KR102587791B1 - Simulation technique for Atomic Layer Deposition for Materials with Narrow Channel or Porous Materials - Google Patents

Abstract

The present invention builds a simulation model by subdividing the model in a wide-area reactor containing many fine areas of interest, and builds a database through machine learning methodology to quickly obtain analysis results and have a micro-channel to implement high-speed computation. This relates to an atomic layer deposition simulation method for a deposited object having a porous material.
The atomic layer deposition simulation method for a deposited object having a microchannel or a porous material of the present invention includes a computational fluid analysis model machine learning step of building a computational fluid analysis model consisting of a global model and a local model and performing machine learning; It includes an atomic layer deposition analysis step of performing atomic layer deposition analysis of a deposited object having at least two or more microchannels or a porous material based on the results of machine learning.

Description

Atomic layer deposition simulation method for deposited materials with microchannels or porous materials {Simulation technique for Atomic Layer Deposition for Materials with Narrow Channel or Porous Materials}

The present invention relates to an atomic layer deposition simulation method for a deposited object having microchannels or a porous material. Specifically, the present invention relates to a method for atomic layer deposition simulation on a deposited object having a large area but patterned microchannels or having porous properties. The goal is to implement computational thermal fluid analysis at high speed for the Atomic Layer Deposition (ALD) process and match it to material processing equipment to perform the work efficiently and smoothly.

In thermal fluid-based material processes such as thin film growth, it requires a lot of trial and error to completely experimentally optimize various parameters such as thermal fluid composition, temperature, fluid velocity, heater power, hot zone composition, and material synthesis location. In particular, in many cases, the reactor area where material processing occurs cannot be visually monitored from the outside, so the process is controlled depending on various process signals mounted on the equipment. Accordingly, simulation technology based on Computational Fluid Dynamics is widely used in the optimization of these thermal fluid reactors.

Meanwhile, computational fluid dynamics simulation techniques use the Navier-Stokes equations, which are nonlinear partial differential equations that describe fluid phenomena, to various numerical analysis techniques, such as the finite difference method, finite element method, and finite volume method ( This refers to converting to algebraic equations by discretizing them using methods such as the Finite Volume Method, and solving and analyzing fluid flow problems using numerical algorithm.

Recently, in addition to fluid analysis, multi-physics analysis that simultaneously analyzes various physical phenomena such as mass transfer, chemical reaction, and heat transfer has become common, and various general-purpose or dedicated simulations support this. The package is commercially available. While dedicated simulation packages that are specialized for specific analysis problems cannot respond to a variety of problems and have limitations in that users cannot directly change or add code, they have the advantage of being easy to use for users who do not have a deep understanding of simulation. A general-purpose simulation package, on the other hand, can add and change various user codes, while developing a solution for analyzing a specific problem requires a lot of cost and effort. The analysis method of this patent therefore refers to an analysis method utilizing a general-purpose simulation package.

In such fluid analysis simulations, the domain that is the object of analysis is divided into a finite number of elements for analysis. In general, fluid analysis requires a fairly dense element network. The mesh must be constructed particularly densely in areas of high interest, but obtaining data in a fine area precisely by configuring the entire area tightly within a large area reactor takes too much calculation time and the possibility of errors during the analysis process is high.

For example, in Republic of Korea Patent Publication No. 10-2007-0013839, the growth mechanism of a thin film is modeled and deposition is performed using a single precursor using Atomic Layer Deposition (ALD), one of the nano thin film deposition technologies. One, the thickness of a single-component thin film and a two-component thin film containing an oxide film, a nitride film, a sulfide film, etc., and the thickness of a multi-component thin film containing at least three or more components deposited sequentially by time-sharing supply of two or more precursors, and This relates to a process simulation method for atomic layer deposition to predict composition.

However, in computational fluid analysis, the mesh must be constructed particularly densely in areas of high interest, but it is necessary to collect data in a fine area within a large area of the reactor, such as when depositing materials with micro channels or porous materials through the ALD process. In order to obtain it precisely, the mesh model of the entire model must also be tightly constructed, so it takes too much calculation time and the possibility of errors in the analysis process is high, so a special approach is required.

