High-throughput screening method, system and medium for self-assembled single-layer thermal interface material

Abstract

The invention relates to a high-throughput screening method and system for a self-assembled single-layer thermal interface material and a medium. The high-throughput screening method comprises the following steps: constructing an interface heat transport simulation system based on a self-assembled monomolecular layer; screening set preparation is carried out on the end group structure, and an end group screening set and a pre-calculation set are constructed; carrying out molecular dynamics simulation on the self-assembled monomolecular layer corresponding to the pre-calculation set to obtain a physical association index; training a machine learning model by taking the physical association index as a machine learning objective function; carrying out machine learning prediction on the uncalculated screening candidate data set based on the trained machine learning model, and carrying out parallel screening of physical association indexes to obtain a simulation verification set; and carrying out molecular dynamics simulation for verification. Compared with the prior art, high-throughput screening of the self-assembled single-layer thermal interface material is achieved, and the method has important significance on solid-liquid interface heat conduction characteristic regulation and control design under data driving.

Description

Technical Field

[0001] The present invention belongs to the field of material genetic engineering, and in particular relates to a high-throughput screening method, system and medium for self-assembled single-layer thermal interface materials. Background Art

[0002] The study of heat transport at solid-liquid interfaces is of great theoretical significance and holds significant value in applications such as micro- and nano-devices, chip thermal interface materials, nanofluid-enhanced heat transfer, and solar thermal conversion. In many applications, the solid phase material is typically a metal, such as gold or copper, which is characterized by high thermal conductivity. In contrast to solids, most liquids (such as water) lack efficient heat transfer mechanisms, with thermal conductivities typically below 1 W / mK.

[0003] As the characteristic dimensions of a system approach the nanoscale, the thermal resistance of the solid-liquid interface determines the thermal conductivity of the entire material. Using self-assembled monolayers for interface modification is a new strategy for enhancing interfacial thermal conductivity by manipulating interfacial heat transport at the molecular level. In general, self-assembled monolayers can functionalize the material surface, thereby improving the physical connection at the solid-water interface, reducing the vibrational mismatch between the two phases, and ultimately lowering the interfacial thermal resistance. Such self-assembled monolayer thermal interface materials have great potential for modulating the interfacial thermal resistance of water-based and other systems.

[0004] At present, the design work of many types of materials has achieved high-throughput screening by preliminarily constructing a calculation data set under a unified system. In terms of material oxygen release performance, CN119049589A discloses a data-driven material oxygen release performance analysis prediction and screening method, system, and medium, which achieves high-throughput screening by constructing a machine learning model to obtain a regression prediction of oxygen release performance. CN116631522A discloses a high-throughput calculation method and screening system for catalytic materials, which carries out screening of catalytic materials by constructing a high-throughput computing framework to implement simulation calculations. CN119380888A discloses a high-throughput screening method and system for polymer thermoelectric materials, which quickly obtains target materials through multiple screening modules combined with first-principles calculations. In short, high-throughput screening is the mainstream method under the current new paradigm of material research and development.

[0005] However, the existing experimental and computational data for self-assembled monolayer thermal interface materials is very limited, which makes it difficult for researchers to develop effective and reliable machine learning models. Secondly, directly obtaining the solid-liquid interface thermal conductivity characteristics of self-assembled monolayers requires extremely high costs both in experiments and simulations, which also hinders the high-throughput screening of self-assembled monolayer thermal interface materials. In addition, the current data characteristics for self-assembled monolayers (organic molecules) are not strongly correlated with interfacial heat transport, resulting in unreliable screening results.

[0006] Therefore, in order to regulate the interfacial thermal conductivity of self-assembled monolayers, it is urgent to develop a high-throughput screening method that can assist in the design of self-assembled monolayer thermal interface materials to assist scientific researchers in carrying out new paradigm research. Summary of the Invention

[0007] The purpose of the present invention is to overcome the current lack of high-throughput screening for self-assembled monolayer thermal interface materials by providing a method, system, and medium for high-throughput screening of self-assembled monolayer thermal interface materials. The high-throughput screening method of the present invention can achieve high thermal conductivity screening of self-assembled monolayer thermal interface materials at solid-liquid interfaces, assisting researchers in conducting data-driven research such as machine learning on complex self-assembled monolayers, and providing guidance for high-throughput screening of self-assembled monolayer thermal interface materials in the field of materials genetic engineering.

[0008] The purpose of the present invention can be achieved by the following technical solutions:

[0009] The present invention first provides a high-throughput screening method for self-assembled monolayer thermal interface materials, comprising the following steps:

[0010] S1: Construct an interfacial heat transport simulation system based on a self-assembled monolayer, a gold substrate, and water molecules; the monomer chain structure of the self-assembled monolayer is -S(CH2) n -X, X is a terminal structure;

[0011] S2: Prepare a screening set for the terminal structure X of S1, construct an terminal screening set and extract some fragments from it as a pre-calculation set;

[0012] S3: Perform molecular dynamics simulations on the self-assembled monolayer corresponding to the pre-calculated set in S2 to obtain physical correlation indicators including interfacial interaction energy and vibrational spectrum coupling;

[0013] S4: Use the physical correlation indicators obtained in S3 as the machine learning objective function and independently carry out machine learning model training;

[0014] S5: Based on the machine learning model trained in S4, machine learning predictions are performed on the physical indicators of the screening candidate data set that has not been subjected to molecular dynamics simulation calculations. Subsequently, parallel screening of the physical correlation indicators of the two sets is performed to obtain a simulation verification set;

[0015] S6: Perform molecular dynamics simulation on the simulation verification set screened in S5 to obtain the thermal conductivity of the solid-liquid interface.

