Solid-solid interface structure global optimization method and device, equipment and storage medium

Abstract

The present invention provides a method and device, equipment and storage medium for global optimization of solid-solid interface structure. The method comprises: generating one or more initial solid-solid interface configurations; using the minimum jump method to perform molecular dynamics simulation on each initial solid-solid interface configuration, recording the structure and interface energy of the solid-solid interface configuration during the molecular dynamics simulation, and forming a data set; constructing a machine learning potential function model according to the data set to describe the relationship between the solid-solid interface configuration and the interface energy; periodically using the minimum jump method to perform molecular dynamics simulation on each initial solid-solid interface configuration, and using the machine learning potential function model to calculate the interface energy of the solid-solid interface configuration obtained by each simulation, obtaining multiple solid-solid interface configurations with local energy minimum values, and outputting the obtained solid-solid interface configuration with the minimum interface energy when the number of iterations is met.

Description

Technical Field

[0001] The present invention belongs to the field of interface structure optimization, and in particular relates to a solid-solid interface structure global optimization method and device, equipment and storage medium. Background Art

[0002] Solid-solid interfaces (hereinafter referred to as solid-solid interfaces) are ubiquitous in various materials, and their structure and properties have a significant impact on the overall performance of the material. Only with a deep understanding of the interface can we more effectively construct an interface with the desired properties. However, it is extremely challenging to determine the interface structure through experimental means, which requires extremely high time and spatial resolution, resulting in the current understanding of the internal interface properties of solids is still quite limited. Due to the lack of experimental data, theoretical research can often only be based on assumptions when exploring interface structures, speculating on the possible structure and chemical composition of the interface.

[0003] A stable interface should not only have low lattice strain but also good atomic matching. The lattice strain generally needs to be low enough, such as <10%. Or the global potential energy surface search method can be directly used to find the interface structure. There are two obvious problems with the manual construction method: (1) It does not fully consider all possible crystal plane orientation relationships when forming the interface. Most theoretical research works only consider a limited number of or even one crystal plane orientation relationship, so the rationality of the interface model cannot be guaranteed; (2) In order to make the interface strain small enough, the two phase surfaces that constitute the interface are usually expanded separately when constructing the model, which results in a large interface model.

[0004] Large-scale first-principles calculations (also known as density functional theory (DFT)) provide a convenient and reliable tool for the construction of global potential energy surfaces, and advances in sampling technology have accelerated the global exploration of potential energy surfaces. However, since most interface models have a large number of atoms and the symmetry of the atomic structure is low when constructing interface models, the number of single-point energies required to construct the interface potential energy surface increases exponentially with the increase in the number of atoms. Therefore, the computational cost of using accurate first-principles calculations to search the potential energy surface is too high (the size of each single-point energy needs to be calculated through DFT). Secondly, there are more complex interface potential energy surfaces in the bulk model, so in the process of structural optimization, it is inevitable to fall into the local minimum of the potential energy surface, resulting in a metastable interface structure. Therefore, it is necessary to study a new global optimization method to improve the search speed and efficiency of constructing the global potential energy surface while maintaining the high accuracy of first-principles calculations. Summary of the invention

[0005] The present invention provides a global optimization method and device, equipment and storage medium for a solid-solid interface structure, which can improve the search efficiency of the interface configuration and avoid the limitation of the local minimum energy.

[0006] On the one hand, a method for global optimization of a solid-solid interface structure is provided, comprising:

[0007] generating one or more initial solid-solid interface configurations;

[0008] The minimum jump method is used to perform molecular dynamics simulation on each initial solid-solid interface configuration, and the structure and interface energy of the solid-solid interface configuration during the molecular dynamics simulation are recorded to form a data set;

[0009] A machine learning potential function model is constructed based on the data set to describe the relationship between the solid-solid interface configuration and interface energy;

[0010] The minimum jump method is used periodically to perform molecular dynamics simulation on each initial solid-solid interface configuration, and the machine learning potential function model is used to calculate the interface energy of the solid-solid interface configuration obtained in each simulation, and multiple solid-solid interface configurations with local energy minima are obtained. When the number of iterations is met, the solid-solid interface configuration with the minimum interface energy is output.

[0011] Optionally, a molecular dynamics simulation is performed on each initial solid-solid interface configuration using a minimum jump method, and the structure of the solid-solid interface configuration and its interface energy during the molecular dynamics simulation are recorded to form a data set, including:

[0012] Step 21, generate a history list, the history list is used to store historical solid-solid interface configurations ;

[0013] Step 22: assigning an initial solid-solid interface configuration Initial kinetic energy ;

[0014] Step 23, initial solid-solid interface configuration Conduct molecular dynamics simulations to obtain multiple solid-solid interface configurations The structure and interface energy of are used as the training set in the data set; the solid-solid interface configuration with local energy minimum in the training set The first solid-solid interface configuration , when the first solid-solid interface configuration The number of meets the preset value , go to the next step;

[0015] Step 24, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain a new structure with a local energy minimum as the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of

[0016] Step 25, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface The initial solid-solid interface configuration and the historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step;

[0017] Step 26, calculate the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After taking its interface energy as the test set in the data set, we proceed to the next step;

[0018] Step 27, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list;

[0019] Step 28, Zoom out , after receiving the next search for the second solid-solid interface configuration When the instruction is received, return to step 23.

