Predicting Etch Characteristics in Thermal and Atomic Layer Etching

Abstract

Predicting etching in thermal etching reactions using machine learning models. Quantum mechanical simulations are used to identify the chemical characteristics and associated energies of etching processes in one or more reaction paths of a given thermal etching reaction. Labels indicating etching characteristics can be associated with the chemical characteristics and associated energies of a given thermal etching reaction. The machine learning model can be trained using the chemical characteristics and associated energies as independent variables and using the labels as dependent variables for many different etching reactions of different types. When the chemical characteristics and associated energies of a new thermal etching reaction are provided as input in the machine learning model, the machine learning model can accurately predict the etching characteristics of the new thermal etching reaction as output.

Description

[0001] Incorporated by Reference

[0002] The PCT application form is filed concurrently with this specification as a part of this application. Each application to which this application claims the benefit or priority as identified in the concurrently filed PCT application form is incorporated herein by reference in its entirety and for all purposes. Background Art

[0003] Semiconductor manufacturing processes include etching of many materials. Different types of etching, including thermal etching and chemical etching, can be used to etch materials from the substrate surface. As feature sizes shrink and the demand for atomic scale processing increases, thermal or chemical etching may be atomic layer etching (ALE). Etching precursors may or may not react with the substrate surface to etch certain materials from the substrate surface. Predicting whether the etching precursor and the material on the substrate surface will be etched is challenging. Summary of the invention

[0004] The present invention provides a method for predicting etching characteristics in a reaction between a surface layer and an etching precursor. The method comprises: (a) using a quantum mechanical model to identify the chemical characteristics and associated energies of one or more reaction paths in a simulated reaction between the surface layer and the etching precursor; (b) providing the chemical characteristics and associated energies in the simulated reaction as input to a machine learning model; and (c) using the machine learning model to determine the etching characteristics of the simulated reaction between the surface layer and the etching precursor.

[0005] In some implementations, the chemical properties and associated energies of the one or more reaction paths in the simulated reaction are determined using a quantum mechanical model configured with a selected surface representation of the surface layer and one or more selected initial conditions. In some implementations, the one or more selected initial conditions include the separation distance between the surface layer and the etching precursor at the start of the simulated reaction, the orientation and / or direction of the surface layer and the etching precursor at the start of the simulated reaction, the form of the initial chemical bridge between the surface layer and the etching precursor at the start of the simulated reaction, the internal energy or kinetic energy of the etching precursor or the surface layer at the start of the simulated reaction, or a combination thereof. In some implementations, the selected surface representation of the surface layer is selected from a group consisting of molecules, small molecule clusters, and macromolecular clusters. In some implementations, the quantum mechanical model includes a density functional theory (DFT) model, a Hartree-Fock model, a semi-empirical model, or a combination thereof. In some implementations, the chemical properties include a bonding configuration or molecular structure of one or more reaction intermediates and / or products produced in the one or more reaction paths. The bonding configuration or molecular structure may include one or more of a single-bridge dimer, a double-bridge dimer, a triple-bridge dimer, and no dimer. In some implementations, the etching characteristics of the simulated reaction include an etching rate of a surface layer. In some implementations, the method further includes organizing the chemical characteristics and associated energies of one or more reaction paths in the simulated reaction into a feature vector. In some implementations, the machine learning model is trained using a training set comprising a plurality of training set members, each training set member comprising: (i) a feature vector comprising chemical characteristics and associated energies for intermediates and / or products generated in one or more reaction paths, the chemical characteristics and associated energies being used for thermal etching reactions simulated by at least one quantum mechanical simulation; and (ii) a label representing the characteristics of the thermal etching reaction. In some implementations, the method further includes: identifying one or more candidate thermal etching reactions using the machine learning model, wherein each of the identified one or more candidate thermal etching reactions includes at least an identified etching precursor. Each of the one or more identified candidate thermal etch reactions may also include an identified material to be etched by an identified etch precursor, an identified etch mask material, and / or an identified chamber material.

[0006] Another aspect relates to a method for generating a machine learning model for predicting etching information in a thermal etching reaction. The method includes: performing at least one quantum mechanical simulation for each of a plurality of thermal etching reactions, each quantum mechanical simulation generating chemical characteristics and associated energies of one or more reaction pathways in a corresponding thermal etching reaction between a surface layer and an etching precursor; determining experimentally determined etching characteristics for each of the plurality of thermal etching reactions; generating a training set comprising a plurality of training set members, each training set member comprising (i) at least one feature vector containing chemical characteristics and associated energies of intermediates and / or products generated in one or more reaction pathways for the corresponding thermal etching reaction; and (ii) a label representing the experimentally determined etching characteristics; and generating a machine learning model trained using the training set, wherein the machine learning model is configured to predict etching information in the thermal etching reaction.

[0007] In some implementations, performing the at least one quantum mechanical simulation includes performing multiple quantum mechanical simulations for each of the multiple thermal etching reactions, and each eigenvector contains one of the multiple quantum mechanical simulations of the corresponding thermal etching reaction, the chemical properties of the intermediates and / or products produced in the one or more reaction paths, and the associated energy. In some implementations, each of the at least one quantum mechanical simulation includes a quantum mechanical model configured with a surface representation of a surface layer and one or more initial conditions. In some implementations, the one or more initial conditions include the separation distance between the surface layer and the etching precursor at the beginning of the quantum mechanical simulation, the orientation and / or direction of the surface layer and the etching precursor at the beginning of the quantum mechanical simulation, the form of the initial chemical bridge between the surface layer and the etching precursor at the beginning of the quantum mechanical simulation, the internal energy or kinetic energy of the etching precursor or the surface layer at the beginning of the quantum mechanical simulation, or a combination thereof. In some implementations, the surface representation of the surface layer is selected from the group consisting of: molecules, small molecule clusters, and macromolecular clusters. In some implementations, the quantum mechanical model includes a density functional theory (DFT) model, a Hartree-Fock model, a semi-empirical model, or a combination thereof. In some implementations, the chemical properties include bonding configurations or molecular structures of intermediates and / or products produced in one or more reaction pathways. The bonding configurations or molecular structures may include one or more of a single-bridge dimer, a double-bridge dimer, a triple-bridge dimer, and no dimer. In some implementations, generating the training set includes: organizing the chemical properties and associated energies of the one or more reaction pathways in the corresponding thermal etching reaction into the at least one feature vector; and organizing the etching properties determined experimentally into the labels.

[0008] These and other aspects are further described below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1A is a schematic diagram of an example of a non-self-limiting thermal etching process.

[0010] Figure 1B is a schematic diagram of an example of a self-limiting thermal etching process.

[0011] Figure 1C Schematic diagram showing different modification mechanisms in the modification operation of atomic layer etching.

[0012] Figure 1D Schematic diagram showing different removal mechanisms in the removal operation of isotropic atomic layer etching.

[0013] Figure 2 A schematic diagram showing an exemplary ligand exchange reaction in atomic layer etching is shown.

[0014] Figure 3A A schematic diagram showing the possible minimum energy isomers in an exemplary reaction between aluminum fluoride and a metal chloride is shown.

[0015] Figure 3B An exemplary energy diagram showing the reaction intermediates and products produced in the reaction pathway between aluminum fluoride and a metal chloride.

[0016] Figure 4 Exemplary reaction patterns for thermal etching reactions between aluminum fluoride and a number of metal chlorides are shown.

[0017] Figure 5 is a flowchart of an exemplary method for generating a machine learning model according to some implementation schemes, wherein the machine learning model is used to predict etching information in a thermal etching reaction.

[0018] Fig. 6A A schematic diagram showing some possible parameters in a quantum mechanical simulation of the reaction between a surface layer and an etch precursor according to some implementations.

[0019] Figure 6B shows that according to some implementation schemes Fig. 6A Table of generated eigenvectors for the reaction between the surface layer and the etching precursor from many quantum mechanical simulations run in .

[0020] Figure 6C A table of feature vectors for different thermal etch reactions and a corresponding table of labels representing etching characteristics of the different thermal etch reactions are shown according to some implementations.

[0021] Figure 7is a flow chart of an exemplary method for predicting etching characteristics in a reaction between a surface layer and an etching precursor, according to some implementations.

[0022] Figure 8 Shown is a schematic diagram for validating a machine learning model from experimental measurements according to some implementation schemes.

