Semiconductor device simulation method and system based on machine learning

Abstract

The invention provides a semiconductor device simulation method and system based on machine learning, and belongs to the technical field of simulation models, and the method comprises the steps: simulating the generation of a rectangular cubic region and the injection of liquid cooling liquid, and simulating the placement of a plurality of groups of semiconductor devices and a plurality of groups of simulation control modules; acquiring simulated geometric data, and generating a plurality of three-dimensional liquid cooling areas; layering processing is carried out, a plurality of three-dimensional layered liquid cooling areas are obtained, and real-time temperature data are obtained; executing a simulation distribution strategy, obtaining a plurality of pieces of real-time distribution data, and importing the real-time distribution data into a simulation control module; obtaining a liquid cooling linear regression model, stopping a simulation distribution strategy, obtaining a plurality of pieces of prediction distribution data, and importing the prediction distribution data into a simulation control module; a simulation information storage architecture is established, and a liquid cooling system and hardware of an actual semiconductor device can be established based on related simulation data; the invention provides a method for simulating regional real-time adjustment and pre-adjustment of the liquid cooling system.

Description

Technical Field

[0001] The present invention belongs to the technical field of simulation models, and more specifically, relates to a semiconductor device simulation method and system based on machine learning. Background Art

[0002] In today's era of rapid technological development, semiconductor technology, as the core of the modern electronics industry, is widely used in many fields such as computers, communications, and artificial intelligence. With the continuous improvement of semiconductor device performance and the continuous increase in integration, the heat generated by chips during operation has also risen sharply. Excessive temperature can seriously affect the performance and reliability of semiconductor devices, and even cause chip damage and shorten its service life. Therefore, efficient heat dissipation solutions are crucial to ensuring the stable operation of semiconductor devices.

[0003] In recent years, machine learning technology has made significant progress, demonstrating powerful capabilities in data analysis, pattern recognition, and prediction. Applying machine learning to liquid cooling model simulation for semiconductor devices can fully leverage its ability to process and analyze complex data, exploring the potential relationships between various parameters in the liquid cooling system, thereby establishing more accurate and efficient simulation models. This machine learning-based simulation method can evaluate and optimize different design solutions in a short period of time, significantly shortening the R&D cycle, reducing R&D costs, and improving the design level and performance of semiconductor device liquid cooling systems. Therefore, developing a machine learning-based semiconductor device simulation method has important practical significance and application value. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a semiconductor device simulation method and system based on machine learning to solve the technical problem in the prior art that traditional semiconductor simulation methods often lack the ability to simulate real-time adjustment and pre-adjustment of liquid cooling systems.

[0005] The purpose and effectiveness of the machine learning-based semiconductor device simulation method and system of the present invention are achieved by the following specific technical means:

[0006] A semiconductor device simulation method based on machine learning includes the following steps:

[0007] Simulating the generation of a rectangular cube region, injecting a liquid coolant into the region, placing multiple semiconductor devices in the region and setting operating parameters so that the multiple semiconductor devices are immersed in the simulated liquid coolant, adding multiple simulated control modules, and simulating the activation of the multiple simulated control modules and the multiple semiconductor devices.

[0008] Acquiring simulation geometry data of a rectangular cube region, establishing a two-dimensional plane based on the simulation geometry data, meshing the two-dimensional plane into sub-planes of equal area, numbering each sub-plane, and extending each sub-plane to the bottom of the rectangular cube region based on the simulation geometry data to form a plurality of three-dimensional liquid cooling regions;

[0009] Performing layered processing on each 3D liquid cooling area to obtain multiple 3D layered liquid cooling areas, where each sub-plane contains three corresponding 3D layered liquid cooling areas. Each 3D layered liquid cooling area is numbered and labeled, and real-time temperature data of each 3D layered liquid cooling area is obtained.