[0020] Optionally, a machine learning potential function model is constructed based on the data set to describe the relationship between the solid-solid interface configuration and the interface energy, including:

[0021] Step 31, using multiple descriptors to describe the solid-solid interface configuration in the training set Atomic coordinates at different regions at the interface to describe the solid-solid interface configuration in the training set The atomic structure of

[0022] Step 32, training a machine learning potential function model based on the descriptor and the potential energy of the atomic structure under the descriptor description area; when training the machine learning potential function model, using the loss function to calculate the weight and bias The gradient of , and the adaptive moment estimation algorithm is used to update the weights in the opposite direction of the gradient and bias ;

[0023] Step 33, using a test set to test whether the machine learning potential function model converges, when the machine learning potential function model does not converge, the minimum jump method is used again to calculate the initial solid-solid interface configuration. Molecular dynamics simulation was performed to simulate the solid-solid interface configuration. and its energy are added to the training set, and the simulated second solid-solid interface configuration is Add the test set; use the updated training set, return to step 31 and retrain the machine learning potential function model until the machine learning potential function model converges.

[0024] Optionally, a molecular dynamics simulation is performed periodically on each initial solid-solid interface configuration using a minimum jump method, and a machine learning potential function model is used to calculate the interface energy of the solid-solid interface configuration obtained by each simulation, to obtain multiple solid-solid interface configurations with local energy minima. When the number of iterations is met, the solid-solid interface configuration with the minimum interface energy is output, including:

[0025] Step 41, generating a candidate list, the candidate list is used to store the second solid-solid interface configuration obtained by simulation ;

[0026] Step 42, initial solid-solid interface configuration Molecular dynamics simulations were performed to obtain The first solid-solid interface configuration with a local energy minimum ; During the molecular dynamics simulation, the simulated solid-solid interface configuration The interface energy of the solid-solid interface is calculated based on the machine learning potential function model. The atomic structure of is calculated;

[0027] Step 43, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of

[0028] Step 44, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface and the initial solid-solid interface configuration , Historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step;

[0029] Step 45, calculating the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After joining the candidate list, proceed to the next step;

[0030] Step 46, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list;

[0031] Step 47, determining whether the number of iterations meets the requirement. If the number of iterations does not meet the requirement, return to step 42 until the number of iterations meets the requirement, and then proceed to the next step.

[0032] Step 48, select the second solid-solid interface configuration with the lowest energy in the candidate list Output as the solid-solid interface configuration with the minimum interface energy.

[0033] On the other hand, a solid-solid interface structure global optimization device is provided, comprising:

[0034] A generation module generates one or more initial solid-solid interface configurations;

[0035] The first interface configuration simulation module is used to perform molecular dynamics simulation on each initial solid-solid interface configuration using the minimum jump method, record the structure of the solid-solid interface configuration and its interface energy during the molecular dynamics simulation process, and form a data set;

[0036] A model building module is used to build a machine learning potential function model based on the data set to describe the relationship between the solid-solid interface configuration and the interface energy;

[0037] The second interface configuration simulation module is used to periodically perform molecular dynamics simulation on each initial solid-solid interface configuration using the minimum jump method, and use the machine learning potential function model to calculate the interface energy of the solid-solid interface configuration obtained in each simulation, to obtain multiple solid-solid interface configurations with local energy minima, and when the number of iterations is met, the solid-solid interface configuration with the minimum interface energy is output.

[0038] Optionally, the first interface configuration simulation module is used to:

[0039] Step 21, generate a history list, the history list is used to store historical solid-solid interface configurations ;

[0040] Step 22: assigning an initial solid-solid interface configuration Initial kinetic energy ;

[0041] Step 23, initial solid-solid interface configuration Conduct molecular dynamics simulations to obtain multiple solid-solid interface configurations The structure and interface energy of are used as the training set in the data set; the solid-solid interface configuration with local energy minimum in the training set The first solid-solid interface configuration , when the first solid-solid interface configuration The number of meets the preset value , go to the next step;

[0042] Step 24, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain a new structure with a local energy minimum as the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of

[0043] Step 25, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface The initial solid-solid interface configuration and the historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step;

[0044] Step 26, calculate the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After taking its interface energy as the test set in the data set, we proceed to the next step;

[0045] Step 27, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list;

[0046] Step 28, Zoom out , after receiving the next search for the second solid-solid interface configuration When the instruction is received, return to step 23.

[0047] Optionally, a model building module for:

[0048] Step 31, using multiple descriptors to describe the solid-solid interface configuration in the training set Atomic coordinates at different regions at the interface to describe the solid-solid interface configuration in the training set The atomic structure of

[0049] Step 32, training a machine learning potential function model based on the descriptor and the potential energy of the atomic structure under the descriptor description area; when training the machine learning potential function model, using the loss function to calculate the weight and bias The gradient of , and the adaptive moment estimation algorithm is used to update the weights in the opposite direction of the gradient and bias ;

[0050] Step 33, using a test set to test whether the machine learning potential function model converges, when the machine learning potential function model does not converge, the minimum jump method is used again to calculate the initial solid-solid interface configuration. Molecular dynamics simulation was performed to simulate the solid-solid interface configuration. and its energy are added to the training set, and the simulated second solid-solid interface configuration is Add the test set; use the updated training set, return to step 31 and retrain the machine learning potential function model until the machine learning potential function model converges.