[0016] Furthermore, in step S2, the specific steps of preparing the screening set are:

[0017] S2.1: Extract short polymer fragments from the ZINC database as end group screening sets;

[0018] S2.2: Visualize the 2D chemical structures of short polymer fragments using UMAP to verify the distribution range and structural diversity of the end group screening set

[0019] S2.3: Randomly select a certain proportion of short polymer fragments from the end group screening set as the pre-calculation set;

[0020] S2.4: Perform statistical analysis on the substructures of the molecular fragments in the pre-calculated set and the end group screening set to verify the number and types of substructures contained in the end group screening set.

[0021] Furthermore, in step S3, the terminal group X structure is in a state of being in a state of being in a state of being in a state of being in a state of being in a state of being n - Use RDKit or STK software to write automated programs to achieve chain splicing.

[0022] Furthermore, in step S3, a mixed force field distribution is used in the molecular dynamics simulation.

[0023] Furthermore, the bond interactions between the sulfur atoms in the self-assembled monolayer and the gold atoms in the gold substrate are simulated using a Morse force field.

[0024] Furthermore, the self-assembled monolayer molecules are simulated using PCFF potential, and the water molecules are simulated using TIP3P potential.

[0025] Furthermore, the non-bonded interactions among the gold atoms, the self-assembled monolayer molecules, and the water molecules are simulated using LJ potential.

[0026] Furthermore, in step S2, the long-range electrostatic interactions of the simulated system in the molecular dynamics simulation are calculated using the PPPM method.

[0027] Furthermore, in step S2, the time step of the simulation system in the molecular dynamics simulation is set to 0.2-0.3 fs, preferably 0.25 fs.

[0028] Furthermore, in step S3, the interface interaction energy E is referred to formula (1):

[0029]

[0030] Where, F ij is the force exerted by atom j in the gold substrate or water molecule on atom i in the self-assembled monolayer, Δd j is the displacement of atom j; F ji is the force exerted by atom i in the self-assembled monolayer on atom j in the gold substrate or water molecule, Δdi is the displacement of atom i.

[0031] Furthermore, in step S3, the screening criterion for the vibration spectrum coupling is specifically the overlapping area ratio of the vibration spectra between the self-assembled monolayer and the water molecules.

[0032] Furthermore, the vibration spectrum VDOS is obtained by Fourier transform of the velocity autocorrelation function, referring to formula (2):

[0033]

[0034] Where ω is the vibration wave number, v i is the atomic velocity, t0 and t are the initial time and correlation time, respectively.

[0035] Furthermore, in step S4, the machine learning model includes an interface interaction energy machine learning model and a vibration spectrum coupling machine learning model, and three types of machine learning models, random forest (RF), extreme gradient boosting (XGB) and gradient boosting (GB) regression, are deployed and 10-fold cross-validation is performed. Each training uses the global optimization tool Bayesian optimization package to adjust the hyperparameters of all machine learning models.

[0036] Furthermore, in step S5, obtaining the simulation verification set specifically includes the following steps:

[0037] S5.1: Use the machine learning model to predict interfacial interaction energy and vibrational spectral coupling, calculate weighted prediction values, and use the weighted prediction values ​​as the ranking basis to select the top 20% of the end-group structures for each of the two physical correlation indicators as candidate structures. In addition, select the top 20% of the pre-calculated end-group structures as candidate structures. These two groups of structures are screened according to the process.

[0038] S5.2: Find the intersection of the two screening results of interfacial interaction energy and vibrational spectrum coupling, and select candidate structures with high interfacial interaction energy and strong vibrational spectrum coupling;

[0039] S5.3: Based on the synthesizable coefficient, eliminate molecular fragments with higher synthesis difficulty from the candidate structures, and finally obtain the simulation verification set.

[0040] Furthermore, in step S6, non-equilibrium molecular dynamics simulation is used in the molecular dynamics simulation to obtain the heat flux density in the simulation system and the temperature gradient obtained by fitting at the interface.

[0041] Furthermore, the interfacial thermal conductivity of the self-assembled monolayer is referred to formula (3);

[0042] G = q / ΔT (3);

[0043] Where G is the interfacial thermal conductivity, q is the heat flux density, and ΔT is the temperature gradient at the interface in the direction of heat transport.

[0044] The present invention also provides a high-throughput screening system for self-assembled monolayer thermal interface materials, the high-throughput screening system comprising:

[0045] Modeling module, used to construct a solid-liquid interface simulation system based on self-assembled monolayers, gold substrates, and water molecules;

[0046] The screening set preparation module is used to construct the end group screening set and extract some fragments from it as the pre-calculation set;

[0047] The molecular dynamics simulation module is used to perform molecular dynamics simulations on the self-assembled monolayer corresponding to the pre-calculated set to obtain physical correlation indicators including interface interaction energy and vibrational spectrum coupling;

[0048] The machine learning model training module is used to independently train the machine learning model using physical correlation indicators as the objective function;

[0049] The multi-objective screening and simulation verification module is used to perform machine learning predictions on the end group set for which molecular dynamics simulation has not been performed to calculate physical indicators, and to perform simulation verification based on the simulation verification set obtained by screening.

[0050] The present invention also provides a storage medium comprising computer executable instructions, wherein the storage medium comprising computer executable instructions is used to perform a high-throughput screening method for self-assembled monolayer thermal interface materials when executed by a computer processor.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] (1) The present invention successfully achieved high-throughput screening of self-assembled monolayer thermal interface materials through the steps of screening set preparation, interface thermal conductivity screening index calculation, machine learning virtual screening model construction, multi-target screening and simulation verification, which is of great significance for the data-driven design of self-assembled monolayer interface thermal conductivity properties regulation.