Republic of Korea Patent Publication No. 10-2007-0013839 (2007.01.31)

The purpose of the present invention is to provide an atomic layer deposition simulation method for a deposited object having a microchannel or a porous material in which a simulation model is constructed by subdividing the model in a wide-area reactor containing a large number of microscopic areas of interest. .

Another purpose of the present invention is to provide an atomic layer deposition simulation method for deposited materials with microchannels or porous materials that quickly obtains analysis results and implements high-speed calculations by building a database through machine learning methodology. There is.

The atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to the present invention involves a computational fluid analysis model machine learning step in which a computational fluid analysis model consisting of a global model and a local model is constructed and machine learning is performed. And it may include an atomic layer deposition analysis step of performing atomic layer deposition analysis of a deposited object having at least two or more microchannels or a porous material based on the results of machine learning.

Here, the computational fluid analysis model machine learning step is to secure the accuracy of the computational fluid analysis model by obtaining experimental results that match the actual data using the computational fluid analysis model, and then using the computational fluid analysis model to secure the accuracy of the computational fluid analysis model at the fluid inlet and outlet of the micro channel or porous material. The fluid flow is derived as a global model and applied to the local model, and the detailed fluid behavior inside the micro channel is precisely analyzed. Meanwhile, machine learning is performed on the fluid flow within the micro channel according to the fluid pressure of the fluid inlet and outlet. The results can be saved to the database.

Additionally, in the atomic layer deposition analysis step, the atomic layer deposition analysis can be performed by extracting the characteristics of the fluid flow in the fluid inlet and outlet of the microchannel or porous material from the database.

Here, to ensure accuracy, a computational fluid analysis model can be adopted in which the crystal growth rate value for each location (growth rate per cycle) matches the experimental data and the reference value or more by considering physical phenomena, mesh size, and surface reaction.

In addition, in the computational fluid analysis model machine learning stage, the global model can perform an overall reactor analysis using the Navier-Stokes equations to derive a flow profile for the boundary of a highly periodic aspect ratio porous medium.

Here, in the computational fluid analysis model machine learning stage, the local model can be analyzed using Darcy's law, an analysis method for porous media, to analyze the microchannel of interest.

In addition, machine learning can use an artificial neural network that applies temperature, pressure, mole fraction of carrier gas, mole fraction of precursor, diffusion rate of carrier gas, diffusion rate of precursor, and permeability as input to the input layer.

Here, machine learning can use an artificial neural network that applies at least two or more hidden layers.

Additionally, machine learning can use an artificial neural network that applies the surface coverage of a deposited object having a microchannel or porous material as an output layer.

Here, the surface coverage may include the surface coverage trend in the hole depth direction depending on the pressure at the starting and ending points of the deposited material having a microchannel or porous material. there is.

The atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to the present invention has the advantage of constructing a simulation model by subdividing the model in a wide area reactor including many fine areas of interest.

In addition, the atomic layer deposition simulation method for deposited objects with microchannels or porous materials according to the present invention has the advantage of quickly obtaining analysis results and implementing high-speed calculations by building a database through machine learning methodology. There is.

Figure 1 is a flowchart showing an atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to an embodiment of the present invention.
FIG. 2 is a diagram showing a showerhead type atomic layer deposition process system as an example to which the computational fluid model of FIG. 1 can be applied.
Figure 3 is a diagram showing an equivalent model using Darcy's law, which is an analysis method for porous media, as an example of an analysis method for the local model of Figure 1.
Figure 4 is a configuration diagram showing the machine learning steps of the computational fluid analysis model of Figure 1 in detail.
Figure 5 is a configuration diagram showing the atomic layer deposition analysis steps of Figure 1 in detail.
Figure 6 is an example of building a global model as an example of the computational fluid model of Figure 1.
Figure 7 is an example of a neural network model for machine learning to be used to speed up the global-local model integrated analysis in Figure 1.
Figure 8 is an example of performing analysis using the results of machine learning in Figure 1.