[0010] Executing a simulation allocation strategy based on the real-time temperature data, obtaining a plurality of real-time allocation data, importing the plurality of real-time allocation data into a plurality of corresponding simulation control modules, and recording the first simulation data and the second simulation data;

[0011] Obtaining a liquid cooling linear regression model, stopping the use of the simulated allocation strategy to obtain real-time allocation data, importing the real-time temperature data into the liquid cooling linear regression model to obtain a plurality of predicted allocation data, importing the plurality of predicted allocation data into a plurality of corresponding simulation control modules, and recording the third simulation data and the fourth simulation data;

[0012] A simulation information storage architecture is established, and each set of simulation data and liquid cooling liquid parameters are stored in the simulation information storage architecture. Users can establish the liquid cooling system and hardware of the actual semiconductor device based on the data in the simulation information storage architecture.

[0013] As a further solution of the present invention, the simulation allocation strategy includes:

[0014] The real-time temperature data is recorded in time, and the real-time temperature data can be expressed as T i,j,k , where i represents the number mark of each three-dimensional layered liquid cooling area, k represents the time point of time recording, the time point interval is 0.05min, and j represents the layer where the three-dimensional layered liquid cooling area is located;

[0015] Based on the real-time temperature data, the real-time temperature gradient of adjacent layers in each three-dimensional layered liquid cooling area is obtained, and the real-time heat load in each three-dimensional layered liquid cooling area is obtained based on the real-time temperature data;

[0016] Acquire multiple real-time distribution data based on real-time temperature gradient and real-time heat load.

[0017] As a further solution of the present invention, the calculation formula of the real-time distribution data is:

[0018]

[0019] Where n represents the number of the sub-plane, It represents the real-time allocation data corresponding to the sub-plane numbered n. It is expressed as the upper limit of the flow rate that the rectangular cube area can withstand, l is the reduction proportional constant, Q n(i,j,k) The real-time heat load of the three 3D layered liquid cooling areas contained in the sub-plane numbered n is represented as follows: It is represented by the real-time temperature gradient in the three 3D layered liquid cooling regions contained in the sub-plane numbered n, η n It represents the flow weight of the sub-plane numbered n, j represents the level at which the three-dimensional layered liquid cooling area is located, j=1 represents the upper level, j=2 represents the middle level, and j=3 represents the lower level.

[0020] As a further solution of the present invention, the real-time temperature gradient calculation formula is:

[0021]

[0022] in, It is represented by the real-time temperature gradient in the three-dimensional layered liquid cooling area numbered i, T (i,j-1,k) Represented as the real-time temperature data of the adjacent upper layer, T (i,j,k) Represented as real-time temperature data of the layer.

[0023] As a further solution of the present invention, the step of establishing a simulated information storage architecture includes:

[0024] The data lake layer is used to store the source data of the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometry data, and the parameters of the liquid cooling liquid. The data lake layer is established based on the distributed system HDFS;

[0025] The data processing layer is used to extract relevant data stored in the data lake layer and perform data cleaning, data conversion, and data standardization operations on the relevant data based on the stream processing platform. After data processing is completed, the processed data is sent to the time series data layer;

[0026] The time series data layer is used to align the timestamps of the processed data to ensure the correct timing of the data, and then send the processed data to the table output layer;

[0027] The table output layer generates the processed relevant data into an interactive simulation data table based on the report generation engine and exports the file;

[0028] The interactive simulation data table is a visual table that supports multi-dimensional cross-queries.

[0029] As a further solution of the present invention, the user can establish a liquid cooling system and hardware for an actual semiconductor device based on the data in the simulated information storage architecture, including:

[0030] The user can establish a liquid cooling system for an actual semiconductor device based on the second simulation data and the fourth simulation data, select the control module actually used based on the first simulation data and the third simulation data, establish an actual rectangular cube area of ​​equal proportion and adjust the depth of the liquid coolant used based on the simulation geometry data, and select the liquid coolant actually used based on the parameters of the liquid coolant.