[0053] (2) The high-throughput screening method of the present invention uses molecular dynamics simulation to obtain relevant interfacial heat transport data, which solves the problem of very limited existing experimental and computational data. By calculating physical indicators related to solid-liquid interface heat transport, such as interfacial interaction energy and vibration spectrum coupling, the method solves the current problems of lack of interfacial heat transport data and inconsistent acquisition methods in the design of self-assembled monolayer structures. The method only requires a relatively low computational cost to achieve high-throughput screening design of any self-assembled monolayer.

[0054] (3) The high-throughput screening method of the present invention unifies the acquisition process in the theoretical calculation process of the thermal conductivity characteristics of the self-assembled monolayer interface, realizes the automation of the program throughout the process, and has guiding significance in the high-throughput screening of self-assembled monolayer thermal interface materials in the field of material genetic engineering.

[0055] (4) The high-throughput screening method of this embodiment is highly applicable. During the simulation calculation process, by changing the relevant parameter settings in accordance with common sense, the physical correlation indicators and interface thermal conductivity associated with the interfacial thermal conductivity characteristics of the self-assembled monolayer can also be obtained. The data set can also be modified in terms of set dimensions (quantity) according to design requirements. In addition, the present invention can also obtain a machine learning virtual screening model related to the physical correlation of interfacial heat transport by changing the relevant framework settings in accordance with common sense.

[0056] (5) The present invention can overcome the problems of high preparation and characterization costs and poor reliability of high-throughput screening under small samples caused by the lack of strongly correlated physical screening indicators in traditional material research and development methods. It can realize high-throughput screening of self-assembled monolayer thermal interface materials, greatly improve the fitting performance and screening reliability of the machine learning virtual screening model of self-assembled monolayer interface thermal transport, and save more costs and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 Schematic diagram of the high-throughput screening method of the present invention.

[0058] Figure 2 Schematic diagram of the structure of the high-throughput screening system of the present invention.

[0059] Figure 3 Schematic diagram of the high-throughput screening method in Example 2 of the present invention.

[0060] Figure 4 Schematic diagram of the end group screening set of the self-assembled monolayer in Example 2 of the present invention.

[0061] Figure 5 Schematic diagram of the data distribution of the interface interaction energy and vibration spectrum overlapping area ratio of the pre-calculated set in Example 2 of the present invention.

[0062] Figure 6 This is a diagram showing the training effect of the machine learning model for interface interaction energy and vibration spectrum overlapping area ratio in Example 2 of the present invention.

[0063] Figure 7 Schematic diagram of the calculation results of interface thermal conductivity in Example 2 of the present invention.

[0064] Figure 8Schematic diagram of the end group structure of the self-assembled monolayer thermal interface material obtained after high-throughput screening in Example 2 of the present invention.

[0065] Description of the marks in the figure:

[0066] 1-Modeling module, 2-Screening set preparation module, 3-Molecular dynamics simulation module, 4-Machine learning model training module, 5-Multi-target screening and simulation verification module. DETAILED DESCRIPTION

[0067] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention and provides a detailed implementation method and specific operating process. However, the scope of protection of the present invention is not limited to the following embodiment. Features such as component models, material names, connection structures, control methods, algorithms, etc. that are not clearly described in this technical solution are considered common technical features disclosed in the prior art.

[0068] The purpose of the present invention is to solve the problem that there is still no high-throughput screening method for self-assembled monolayer thermal interface materials that can regulate the thermal conductivity of the solid-liquid interface in the current self-assembled monolayer structure design. In view of the fact that existing experimental and computational data are very limited, the present invention adopts molecular dynamics simulation to obtain relevant interface heat transport data. In order to solve the problem of high computational cost, the present invention effectively avoids the cost of fully calculating the solid-liquid interface thermal conductivity of the entire material by calculating physical process data such as interface interaction energy and vibration spectrum coupling. Through the high-throughput screening method of the present invention, high-throughput screening of self-assembled monolayer thermal interface materials can be achieved, assisting scientific researchers to carry out data-driven research such as machine learning on self-assembled monolayers under complex structures, filling this research gap, and providing theoretical support and experimental basis for related applications.

[0069] Specifically, the first aspect of the present invention provides a high-throughput screening method for self-assembled monolayer thermal interface materials, such as Figure 1 As shown, the following steps are included:

[0070] S1: Construct an interfacial heat transport simulation system based on a self-assembled monolayer, a gold substrate, and water molecules; the monomer chain structure of the self-assembled monolayer is -S(CH2) n -X, X is a terminal structure;

[0071] S2: Prepare a screening set for the terminal structure X of S1, construct an terminal screening set and extract some fragments from it as a pre-calculation set;

[0072] S3: Perform molecular dynamics simulations on the self-assembled monolayer corresponding to the pre-calculated set in S2 to obtain physical correlation indicators including interfacial interaction energy and vibrational spectrum coupling;

[0073] S4: Use the physical correlation indicators obtained in S3 as the machine learning objective function and independently carry out machine learning model training;

[0074] S5: Based on the machine learning model trained in S4, machine learning predictions are performed on the physical indicators of the screening candidate data set that has not been subjected to molecular dynamics simulation calculations. Subsequently, parallel screening of the physical correlation indicators of the two sets is performed to obtain a simulation verification set;

[0075] S6: Perform molecular dynamics simulation on the simulation verification set screened in S5 to obtain the thermal conductivity of the solid-liquid interface.

[0076] In some specific embodiments, in step S2, the specific steps of preparing the screening set are:

[0077] S2.1: Extract short polymer fragments from the ZINC database as end group screening sets;

[0078] S2.2: Visualize the 2D chemical structures of short polymer fragments using UMAP to verify the distribution range and structural diversity of the end group screening set

[0079] S2.3: Randomly select a certain proportion of short polymer fragments from the end group screening set as the pre-calculation set;

[0080] S2.4: Perform statistical analysis on the substructures of the molecular fragments in the pre-calculated set and the end group screening set to verify the number and types of substructures contained in the end group screening set.