[0051] Optionally, the second interface configuration simulation module is used to:

[0052] Step 41, generating a candidate list, the candidate list is used to store the second solid-solid interface configuration obtained by simulation ;

[0053] Step 42, initial solid-solid interface configuration Molecular dynamics simulations were performed to obtain The first solid-solid interface configuration with a local energy minimum ; During the molecular dynamics simulation, the simulated solid-solid interface configuration The interface energy of the solid-solid interface is calculated based on the machine learning potential function model. The atomic structure of is calculated;

[0054] Step 43, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of

[0055] Step 44, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface and the initial solid-solid interface configuration , Historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step;

[0056] Step 45, calculating the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After joining the candidate list, proceed to the next step;

[0057] Step 46, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list;

[0058] Step 47, determining whether the number of iterations meets the requirement. If the number of iterations does not meet the requirement, return to step 42 until the number of iterations meets the requirement, and then proceed to the next step.

[0059] Step 48, select the second solid-solid interface configuration with the lowest energy in the candidate list Output as the solid-solid interface configuration with the minimum interface energy.

[0060] On the other hand, an electronic device is provided, comprising the global optimization device for the solid-solid interface structure as described in any one of the above items.

[0061] On the other hand, a computer-readable storage medium is provided, wherein at least one program code is stored in the computer-readable storage medium, and the program code is executed by a processor to implement the global optimization method of the solid-solid interface structure as described in any one of the above items.

[0062] The technical solution provided by the embodiments of the present disclosure brings the following beneficial effects:

[0063] In the disclosed embodiment, a global optimization method for solid-solid interface structure is provided, and an adaptive minimum jump method is formed by combining the minimum jump method and the machine learning potential function model. On the one hand, the lowest potential energy structure of the material interface can be accurately predicted and the local minimum can be jumped out; on the other hand, the mapping relationship between the interface atomic configuration and the interface energy is established through the machine learning potential function model, which can quickly and accurately predict the energy of the interface structure and reduce the calculation cost. There is no need to use the first principle to calculate the interface energy (using the first principle, that is, using DFT, it is necessary to calculate the potential energy of multiple regions of the interface structure separately, and finally obtain the interface energy of the interface structure, and the calculation process is relatively cumbersome), so that the search efficiency of the interface structure can be greatly improved. In addition, the interface energy in the data set used by the machine learning potential function model is obtained by DFT calculation. Therefore, the accuracy of the interface energy predicted by the machine learning potential function model based on the interface structure will be closer to DFT, which can ensure the accuracy of the solid-solid interface configuration with the minimum interface energy obtained in the end. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0065] Figure 1 A method flow chart of a global optimization method for a solid-solid interface structure provided by an embodiment of the present disclosure;

[0066] Figure 2 A flowchart of a minimum jump method provided by an embodiment of the present disclosure;

[0067] Figure 3 A schematic diagram of a machine learning potential function model construction process provided by an embodiment of the present disclosure;

[0068] Figure 4 A flowchart of an adaptive minimum jump method provided by an embodiment of the present disclosure;

[0069] Figure 5 A flow chart of searching for the lowest interface energy configuration using an adaptive minimum jump method provided in an embodiment of the present disclosure;

[0070] Figure 6 A schematic diagram of an adaptive minimum jump method for exploring the interface potential energy surface provided by an embodiment of the present disclosure;

[0071] Figure 7A structural block diagram of a solid-solid interface structure global optimization device provided by an embodiment of the present disclosure;

[0072] Figure 8 A structural block diagram of an electronic device provided in an embodiment of the present disclosure.

[0073] The reference numerals are as follows:

[0074] 101: generation module; 102: first interface configuration simulation module; 103: model building module; 104: second interface configuration simulation module;

[0075] 201: processor; 202: memory. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] Figure 1 A method flow chart of a method for global optimization of a solid-solid interface structure provided by an embodiment of the present disclosure. Figure 1 ,include:

[0078] S1. Generate one or more initial solid-solid interface configurations.

[0079] For example, initialize a β-Ga2O3 / AlN interface configuration as the initial solid-solid interface configuration, denoted as .

[0080] S2. Use the minimum jump method to perform molecular dynamics simulation on each initial solid-solid interface configuration, record the structure of the solid-solid interface configuration and its interface energy during the molecular dynamics simulation, and form a data set.

[0081] In one example, step S2 includes:

[0082] Step 21, generate a history list, the history list is used to store historical solid-solid interface configurations .

[0083] Step 22: assigning an initial solid-solid interface configuration Initial kinetic energy .

[0084] Among them, the initial kinetic energy For the kinetic energy of atoms in molecular dynamics simulation, giving initial kinetic energy can also be considered as giving initial temperature .

[0085] For example, the initial temperature It can be 300K.

[0086] Step 23, initial solid-solid interface configuration Conduct molecular dynamics simulations to obtain multiple solid-solid interface configurations The structure and interface energy of are used as the training set in the data set; the solid-solid interface configuration with local energy minimum in the training set The first solid-solid interface configuration , when the first solid-solid interface configuration The number of meets the preset value , go to the next step.

[0087] In the disclosed embodiment, the molecular dynamics simulation is implemented by the Verlet algorithm based on the microcanonical ensemble, and the Maxwell-Boltzmann distribution is used to thermalize the system to ensure that the initial state of the system is close to the equilibrium state.

[0088] In step 23, the solid-solid interface configuration is obtained by density functional theory (DFT) calculation. The interface energy.

[0089] In the disclosed embodiment, all solid-solid interface configurations obtained by molecular dynamics simulation are Part of it is the first solid-solid interface configuration with a local energy minimum For example, the number of solid-solid interface configurations is 20, and the first solid-solid interface configuration The number of them is 4. The first solid-solid interface configuration The preset value of the number of It can be set as needed. When performing molecular dynamics simulation, it will always look for the first solid-solid interface configuration with a local energy minimum. , by default A certain number of first solid-solid interface configurations can be found Then stop the molecular dynamics simulation.