Specific embodiments for carrying out the present invention will be described with reference to the attached drawings.

Since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. This is not intended to limit the present invention to specific embodiments, and may be understood to include all changes, equivalents, and substitutes included in the spirit and technical scope of the present invention.

Hereinafter, an atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to the present invention will be described in detail with reference to the attached drawings.

Figure 1 is a flowchart showing an atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to an embodiment of the present invention, and Figures 2 to 8 are detailed drawings for explaining Figure 1 in detail. Configuration diagram and example.

Hereinafter, an atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to an embodiment of the present invention will be described with reference to FIGS. 1 to 8.

First, referring to FIG. 1, the atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to an embodiment of the present invention is a computational method consisting of a global model 100 and a local model 200. It includes a step of building a fluid analysis model and performing machine learning (S100), and a step of performing atomic layer deposition analysis of a deposited object having at least two or more microchannels or a porous material based on the results of the machine learning (S200). can do.

Here, in the computational fluid analysis model machine learning step (S100), the accuracy of the computational fluid analysis model is secured by obtaining experimental results that match the actual data using the computational fluid analysis model, and then the fluid inlet of the micro channel or porous material is The fluid flow in the discharge part is derived as a global model (100) and applied to the local model (200), and the detailed fluid behavior inside the micro channel is precisely analyzed, while the fluid flow in the micro channel is measured at the fluid inlet and outlet. The results of machine learning based on fluid pressure are stored in the database 300.

In addition, in the atomic layer deposition analysis step (S200), the characteristics of the fluid flow in the fluid inlet and outlet of the microchannel or porous material are extracted from the database 300 and the atomic layer deposition analysis is performed.

Here, to ensure accuracy, a computational fluid analysis model is adopted in which the crystal growth rate value for each location (growth rate per cycle) matches the experimental data and the reference value or more, taking into account physical phenomena, mesh size, and surface reaction.

Artificial intelligence using machine learning has also begun to be applied to thermal fluid analysis, and when machine learning is applied to thermal fluid analysis, the calculation speed can be dramatically improved in multiphysics analysis based on existing computational fluid analysis. It is reported that there is. Tsunooka et al. have implemented a high-speed simulation technique that is 107 (10 million) times faster than the existing simulation method using Computational Fluid Dynamics (CFD) based on the solution growth process of a SiC single crystal using machine learning, and this method was determined. It was suggested that it can be widely used in the design and control of the growth process (Y. Tsunooka, N. Kokubo, G. Hatasa, S. Harada, M. Tagawa, T. Ujihara, CrystEngComm 2018, 20, 6546).

Therefore, the present invention focuses on improving the calculation speed using machine learning, builds a computational fluid analysis model consisting of a global model (100) and a local model (200), and uses this to obtain experimental results that match the actual measurement data. After ensuring the accuracy of the model, in the computerized fluid analysis model machine learning step (S100), the fluid flow in the fluid inlet and outlet of each micro channel or porous material is derived as a global model (100) and used as a local model ( 200) and precisely analyze the detailed fluid behavior inside the channel, while converting the fluid flow within the micro channel into a database through machine learning according to the fluid pressure of the fluid inlet and outlet in the atomic layer deposition analysis step (S200). , it is possible to speed up atomic layer deposition analysis of patterned microchannels or deposited objects with porous materials in a short period of time just by analyzing the global model 100.

For example, in the computational fluid analysis model machine learning step (S100), in the case of an atomic layer deposition simulation for an aluminum plate with a micro channel, input parameters and process signals (process data) for comparison with the simulation data are used for the micro channel. It is obtained through an atomic layer deposition experiment in which an Al ₂ O ₃ thin film is deposited on a large-area aluminum plate.

Using the measured temperature, process pressure, and precursor cycle as input values, the thickness of the thin film applied on silicon is measured using an ellipsometer, and this is used as process data and compared with the input parameters of the simulation.

At this time, considering the applied physical phenomenon, mesh size, and surface reaction, a simulation model in which the crystal growth rate value (growth rate per cycle) for each location matches the experimental data by more than 90% is obtained. The above 90% is an exemplary figure, and these figures can be changed as needed.