[0081] In some specific embodiments, in step S3, the terminal group X structure is -S(CH2) n - Use RDKit or STK software to write automated programs to achieve chain splicing.

[0082] In some specific embodiments, in step S3, a mixed force field distribution is used in the molecular dynamics simulation.

[0083] In some more specific embodiments, the bond interactions between the sulfur atoms in the self-assembled monolayer and the gold atoms in the gold substrate are simulated using a Morse force field.

[0084] In some more specific embodiments, the self-assembled monolayer molecules are simulated using PCFF potential, and the water molecules are simulated using TIP3P potential.

[0085] In some more specific embodiments, the non-bonded interactions among the gold atoms, the self-assembled monolayer molecules, and the water molecules are simulated using LJ potential.

[0086] In some specific embodiments, in step S2, the long-range electrostatic interactions of the simulated system in the molecular dynamics simulation are calculated using the PPPM method.

[0087] In some specific embodiments, in step S2, the time step of the simulation system in the molecular dynamics simulation is set to 0.2-0.3 fs, preferably 0.25 fs.

[0088] In some specific embodiments, in step S3, the interfacial interaction energy E is referred to formula (1):

[0089]

[0090] Where, F ij is the force exerted by atom j in the gold substrate or water molecule on atom i in the self-assembled monolayer, Δd j is the displacement of atom j; F ji is the force exerted by atom i in the self-assembled monolayer on atom j in the gold substrate or water molecule, Δd i is the displacement of atom i.

[0091] In some specific embodiments, in step S3, the screening criterion for the vibrational spectrum coupling is specifically the overlapping area ratio of the vibrational spectra between the self-assembled monolayer and the water molecules.

[0092] In some more specific embodiments, the vibration spectrum VDOS is obtained by Fourier transform of the velocity autocorrelation function, referring to formula (2):

[0093]

[0094] Where ω is the vibration wave number, v i is the atomic velocity, t0 and t are the initial time and correlation time, respectively.

[0095] In some specific embodiments, in step S4, the machine learning model includes an interface interaction energy machine learning model and a vibration spectrum coupling machine learning model, and three types of machine learning models, namely random forest (RF), extreme gradient boosting (XGB) and gradient boosting (GB) regression, are deployed and 10-fold cross validation is performed. Each training uses the global optimization tool Bayesian optimization package to adjust the hyperparameters of all machine learning models.

[0096] In some specific implementations, in step S5, obtaining the simulation verification set specifically includes the following steps:

[0097] S5.1: Use the machine learning model to predict interfacial interaction energy and vibrational spectral coupling, calculate weighted prediction values, and use the weighted prediction values ​​as the ranking basis to select the top 20% of the end-group structures for each of the two physical correlation indicators as candidate structures. In addition, select the top 20% of the pre-calculated end-group structures as candidate structures. These two groups of structures are screened according to the process.

[0098] S5.2: Find the intersection of the two screening results of interfacial interaction energy and vibrational spectrum coupling, and select candidate structures with high interfacial interaction energy and strong vibrational spectrum coupling;

[0099] S5.3: Based on the synthesizable coefficient, eliminate molecular fragments with higher synthesis difficulty from the candidate structures, and finally obtain the simulation verification set.

[0100] In some specific embodiments, in step S6, the molecular dynamics simulation uses non-equilibrium molecular dynamics simulation to obtain the heat flux density in the simulation system and the temperature gradient obtained by fitting at the interface.

[0101] In some more specific embodiments, the non-equilibrium molecular dynamics simulation employs an NVE system.

[0102] In some more specific embodiments, the non-equilibrium molecular dynamics simulation uses a Langevin heat bath as a heat source at 350K and a cold source at 250K, and the duration is 2.0-5.0 ns.

[0103] In some specific embodiments, the interfacial thermal conductivity of the self-assembled monolayer is referred to formula (3);

[0104] G = q / ΔT (3);

[0105] Where G is the interfacial thermal conductivity, q is the heat flux density, and ΔT is the temperature gradient at the interface in the direction of heat transport.

[0106] The present invention also provides a high throughput screening system for self-assembled monolayer thermal interface materials. Figure 2As shown, the high-throughput screening system includes: a modeling module 1, which is used to construct a solid-liquid interface simulation system based on a self-assembled monolayer, a gold substrate and water molecules; a screening set preparation module 2, which is used to construct an end group screening set and extract some fragments therefrom as a pre-calculated set; a molecular dynamics simulation module 3, which is used to perform molecular dynamics simulation on the self-assembled monolayer corresponding to the pre-calculated set to obtain physical correlation indicators including interface interaction energy and vibration spectrum coupling; a machine learning model training module 4, which is used to independently carry out the training of the machine learning model with the physical correlation indicators as the objective function; a multi-target screening and simulation verification module 5, which is used to perform machine learning prediction on the end group set for which molecular dynamics simulation has not been carried out to calculate the physical indicators, and to perform simulation verification based on the simulation verification set obtained by the screening.

[0107] The present invention also provides a storage medium comprising computer executable instructions, wherein the storage medium comprising computer executable instructions is used to perform a high-throughput screening method for self-assembled monolayer thermal interface materials when executed by a computer processor.

[0108] The storage medium of the present invention can be an electronic medium, a magnetic medium, an optical medium, an electromagnetic medium, an infrared medium, or a semiconductor system or a propagation medium. The storage medium can also include a semiconductor or solid-state memory, a magnetic tape, a removable computer disk, a random access memory (RAM), a read-only memory (ROM), a hard disk, and an optical disk. Optical disks can include compact disk-read only memory (CD-ROM), compact disk-read-write (CD-RW), and DVD.

[0109] The above embodiments can be implemented individually or in any combination of two or more. The above embodiments will be described in more detail below with reference to specific examples.