[0023] Fig. 9 A table showing feature vectors for different thermal etch reactions according to some implementations, wherein identified etch precursors and identified surface layers are used to design candidate thermal etch reactions. DETAILED DESCRIPTION

[0024] In the present disclosure, the terms "semiconductor wafer", "wafer", "substrate", "wafer substrate" and "partially manufactured integrated circuit" are used interchangeably. It should be understood by those of ordinary skill in the art that the term "partially manufactured integrated circuit" can refer to a silicon wafer during any of the many stages of integrated circuit manufacturing. Wafers or substrates used in the semiconductor device industry typically have a diameter of 200 mm, or 300 mm, or 450 mm. The following detailed description assumes that the present disclosure is implemented on a wafer. However, the present disclosure is not limited to this. The workpiece can have a variety of shapes, sizes, and materials. In addition to semiconductor wafers, other workpieces that can utilize the present disclosure include various objects such as printed circuit boards.

[0025] Thermal Etching

[0026] Thermal etching uses a gas phase reagent to react with the substrate surface and chemically and / or thermally etch material from the substrate surface. As used herein, thermal etching may also be referred to as "chemical etching" and vice versa. Thermal etching can be applied to a variety of materials. Selectivity may be desired in the thermal etching reaction so that one material is removed in the presence of other materials. Selectivity in etching has many applications in semiconductor device manufacturing and surface cleaning. Selectivity in thermal etching may be higher than other conventional etching techniques such as plasma etching or sputtering.

[0027] Thermal etching generally involves a solid etching material and a gaseous etching precursor. The etching precursor is any reactive gas that reacts with the solid etching material in a thermal etching reaction to facilitate removal of the solid etching material from the substrate surface. In some implementations, the etching precursor may modify a surface layer of the substrate surface, and thermal energy may be applied to etch the modified surface layer. Examples of this process include Figure 1A In some implementations, the etching precursor may react with the modified surface layer and etch the modified surface layer by thermal and / or chemical energy. An example of this process is shown in Figure 1B middle.

[0028] Thermal etching processes can be broadly categorized as continuous (non-self-limiting) or discontinuous (self-limiting). A self-limiting thermal etching process can include a thermal atomic layer etching (ALE) process, where an exemplary thermal ALE process can involve a ligand exchange reaction.

[0029] Figure 1A is a schematic diagram of an example of a non-self-limiting thermal etching process. The initial step of the non-self-limiting thermal etching process involves delivering an etching precursor 102 toward a solid material 101 of a surface layer to be etched. During the modification step of the non-self-limiting thermal etching process, the etching precursor 102 is adsorbed onto the solid material 101 of the surface layer to form a modified surface layer 103. Subsequently, during the removal step of the non-self-limiting thermal etching process, thermal energy is applied to the solid material 101 to cause desorption of volatile reaction products 104 and remove the modified surface layer 103.

[0030] Figure 1B is a schematic diagram of an example of a self-limiting thermal etching process. Figure 1B The self-limiting thermal etching process in can be described as a thermal ALE process, which may involve multiple ALE cycles. ALE is a technique for removing thin layers of material using sequential self-limiting reactions. Typically, an "ALE cycle" is a minimum set of operations for performing an etching process once, such as etching a single layer. The result of one cycle is etching at least some surface layers of the substrate. Typically, an ALE cycle includes a modification operation and a removal operation. An ALE cycle may include auxiliary operations, such as sweeping one of the reactants or byproducts. As an example, an ALE cycle may include the following operations: (i) delivering a modification precursor or a reaction gas into the chamber, (ii) sweeping a modification precursor from the chamber, (iii) delivering an etching precursor or a removal gas into the chamber, and (iv) sweeping the chamber.

[0031] The initial step of the self-limiting thermal etching process involves delivering a modified precursor 112 toward a solid material 111 of a surface layer to be etched. For example, the solid material 111 may include crystalline aluminum oxide (A12O3), and the modified precursor 112 may include hydrogen fluoride (HF). In some implementations, the modified precursor 112 may include a gaseous substance or a plasma substance, wherein the plasma substance may include fluorine radicals in the modified precursor 112. Accordingly, the modification may be performed in the presence of a plasma or other energy source, rather than just thermal energy. The modified precursor 112 is adsorbed onto the surface of the solid material 111 to form a modified surface layer 113. The material of the modified surface layer 113 may not be volatile or at least not volatile at a given temperature. The initial step of the self-limiting thermal etching process may be referred to as a "modification" step. In some implementations, the modification step may be followed by a sweep step to sweep the modified precursor 112.

[0032] The subsequent step of the self-limiting thermal etching process involves etching the modified surface layer 113 using an etching precursor 114. The etching precursor 114 reacts with the modified surface layer 113 to convert at least some of the modified surface layer 113 into a volatile product 115. For example, the etching precursor 114 includes trimethylaluminum (TMA), and the modified surface layer 113 includes aluminum trifluoride (AlF3). TMA can accept fluorine atoms from AlF3, and TMA provides CH3 molecules to produce volatile products 115 including dimethylaluminum fluoride (AlF(CH3)2). Although AlF3 requires high temperatures to desorb from the modified surface layer 113, AlF(CH3)2 requires lower temperatures (e.g., room temperature) to desorb from the modified surface layer 113. In other words, the introduction of the etching precursor 114 that reacts with the modified surface layer 113 can reduce the process temperature at which desorption occurs. Removal of the modified surface layer 113 using the etching precursor 114 occurs in a self-limiting manner. The appropriate temperature and etching precursor 114 are selected to selectively react with and remove the modified surface layer 113 without reacting with the bulk layer or the rest of the solid material 111. The subsequent steps in the thermal etching process may be referred to as "removal" steps or "etching" steps. In some implementations, the removal step may be followed by a sweep step to remove excess etching precursor 114 and volatile products 115. The modification step and removal step may be repeated to remove the solid material 111 on a layer-by-layer basis until the desired depth or amount is reached.

[0033] A typical thermal ALE process may include a modification step and a removal step. Figure 1C Schematic diagram showing different modification mechanisms in the modification operation of ALE. Figure 1D Schematic diagram showing different removal mechanisms in an isotropic ALE removal operation. Figure 1C-1D , schematic diagrams 151-154 show a general ALE cycle. In 151, a substrate is provided. In 152, the surface of the substrate is modified. In 153, the next step is prepared. In 154, the modified layer is etched and finally removed. A cycle may only locally etch about 0.1nm to about 50nm of material, or about 0.1nm and about 5nm of material, or about 0.2nm and about 50nm of material, or about 0.2nm and about 5nm of material. The amount of material etched in a cycle can vary. For example, if ALE is integrated with atomic layer deposition (ALD) for protecting the sidewalls, the ALE cycle may remove less material than if ALE is not integrated with ALD.

[0034] Figure 1CDifferent mechanisms for modifying surfaces in ALE are illustrated: (a) chemisorption, (b) deposition, and (c) conversion. In chemisorption, a modifying precursor can react with the surface of a substrate to form an adsorption-limited amount of material that is bonded to the substrate surface. In deposition, a thin layer of material can be deposited on the surface of a substrate before removal. In conversion, a salt is atomically formed on top of the substrate to essentially form a barrier layer, preventing further reaction. The salt can then be removed by subsequent thermal or chemical manipulation. Figure 1C In the present invention, the reactants used to modify the surface can be delivered by gas, plasma, wet (liquid) chemicals, or other sources.

[0035] Figure 1D Different mechanisms for isotropically removing a modified surface in ALE are described: (a) thermal energy, and (b) chemical energy. Using thermal energy, the removal step of ALE can be accomplished purely by adjusting the temperature to cause desorption of the modified surface layer without any further chemical reaction. Using chemical energy, the removal step of ALE can be accomplished by reacting a gaseous precursor with the modified surface layer to cause desorption of the modified surface layer without changing the temperature. Even though the temperature may not change with the chemical energy, the ALE reaction is understood to be a thermal etching reaction. Using both thermal energy and chemical energy, the removal step of ALE can be accomplished by reacting a gaseous precursor with the modified surface layer and applying the desired temperature to cause desorption of the modified surface layer.

[0036] Resources may be limited in experimenting, exploring, calculating, studying, and manually determining etching characteristics in thermal etching reactions. The machine learning model of the present disclosure is configured to predict etching characteristics in thermal etching reactions. The thermal etching reaction includes at least an etching precursor and a surface layer or a modified surface layer. It should be understood that the machine learning model of the present disclosure can be applied to continuous or discontinuous thermal etching. It should also be understood that the machine learning model of the present disclosure can be applied to ALE reactions, specifically to the removal step in the ALE reaction, regardless of whether thermal energy and / or chemical energy is used to remove the modified surface layer. The machine learning model of the present disclosure can not only reduce time, cost, and resources, but also promote the design of new thermal etching reactions and chemical processes.

[0037] Representation of thermal etching reaction

[0038] Generally, the thermal etching reaction is considered to be the conversion of reactants (including surface layer and etching precursor) into volatile products. If the product formed is volatile, the etching precursor etches the surface layer. The etching results can be explained by the stability and volatility of the product. The stability of the volatile product can be determined by understanding the reaction path and energy associated with the reactants, intermediates and / or products in the reaction path.