Therefore, by verifying the simulation model based on experimental data, there is an advantage in being able to reliably build a computational fluid analysis model, and by storing and utilizing the results of machine learning in the database 300, patterned fluids can be created in a short time. Atomic layer deposition analysis of deposited objects with microchannels or porous materials can be speeded up.

FIG. 2 is a diagram showing a showerhead type atomic layer deposition process system as an example to which the computational fluid model of FIG. 1 can be applied. As can be seen in FIG. 2, the showerhead-type atomic layer deposition process system includes a fluid inlet 410 through which the fluid enters, a showerhead nozzle 460 through which the fluid passes, and a fluid through which the fluid flows. It consists of a discharge portion 430, a top wall 450 of the chamber equipped with a showerhead nozzle 460, and a hot plate 440 located at the bottom of the top wall 450 of the chamber. The microchannel 470 of the actual substrate shows a cross section of the actual showerhead nozzle 460.

The equivalent porous substrate 420 is an equivalent representation of a porous substrate for simulation.

That is, the fluid must be introduced into the fluid inlet 410 and pass through a plurality of showerhead nozzles 460. At this time, the pressure distribution is adjusted according to the distance between the inlet of the showerhead nozzle 460 and the fluid inlet 410. Because they are all different, simulations are very complex.

The present invention compares the experimental results based on the pressure at the fluid inlet 410 and the fluid outlet 430 for one showerhead nozzle 460, performs machine learning, and creates a database, thereby creating a pattern in a short time. It has the advantage of speeding up atomic layer deposition analysis of deposited objects with microchannels or porous materials.

FIG. 3 is a diagram showing an equivalent model using Darcy's law, which is an analysis method for porous media, as an example of an analysis method of the local model 200 of FIG. 1. As can be seen in FIG. 3, it is necessary to simplify it in order to analyze the porous medium. In the local model 200, the spacing between the plurality of showerhead nozzles 460, the diameter of the showerhead nozzles 460, and the showerhead By analyzing using Darcy's law, which is an analysis method for porous media, based on the length of the nozzle 460, using an equivalent model and interacting with the global model 100, it is possible to have patterned microchannels or porous materials in a short time. It has the advantage of speeding up the atomic layer deposition analysis of the deposited material.

Figure 4 is a configuration diagram showing the computerized fluid analysis model machine learning step (S100) of Figure 1 in detail.

As can be seen in Figure 4, in the computational fluid analysis model machine learning step (S100), the global model (100) performs the overall reactor analysis using the Navier-Stokes equations to analyze the high periodicity aspect ratio porous medium. Derive the flow profile for the boundary of .

Here, in the computational fluid analysis model machine learning step (S100), the local model 200 is analyzed using Darcy's law, which is an analysis method for porous media, to analyze the microchannel of interest.

In the computational fluid analysis model machine learning step (S100), the global model (100) performs an overall reactor analysis using the Navier-Stokes equations to derive a fluid flow profile for the boundary of a highly periodic aspect ratio porous medium. This can be done, and the local model 200 can perform analysis using Darcy's law, which is an analysis method for porous media, to analyze the microchannel of interest.

At this time, the model is updated by observing the surface reaction with experimental results, machine learning is performed, and the results are stored in the database 300, thereby performing an atomic layer deposition process on a large-area deposited object with patterned microchannels or porous properties. In computational fluid dynamics analysis of (Atomic Layer Deposition), there is an advantage in that atomic layer deposition analysis can be performed reliably.

Figure 5 is a configuration diagram showing the atomic layer deposition analysis step (S200) of Figure 1 in detail. As can be seen in Figure 5, the atomic layer deposition analysis step (S200) is a step of simulating the atomic layer deposition process using the results of machine learning in the computational fluid analysis model machine learning step (S100). As in the analysis model machine learning step (S100), the global model (100) performs an overall reactor analysis using the Navier-Stokes equations to determine the fluid flow profile for the boundary of a highly periodic high aspect ratio porous medium. It can be derived, and the local model 200 is already analyzed based on the results of machine learning stored in the database 300 according to the pressure distribution of the fluid inlet 410 and outlet 430 of the micro channel of interest. There is an advantage in that simulation of the entire reactor can be performed at high speed by using the obtained results.