[0110] Example 1:

[0111] This embodiment provides a high-throughput screening method for self-assembled monolayers for regulating thermal conductivity at a solid-liquid interface, comprising the following steps:

[0112] (1) Construct an interfacial heat transport simulation system based on a self-assembled monolayer, a gold substrate, and water molecules; wherein the monomer chain structure of the self-assembled monolayer is -S(CH2) n -X, X is a terminal structure.

[0113] (2) Prepare a screening set for the end groups of the self-assembled monolayer, extract short polymer fragments from the ZINC database as the end group screening set, and randomly select a certain proportion of candidate fragments as the pre-calculated set;

[0114] (3) Molecular dynamics simulations were performed on the pre-calculated self-assembled monolayers to obtain physical correlation indicators such as interface interaction energy and vibrational spectrum overlap area ratio;

[0115] (4) Taking the pre-calculated set of self-assembled monolayer interface interaction energy and vibration spectrum overlap area ratio as the machine learning objective function, the machine learning model training work is carried out independently;

[0116] (5) The machine learning model that has been trained on the pre-calculated set is used to perform machine learning predictions on the associated physical indicators of the remaining end group candidate sets that have not been simulated. Then, the physical correlation indicators of the two sets are screened in parallel, and finally, the final simulation verification set is obtained by combining the synthesis coefficient screening;

[0117] (6) Non-equilibrium molecular dynamics simulation is performed on the self-assembled monolayer obtained by the final screening to obtain the thermal conductivity of the solid-liquid interface.

[0118] In step (1), the self-assembled monolayer chain consists of three parts: the head group -S-, the chain -(CH2) n - and terminal group X. The terminal group X structure is the same as -S(CH2) n Chain splicing was performed by writing an automated program using RDKit or STK software. Short polymer fragments were extracted from the ZINC database as candidate end group screening datasets. Two-dimensional chemical structures were visualized using UMAP to verify the distribution range and structural diversity of the end group screening set. Due to the scarcity of thermal transport data, a certain proportion of input structures was randomly selected as the initial computational training dataset. Finally, a statistical analysis of the molecular fragment substructures of the precomputed end group set and the total candidate end group set was performed to verify the number and type of substructures contained in the end group set.

[0119] In step (2), a mixed force field distribution was used in the molecular dynamics simulation, in which the Morse force field was used to simulate the bond interactions between the sulfur atoms in the sulfur-based self-assembled layer and the gold atoms in the gold substrate. The PCFF potential was used for the self-assembled monolayer molecules, and the TIP3P potential was used for the water molecules. The LJ potential was used to simulate the non-bonded interactions between the gold atoms, the self-assembled monolayer molecules, and the water molecules. Except for the sulfur atoms covalently bound to the gold atoms, the modified universal force field (UFF) was selected as the parameter setting of the LJ potential. The long-range electrostatic interactions of the entire system were calculated using the PPPM method with an accuracy of 1×10 -5 Considering the presence of lighter hydrogen atoms in the system, 0.25 fs is selected as the system time step. The cutoff radius of the Morse force field in the calculation is set to The cutoff radius of the LJ potential is set to

[0120] In step (3), in the calculation of the interface thermal conductivity screening index, using physical factors that are strongly correlated with interface heat transfer as screening criteria can more effectively screen out end-group structures with high interface thermal conductivity. Generally, strong interface bonding is conducive to interface heat transfer, while vibration mismatch between interfaces hinders the enhancement of interface thermal conductivity. Therefore, this embodiment selects interface interaction energy and vibration spectrum coupling as physical correlation indicators.

[0121] In this example, the gold substrate, the self-assembled monolayer, and water are grouped and the total energy between different groups of atoms at the interface is calculated. The calculation method of the interfacial interaction energy E is referred to formula (1).

[0122]

[0123] Where, F ij is the force exerted by atom j in the gold substrate or water molecule on atom i in the self-assembled monolayer, Δd j is the displacement of atom j; F ji is the force exerted by atom i in the self-assembled monolayer on atom j in the gold substrate or water molecule, Δd i is the displacement of atom i.

[0124] The vibration spectrum can be calculated by Fourier transform of the velocity autocorrelation function, referring to formula (2),

[0125]

[0126] Where ω is the vibration wave number, v i is the atomic velocity, t0 and t are the initial time and correlation time, respectively.

[0127] To quantify the strength of vibrational spectral coupling, the overlapping area ratio of the vibrational spectra between the self-assembled monolayer and water molecules was used as a screening criterion for vibrational spectral coupling.

[0128] In step (4), the interface interaction energy machine learning model and the vibration spectrum coupling machine learning model were built respectively. A total of three types of machine learning models were deployed, namely random forest (RF), extreme gradient boosting (XGB) and gradient boosting (GB) regression, and 10-fold cross validation was performed. Each model was trained 100 times to evaluate its robustness. The global optimization tool Bayesian optimization package was used to adjust the hyperparameters of all machine learning models in each training to improve the prediction accuracy. In the hyperparameter optimization process, 20 pairs of random parameters were initially used to train the Gaussian process and acquisition function. After 100 optimization iterations, the ideal parameters of each machine learning model were determined.

[0129] In step (5), two physical correlation indicators at the solid-liquid interface are predicted for the remaining unknown self-assembled monolayer structures. The specific steps are as follows: First, the interface interaction energy is predicted using three models, and the weighted prediction value is calculated. Using the weighted prediction value as the ranking basis, the self-assembled monolayer end group structures ranked in the top 20% in terms of interface interaction energy are selected as candidate structures. In addition, among the pre-calculated self-assembled monolayer end group structures, the top 20% are selected as candidate structures, and the two groups of structures are screened according to the process. Secondly, screening is performed based on vibration spectrum coupling. The screening process is the same as the interface interaction energy, and these two physical screening indicators are at the parallel screening level. Furthermore, the intersection of the two groups of screening results is found, that is, candidate structures with high interface interaction energy and strong vibration spectrum coupling are found. Finally, molecular fragments with high synthesis difficulty are excluded based on the synthesizable coefficient.