[0090] In the embodiment of the present disclosure, when all solid-solid interface configurations obtained by molecular dynamics simulation The interface energy of one of them is lower than the initial solid-solid interface configuration When the interface energy is The first solid-solid interface configuration .

[0091] Step 24, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain a new structure with a local energy minimum as the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy.

[0092] The first solid-solid interface configuration obtained by simulation in step 23 , its energy will still be higher ( Only a small possibility is the true local energy minimum). Therefore, the present invention uses the BFGS quasi-Newton algorithm to further analyze the first solid-solid interface configuration. When the quasi-Newton algorithm is used for optimization, the atomic structure of the interface configuration is modified, and the interface energy of the modified interface configuration is calculated. If the interface energy reduction value before and after the modification is lower than the threshold value and the interatomic force is lower than the threshold value (for example, the threshold value of the interface energy reduction value can be 10 -5 eV, the threshold of the interatomic force can be 0.01 eV / angstrom), then the modified interface configuration is considered to be a solid-solid interface configuration with a local energy minimum (i.e., the second solid-solid interface configuration). ).

[0093] By further configuring the first solid-solid interface Perform structural optimization to ensure the second candidate solid-solid interface configuration The interface energy is the true local energy minimum.

[0094] Step 25, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface The initial solid-solid interface configuration and the historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step.

[0095] In step 25, the second solid-solid interface configuration is determined by comparison Is it a new solid-solid interface configuration? If the second solid-solid interface configuration If it is not a new solid-solid interface configuration, adjust After that, search for the second solid-solid interface configuration again and adjust It is more conducive to finding new solid-solid interface configurations.

[0096] In the embodiment of the present disclosure, by amplifying the factor and If the second solid-solid interface configuration is enlarged The structure and initial solid-solid interface configuration The structure is the same as that of enlarge If the second solid-solid interface configuration The structure and history of solid-solid interface configurations The structure is the same as that of enlarge .

[0097] In the disclosed embodiment, and are all greater than 1. For example, is 1.1, is 1.1.

[0098] certainly, and The specific value of can be different and can be set as needed.

[0099] If the second solid-solid interface configuration is a new solid-solid interface configuration, then by reducing the factor Zoom out Then proceed to the next step.

[0100] In the disclosed embodiment, is less than 1. For example, It is 1 / 1.1.

[0101] In the embodiment of the present disclosure, by reducing the factor control size, avoiding infinite magnification of the magnification factor .

[0102] Step 26, calculate the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After taking the interface energy as the test set in the data set, we proceed to the next step.

[0103] In one example, It can be 0.1 eV.

[0104] In the embodiment of the present disclosure, by amplifying the factor enlarge , amplification factor Greater than 1. Exemplarily, the amplification factor is 1.1.

[0105] Step 27, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list.

[0106] Step 28, Zoom out , after receiving the next search for the second solid-solid interface configuration When the instruction is received, return to step 23.

[0107] In the embodiment of the present disclosure, by reducing the factor Zoom out .

[0108] In the embodiment of the present disclosure, the reduction factor Less than 1. Exemplarily, the reduction factor is 1 / 1.1.

[0109] In the disclosed embodiment, the internal loop of the minimum hopping method (MH) is used to jump out of the local minimum of the potential energy surface (PES) of the β-Ga2O3 / AlN interface through a shorter molecular dynamics (MD) simulation, and continue to perform structural optimization to find a new local minimum. The external loop of the MH algorithm is used to decide whether to accept the local minimum of the potential energy surface (PES) of the new interface based on an adaptive threshold. By tracking the previously searched minimum of the potential energy surface (PES) of the β-Ga2O3 / AlN interface and adjusting the initial kinetic energy value of MD accordingly, MH can escape from the explored potential energy surface (PES) area, thereby effectively searching for new interface configurations to find new local minima. Through the above method, it can be ensured that the interface energy of the interface structure with minimum energy obtained by the minimum hopping method each time will become smaller and smaller.

[0110] Figure 2A flowchart of a minimum jump method provided by an embodiment of the present disclosure. Figure 2 The minimum hopping method is shown in the search for the second solid-solid interface configuration The specific process.

[0111] S3. Construct a machine learning potential function model based on the data set to describe the relationship between the solid-solid interface configuration and interface energy.

[0112] In one example, step S3 includes:

[0113] Step 31, using multiple descriptors to describe the solid-solid interface configuration in the training set Atomic coordinates at different regions at the interface to describe the solid-solid interface configuration in the training set of the atomic structure.

[0114] Step 32, training a machine learning potential function model based on the descriptor and the potential energy of the atomic structure under the descriptor description area; when training the machine learning potential function model, using the loss function to calculate the weight and bias The gradient of , and the adaptive moment estimation algorithm is used to update the weights in the opposite direction of the gradient and bias .

[0115] In the disclosed embodiment, the loss function is:

[0116]

[0117] in, , and denote the errors between the energy, force and virial tensor predicted by NNIP and calculated by DFT, respectively. represents the serial number of the atom at the interface, , and are the adjustable weights for the energy, force and virial tensors, respectively.

[0118] In the embodiment of the present disclosure, the adaptive moment estimation algorithm is used to optimize the parameters of the Neural Network Interaction Potential (NNIP) model. The size of the embedded Neural Network (NN) is (25, 50, 100), the size of the fitted NN is (240, 240, 240), and the initial value of the learning rate is 1×10 -3 , the final value is set to 3.5×10 -8 , the number of training steps is 2×10 6 During the training process, , and The value of the factor will change as the number of training steps increases. From 1 to 10, From 10 to 1, Reduced from 0.1 to 0.01.

