a single hard mask layer HMi could be monitored. In this embodiment, the machine learning model 130, trained using big data, can arbitrarily determine which hard mask layers HMi (or even all of them) the output parameter setting information PRi* targets, resulting in more stable control of the etching width CD1 of the target layer TG.

[0026] In summary, although the present disclosure has been described above with reference to the embodiments, these are not intended to limit the present disclosure. Those skilled in the art will readily appreciate that various modifications and variations can be made without departing from the spirit and scope of the present disclosure. Therefore, the scope of protection of the present disclosure shall be determined by the appended patent claims.

[0027] 100: Processing device 110: Multi-layer setting unit 120: Goal Setting Unit 130: Machine Learning Model 200: Semiconductor machine 900: Wafer CD1: Etching width EC: Electrostatic Chuck HMi: Hard Mask Layer PRi: Process parameters PRi*: parameter setting information RS: Process Results RS*: Etch target condition S110, S120, S130: Steps ST, ST*: Stacking composition information TG: Target layer Z1, Z2, Z3, Z4: Zones

Claims

1. A method for controlling a multi-stage etching process, comprising: The system receives stack composition information of one of a plurality of hard mask layers; receives an etching target condition; and, through a machine learning model, receives the stack composition information and the etching target condition to generate parameter setting information of one of the hard mask layers through machine learning technology, wherein the machine learning model is trained based on the stack composition information of the hard mask layers, a plurality of process parameters and a process result.

2. The method for controlling a multi-stage etching process as described in claim 1, wherein the etching target condition is an etching rate, an edge profile, or a critical dimension.

3. The control method for the multi-stage etching process as described in claim 1, wherein the parameter setting information is a gas flow rate, a bias RF power, or an E-chuck temperature.

4. The method for controlling a multi-stage etching process as described in claim 1, wherein the etching target conditions are for a single layer of the hard mask layers.

5. A method for controlling a multi-stage etching process as described in claim 1, wherein the etching target conditions are for several adjacent and stacked layers of the hard mask layers.

6. A method for controlling a multi-stage etching process as described in claim 1, wherein the etching target conditions are for all of the hard mask layers.

7. The method for controlling a multi-stage etching process as described in claim 1, wherein the parameter setting information is for a single layer of the hard mask layers.

8. The method for controlling a multi-stage etching process as described in claim 1, wherein the parameter setting information is for several adjacent and stacked layers of the hard mask layers.

9. A method for controlling a multi-stage etching process as described in claim 1, wherein the parameter setting information is for all of the hard mask layers.

10. A processing apparatus for controlling a multi-stage etching process, the processing apparatus comprising: A multi-layer setting unit is used to set the stacking composition information of one of a plurality of hard mask layers; A target setting unit is used to set an etching target condition; and a machine learning model is used to receive the stack composition information and the etching target condition to generate parameter setting information (parameter setting recipe) for the hard mask layers through machine learning technology, wherein the machine learning model is trained based on the stack composition information of the hard mask layers, a plurality of process parameters and a process result.

11. The processing apparatus as claimed in claim 10, wherein the etching target condition is an etching rate, an edge profile, or a critical dimension.

12. The processing apparatus as described in claim 10, wherein the parameter setting information is a gas flow rate, a bias RF power, or an E-chuck temperature.

13. The processing apparatus as described in claim 10, wherein the etching target conditions are for a single layer of the hard masking layers.

14. The processing apparatus as claimed in claim 10, wherein the etching target conditions are for adjacent and stacked layers of the hard mask layers.

15. The processing apparatus as described in claim 10, wherein the etching target conditions are applied to all of the hard mask layers.

16. The processing apparatus as described in claim 10, wherein the parameter setting information is for a single layer of the hard masking layers.

17. The processing apparatus as described in claim 10, wherein the parameter setting information is for several adjacent and stacked layers of the hard masking layers.

18. The processing apparatus as described in claim 10, wherein the parameter setting information is for all of the hard mask layers.

Images