[0031] As a further solution of the present invention, the step of obtaining a liquid-cooled linear regression model includes:

[0032] Acquire a liquid cooling data set, and establish a liquid cooling simulation database based on the liquid cooling data set;

[0033] A linear regression model is established, and the liquid-cooled dataset contained in the liquid-cooled database is imported into the linear regression model as a training set. The linear regression model performs model iteration. The linear regression model is trained based on the loss function and the regularization function, and optimized based on the gradient descent method. When the maximum number of model iterations is reached, the linear regression model stops iterating, obtains the liquid-cooled linear regression model, and exports the liquid-cooled linear regression model.

[0034] As a further solution of the present invention, obtaining a liquid cooling data set and establishing a liquid cooling simulation database based on the liquid cooling data set includes:

[0035] Liquid-cooled database, when relevant data is stored in the liquid-cooled database, the timestamp is matched with the data;

[0036] The liquid cooling data set includes the real-time temperature data of each three-dimensional layered liquid cooling area, real-time allocation data, the real-time temperature gradient of adjacent layers of each three-dimensional layered liquid cooling area obtained when executing the simulated allocation strategy, and the real-time heat load in each three-dimensional layered liquid cooling area. The liquid cooling data set is updated every 0.05 minutes, and the updated liquid cooling data set is continuously imported into the liquid cooling database until the linear regression model derives the liquid cooling linear regression model.

[0037] As a further embodiment of the present invention, the method further comprises:

[0038] The simulation geometric data includes the length, width and depth of the rectangular cube area and the depth of the liquid coolant;

[0039] Establishing a two-dimensional plane based on the length and width in the simulated geometric data, the two-dimensional plane being at the highest point of the depth of the liquid coolant, and extending each sub-plane in a vertical direction to the bottom of the rectangular cube region based on the depth of the liquid coolant in the simulated geometric data;

[0040] The plurality of sub-planes respectively correspond to the plurality of simulation control modules, each sub-plane has a plane boundary, and each three-dimensional liquid cooling region and each three-dimensional layered liquid cooling region have a spatial boundary;

[0041] The stratification process is specifically to divide each three-dimensional liquid cooling area into an upper layer, a middle layer and a bottom layer in the vertical direction.

[0042] A semiconductor device simulation system based on machine learning, comprising:

[0043] Region simulation module, used to simulate and generate rectangular cube regions;

[0044] a placement simulation module, for simulating the placement of multiple groups of semiconductor devices in a rectangular cube area, and immersing the multiple groups of semiconductor devices in a simulated liquid cooling liquid;

[0045] A start-up simulation module is used to simulate and start multiple groups of semiconductor devices placed in simulation and multiple groups of simulation control modules placed in simulation;

[0046] A simulation control module is used to simulate the real-time distribution or real-time pre-distribution of the liquid cooling liquid flow, and the simulation control modules are arranged in multiple groups, with a 3×3 grid arrangement of simulation control modules provided at the bottom of the rectangular cube area;

[0047] a processing module, configured to execute a simulated allocation strategy to obtain a plurality of real-time allocation data, and to establish a linear regression model, derive a liquid-cooled linear regression model, and obtain a plurality of predicted allocation data based on the liquid-cooled linear regression model;

[0048] A capture module is used to establish a two-dimensional plane, mesh the two-dimensional plane into sub-planes of equal area, and extend each sub-plane to generate multiple three-dimensional liquid cooling areas;

[0049] The recording module is used to record the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometric data and the parameters of the liquid coolant, and store the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometric data and the parameters of the liquid coolant based on the simulation information storage architecture.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] This method can establish sub-regions through grid division and layered processing of the two-dimensional plane. It can not only simulate the liquid cooling system with real-time adjustment of the sub-region based on the simulation classification strategy, but also simulate the liquid cooling system with pre-adjustment of the sub-region based on the liquid cooling linear regression model. Through this machine learning-based simulation method, users can evaluate and optimize different design schemes more quickly and accurately, shorten the R&D cycle, and reduce R&D costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flowchart of the steps of a semiconductor device simulation method based on machine learning of the present invention;