[0119] In the disclosed embodiment, the trained NNIP model is also a machine learning potential function model.

[0120] Step 33, using a test set to test whether the machine learning potential function model converges, when the machine learning potential function model does not converge, the minimum jump method is used again to calculate the initial solid-solid interface configuration. Molecular dynamics simulation was performed to simulate the solid-solid interface configuration. and its energy are added to the training set, and the simulated second solid-solid interface configuration is Add the test set; use the updated training set, return to step 31 and retrain the machine learning potential function model until the machine learning potential function model converges.

[0121] In the disclosed embodiment, the NNIP model is trained by the above method to ensure that the prediction accuracy of the NNIP model finally trained is close to or the same as the calculation accuracy of DFT.

[0122] Figure 3 A schematic diagram of a machine learning potential function model construction process provided by an embodiment of the present disclosure. Figure 3 In the disclosed embodiment, the machine learning potential function model (i.e., the trained NNIP model) mainly consists of three parts: data set, descriptor, and neural network. By determining the atomic coordinates in the localized ring of the interface configuration , and through the descriptor To describe NNIP model is based on the descriptor and the energy corresponding to each descriptor Trained. Among them, is the cutoff radius Each atom and its neighbors establish a local coordinate system ,Will The information of atoms in the descriptor is embedded into the descriptor through a fully connected neural network (NN). Constructing descriptors by fitting NN to interface energy The mapping of Passed as input to the hidden layer to calculate the total energy of the atoms in the local environment , interface energy is the sum of the total energy of atoms in each local environment in the interface configuration, that is, .

[0123] In the embodiments of the present disclosure, the descriptors need to satisfy the following requirements: (1) translation and rotation of the interface structure; (2) arrangement of the unit cell axes and atoms in the unit cell; and (3) invariance of the inversion and mirror image of the interface structure to ensure that the total energy of the system remains unchanged when the relative positions of the atoms change.

[0124] S4. Periodically use the minimum jump method to perform molecular dynamics simulation on each initial solid-solid interface configuration, and use the machine learning potential function model to calculate the interface energy of the solid-solid interface configuration obtained in each simulation, and obtain multiple solid-solid interface configurations with local energy minima. When the number of iterations is met, the solid-solid interface configuration with the minimum interface energy is output.

[0125] In one example, step S4 includes:

[0126] Step 41, generating a candidate list, the candidate list is used to store the second solid-solid interface configuration obtained by simulation .

[0127] Step 42, initial solid-solid interface configuration Molecular dynamics simulations were performed to obtain The first solid-solid interface configuration with a local energy minimum ; During the molecular dynamics simulation, the simulated solid-solid interface configuration The interface energy of the solid-solid interface is calculated based on the machine learning potential function model. The atomic structure of is calculated.

[0128] Step 43, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy.

[0129] Step 44, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface and the initial solid-solid interface configuration , Historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step.

[0130] Step 45, calculating the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After joining the candidate list, proceed to the next step.

[0131] Step 46, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list.

[0132] Step 47, determine whether the number of iterations meets the requirement. If the number of iterations does not meet the requirement, return to step 42 until the number of iterations meets the requirement, and then proceed to the next step.

[0133] Step 48, select the second solid-solid interface configuration with the lowest energy in the candidate list Output as the solid-solid interface configuration with the minimum interface energy.

[0134] After the iteration, the solid-solid interface configuration with the minimum interface energy is determined by the hybrid functional HSE06 complete relaxation iteration process to determine the ground state configuration of the interface.

[0135] In the embodiment of the present disclosure, the interface configuration with the minimum energy is searched by combining the NNIP model and the minimum jump method as the solid-solid interface configuration with the minimum interface energy. This improves the efficiency of energy calculation to ensure the search efficiency of the interface configuration. In step 45, by adding the initial solid-solid interface configuration The initial kinetic energy makes the initial solid-solid interface configuration When performing molecular dynamics simulations, it is possible to escape from the local minimum. In addition, the initial solid-solid interface configuration is the second solid-solid interface configuration with local energy minimum obtained last time , by continuously increasing the initial kinetic energy and the second solid-solid interface configuration Molecular dynamics simulation is performed for the initial solid-solid interface configuration, so as to finally break out of the limitation of the local minimum and find the real solid-solid interface configuration with the energy minimum.

[0136] In the disclosed embodiment, a global optimization method for solid-solid interface structure is provided by combining the minimum jump method and the machine learning potential function model. On the one hand, the lowest potential energy structure of the material interface can be accurately predicted and the local minimum can be jumped out; on the other hand, the mapping relationship between the interface atomic configuration and the interface energy is established through the machine learning potential function model, which can quickly and accurately predict the energy of the interface structure and reduce the calculation cost. There is no need to use the first principle to calculate the interface energy (using the first principle, that is, using DFT, it is necessary to calculate the potential energy of multiple regions of the interface structure separately, and finally obtain the interface energy of the interface structure, and the calculation process is relatively cumbersome), so that the search efficiency of the interface structure can be greatly improved. In addition, the interface energy in the data set used by the machine learning potential function model is obtained by DFT calculation. Therefore, the accuracy of the interface energy predicted by the machine learning potential function model based on the interface structure will be closer to DFT, which can ensure the accuracy of the solid-solid interface configuration with the minimum interface energy obtained in the end.

[0137] Figure 4 A flowchart of an adaptive minimum jump method provided by an embodiment of the present disclosure. Figure 4 The Adaptive Minimum Hoping Method (AMHM), which is also the global optimization method for the solid-solid interface structure provided by the present disclosure, is obtained by combining the minimum hopping method and the NNIP model.