Figures 6 to 8 show examples of metal oxide ALD deposition experiments using the methodology presented above, which are explained in detail in Figures 6 to 8.

FIG. 6 is an example illustrating a case of building a global model 100 as an example of the computational fluid model of FIG. 1. As can be seen in FIG. 6, FIG. 6 shows an example of construction of a global model 100 when TMA (TriaMethylAluminum), which is widely used as an Al source, is applied when growing an Al ₂ O ₃ layer through an ALD process. .

This is an experiment to verify the accuracy of the simulation model, which is influenced by various variables such as vapor pressure, diffusion rate, and growth characteristics of the precursor. The precursor injection time (pulse time) of 0.3 s is applied to the deposited object with the same pores as suggested based on TMA. A global model simulation was performed for the experiment applying and the pressure distribution was shown.

Here, the TMA partial pressure can be expressed in rainbow colors, with purple indicating lower pressure and red indicating higher pressure. Additionally, the TMA mole fraction can also be expressed in rainbow colors, with the color purple indicating a lower mole fraction and the more red indicating a higher mole fraction.

Meanwhile, in the global model 100, referring to the color representing the TMA mole fraction, when TMA flows in from the fluid inlet 410, the internal pressure of the showerhead nozzle 460 near the fluid inlet 410 is Although it is high, it can be seen that the internal pressure of the showerhead nozzle 460 is lower as the distance from the fluid inlet 410 increases. That is, it can be seen that the TMA mole fraction decreases as the distance R from the fluid inlet 410 increases to 0 mm, 35 mm, and 70 mm.

In addition, in the local model 200, the distance (R) with the fluid inlet 410 is located in the depth direction (hole) according to the pressure at the starting and ending points of the showerhead nozzle 460 located at 0 mm, 35 mm, and 70 mm. The trend of surface coverage in terms of depth is shown as an example of simulation results obtained for each location of the global model 100. The data of the local model 200 obtained in this way can be used as training data for machine learning based on various variables.

Figure 7 is an example of a neural network model for machine learning to be used to speed up the global-local model integrated analysis in Figure 1.

As can be seen in Figure 7, machine learning is an artificial learning method that applies temperature, pressure, mole fraction of carrier gas, mole fraction of precursor, diffusion rate of carrier gas, diffusion rate of precursor, and permeability as inputs of the input layer. Use a neural network.

Here, machine learning uses an artificial neural network that applies at least two or more hidden layers.

In addition, machine learning uses an artificial neural network that applies the surface coverage of the deposited material with microchannels or porous materials as an output layer.

Here, the surface coverage includes the surface coverage trend in the hole depth direction according to the pressure at the starting and ending points of the deposited material having a microchannel or porous material.

In other words, the artificial neural network is connected to an input layer, a hidden layer, and an output layer. The inputs of the input layer include temperature, pressure, mole fraction of carrier gas, mole fraction of precursor, diffusion rate of carrier gas, diffusion rate of precursor, and permeability (Permeability). ) is used, and a model that applies a two-layer hidden layer and obtains the surface coverage as an output layer is exemplified.

Here, the output layer is the surface coverage of the deposited object, and may include a surface coverage trend in the depth direction depending on the pressure at the starting and ending points of the deposited object having a microchannel and a porous material.

Meanwhile, the artificial neural network according to the present invention can change input data or add or subtract the number of hidden layers for each output data to be derived.

Figure 8 is an example of performing analysis using the results of machine learning in Figure 1. As can be seen in Figure 8, it shows an example of the analysis step that can be achieved when the high-speed local model results obtained through machine learning are converted into a database. The global model (100) and the local model (200) before applying machine learning are shown. By omitting the complex and time-consuming interdependent repetitive calculation process and simply analyzing the global model 100, it is converted into a numerical database and the data of the local model 200 stored in the database 300 is read. Interpretation of the material is possible.