[0039] For example, the thermal etching reaction may include one or more ligand exchange reactions. A ligand exchange reaction is a reaction pathway in which a ligand in a compound is replaced by another ligand via ligand substitution. Figure 2 Schematic diagram showing an example ligand exchange reaction in ALE. A substrate may include a host material 201 such as a metal oxide (e.g., aluminum oxide). The surface of the substrate may be fluorinated to form a modified surface layer, where the modified surface layer includes a metal fluoride 202 (e.g., aluminum fluoride). A metal precursor 203 in a vapor phase may flow toward the substrate to react with the modified surface layer, wherein the metal precursor 203 includes tin (II) acetylacetonate (Sn(acac)2). The metal precursor 203 is used as an etching precursor. The metal precursor 203 may accept fluorine atoms from the metal fluoride 202 and donate acetylacetonate (acac) molecules to form a volatile product. In other words, the metal precursor 203 exchanges ligands with the metal fluoride 202. The ligand exchange forms a first volatile product 204, which is fluorine-tin(II) acetylacetonate (SnF(acac)), and a second volatile product 205, which is metal(III) acetylacetonate (M(acac)3). The first volatile product 204 and the second volatile product 205 are desorbed from the substrate surface, allowing the host material 201 to be etched in a self-limiting manner. The ligand exchange is facilitated by the ability of fluorine to form a stable bridge with tin and the ability of the acetylacetonate ligand to form a stable metal acetylacetonate compound.

[0040] The conversion of reactants to one or more volatile products involves one or more reaction paths. Each reaction path includes two or more reactants, one or more intermediates, and one or more products. Each reaction path can represent a possible reaction mechanism that causes the entire thermal etching reaction to occur. Specifically, the reaction path can be represented by discrete intermediate compounds and their associated energy. Additionally or alternatively, the reaction path can be represented by a reaction coordinate, which includes not only discrete intermediate compounds, but also all intermediate states between discrete intermediate compounds. These discrete intermediate states can represent many molecular configurations, intermolecular conformations, intermolecular orientations, bond lengths, etc. Discrete intermediate compounds can be formed at different steps in the reaction path, and each step in this reaction path can have its own associated energy change (ΔE). Energy can be considered as a thermodynamic property, such as ground state energy.

[0041] Predicting etching characteristics in a thermal etching reaction may depend on the reaction path of the thermal etching reaction and whether certain intermediate states or products are thermodynamically or energetically favorable in the reaction path. In other words, the reaction path and the associated energies of reactants, intermediates, and products in the reaction path may determine whether reactants in the thermal etching reaction are converted into volatile products.

[0042] Figure 3AA schematic diagram of possible minimum energy isomers in an exemplary reaction between aluminum fluoride and a metal chloride is shown. Possible minimum energy isomers can be generated by performing a simulation of an exemplary reaction between aluminum fluoride and a metal chloride. The reaction between the two molecules (i.e., aluminum fluoride and a metal chloride) can result in discrete intermediates and / or product compounds in one or more reaction pathways, wherein each discrete intermediate and / or product compound has a unique bonding configuration or molecular structure. The exemplary reaction can be simulated in a first simulation 310 and a second simulation 311. Figure 3A The first simulation 310 in allows for a triple bridge bonding configuration between aluminum atoms in the aluminum fluoride and metal atoms in the metal chloride. Figure 3A The second simulation 311 in allows for a double bridge bonding configuration between the aluminum atoms in the aluminum fluoride and the metal atoms in the metal chloride. Figure 3A In the above, "triple bridge" indicates that three halogen atoms (including fluorine atoms and / or chlorine atoms) form a bridge connecting the aluminum atom and the metal atom. "Double bridge" indicates two halogen atoms. "Single bridge" indicates one halogen atom. "No dimer" indicates that there is no bridge connecting the aluminum atom and the metal atom. Figure 3A As shown in FIG. 3 , a first simulation 310 allowing a three-bridge configuration generates a three-bridge isomer 312, a two-bridge isomer 313, a single-bridge isomer 314, and no dimer 315 as possible intermediate and / or product compounds. A second simulation 311 allowing a two-bridge configuration generates a two-bridge isomer 313, a single-bridge isomer 314, and no dimer 315 as possible intermediate and / or product compounds. For example, a simulation of a quantum mechanically simulated thermal etching reaction discussed below can generate possible intermediate and / or product compounds produced in one or more reaction pathways.

[0043] The stabilities of these possible intermediate and / or product compounds can be determined from an energy diagram, which displays the energies associated with intermediates and / or product compounds in a reaction pathway. Figure 3B Example energy diagram showing reaction intermediates and products produced in the reaction pathway between aluminum fluoride and a metal chloride. Figure 3B The energy diagram in FIG. 1 shows four different reaction paths corresponding to four different reactions: (1) a reaction path for the reaction between aluminum fluoride and silicon tetrachloride, (2) a reaction path for the reaction between aluminum fluoride and germanium tetrachloride, (3) a reaction path for the reaction between aluminum fluoride and tin tetrachloride, and (4) a reaction path for the reaction between aluminum fluoride and titanium tetrachloride. The energy diagram shows the energy change (ΔE) as a function of the reaction coordinates representing the steps along each reaction path. The minimum energy in the energy diagram can be associated with different reaction intermediates or products. Accordingly, each different reaction intermediate and product may have an associated energy change.

[0044] As in Figure 3BAs shown in, each reaction path starts with two reactants 320, namely aluminum fluoride (AlF3) and metal chloride (MCl4), wherein the two reactants 320 have zero correlation energy. The two reactants 320 can represent etching precursors and surface layers in the hot etching reaction. As each reaction path progresses, a first reaction intermediate 321 is formed with correlation energy at the first minimum of the energy diagram, and a second reaction intermediate 322 is formed with correlation energy at the second minimum of the energy diagram. The first reaction intermediate 321 can represent a dimer between aluminum fluoride and a metal chloride with a fluorine single bridge or a fluorine and chlorine double bridge. The second reaction intermediate 322 can represent a dimer between aluminum fluoride and a metal chloride with a fluorine and fluorine double bridge. The second reaction intermediate 322 has a lower energy minimum than the first reaction intermediate 321. As each reaction path progresses further, a reaction product 323 including aluminum chloride fluoride (AlF2Cl) and metal fluoride chloride (MFCl3) is formed after the ligand exchange reaction. Reaction product 323 has an associated energy greater than each of first reactive intermediate 321 and second reactive intermediate 322. Thus, the energy diagram shows that second reactive intermediate 322 is the most stable compound along the reaction pathway.

[0045] Energy diagrams (e.g. Figure 3B ) can be generated by simulation of the reaction between aluminum fluoride and the selected metal chloride. Figure 3B Only the reaction path for each thermal etching reaction is shown, but there are usually multiple reaction paths for thermal etching reactions. In fact, simulation can generate multiple reaction paths for thermal etching reactions. These different reaction paths have their own related reaction intermediates / products, related energy changes and related probability of occurrence. Related reaction intermediates / products, changes in related energy and related probability of occurrence generated by simulation can provide useful information in determining the stability of some reaction intermediates / products in the reaction path. This can be represented in the reaction pattern.

[0046] The chemical properties of one or more reaction paths of thermal etching reaction can be presented with reaction patterns. "Reaction pattern" represents the potential contribution of multiple reaction paths in thermal etching reaction. In some implementations, reaction pattern can also be referred to as "reaction fingerprint" or "reaction tag". The reaction pattern of thermal etching reaction can include many chemical property dimensions and energy dimensions. Many chemical properties of thermal etching reaction include reaction intermediates and / or products, and their bonding configuration and / or molecular structure. For example, bonding configuration and / or molecular structure can include one or more of single bridge dimer, double bridge dimer, three bridge dimer and no dimer as possible bonding configuration and / or molecular structure in reaction path. In some implementations, reaction pattern can display chemical properties along x-axis and display energy along y-axis. However, it will be understood that more complex reaction patterns can display additional dimensions, such as additional dimensions of temperature and / or pressure.