[0138] Figure 5 A flowchart of a method for searching for the lowest interface energy configuration using an adaptive minimum jump method is provided in an embodiment of the present disclosure. Figure 5 , Figure 5 The process of AMHM searching for the lowest energy configuration of the β-Ga2O3 / AlN interface is demonstrated. The interface structure is used as the initial structure at step 0 to start the MD simulation, and its initial energy is −823.24 eV. The energy change during the first MD simulation is the path between steps 0 and 1. When the energy of the interface reaches the local minimum for the second time, the MD simulation is stopped, and the interface structure at this time is optimized. It is found that its lowest energy value is −823.24 eV, which is the same as the energy of the initial interface structure. The interface structure obtained in this cycle is discarded and the next cycle is entered. It was not until the fourth cycle that AMHM found a new minimum energy structure with an energy of −823.31 eV, and the energy difference with the initial structure was 0.07 eV, which is less than 2.50eV, so accept this minimum and add it to the minimum list, while reducing and T. In the subsequent cycles, AMHM did not find a new interface configuration with an energy minimum until the temperature rose to nearly 2000K, and then a new minimum energy interface configuration was found in the 19th and 20th cycles, with energies of −823.06 eV and −822.94 eV, respectively. It can be found that Figure 5 There are four "√" marks in (c), indicating that four local energy minima of the β-Ga2O3 / AlN interface structures were found during the AMHM iteration. However, due to the high temperature at the 19th iteration, the atomic structure of the β-Ga2O3 / AlN interface changed significantly. When the AMHM method reaches the maximum number of iterations, the structure search process is stopped. Figure 5 (a) and Figure 5 (b) shows the temperature T and energy of the MD simulation. changes, Figure 5 (d) and Figure 5 (e) in the figure shows the atomic structures obtained at the 4th and 19th iterations respectively.

[0139] Figure 6 A schematic diagram of an adaptive minimum jump method for exploring the interface potential energy surface provided by an embodiment of the present disclosure. Figure 6 The potential energy surface is a potential energy surface of a structure, such as the potential energy surface of β-Ga2O3 / AlN. An energy point on the potential energy surface represents a β-Ga2O3 / AlN interface configuration, and each energy point represents a different atomic position in the β-Ga2O3 / AlN interface configuration.

[0140] exist Figure 6 In the example, the initial point can be identified as the first local energy minimum point we found. Opt1 indicates the first structural optimization (that is, structural optimization using the BFGS algorithm), and a structure with lower energy is found near the initial point.

[0141] Taking the solid-solid interface configuration optimized by Opt1 as the initial solid-solid interface configuration, molecular dynamics simulation (MD2) is performed after increasing the initial kinetic energy (which can also be considered as increasing the temperature T). After optimizing the solid-solid interface configuration obtained by the MD2 simulation, the temperature is increased again and MD3 is entered. The process of molecular dynamics simulation (MD) and structural optimization (Opt) is repeated until the cycle is stopped after the limit on the number of iterations is met.

[0142] Figure 7 A structural block diagram of a solid-solid interface structure global optimization device provided in an embodiment of the present disclosure. It includes:

[0143] A generation module 101 generates one or more initial solid-solid interface configurations;

[0144] The first interface configuration simulation module 102 is used to perform molecular dynamics simulation on each initial solid-solid interface configuration using a minimum jump method, record the structure of the solid-solid interface configuration and its interface energy during the molecular dynamics simulation, and form a data set;

[0145] A model building module 103 is used to build a machine learning potential function model based on the data set to describe the relationship between the solid-solid interface configuration and the interface energy;

[0146] The second interface configuration simulation module 104 is used to periodically perform molecular dynamics simulation on each initial solid-solid interface configuration using the minimum jump method, and use a machine learning potential function model to calculate the interface energy of the solid-solid interface configuration obtained in each simulation, to obtain multiple solid-solid interface configurations with local energy minima, and when the number of iterations is met, output the solid-solid interface configuration with the minimum interface energy.

[0147] Optionally, the first interface configuration simulation module 102 is used to:

[0148] Step 21, generate a history list, the history list is used to store historical solid-solid interface configurations ;

[0149] Step 22: assigning an initial solid-solid interface configuration Initial kinetic energy ;

[0150] Step 23, initial solid-solid interface configuration Conduct molecular dynamics simulations to obtain multiple solid-solid interface configurations The structure and interface energy of are used as the training set in the data set; the solid-solid interface configuration with local energy minimum in the training set The first solid-solid interface configuration , when the first solid-solid interface configuration The number of meets the preset value , go to the next step;

[0151] Step 24, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain a new structure with a local energy minimum as the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of

[0152] Step 25, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface The initial solid-solid interface configuration and the historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step;

[0153] Step 26, calculate the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After taking its interface energy as the test set in the data set, we proceed to the next step;

[0154] Step 27, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list;

[0155] Step 28, Zoom out , after receiving the next search for the second solid-solid interface configuration When the instruction is received, return to step 23.