Therefore, the atomic layer deposition simulation method for a deposited object having a micro channel or a porous material according to the present invention has the advantage of not only improving and speeding up the analysis, but also monitoring the deposition behavior of the equipment and the deposited object.

As described above, the atomic layer deposition simulation method for a deposited object having a microchannel or a porous material according to the present invention has the advantage of constructing a simulation model by subdividing the model in a wide area reactor including many fine areas of interest. It has the advantage of being able to quickly obtain analysis results and implement high-speed calculations by building a database through machine learning methodology.

What has been described above includes examples of one or more embodiments. Of course, for the purposes of describing the above-described embodiments, it is not possible to describe every possible combination of components or methods, and those skilled in the art will recognize that many additional combinations and permutations of the various embodiments are possible. Accordingly, the described embodiments are intended to include all alternatives, modifications and alterations that fall within the spirit and scope of the appended claims.

410: fluid inlet, 420: porous substrate
430: discharge unit 440: hot plate
450: Top wall 460: Showerhead nozzle
470: micro channel

Claims (10)

A computational fluid analysis model machine learning step in which a computational fluid analysis model consisting of a global model and a local model is built and machine learning is performed; and
An atomic layer deposition analysis step of performing atomic layer deposition analysis of a deposited object having at least two or more microchannels or a porous material based on the results of performing the machine learning; Atomic layer deposition simulation method. According to clause 1,
In the computational fluid analysis model machine learning step, after securing the accuracy of the computational fluid analysis model by obtaining experimental results that match actual data using the computational fluid analysis model, the fluid inlet of the micro channel or the porous material The fluid flow in the microchannel is derived from the global model and applied to the local model, and the detailed fluid behavior inside the microchannel is precisely analyzed, while the fluid flow in the microchannel is calculated from the fluid inlet and the outlet. An atomic layer deposition simulation method for a deposited object having a microchannel or a porous material, characterized in that the results of performing the machine learning according to negative fluid pressure are stored in a database. According to clause 2,
In the atomic layer deposition analysis step, the characteristics of the fluid flow in the fluid inlet and outlet of the micro channel or the porous material are extracted from the database and the atomic layer deposition analysis is performed. Atomic layer deposition simulation method for deposited materials or porous materials. According to clause 2,
Securing the accuracy is characterized by adopting the computational fluid analysis model in which the crystal growth rate value for each location (growth rate per cycle) matches the experimental data and the reference value or more, taking into account physical phenomena, mesh size, and surface reaction. Atomic layer deposition simulation method for deposited materials with channels or porous materials. According to clause 1,
In the computational fluid analysis model machine learning step, the global model performs an overall reactor analysis using Navier-Stokes equations to derive a flow profile for the boundary of a highly periodic aspect ratio porous medium. Atomic layer deposition simulation method for a deposited object having a microchannel or a porous material. According to clause 1,
In the computational fluid analysis model machine learning step, the local model has a microchannel or a porous material, characterized in that it is analyzed using Darcy's law, which is an analysis method for porous media, to analyze the microchannel of interest. Atomic layer deposition simulation method for deposited materials. According to clause 1,
The machine learning is characterized by using an artificial neural network that applies temperature, pressure, mole fraction of carrier gas, mole fraction of precursor, diffusion rate of carrier gas, diffusion rate of precursor, and permeability as input to the input layer. Atomic layer deposition simulation method for a deposited object having a microchannel or a porous material. According to clause 1,
The machine learning is an atomic layer deposition simulation method for a deposited object having a microchannel or a porous material, characterized in that it uses an artificial neural network that applies at least two or more hidden layers. According to clause 1,
The machine learning is performed on a deposited object having a microchannel or a porous material, characterized in that it uses an artificial neural network that applies the surface coverage of the deposited object having the microchannel or the porous material as an output layer. Simulation method for atomic layer deposition. According to clause 9,
The surface coverage includes a trend of surface coverage in the depth direction (hole depth) depending on the pressure at the starting and ending points of the microchannel or the deposited object having the porous material. An atomic layer deposition simulation method for a deposited object having a microchannel or a porous material.