[0047] Figure 4 Exemplary reaction patterns of thermal etching reactions between aluminum fluoride and a number of metal chlorides are shown. The x-axis shows the chemical property dimension 430, and the y-axis shows the energy dimension 440. The first reaction pattern 420 presents the chemical properties and associated energies in the thermal etching reaction between aluminum fluoride and titanium tetrachloride. The second reaction pattern 421 presents the chemical properties and associated energies in the thermal etching reaction between aluminum fluoride and silicon tetrachloride. The third reaction pattern 422 presents the chemical properties and associated energies in the thermal etching reaction between aluminum fluoride and tin tetrachloride. The fourth reaction pattern 423 presents the chemical properties and associated energies in the thermal etching reaction between aluminum fluoride and germanium tetrachloride. Simulations such as quantum mechanical simulations are performed to generate each reaction pattern 420-423 for each thermal etching reaction. The quantum mechanical simulation tracks multiple different reaction paths and their potential contributions in the thermal etching reaction, providing reaction intermediates and / or products from each of the multiple reaction paths and their associated energies, which can be represented by reaction patterns. The multiple reaction pathways may contribute multiple data sets for reaction intermediates and / or products and their associated energies, and the multiple data sets are used to provide the mean, median, minimum / maximum, and mean ± 1 standard deviation along the energy dimension 440 for each chemical property in the chemical property dimension 430. In some implementations, each reaction pattern 420-423 may be generated from multiple quantum mechanical simulations of a single thermal etch, where the multiple quantum mechanical simulations of the single etch reaction are run with different parameters (e.g., surface characterization, initial conditions, etc.).

[0048] For different etch precursors, commonalities in patterns and trends can be seen. Figure 4The various reaction modes 420-423 in the first reaction mode 420 determine and identify patterns and trends. For example, titanium tetrachloride (TiCl4) in the first reaction mode 420 and tin tetrachloride (SnCl4) in the third reaction mode 422 share similar "fingerprints" in terms of the formation of fluorine triple bridge configuration, fluorine double bridge configuration, and fluorine and chlorine double bridge configuration. Silicon tetrachloride (SnCl4) in the second reaction mode 421 and germanium tetrachloride (GeCl4) in the fourth reaction mode 423 share similar "fingerprints" in terms of the formation of fluorine double bridge configuration and fluorine / chlorine double bridge and the absence of any dimer indicated by the X configuration. Fluorine triple bridge configuration and fluorine double bridge configuration usually have the lowest minimum energy of etching precursors in the first reaction mode 420 and the third reaction mode 422. Fluorine double bridge configuration usually has the lowest minimum energy for each etching precursor in the reaction mode 420-423. There is no fluorine single bridge configuration in the first reaction mode 420 and the third reaction mode 422. Additionally, the energies associated with the chlorine triple bridge configuration and the chlorine double bridge configuration are generally high for reaction modes 420-422. Figure 4 Some reaction modes in the Figure 4 Some of the reaction modes in the reaction patterns do not result in etching. Experimentally, the first reaction mode 420 and the third reaction mode 422 result in etching, while the second reaction mode 421 and the fourth reaction mode 423 do not result in etching. Humans can observe some trends and patterns in the reaction modes 420-423, but with more etching precursors and more surface materials mixed, identifying trends and patterns becomes increasingly complex. In other words, the human brain can recognize what is happening. Figure 4 However, with the breadth of many different responses of different types, it would be very difficult, if not impossible, for the human brain to discern the connections between specific combinations of data points and etched information.

[0049] Adding to the complexity of discerning the connections between specific combinations of data points and etching information across many different types is obtaining the correct assumptions about the reaction pathways, surfaces, and starting conditions. Quantum mechanical simulations can utilize specific quantum mechanical models, specific assumptions about the surfaces of the reactant materials used to represent the surface modification, and specific assumptions about the initial conditions of the thermal etching reactions. Certain specific combinations of quantum mechanical models, surface characterizations, and initial conditions may be applicable to certain types of thermal etching reactions, and may not be applicable to other types of thermal etching reactions in terms of their predictive power.

[0050] There are many quantum mechanical simulation models for computationally generating reaction pathways and associated energy changes. Quantum mechanical simulation models are typically used in physics, chemistry, and materials science to study the electronic structure of many-body systems, including atoms, molecules, and condensed phases. Quantum mechanical simulation models can be used to represent the electron density using the Schrödinger (wave) equation and calculate the associated energy or potential.

[0051] In the context of the present disclosure, the quantum mechanical simulation model can identify multiple reaction paths that may exist for a given combination of etching precursors and surface layers. In some implementations, the surface layer is a modified surface layer. For each identified reaction path, the quantum mechanical simulation model calculates the energy of one or more reactants, products, and intermediates. In this way, the quantum mechanical simulation model can determine the energy change (ΔE) of each step in the reaction path.

[0052] There are many different quantum mechanics simulation tools for forming multi-body systems and identifying models of many reaction paths. These quantum mechanics simulation tools use different approximations in Schrodinger (wave) equation. These quantum mechanics simulation tools can exist as software packages, which may or may not be commercially available. Examples include but are not limited to ABINIT, ACES, AMPAC, ADF, Atomisix ToolKit, BigDFT, CADPAC, CASINO, CASTEP, CFOUR, COLUMBUS, CONQUEST, CP2K, CPMD, CRYSTAL, DACAPO, Dalton, deMon2k, DFTB+, DFT++, DIRAC, DMol3, ELK, Empire, EPW, ErgoSCF, ERKALE, EXCITING, FLEUR, FHI-aims, FPLO, FreeON, Firefly, GAMESS, Gaussian, GPAW, HiLAPW, HORTON, HyperChem, Jagua r, JDFTx, LOWDIN, MADNESS, MISSTEP, MOLCAS, MoIDS, MOLGW, MOLPRO, MONSTERGAUSS, MOPAC, MPQC, NRLMOL, NTChem, NWChem, Octopus, ONETEP, OpenAto m, OpenMX, ORCA, phase0, PLATO, PQS, Priroda-06, PSI, PUPIL, PWmat, PWscf, PyQuante, PySCF, Qbox, Q-Chem, QMCPACK, Quantemol-N, QSite, Quantum ESPRESSO, RMG, RSPt, Scigress, Spartan, Siam Quantum, SIESTA, TB-LMTO, TeraChem, TURBOMOLE, VASP, WIEN2k and Yambo Code. Some quantum mechanics simulation tools may include density functional theory (DFT) tools, Hartree-Fock tools, and / or semi-empirical tools. Many of the above quantum mechanics simulation tools use some categories or mixtures of DFT tools, Hartree-Fock tools, and semi-empirical tools, and some of the above quantum mechanics simulation tools are publicly available. For example, Gaussian uses DFT tools, Hartree-Fock tools, and semi-empirical tools. MOLPRO uses Hartree-Fock and DFT tools, but does not use semi-empirical tools. MONSTERGAUSS uses Hartree-Fock tools, but does not use DFT and semi-empirical tools.Atomistix ToolKit uses semi-empirical and DFT tools, but not Hartree-Fock tools. Different quantum mechanical simulation tools and different parameterizations of quantum mechanical simulation tools may affect the calculations and results, including the identification of specific reaction pathways and the energies associated with the many species from a specific reaction pathway.

[0053] In addition to selecting from among different quantum mechanical simulation models in a quantum mechanical simulation of a thermal etching reaction, one of a variety of surface representations of the surface layer can also be selected in the quantum mechanical simulation. The surface representation can describe the form of chemical species on the surface layer that participate in the thermal etching reaction with the etching precursor. As an example, the surface representation can be a single molecule (e.g., aluminum fluoride (AlF3)) or a cluster of molecules (e.g., [AlF3] x ) surface characterization. Molecular clusters can be a wide range of sizes, such as large and small. As used herein, small molecular clusters can be less than about 20 atoms, or between about 10 atoms and about 20 atoms. For example, a small AlF3 molecular cluster will be between 3 molecules and 5 molecules. As used herein, a macromolecular cluster can be greater than about 20 atoms, greater than about 30 atoms, or greater than about 50 atoms.

[0054] Different surface characterizations can affect the results of a quantum mechanical simulation of a given thermal etching reaction. In some implementations, a thermal etching reaction simulated using a first surface characterization may provide a different reaction pattern than when simulated using a second surface characterization, even if all other aspects of the reaction are exactly the same.

[0055] In addition to selecting from different quantum mechanical simulation models and selecting from different surface representations of the surface layer, one of various initial conditions can also be selected in the quantum mechanical simulation of the thermal etching reaction. The initial condition can represent the initial interaction between the surface layer and the etching precursor in the quantum mechanical simulation of the thermal etching reaction. Examples of initial conditions in the quantum mechanical simulation include: the separation distance between the surface layer and the etching precursor at the beginning of the simulated reaction, the orientation and direction of the surface layer and the etching precursor at the beginning of the simulated reaction, the form of the initial chemical bridge or bond between the surface layer and the etching precursor at the beginning of the simulated reaction, and the internal energy or kinetic energy of the etching precursor and / or the surface layer at the beginning of the simulated reaction. The above initial conditions can be specified for each quantum mechanical simulation. For example, when the initial condition assumes that the two reactants have not yet begun to interact chemically, the initial condition can assume that there is no chemical bridge or bond between the two reactants. In some implementations, the initial condition can assume that there is no separation distance between the two reactants. For example, the initial condition can assume that the etching precursor has been physically adsorbed and diffused on the surface of the surface layer before the reaction forms a chemical bridge or bond.