[0156] Optionally, the model building module 103 is used to:

[0157] Step 31, using multiple descriptors to describe the solid-solid interface configuration in the training set Atomic coordinates at different regions at the interface to describe the solid-solid interface configuration in the training set The atomic structure of

[0158] Step 32, training a machine learning potential function model based on the descriptor and the potential energy of the atomic structure under the descriptor description area; when training the machine learning potential function model, using the loss function to calculate the weight and bias The gradient of , and the adaptive moment estimation algorithm is used to update the weights in the opposite direction of the gradient and bias ;

[0159] Step 33, using a test set to test whether the machine learning potential function model converges, when the machine learning potential function model does not converge, the minimum jump method is used again to calculate the initial solid-solid interface configuration. Molecular dynamics simulation was performed to simulate the solid-solid interface configuration. and its energy are added to the training set, and the simulated second solid-solid interface configuration is Add the test set; use the updated training set, return to step 31 and retrain the machine learning potential function model until the machine learning potential function model converges.

[0160] Optionally, the second interface configuration simulation module 104 is used to:

[0161] Step 41, generating a candidate list, the candidate list is used to store the second solid-solid interface configuration obtained by simulation ;

[0162] Step 42, initial solid-solid interface configuration Molecular dynamics simulations were performed to obtain The first solid-solid interface configuration with a local energy minimum ; During the molecular dynamics simulation, the simulated solid-solid interface configuration The interface energy of the solid-solid interface is calculated based on the machine learning potential function model. The atomic structure of is calculated;

[0163] Step 43, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of

[0164] Step 44, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface and the initial solid-solid interface configuration , Historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step;

[0165] Step 45, calculating the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After joining the candidate list, proceed to the next step;

[0166] Step 46, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list;

[0167] Step 47, determining whether the number of iterations meets the requirement. If the number of iterations does not meet the requirement, return to step 42 until the number of iterations meets the requirement, and then proceed to the next step.

[0168] Step 48, select the second solid-solid interface configuration with the lowest energy in the candidate list Output as the solid-solid interface configuration with the minimum interface energy.

[0169] Figure 8 This is a structural block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 8 , electronic devices may include Figure 7 The solid-solid interface structure global optimization device. Generally, the electronic device includes: a processor 201 and a memory 202.

[0170] The processor 201 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 201 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 201 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state.

[0171] The memory 202 may include one or more computer-readable storage media, which may be non-transitory. The memory 202 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 202 is used to store at least one instruction, which is used to be executed by the processor 201 to implement the solid-solid interface structure global optimization method performed by an electronic device provided in the method embodiment of the present application.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A global optimization method for solid-solid interface structure, characterized in that: include: generating one or more initial solid-solid interface configurations; The minimum jump method is used to perform molecular dynamics simulation on each initial solid-solid interface configuration, and the structure and interface energy of the solid-solid interface configuration during the molecular dynamics simulation are recorded to form a data set; The steps of constructing a machine learning potential function model based on the data set to describe the relationship between the solid-solid interface configuration and the interface energy include: Step 31, using multiple descriptors to describe the solid-solid interface configuration in the training set Atomic coordinates at different regions at the interface to describe the solid-solid interface configuration in the training set The atomic structure of Step 32, training a machine learning potential function model based on the descriptor and the potential energy of the atomic structure under the descriptor description area; when training the machine learning potential function model, using the loss function to calculate the weight and bias The gradient of , and the adaptive moment estimation algorithm is used to update the weights in the opposite direction of the gradient and bias ; Step 33, using a test set to test whether the machine learning potential function model converges, when the machine learning potential function model does not converge, the minimum jump method is used again to calculate the initial solid-solid interface configuration. Molecular dynamics simulation was performed to simulate the solid-solid interface configuration. and its energy are added to the training set, and the simulated second solid-solid interface configuration is Add the test set; use the updated training set, return to step 31 and retrain the machine learning potential function model until the machine learning potential function model converges; The minimum jump method is used periodically to perform molecular dynamics simulation on each initial solid-solid interface configuration, and the machine learning potential function model is used to calculate the interface energy of the solid-solid interface configuration obtained in each simulation, and multiple solid-solid interface configurations with local energy minima are obtained. When the number of iterations is met, the solid-solid interface configuration with the minimum interface energy is output.

2. The method for global optimization of solid-solid interface structure according to claim 1, characterized in that: The minimum jump method is used to perform molecular dynamics simulation on each initial solid-solid interface configuration, and the structure and interface energy of the solid-solid interface configuration during the molecular dynamics simulation are recorded to form a data set, including: Step 21, generate a history list, the history list is used to store historical solid-solid interface configurations ; Step 22, assigning an initial solid-solid interface configuration Initial kinetic energy ; Step 23, initial solid-solid interface configuration Conduct molecular dynamics simulations to obtain multiple solid-solid interface configurations The structure and interface energy of are used as the training set in the data set; the solid-solid interface configuration with local energy minimum in the training set The first solid-solid interface configuration , when the first solid-solid interface configuration The number of meets the preset value , go to the next step; Step 24, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain a new structure with a local energy minimum as the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of Step 25, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface The initial solid-solid interface configuration and the historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step; Step 26, calculate the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After taking its interface energy as the test set in the data set, we proceed to the next step; Step 27, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list; Step 28, Zoom out , after receiving the next search for the second solid-solid interface configuration When the instruction is received, return to step 23.