[0130] In step (6), non-equilibrium molecular dynamics simulation is used to solve the interface thermal conductivity of the determined screening target. The force field distribution is consistent with the physical correlation index calculated in the previous part. The interface thermal conductivity is calculated according to formula (3).

[0131] G=q / ΔT, (3)

[0132] where q is the heat flux density and ΔT is the temperature gradient at the interface in the direction of heat transport.

[0133] Non-equilibrium molecular dynamics simulation is closer to the experimental process of heat transport measurement. By applying a fixed thermostat (heat source and heat sink) to the material heat transport system, a heat flow and a temperature gradient can be formed inside the material, thereby directly solving the interface thermal conductivity G. This embodiment system carries out non-equilibrium molecular dynamics simulation in a microcanonical ensemble (NVE), in which a Langevin heat bath is used for a heat source at a constant temperature of 350K and a cold source at a constant temperature of 250K. When the non-equilibrium state becomes stable, the interface thermal conductivity can be obtained by periodically obtaining the heat flux density in the simulation system and the temperature gradient obtained by fitting at the interface.

[0134] The high-throughput screening method of this embodiment is highly applicable. By modifying the calculation settings involved in the simulation calculation process, such as the time step, number of steps, cutoff radius, and force field parameters, in accordance with common sense calculations, physical correlation indicators and interface thermal conductivity associated with the interfacial thermal conductivity characteristics of the self-assembled monolayer can also be obtained. The data set can also be modified in terms of set dimensions (quantity) according to design requirements. Furthermore, by modifying the frameworks involved in the machine learning training process, such as the model type, training parameters, number of training sessions, and training procedures, in accordance with common sense training, machine learning virtual screening models related to the physical association of interfacial thermal transport can also be obtained.

[0135] Example 2:

[0136] This embodiment provides a high-throughput screening method for self-assembled monolayers for regulating thermal conductivity at a solid-liquid interface, comprising the following steps:

[0137] (1) Prepare a screening set for the end groups of self-assembled monolayers.

[0138] This example first constructs an interface heat transport simulation system based on a self-assembled monolayer, a gold substrate, and water molecules. Figure 3 As shown, the monomer chain structure of the self-assembled monolayer is -S(CH2) n -X, where X is the end group structure. Specifically, the self-assembled monolayer is arranged in an 8×8 array on the surface of a gold (111) substrate with a face-centered cubic lattice structure. Each side of the substrate is composed of 25 layers of gold atoms, and 64 single chains of the self-assembled monolayer are distributed at the interface, with the distance between sulfur atoms being 0.497 nanometers. In addition, 2000 water molecules are compacted in the space between the two gold substrates and the self-assembled monolayer at the solid-liquid interface.

[0139] 1000 short polymer fragments were extracted from the ZINC database as a candidate data set for screening the terminal structure X. Considering the high degree of freedom in molecular structure design, the candidate data set ensures the diversity of terminal functional groups. The two-dimensional chemical structure was visualized using UMAP to verify that the extracted terminal group screening set has a wide distribution range and structural diversity, such as Figure 4 As shown in a. Due to the scarcity of interfacial heat transport data, 250 input structures were randomly selected as preliminary calculation and training data. Finally, a statistical analysis was performed on the molecular fragment substructures of the pre-calculated end group set and the total candidate end group set to check the number and type of substructures contained in the end group set, as shown in Figure 4 As shown in b.

[0140] (2) Molecular dynamics simulations were performed on 250 self-assembled monolayers in the pre-calculated set to obtain physical indicators such as interface interaction energy and vibration spectrum overlap area ratio.

[0141] Molecular dynamics simulations employed a hybrid force field distribution, with the Morse force field modeling the bond interactions between sulfur atoms in the sulfur-based self-assembled layer and gold atoms in the gold substrate. The PCFF potential was used for the self-assembled monolayer molecules, while the TIP3P potential was used for the water molecules. The LJ potential was used to model the non-bonded interactions between the gold atoms, the self-assembled monolayer molecules, and the water molecules. Except for the sulfur atoms covalently bound to the gold atoms, the LJ potential was parameterized using the modified universal force field (UFF). The long-range electrostatic interactions of the entire system were calculated using the PPPM method.

[0142] (2.1) The gold substrate, self-assembled monolayer, and water were grouped and the total energy between different groups of atoms at the interface was calculated. The calculation method of the interfacial interaction energy E was referred to formula (1). The interfacial interaction energy data distribution of 250 different self-assembled monolayers is shown in Figure 5 a.

[0143] (2.2) The vibration spectrum is calculated by Fourier transform of the velocity autocorrelation function, referring to formula (2). In order to quantify the strength of vibration spectrum coupling, the overlapping area ratio of the vibration spectrum between the self-assembled monolayer and the water molecule is used as a screening criterion for vibration spectrum coupling. The data distribution of the vibration spectrum overlapping area ratio of 250 different self-assembled monolayers is shown in Figure 5 b.

[0144] (3) The interfacial interaction energy of the self-assembled monolayer of the pre-calculated set (model training effect see Figure 6 a) Vibration spectrum overlapping area ratio (model training effect see Figure 6 b) Physical indicators are used as machine learning objective functions to independently train machine learning models.

[0145] Specifically, the machine learning virtual screening model construction contents include the interface interaction energy machine learning model and the vibration spectrum coupling machine learning model. A total of three types of machine learning models were deployed, namely random forest (RF), extreme gradient boosting (XGB) and gradient boosting (GB) regression, and 10-fold cross-validation was performed. Each training uses the global optimization tool Bayesian optimization package to adjust the hyperparameters of all machine learning models to improve prediction accuracy. During the hyperparameter optimization process, the Gaussian process and acquisition function were initially trained using 20 pairs of random parameters. After 100 optimization iterations, the ideal parameters of each independent machine learning model were determined.