[0056] Different initial conditions may affect the results of a quantum mechanical simulation of a given thermal etching reaction. As with surface characterization, the energy of the species or the energy change (ΔE) in the reaction path can be a function of the initial conditions of the reaction. In some implementations, a thermal etching reaction simulated using a first initial condition may provide a different reaction pattern than when simulated using a second initial condition, even if all other aspects of the reaction are exactly the same.

[0057] Machine Learning Models

[0058] The machine learning model of the present disclosure is a trained computational model that takes as input the chemical characteristics and associated energies of one or more reaction pathways of the simulated reaction between the surface layer and the etching precursor, and provides as output the etching characteristics of the simulated reaction. The quantum mechanical simulation model as described above can be used to identify the chemical characteristics and associated energies of one or more reaction pathways of the simulated reaction. In some implementations, the chemical characteristics and associated energies can be as follows: Figure 4 In the example shown in FIG. 1 , the energy of ...

[0059] Whether or not represented by a reaction pattern, the chemical properties and associated energies can provide data points in the simulated reaction. In some implementations, etching properties that may or may not be determined experimentally can provide additional data points to match the simulated reaction. As there are more reactions of different types, more data points can be provided. The machine learning model of the present disclosure can be trained to recognize and find patterns between data points across many different reactions of different types in order to accurately predict the etching properties of new reactions.

[0060] Figure 5 5 is a flow chart of an exemplary method of generating a machine learning model for predicting etching information in a thermal etching reaction according to some implementations. The operations in process 500 may be performed in a different order and / or with different, fewer, or additional operations. The machine learning model may be supervised, unsupervised, or partially supervised.

[0061] At block 510 of process 500, at least one quantum mechanical simulation is performed for each of a plurality of thermal etching reactions. Each quantum mechanical simulation generates chemical characteristics and associated energies of one or more reaction paths in a corresponding thermal etching reaction between a surface layer and an etching precursor. In some implementations, the surface layer is a modified surface layer. The generated chemical characteristics and associated energies are used as part of a training set member in a training set for training the machine learning model.

[0062] The characteristics of the etching process are typically the characteristics of the reaction pathway, including the chemical characteristics of the reaction and / or the various chemical species involved in the reaction. Describing these characteristics in the reaction pathway are the chemical properties and associated energies (stability, lifetime, etc.) of the chemical species produced in the reaction pathway. The chemical properties may include the bonding configuration or molecular structure of one or more intermediates and / or products produced in each of one or more reaction pathways. The chemical properties may reflect the type of interaction between the surface layer and the etching precursor. The type of interaction may be a chemical bridge or bonding between two reactants (i.e., the surface layer and the etching precursor). In some implementations, the bonding configuration or molecular structure includes one or more of a single-bridge dimer, a double-bridge dimer, a triple-bridge dimer, and no dimer. Return to reference Figure 3B In an example thermal etching reaction between aluminum fluoride (AlF3) and metal chloride (MCl4) in the embodiment of the present invention, the reaction intermediate may include an AlF3-MCl4 dimer with a single fluorine bridge, an AlF3-MCl4 dimer with a double bridge of fluorine and chlorine, or an AlF3-MCl4 dimer with a double bridge of fluorine and fluorine. The reaction product may not include dimers in the form of AlF2Cl and MFCl3.

[0063] The quantum mechanical simulation identifies a variety of possible reaction pathways. For each identified pathway, the quantum mechanical simulation identifies the chemical properties (e.g., bonding configuration or molecular structure) of the reactants, one or more intermediates, and one or more products. Furthermore, the quantum mechanical simulation calculates the energies associated with the reactants, one or more intermediates, and one or more products and / or the energy change (ΔE) used to reach the one or more intermediates and one or more products from the reactants.

[0064] Such data for quantum mechanical simulations can be expressed as feature vectors. Each feature vector contains the chemical properties and associated energies of intermediates and / or products produced in one or more reaction pathways in a corresponding thermal etching reaction. In some implementations, the feature vectors include Figure 4 , but it should be understood that more complex feature vectors may include additional information. For example, this additional information may include one or more of the temperature and pressure of the thermal etching reaction. In some implementations, the feature vector also contains one or both of the temperature and pressure of the thermal etching reaction. In some implementations, process 500 includes organizing the chemical properties and associated energies of intermediates and / or products produced in one or more reaction paths in the corresponding thermal etching reaction into at least one feature vector.

[0065] There may be multiple eigenvectors for a single thermal etching reaction. Specifically, independent eigenvectors may be provided from different quantum mechanical simulations of the thermal etching reaction. Different quantum mechanical simulations for thermal etching reactions may be derived from different quantum mechanical models, different surface representations, and / or different initial conditions. Therefore, each of at least one quantum mechanical simulation includes a quantum mechanical model configured with a surface representation of the surface layer and one or more initial conditions. Accordingly, each eigenvector of the quantum mechanical simulation may have its own unique combination of quantum mechanical model, surface representation, and initial conditions.

[0066] Fig. 6A A schematic diagram showing some possible parameters in a quantum mechanical simulation of the reaction between the surface layer and the etching precursor according to some implementations. It should be understood that the quantum mechanical simulation 610 of the thermal etching reaction is not limited to Fig. 6A Instead of the three parameters shown in FIG. 1 , different, fewer, or additional parameters may be included. Additional parameters may include, for example, temperature or kinetic energy variations and pressure variations.

[0067] exist Fig. 6A , the first parameter 611 of the quantum mechanical simulation 610 includes a quantum mechanical model. The first parameter 611 may include a quantum mechanical model selected from a DFT model, a Hartree-Fock model, a semi-empirical model, or a combination thereof. It should be understood that the quantum mechanical model is not limited to the above models, and other appropriate quantum mechanical models may be used. Independent eigenvectors for a given thermal etching reaction may be provided by simulating the thermal etching reaction using different types of quantum mechanical models. For example, one eigenvector may be generated using a DFT model, and another eigenvector may be generated using a Hartree-Fock model. Each quantum mechanical simulation 610 provides its own set of chemical properties and associated energies for the thermal etching reaction.

[0068] exist Fig. 6A In the quantum mechanical simulation 610, the second parameter 612 includes a surface characterization. The second parameter 612 may include a surface characterization of a surface layer selected from molecules, small molecule clusters, and macromolecular clusters. It should be understood that the surface characterization is not limited to the above characterization, and other appropriate surface characterizations may be utilized. Independent feature vectors for a given thermal etching reaction may be provided by simulating the thermal etching reaction using different surface characterizations. For example, a single molecule representing the surface layer may be used to generate a feature vector, and a small molecule cluster representing the surface layer may be used to generate another feature vector. These variations are provided in the quantum mechanical simulation 610.

[0069] exist Fig. 6AIn the example, the third parameter 613 of the quantum mechanical simulation 610 includes an initial condition. The third parameter 613 may include the separation distance between the surface layer and the etching precursor at the beginning of the quantum mechanical simulation, the orientation and / or direction of the surface layer and the etching precursor at the beginning of the quantum mechanical simulation, the form of the initial chemical bridge between the surface layer and the etching precursor at the beginning of the quantum mechanical simulation, the internal energy or kinetic energy of the etching precursor or the surface layer at the beginning of the quantum mechanical simulation, or a combination thereof. It should be understood that the initial condition is not limited to the above initial condition, and other appropriate initial conditions can be used. An independent characteristic vector of a given thermal etching reaction can be provided by simulating the thermal etching reaction using different initial conditions. For example, an initial chemical bridge of a single-bridge dimer can be used to generate a characteristic vector, and another characteristic vector can be generated by an initial chemical bridge of a double-bridge dimer. In another example, different characteristic vectors can be generated by different orientations of the etching precursor. A plurality of different initial conditions are provided to a quantum mechanical model having a specific surface representation for running independent simulations. In other words, the simulation can be run several times using the same simulation tool or model, so that each quantum mechanical simulation 610 is run with different initial conditions. Each initial condition provides its own characteristic vector.

[0070] The quantum mechanical simulation 610 of the thermal etching reaction may include Fig. 6A In some implementations, independent feature vectors for a given thermal etching reaction may be provided by simulating the thermal etching reaction using different temperatures or kinetic energies. In some implementations, independent feature vectors for a given thermal etching reaction may be provided by simulating the thermal etching reaction using different pressures. These variations are provided in the quantum mechanical simulation 610.