3. The global optimization method for solid-solid interface structure according to claim 2, characterized in that: The minimum jump method is used periodically to perform molecular dynamics simulation on each initial solid-solid interface configuration, and the machine learning potential function model is used to calculate the interface energy of the solid-solid interface configuration obtained in each simulation, and multiple solid-solid interface configurations with local energy minima are obtained. When the number of iterations is met, the solid-solid interface configuration with the minimum interface energy is output, including: Step 41, generating a candidate list, the candidate list is used to store the second solid-solid interface configuration obtained by simulation ; Step 42, initial solid-solid interface configuration Molecular dynamics simulations were performed to obtain The first solid-solid interface configuration with a local energy minimum ; During the molecular dynamics simulation, the simulated solid-solid interface configuration The interface energy of the solid-solid interface is calculated based on the machine learning potential function model. The atomic structure of is calculated; Step 43, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of Step 44, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface and the initial solid-solid interface configuration , Historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step; Step 45, calculating the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After joining the candidate list, proceed to the next step; Step 46, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list; Step 47, determining whether the number of iterations meets the requirement. If the number of iterations does not meet the requirement, return to step 42 until the number of iterations meets the requirement, and then proceed to the next step. Step 48, select the second solid-solid interface configuration with the lowest energy in the candidate list Output as the solid-solid interface configuration with minimum interface energy.

4. A global optimization device for solid-solid interface structure, characterized in that: include: A generation module generates one or more initial solid-solid interface configurations; The first interface configuration simulation module is used to perform molecular dynamics simulation on each initial solid-solid interface configuration using the minimum jump method, record the structure of the solid-solid interface configuration and its interface energy during the molecular dynamics simulation process, and form a data set; A model building module is used to build a machine learning potential function model based on the data set to describe the relationship between the solid-solid interface configuration and the interface energy; Model building modules for: Step 31, using multiple descriptors to describe the solid-solid interface configuration in the training set Atomic coordinates at different regions at the interface to describe the solid-solid interface configuration in the training set The atomic structure of Step 32, training a machine learning potential function model based on the descriptor and the potential energy of the atomic structure under the descriptor description area; when training the machine learning potential function model, using the loss function to calculate the weight and bias The gradient of , and the adaptive moment estimation algorithm is used to update the weights in the opposite direction of the gradient and bias ; Step 33, using a test set to test whether the machine learning potential function model converges, when the machine learning potential function model does not converge, the minimum jump method is used again to calculate the initial solid-solid interface configuration. Molecular dynamics simulation was performed to simulate the solid-solid interface configuration. and its energy are added to the training set, and the simulated second solid-solid interface configuration is Add the test set; use the updated training set, return to step 31 and retrain the machine learning potential function model until the machine learning potential function model converges; The second interface configuration simulation module is used to periodically perform molecular dynamics simulation on each initial solid-solid interface configuration using the minimum jump method, and use a machine learning potential function model to calculate the interface energy of the solid-solid interface configuration obtained in each simulation, to obtain multiple solid-solid interface configurations with local energy minima, and when the number of iterations is met, the solid-solid interface configuration with the minimum interface energy is output.

5. The solid-solid interface structure global optimization device according to claim 4, characterized in that: The first interface configuration simulation module is used for: Step 21, generate a history list, the history list is used to store historical solid-solid interface configurations ; Step 22: assigning an initial solid-solid interface configuration Initial kinetic energy ; Step 23, initial solid-solid interface configuration Conduct molecular dynamics simulations to obtain multiple solid-solid interface configurations The structure and interface energy of are used as the training set in the data set; the solid-solid interface configuration with local energy minimum in the training set The first solid-solid interface configuration , when the first solid-solid interface configuration The number of meets the preset value , go to the next step; Step 24, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain a new structure with a local energy minimum as the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of Step 25, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface The initial solid-solid interface configuration and the historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step; Step 26, calculate the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 23 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After taking its interface energy as the test set in the data set, we proceed to the next step; Step 27, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list; Step 28, Zoom out , after receiving the next search for the second solid-solid interface configuration When the instruction is received, return to step 23.

6. The solid-solid interface structure global optimization device according to claim 5, characterized in that: The second interface configuration simulation module is used for: Step 41, generating a candidate list, the candidate list is used to store the second solid-solid interface configuration obtained by simulation ; Step 42, initial solid-solid interface configuration Molecular dynamics simulations were performed to obtain The first solid-solid interface configuration with a local energy minimum ; During the molecular dynamics simulation, the simulated solid-solid interface configuration The interface energy of the solid-solid interface is calculated based on the machine learning potential function model. The atomic structure of is calculated; Step 43, using a quasi-Newton algorithm to calculate the first solid-solid interface configuration with the lowest interface energy Perform local structural optimization to obtain the second solid-solid interface configuration , the second solid-solid interface configuration The interface energy is less than or equal to the first solid-solid interface configuration The interface energy of Step 44, comparing the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration The structure at the second solid-solid interface and the initial solid-solid interface configuration , Historical solid-solid interface configuration When the structures are the same, the initial kinetic energy is amplified by the amplification factor , and return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration , Historical solid-solid interface configuration If the structures are different, go to the next step; Step 45, calculating the second solid-solid interface configuration and the initial solid-solid interface configuration Is the energy difference below the threshold? , when the energy difference is not less than the threshold When zooming in Then return to step 42 until the second solid-solid interface configuration and the initial solid-solid interface configuration The energy difference is below the threshold When the second solid-solid interface configuration After joining the candidate list, proceed to the next step; Step 46, forming a second solid-solid interface As the initial solid-solid interface configuration After that, As Add to history list; Step 47, determining whether the number of iterations meets the requirement. If the number of iterations does not meet the requirement, return to step 42 until the number of iterations meets the requirement, and then proceed to the next step. Step 48, select the second solid-solid interface configuration with the lowest energy in the candidate list Output as the solid-solid interface configuration with minimum interface energy.

7. An electronic device, characterized in that: It comprises the solid-solid interface structure global optimization device as described in any one of claims 4 to 6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one program code, and the program code is executed by a processor to implement the global optimization method for the solid-solid interface structure according to any one of claims 1 to 3.

Images