[0146] (4) Based on the machine learning model trained on the pre-calculated set (250 self-assembled monolayer data sets), the associated physical indicators of the remaining end group candidate set (750 self-assembled monolayer data sets) that have not been simulated are predicted by machine learning. Then, the physical correlation indicators of the two types of sets are screened in parallel, and finally, the final simulation verification set is obtained by combining the screening of synthesizable coefficients.

[0147] (4.1) Predict interfacial interaction energies using the three models and calculate weighted predictions. Using the weighted predictions as the ranking criterion, select the top 20% of SAM end-group structures with the highest interfacial interaction energies as candidate structures. Additionally, select the top 20% of the pre-calculated SAM end-group structures as candidate structures. Both sets of structures are screened according to the same process.

[0148] (4.2) The vibration spectrum coupling is screened. The screening process is the same as the interface interaction energy. These two physical screening indicators are at the parallel screening level.

[0149] (4.3) Find the intersection of the two sets of screening results, that is, find candidate structures with high interface interaction energy and strong vibrational spectrum coupling.

[0150] (4.4) Eliminate molecular fragments with high synthesis difficulty based on the synthesizable coefficient. SA It is a comprehensive evaluation based on fragment contribution and complexity penalty, and is subsequently used as a reference indicator for the feasibility of organic molecule synthesis. SA The coefficient ranges from 1 to 10, and the higher the score, the more difficult it is to synthesize. Among all the polymer structures, 14 structures have a synthesizable coefficient of no more than 3.00, of which 7 were obtained through pre-simulation calculations and the other 7 were obtained through virtual screening.

[0151] (5) Finally, the input file is written through the ITC-SAM program and automatically connected to the LAMMPS software. The 14 self-assembled monolayers obtained by screening are simulated in a microcanonical ensemble (NVE) to solve the solid-liquid interface thermal conductivity. The force field distribution is consistent with the physical correlation index calculated in the previous part. The non-equilibrium molecular dynamics simulation of this embodiment adopts the NVE system, using a Langevin heat bath as a heat source of 350K and a cold source of 250K, and the duration is 2.0-5.0ns. The ITC-SAM program is combined with the non-equilibrium molecular dynamics simulation results to automatically obtain the heat flux density in the simulation system and the temperature gradient obtained by fitting at the interface, and the interface thermal conductivity reference is calculated according to formula (3).

[0152] Taking the interfacial thermal conductivity calculation of the self-assembled monolayer of single-chain -S-(CH2)6-OH as an example, the absolute values ​​of energy input and output on both sides of the heat source and heat sink of the simulation system in a stable state are as follows: Figure 7 As shown in a, its heat flux density q (4222.63MW / m 2 ) is directly simulated by the mean energy E (587.6eV) of the heat source and heat sink parts along the cross-sectional area S of the heat flow direction. The unit conversion is obtained by dividing the product of the simulation running time (1.6ns); at the same time, the temperature gap obtained by fitting at the interface is obtained. The cross-sectional diagram of the temperature distribution of the self-assembled monolayer along the heat transport direction is shown in FIG. Figure 7 b. ΔT1 and ΔT2 are 27.512K and 27.881K, respectively, and their average ΔT is 27.697K. Finally, the heat flux q is divided by the temperature gap ΔT to obtain the interface thermal conductivity G, that is, the interface thermal conductivity of the self-assembled monolayer of the single chain -S-(CH2)6-OH is 152.458MW / m 2 K.

[0153] In the pre-calculated data set, the interface thermal conductivity of three self-assembled monolayer structures is similar to that of the reference self-assembled monolayer (-S-(CH2)6-OH, ~150MW / m 2 K) equal to or exceeding it, e.g. Figure 8 As shown in a. For the 7 structures obtained by virtual screening, Figure 8 b shows that 6 of them meet the interface thermal conductivity requirements (>150MW / m 2 K), successfully demonstrated that the high-throughput screening method proposed in the present invention has extremely high reliability.

[0154] According to the traditional materials research and development model, developing new materials that meet performance requirements often requires extremely high experimental preparation and characterization costs, and the lack of strongly correlated physical screening indicators often leads to extremely poor reliability of high-throughput screening with small sample sizes. The high-throughput screening method of this embodiment can overcome the shortcomings of traditional materials research and development methods, enabling high-throughput screening of self-assembled monolayer thermal interface materials, greatly improving the fitting performance and screening reliability of the machine learning virtual screening model for self-assembled monolayer interface thermal transport, and further saving costs and time.

[0155] The above description of the embodiments is intended to facilitate understanding and use of the invention by those skilled in the art. It will be apparent that those skilled in the art can readily make various modifications to these embodiments and apply the general principles described herein to other embodiments without requiring inventive effort. Therefore, the present invention is not limited to the above-described embodiments. Improvements and modifications made by those skilled in the art based on the disclosure of the present invention, without departing from the scope of the present invention, should be within the scope of protection of the present invention.

Claims

1. A high-throughput screening method for self-assembled monolayer thermal interface materials, characterized in that: The following steps are involved: S1: Construct an interfacial heat transport simulation system based on a self-assembled monolayer, a gold substrate, and water molecules; the monomer chain structure of the self-assembled monolayer is -S(CH2) n -X, X is a terminal structure; S2: Prepare a screening set for the terminal structure X of S1, construct an terminal screening set and extract some fragments from it as a pre-calculation set; S3: Perform molecular dynamics simulations on the self-assembled monolayer corresponding to the pre-calculated set in S2 to obtain physical correlation indicators including interfacial interaction energy and vibrational spectrum coupling; S4: Use the physical correlation indicators obtained in S3 as the machine learning objective function and independently carry out machine learning model training; S5: Based on the machine learning model trained in S4, machine learning predictions are performed on the physical indicators of the screening candidate data set that has not been subjected to molecular dynamics simulation calculations. Subsequently, parallel screening of the physical correlation indicators of the two sets is performed to obtain a simulation verification set; S6: Perform molecular dynamics simulation on the simulation verification set screened in S5 to obtain the thermal conductivity of the solid-liquid interface.