[0071] Figure 6B shows the output from some implementations Fig. 6A Table of generated eigenvectors of multiple quantum mechanical simulations run in, which are used for the reaction between the surface layer and the etching precursor. Quantum mechanical simulations with different parameters generate different eigenvectors, wherein the eigenvectors contain at least reaction intermediates and / or products and their associated energy changes. The different parameters of the quantum mechanical simulations are represented by three-digit (abc) combinations in brackets. For example, (111) corresponds to the DFT model, a single molecule for representing the surface layer and the first initial condition, (211) corresponds to the Hartree-Fock model, a single molecule for representing the surface layer and the first initial condition, (121) corresponds to the DFT model, a small molecule cluster for representing the surface layer and the first initial condition, and so on.

[0072] Each row in the table 620 corresponds to a reaction intermediate or product 630, and each column in the table 620 corresponds to an energy change (ΔE) 640 associated with the reaction intermediate or product 630 for a given quantum mechanical simulation (abc). The last row in the table 620 corresponds to a feature vector 650 associated with each quantum mechanical simulation (abc). The feature vector 650 can rank the reaction intermediates or products 630 by ΔE 640 to reflect which reaction intermediates or products 630 are stable. In some implementations, a coefficient can be assigned to each reaction intermediate or product 630 in a particular feature vector 650, so that some reaction intermediates or products 630 have more weight than other reaction intermediates or products 630.

[0073] return Figure 5 , in box 520 of process 500, an experimentally determined etching characteristic is determined for each of a plurality of thermal etching reactions. The experimentally determined etching characteristic for the thermal etching reaction between the surface layer and the etching precursor may represent the desired output of the thermal etching reaction in the supervised machine learning model. In some implementations, the etching characteristic includes whether the etching precursor etches the surface layer. This may be indicated by a binary output. In some implementations, the etching characteristic includes an etching rate of the surface layer by the etching precursor. This may be indicated by a numerical value reflecting the etching rate, wherein a high etching rate value indicates that the etching precursor etches, and a low etching rate value or a zero value indicates that the etching precursor does not etch. In some implementations, the etching characteristic includes some other numerical characteristics regarding the effectiveness of the etching precursor in etching the surface layer.

[0074] It should be understood that in some other implementations, the etching characteristics of the thermal etching reaction are not determined experimentally. For example, a reliable simulator can be used to determine the etching characteristics via simulation. In some implementations, for some but not all of the thermal etching reactions in a plurality of thermal etching reactions, experimentally determined etching characteristics are determined. The machine learning model may or may not be trained using experimentally determined values. Therefore, the machine learning model of the present disclosure need not be limited to a supervised machine learning model.

[0075] The experimentally determined etching characteristics can be used as part of a training set member in a training set for training a machine learning model. In some implementations, process 500 also includes organizing the experimentally determined etching characteristics into labels. The labels can be provided as an indication of whether etching will occur, etching rate, or some other numerical characteristic regarding the effectiveness of thermal etching.

[0076] The labels in the training set contain dependent variables that indicate experimentally determined etching characteristics for multiple thermal etching reactions. The feature vectors in the training set contain independent variables, where each feature vector represents a single thermal etching reaction. In some implementations, each feature vector represents a single thermal etching reaction and a single quantum mechanical simulation. In some other implementations, each feature vector represents a single thermal etching reaction and multiple quantum mechanical simulations of the single thermal etching reaction. A training set for training a machine learning model can be generated from feature vectors and associated labels for multiple thermal etching reactions.

[0077] At block 530 of process 500, a training set is generated comprising a plurality of training set members. Each training set member comprises (i) at least one feature vector containing chemical properties and associated energies of intermediates and / or products generated in one or more reaction pathways corresponding to the thermal etching reaction; and (ii) a label representing an experimentally determined etching property.

[0078] One or more training set members may be provided for each thermal etching reaction. That is, one or more combinations of feature vectors and associated labels may be provided for a single thermal etching reaction. For example, a feature vector and its associated label may be provided for aluminum fluoride and silicon tetrachloride reactions. Another feature vector and its associated label may be provided for aluminum fluoride and germanium tetrachloride reactions. Another feature vector and its associated label may be provided for aluminum fluoride and tin tetrachloride reactions. And another feature vector and its associated label may also be provided for aluminum fluoride and titanium tetrachloride reactions. Aluminum fluoride represents a surface layer, and each metal chloride represents an etching precursor in a thermal etching reaction. In the case of a thermal ALE reaction cycle, each of these fluorides in aluminum fluoride may be produced in advance by a first ALE stage (e.g., converting aluminum oxide to aluminum trifluoride). Therefore, aluminum fluoride is a modified surface layer in a thermal ALE reaction. Each feature vector and associated label may be provided as a training set member in a training set.

[0079] Figure 6C A feature vector table for different thermal etching reactions and a corresponding label table representing etching characteristics of different thermal etching reactions according to some implementations are shown. The feature vector table 660 and the corresponding label table 670 may be provided as training set members in a training set for training a machine learning model.

[0080] The feature vector table 660 can be obtained from Figure 6A-6B The multiple quantum mechanical simulations 610 shown in FIG. 6 are generated, which are run across different thermal etching reactions. Each row in the feature vector table 660 includes the etching precursor M in the thermal etching reaction. x N y 661, and each column in the feature vector table 660 includes the surface layer S corresponding to the thermal etching reaction x T y662. Each cell in the feature vector table 660 includes the etching precursor M x N y 661 and surface layer S x T y 662, where each cell represents a different characteristic vector for a different thermal etching reaction. Figure 6A-6B At least one of the quantum mechanical simulations 610 run in generates a feature vector 663. In other words, a particular feature vector 663 may include Figure 6B One of the feature vectors 650 in, or Figure 6B Characteristic vector 663 contains the chemical identities and associated energies of reaction intermediates and / or products produced in one or more reaction pathways of a single thermal etching reaction.

[0081] The corresponding label table 670 can be generated by experiment or can be generated via a reliable simulator. Each row in the corresponding label table 670 includes the etching precursor M in the thermal etching reaction. x N y 671, and each column in the corresponding label table 670 includes the surface layer S in the corresponding thermal etching reaction x T y 672. Each cell in the corresponding tag table 670 includes an etching precursor M x N y 671 and surface layer S x T y 672, where each cell represents a different label for a different thermal etching reaction. Label 673 may indicate an etching characteristic determined experimentally or a predicted etching characteristic. Label 673 may include a binary value indicating whether etching will occur, an etching rate, or some other numerical characteristic regarding the effectiveness of etching. Figure 6C As shown in FIG. 6, label 673 shows an indication of the etching precursor M x N y 671 and surface layer S x T y 672 between zero values ​​indicating no etching and a value indicating etching precursor M x N y 671 and surface layer S x T y672. In the case where etching is indicated, the feature vector 663 in the feature vector 660 table is indicated by (i'j'k'), and in the case where etching is not indicated, the feature vector 663 in the feature vector 660 table is indicated by (ijk). The feature vectors 663 and corresponding labels 673 for multiple thermal etching reactions can be used as training set members in a training set.

[0082] In some implementations, training set members can be selected in the training set over the range of thermal etching reactions (i.e., etching reaction space), where the machine learning model is expected to make accurate predictions. The accuracy and range of the machine learning model in predicting etching characteristics may depend on the selection of appropriate training set members that are expected to make accurate models. For example, a training set member with a halogenated etching precursor may be more accurate than an etching precursor involving trimethylaluminum (TMA) in predicting etching characteristics in a thermal etching reaction involving an etching precursor of silicon tetrachloride (SiCl4). In some implementations, certain quantum mechanical models, surface characterizations, and initial conditions may provide more accurate predictions in some categories of thermal etching reactions than other types. The training set members in the training set may be optimized for accurate predictions. Eliminate training set members with poor predictions. Methods such as experimental design (DOE) based on statistical principles may be used to determine which factors are more important than other factors to select training set members for the optimization. The DOE does not necessarily include a surface modification step that converts, for example, Al2O3 into AlF3. On the contrary, Al2O3 and AlF3 samples may be exposed to etching precursors in independent experiments.

[0083] Back to Figure 5 At block 540 of process 500, a machine learning model trained using the training set is generated, wherein the machine learning model is configured to predict etching information in a thermal etching reaction. Using data points from a feature vector containing independent variables and a label containing a dependent variable, the machine learning model can be trained to recognize patterns and predict etching information in a thermal etching reaction.

[0084] An appropriate training algorithm may be used to train a machine learning model using a training set. The training algorithm may be used to identify patterns in data points between independent variables (inputs) and dependent variables (outputs) so as to accurately predict etching information (new outputs) when a new thermal etching reaction (new input) is presented. The training algorithm may be based on one of several machine learning algorithms. Machine learning algorithms may be divided into three major categories: supervised learning, unsupervised learning, and reinforcement learning. Although the present disclosure focuses on supervised learning, it should be understood that unsupervised learning, reinforcement learning, or other suitable forms of learning may be used to train a machine learning model.

[0085] Supervised learning is useful where the attributes (labels) of a certain data set (training set) are available. Examples of supervised machine learning algorithms include, but are not limited to, linear regression, logistic regression, decision trees, learning vector quantization, support vector machines (SVMs), Naive Bayes, k-nearest neighbor algorithms, random forests, and gradient enhancement. Semi-supervised learning is a supervised learning method with a small amount of labeled data and a large amount of unlabeled data for certain data sets. Unsupervised learning is useful in situations where implicit relationships in a given unlabeled data set (projects are not pre-assigned) have not yet been found. Examples of unsupervised machine learning algorithms include k-means. Reinforcement learning falls between supervised and unsupervised learning, where some feedback can be used for each prediction step or action, but there are no precise labels. Rather than presenting the correct input / output pair as in supervised learning, a given input is mapped to a reward function that the agent attempts to maximize. Examples of reinforced machine learning algorithms include Markov decision processes. Other types of learning that can fall into one or more of the above categories include, for example, deep learning and artificial neural networks (e.g., convolutional neural networks).

[0086] There may be many training tools or frameworks to train machine learning models. Examples of proprietary training tools include, but are not limited to, Amazon Machine Learning, Microsoft Azure Machine Learning Studio, DistBelief, Microsoft Cognitive Toolkit. Examples of open source training tools include, but are not limited to, Apache Singa, Caffe, H2O, PyTorch, MLPACK, Google TensorFlow, Torch, and Accord.Net.

[0087] The trained machine learning model can be used to predict etching information of a thermal etching reaction between an etching precursor and a surface layer. The trained machine learning model can take as input the chemical characteristics and associated energies of intermediates and / or products produced in one or more reaction paths in the thermal etching reaction, and provide as output the etching characteristics of the thermal etching reaction. In some implementations, such chemical characteristics and associated energies can be organized into a feature vector for use as input. In some implementations, such as Figure 4 The reaction pattern shown in for a single thermal etch reaction can be used as input.

[0088] The trained machine learning model can take one of several forms. In some implementations, the trained machine learning model is a classification and regression tree or a random forest tree. In some implementations, the trained machine learning model is an artificial neural network, such as a convolutional neural network. In some implementations, the trained machine learning model is a linear classifier, such as a linear regression, a logistic regression, or a support vector machine.

[0089] In the case where it is unclear or unknown whether a given etch precursor etches a surface layer, how fast or slow a given etch precursor etches a surface layer, how selectivity a given etch precursor has when etching a surface layer, or how effective a given etch precursor is when etching a surface layer, a machine learning model of the present disclosure can be trained to accurately predict the etching information. The etching information may include, for example, whether an etch precursor etches a surface layer, the etching rate of an etch precursor on a surface layer, or some other value indicating the effectiveness of an etch precursor in etching a surface layer. This can be used to determine the selectivity of an etch precursor when there are many materials on the substrate surface.

[0090] In some implementations, the machine learning model of the present disclosure can be used to identify candidate thermal etching reactions, including candidate thermal ALE reactions. For example, in the case where a substrate contains multiple different materials that will be exposed to an etching precursor at the same time, the machine learning model can identify what combination of etching precursors and / or surface layers will provide the desired selectivity in a given application. In some implementations, the machine learning model of the present disclosure can be used to identify candidate etching masks. For example, the machine learning model can identify which materials on the substrate will resist etching and can be used as an etching mask for a given etching precursor. In some implementations, the machine learning model of the present disclosure can be used to identify chamber materials that resist etching for a given etching precursor. In this way, the chamber walls and other components are not unintentionally etched by the etching precursor. In some implementations, the machine learning model can design a new etching process or reactor design by identifying one or more of the following: (i) candidate thermal etching reactions, (ii) candidate etching mask materials, and (iii) candidate chamber materials.

[0091] Figure 7 is a flow chart of an example method for predicting etching characteristics in a reaction between a surface layer and an etching precursor according to some implementations. The operations of process 700 may be performed in a different order and / or with different, fewer, or additional operations.

[0092] At block 710 of process 700, a quantum mechanical model is used to identify chemical properties and associated energies of one or more reaction paths in a simulated reaction between a surface layer and an etch precursor. In some implementations, the surface layer is a modified surface layer after a first stage of a thermal ALE reaction. In some implementations, the chemical properties include a bonding configuration or molecular structure of one or more reaction intermediates and / or products produced in the one or more reaction paths. In some implementations, the bonding configuration or molecular structure includes one or more of a single-bridge dimer, a double-bridge dimer, a triple-bridge dimer, and no dimer. The bonding configuration or molecular structure is shown in FIG. Figure 3A Possible reaction intermediates and / or products produced along the reaction pathway and their associated energies can be shown in an energy diagram. Figure 3B An example of an energy diagram is shown in FIG. 1 , which shows four different reaction pathways for four different thermal etching reactions. In some implementations, the chemical properties and associated energies can be expressed as reaction patterns. Examples of four different reaction patterns are shown in FIG. Figure 4 as shown in .

[0093] In some implementations, process 700 also includes organizing the chemical properties and associated energies of one or more reaction pathways of the simulated reaction into a feature vector. As described earlier, the feature vector contains at least the chemical properties and associated energies of one or more reaction pathways for the simulated reaction.

[0094] Using a quantum mechanical model configured with a selected surface representation of a surface layer and one or more selected initial conditions, determine the chemical properties and associated energy of one or more reaction paths in the simulated reaction. However, it should be understood that the quantum mechanical model can be configured with different, fewer or additional parameters. In some implementations, one or more selected initial conditions include the separation distance between the surface layer and the etching precursor at the beginning of the simulated reaction, the orientation and / or direction of the surface layer and the etching precursor at the beginning of the simulated reaction, the form of the initial chemical bridge between the surface layer and the etching precursor at the beginning of the simulated reaction, the internal energy or kinetic energy of the etching precursor or the surface layer at the beginning of the simulated reaction, or a combination thereof. However, it should be understood that the initial conditions are not limited to the aforementioned initial conditions, and other initial conditions can be configured or adjusted for the quantum mechanical model. In some implementations, the selected surface representation of the surface layer is selected from the group consisting of: molecules, small molecule clusters, and macromolecular clusters. In some implementations, the quantum mechanical model includes a density functional theory (DFT) model, a Hartree-Fock model, a semi-empirical model, or a combination thereof. For each quantum mechanical simulation, a unique combination of a quantum mechanical model, a selected surface representation, and one or more selected initial conditions can generate a separate eigenvector. In some implementations, the quantum mechanical model, surface representation, and initial conditions can be selected to optimize the accuracy of the prediction.

[0095] At block 720 of process 700, the chemical properties and associated energies in the simulated reaction are provided as inputs to the machine learning model. Other properties of the simulated reaction may be provided as inputs to the machine learning model. In some implementations, the inputs may also include one or both of the temperature and pressure of the simulated reaction.

[0096] In some implementations, the Figure 5 The process 500 described in is used to train the machine learning model. In some implementations, a training set comprising a plurality of training set members is used to train the machine learning model, each training set member comprising (i) a feature vector comprising chemical properties and associated energies for intermediates and / or products produced in one or more reaction pathways, the chemical properties and associated energies for thermal etching reactions simulated by at least one quantum mechanical simulation; and (ii) a label representing the properties of the thermal etching reaction. Each training set member includes at least one feature vector for a given thermal etching reaction and an associated label. In some implementations, for a given thermal etching reaction, a particular training set member may include a plurality of feature vectors across a plurality of quantum mechanical simulations. Each feature vector may include a unique combination of: a specified etching precursor, a specified surface layer, a specified quantum mechanical model, a specified surface representation of a specified surface layer, and one or more specified initial conditions. The machine learning model may be trained to recognize patterns across the training set members in the training set so as to be able to accurately predict the etching properties of the simulated reaction. The machine learning model may be trained using any suitable training algorithm including any of the training algorithms described above.

[0097] At block 730 of process 700, an etching characteristic of a simulated reaction between the surface layer and the etching precursor is determined using a machine learning model. In some implementations, the etching characteristic of the simulated reaction includes an etching rate of the surface layer. In some implementations, the etching characteristic of the simulated reaction includes an indication of whether the surface layer is etched.

[0098] In some implementations, the machine learning model can be validated against experimental measurements. In some implementations, process 700 also includes validating the determined etching characteristics of the simulated reaction based on experimental measurements of the etching characteristics. This can be done by comparing the etching characteristics determined at block 730 of process 700 to the experimental measurements of the etching characteristics. The machine learning model can continue to learn and be trained with this feedback in order to improve its prediction accuracy.