2. The high-throughput screening method for self-assembled monolayer thermal interface materials according to claim 1, characterized in that: In step S2, the specific steps of preparing the screening set are: S2.1: Extract short polymer fragments from the ZINC database as end group screening sets; S2.2: Visualize the 2D chemical structures of short polymer fragments using UMAP to verify the distribution range and structural diversity of the end group screening set S2.3: Randomly select a certain proportion of short polymer fragments from the end group screening set as the pre-calculation set; S2.4: Perform statistical analysis on the substructures of the molecular fragments in the pre-calculated set and the end group screening set to verify the number and types of substructures contained in the end group screening set.

3. The high-throughput screening method for self-assembled monolayer thermal interface materials according to claim 1, characterized in that: In step S3, a mixed force field distribution is used in the molecular dynamics simulation; The bond interactions between the sulfur atoms in the self-assembled monolayer and the gold atoms in the gold substrate were simulated using the Morse force field; the self-assembled monolayer molecules were simulated using the PCFF potential, and the water molecules were simulated using the TIP3P potential; the non-bonded interactions between the gold atoms, the self-assembled monolayer molecules, and the water molecules were simulated using the LJ potential; The long-range electrostatic interactions of the simulated system in the molecular dynamics simulation were calculated using the PPPM method, and the time step of the simulation system was set to 0.2-0.3 fs.

4. The high-throughput screening method for self-assembled monolayer thermal interface materials according to claim 1, characterized in that: In step S3, the interface interaction energy E is calculated by referring to formula (1): Where, F ij is the force exerted by atom j in the gold substrate or water molecule on atom i in the self-assembled monolayer, Δd j is the displacement of atom j; F ji is the force exerted by atom i in the self-assembled monolayer on atom j in the gold substrate or water molecule, Δd i is the displacement of atom i.

5. The high-throughput screening method for self-assembled monolayer thermal interface materials according to claim 1, characterized in that: In step S3, the screening criterion for the vibrational spectrum coupling is specifically the overlapping area ratio of the vibrational spectra between the self-assembled monolayer and the water molecules; The vibration spectrum VDOS is obtained by Fourier transform of the velocity autocorrelation function, referring to formula (2): Where ω is the vibration wave number, v i is the atomic velocity, t0 and t are the initial time and correlation time, respectively.

6. The high-throughput screening method for self-assembled monolayer thermal interface materials according to claim 1, characterized in that: In step S4, the machine learning models include an interface interaction energy machine learning model and a vibration spectrum coupling machine learning model, and three types of machine learning models, namely random forest RF, extreme gradient boosting XGB and gradient boosting GB regression, are deployed and 10-fold cross-validation is performed. Each training uses the global optimization tool Bayesian optimization package to adjust the hyperparameters of all machine learning models.

7. The high-throughput screening method for self-assembled monolayer thermal interface materials according to claim 1, characterized in that: In step S5, the acquisition of the simulation verification set specifically includes the following steps: S5.1: Use the machine learning model to predict interfacial interaction energy and vibrational spectral coupling, calculate weighted prediction values, and use the weighted prediction values ​​as the ranking basis to select the top 20% of the end-group structures for each of the two physical correlation indicators as candidate structures. In addition, select the top 20% of the pre-calculated end-group structures as candidate structures. These two groups of structures are screened according to the process. S5.2: Find the intersection of the two screening results of interfacial interaction energy and vibrational spectrum coupling, and select candidate structures with high interfacial interaction energy and strong vibrational spectrum coupling; S5.3: Based on the synthesizable coefficient, eliminate molecular fragments with higher synthesis difficulty from the candidate structures, and finally obtain the simulation verification set.

8. The high-throughput screening method for self-assembled monolayer thermal interface materials according to claim 1, characterized in that: In step S6, the molecular dynamics simulation uses non-equilibrium molecular dynamics simulation to obtain the heat flux density in the simulation system and the temperature gradient obtained by fitting at the interface; The interfacial thermal conductivity of the self-assembled monolayer is referred to formula (3); G = q / ΔT (3); Where G is the interfacial thermal conductivity, q is the heat flux density, and ΔT is the temperature gradient at the interface in the direction of heat transport.

9. A high-throughput screening system for self-assembled monolayer thermal interface materials, characterized in that: include: Modeling module, used to construct a solid-liquid interface simulation system based on self-assembled monolayers, gold substrates, and water molecules; The screening set preparation module is used to construct the end group screening set and extract some fragments from it as the pre-calculation set; The molecular dynamics simulation module is used to perform molecular dynamics simulations on the self-assembled monolayer corresponding to the pre-calculated set to obtain physical correlation indicators including interface interaction energy and vibrational spectrum coupling; The machine learning model training module is used to independently train the machine learning model using physical correlation indicators as the objective function; The multi-objective screening and simulation verification module is used to perform machine learning predictions on the end group set for which molecular dynamics simulation has not been performed to calculate physical indicators, and to perform simulation verification based on the simulation verification set obtained by screening.

10. A storage medium containing computer-executable instructions, characterized in that: When the storage medium of the computer executable instructions is executed by a computer processor, it is used to perform the high-throughput screening method of the self-assembled monolayer thermal interface material according to any one of claims 1 to 8.

Images