[0099] Figure 8A schematic diagram of validating a machine learning model from experimental measurements according to some implementations is shown. The machine learning model can use a feature vector table 850 and a corresponding label table 860 to predict an etching characteristic 863 of a new thermal etching reaction between a new etching precursor and a new surface layer. The feature vector table 850 and the corresponding label table 860 may include a training set, which includes a specific etching reaction span. A new feature vector 853 may be generated from at least one quantum mechanical simulation of a new thermal etching reaction, and the new feature vector 853 may be provided in the feature vector table 850. The machine learning model predicts the expected label indicated by the etching characteristic 863 in the corresponding label table 860 as follows. The etching characteristic 863 is compared with an experimentally measured value or an experimentally determined label 873 to confirm the accuracy of the prediction. Based on the feedback provided after comparing the etching characteristic 863 with the experimentally determined label 873, the set of feature vectors 880 used to generate the feature vector table 850 is adjusted to maximize the prediction success. The set of feature vectors 880 may be generated from multiple quantum mechanical simulations of different parameters for each thermal etching reaction. What parameters are used and what quantum mechanical simulations are used in the set of feature vectors 880 used to generate the training set will be adjusted based on feedback from the experimentally determined labels 873.

[0100] The machine learning model can receive ongoing training or improvement using feedback from new measurements of etching that confirm or reject the results predicted by the machine learning model. Feedback to the machine learning model can appear not only in the form of new measurements of etching, but also in the form of reaction intermediates and / or products identified in the thermal etching reaction. For example, the feedback can be received from a mass spectrometer or other instrument equipped in a reactor or chamber. The feedback can be used to further improve the parameters and quantum mechanical simulations used to generate the feature vectors of a given training set. In other words, the parameters and quantum mechanical simulations used to generate the feature vectors can be improved based on experimentally determined reaction intermediates and / or products. In some implementations, a mass spectrometer or other instrument equipped with a reactor can be integrated with software for the machine learning model to provide continuous training.

[0101] In some implementations, the machine learning model can be used to design a thermal etching process or a thermal ALE process. Specifically, the machine learning model can be used to identify candidate thermal etching reactions, candidate etching mask materials, and candidate chamber materials with etching precursors and / or surface layers that achieve a desired selectivity and etching rate to resist etching for a given etching chemistry. In some implementations, Figure 7The process 700 in also includes identifying one or more candidate thermal etching reactions using a machine learning model, wherein each of the one or more identified candidate thermal etching reactions includes at least an identified etching precursor. Each of the one or more identified candidate thermal reactions also includes an identified material (which is to be etched by the identified etching precursor), an identified etching mask material, and / or an identified chamber material. The machine learning model can identify a combination of materials for thermal etching at a desired selectivity without undergoing excessive experimentation.

[0102] Fig. 9 A feature vector table for different thermal etching reactions according to some implementations is shown, wherein the identified etching precursors and the identified surface layers are used to design candidate thermal etching reactions. Feature vector table 950 includes rows of etching precursors 951 and columns of surface layers 952. For a given etching precursor 951, a pair of surface layers 960 can be identified: one surface layer that is etched, and one surface layer that is not etched. For example, if a pair of surface layers 960 are varied by a single component (e.g., Al2O3 and AlF3), a surface modification step can be designed to produce a pair of surface layers 960 such that one surface layer is modified and the other surface layer remains the main material. In this way, the etching precursor 951 can selectively etch the modified surface layer without etching the main material.

[0103] in conclusion

[0104] In the foregoing description, numerous specific details are set forth to provide a thorough understanding of the implementations presented. The disclosed implementations may be practiced without some or all of these specific details. In other cases, well-known process operations are not described in detail so as not to unnecessarily obscure the disclosed implementations. Although the disclosed implementations are described in conjunction with the implementations, it should be understood that they are not intended to limit the disclosed implementations.

[0105] Although the foregoing implementation has been described in some detail for the purpose of clarity of understanding, it is apparent that certain changes and modifications may be practiced within the scope of the appended claims. It should be noted that there are many alternative ways of implementing the processes, systems, and devices of the present implementation. Therefore, the present implementation is to be considered illustrative rather than restrictive, and the implementation is not limited to the details given herein.

Claims

1. A method for predicting etching characteristics in a reaction between a surface layer and an etching precursor, the method comprising: (a) using a quantum mechanical model to identify the chemical nature and associated energies of one or more reaction pathways in a simulated reaction between the surface layer and the etch precursor; (b) providing the chemical properties and associated energies in the simulated reaction as input to a machine learning model; as well as (c) determining an etching characteristic of the simulated reaction between the surface layer and the etching precursor using the machine learning model.

2. The method of claim 1 , wherein the chemical properties and associated energies of the one or more reaction pathways in the simulated reaction are determined using a quantum mechanical model configured with a selected surface representation of the surface layer and one or more selected initial conditions.

3. The method of claim 2, wherein the one or more selected initial conditions include a separation distance between the surface layer and the etching precursor when the simulated reaction starts, an orientation and / or direction of the surface layer and the etching precursor when the simulated reaction starts, a form of an initial chemical bridge between the surface layer and the etching precursor when the simulated reaction starts, an internal energy or kinetic energy of the etching precursor or the surface layer when the simulated reaction starts, or a combination thereof.

4. The method of claim 2, wherein the selected surface characterization of the surface layer is selected from the group consisting of molecules, small molecule clusters, and large molecule clusters, wherein the small molecule clusters are equal to or less than 20 atoms and the large molecule clusters are greater than 20 atoms. 5 . The method of claim 1 , wherein the quantum mechanical model comprises a density functional theory (DFT) model, a Hartree-Fock model, a semi-empirical model, or a combination thereof.

6. The method of any one of claims 1 to 5, wherein the input comprises one or both of a temperature and a pressure of the simulated reaction.

7. The method of any one of claims 1 to 5, wherein the chemical characteristics include the bonding configuration or molecular structure of one or more reaction intermediates and / or products produced in the one or more reaction pathways.

8. The method according to claim 7, wherein: The machine learning model is trained using a training set comprising a plurality of training set members, each training set member comprising: (i) a feature vector comprising chemical properties and associated energies for intermediates and / or products produced in one or more reaction pathways for a thermal etching reaction simulated by at least one quantum mechanical simulation; and (ii) a label representing the identity of the thermal etching reaction.

9. The method of claim 8, wherein each of the feature vectors comprises a unique combination of a specified etch precursor, a specified modified surface layer, a specified quantum mechanical model, a specified surface characterization of the specified modified surface layer, and one or more specified initial conditions.

10. The method according to any one of claims 1 to 5, further comprising: One or more candidate thermal etch reactions are identified using the machine learning model, wherein each of the one or more identified candidate thermal etch reactions includes at least an identified etch precursor.

11. A method for generating a machine learning model for predicting etching information in a thermal etching reaction: performing at least one quantum mechanical simulation for each of a plurality of thermal etching reactions, each quantum mechanical simulation generating chemical characteristics and associated energies of one or more reaction pathways in a corresponding thermal etching reaction between the surface layer and the etching precursor; determining an experimentally determined etch characteristic for each of the plurality of thermal etch reactions; Generating a training set comprising a plurality of training set members, each training set member comprising (i) at least one feature vector containing chemical properties and associated energies of intermediates and / or products generated in one or more reaction pathways for the corresponding thermal etching reaction; and (ii) a label representing the experimentally determined etching property; as well as A machine learning model trained using the training set is generated, wherein the machine learning model is configured to predict etching information in the thermal etching reaction.

12. A method according to claim 11, wherein performing the at least one quantum mechanical simulation includes performing multiple quantum mechanical simulations for each of the multiple thermal etching reactions, and wherein each eigenvector contains the chemical properties and associated energies of intermediates and / or products produced in the one or more reaction pathways by one of the multiple quantum mechanical simulations of the corresponding thermal etching reaction.

13. The method of claim 11, wherein each of the at least one quantum mechanical simulation comprises a quantum mechanical model configured with a surface representation of a surface layer and one or more initial conditions.

14. The method of claim 13, wherein the one or more initial conditions include a separation distance between the surface layer and the etching precursor at the start of the quantum mechanical simulation, an orientation and / or direction of the surface layer and the etching precursor at the start of the quantum mechanical simulation, a form of an initial chemical bridge between the surface layer and the etching precursor at the start of the quantum mechanical simulation, an internal energy or kinetic energy of the etching precursor or the surface layer at the start of the quantum mechanical simulation, or a combination thereof.

15. The method of claim 13, wherein the surface characterization of the surface layer is selected from the group consisting of molecules, small molecule clusters, and large molecule clusters, wherein the small molecule clusters are equal to or less than 20 atoms and the large molecule clusters are greater than 20 atoms.

16. The method according to any one of claims 11 to 15, wherein generating the training set comprises: organizing chemical characteristics and associated energies of the one or more reaction pathways in the corresponding thermal etching reaction into the at least one feature vector; and The experimentally determined etching characteristics are organized into the tags.

Images