[0053] Figure 2 It is a flowchart of the steps of simulating allocation strategy in a semiconductor device simulation method based on machine learning of the present invention;

[0054] Figure 3 It is a schematic diagram of simulating the placement of multiple groups of semiconductor devices in a semiconductor device simulation system based on machine learning of the present invention;

[0055] Figure 4 It is a schematic diagram of meshing a two-dimensional plane in a semiconductor device simulation method based on machine learning of the present invention;

[0056] Figure 5 This is a schematic diagram of simulating the placement of an analog control module in a semiconductor device simulation system based on machine learning according to the present invention. DETAILED DESCRIPTION

[0057] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the technical solutions of the present invention, but are not intended to limit the scope of protection of the present invention.

[0058] Example 1:

[0059] As attached Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 As shown:

[0060] The present invention provides a semiconductor device simulation method based on machine learning, which is applicable to semiconductor simulation and includes the following steps:

[0061] S101: simulate and generate a rectangular cube area, simulate the injection of liquid coolant into the rectangular cube area, record the parameters of the liquid coolant, set the operating parameters of the semiconductor device, simulate the placement of multiple groups of semiconductor devices in the rectangular cube area, and immerse the multiple groups of semiconductor devices in the simulated liquid coolant. At the same time, add multiple groups of simulated control modules. When adding multiple groups of simulated control modules, simulate the startup of the multiple groups of simulated control modules and the multiple groups of semiconductor devices to make them work, thereby simulating the operating environment of the actual liquid cooling system.

[0062] Specifically, when simulating the placement of multiple groups of semiconductor devices in a rectangular cube area, a three-dimensional coordinate system is established based on the rectangular cube area, and the liquid cooling liquid depth is equidistantly divided into three groups of placement areas along the vertical direction of the Z axis. Three groups of semiconductor devices are simulated and placed equidistantly along the X axis and Y axis directions respectively, and the placement position is in the middle position of the top placement area. The simulated semiconductor devices are arranged in a 3*3 grid format. Subsequently, the semiconductor devices are simulated and placed equidistantly downward along the vertical direction of the Z axis. The semiconductor devices simulated and placed equidistantly downward are in the middle position of the middle placement area and the bottom placement area, and the simulated semiconductor devices are arranged in a 3*3*3 three-dimensional grid format.

[0063] Furthermore, in actual use, the user can set relevant parameters of the simulation control module, liquid cooling liquid, and rectangular cube area, including relevant operating parameters such as the power of the simulation control module, the specific heat capacity of the liquid cooling liquid, the length, width and depth of the rectangular cube area. When simulating the placement of semiconductor devices, the method of simulating the placement of semiconductor devices can be adjusted, and an N*N*N three-dimensional grid arrangement can be adopted, where N represents the number of semiconductor devices simulated by a single coordinate axis in the three-dimensional coordinate system, and the number of simulation control modules will also change with the number of semiconductor devices placed. An N*N grid-arranged simulation control module is arranged at the bottom of the rectangular cube area, allowing users to simulate and evaluate different liquid cooling system design schemes during simulation.

[0064] S102: Acquire simulation geometric data of the rectangular cube area, record the simulation geometric data, establish a two-dimensional plane based on the simulation geometric data, mesh the two-dimensional plane, divide the two-dimensional plane into sub-planes of equal area, number each sub-plane, and extend each sub-plane to the bottom of the rectangular cube area based on the simulation geometric data to form multiple three-dimensional liquid cooling areas.

[0065] Specifically, the simulation geometric data includes the length, width and depth of the rectangular cube area and the depth of the liquid cooling liquid. A two-dimensional plane is established based on the length and width in the simulation geometric data. The two-dimensional plane is at the highest point of the depth of the liquid cooling liquid. Based on the depth of the liquid cooling liquid in the simulation geometric data, each sub-plane is extended vertically to the bottom of the rectangular cube area. Multiple sub-planes correspond to multiple simulation control modules respectively. Each sub-plane has a plane boundary. Each three-dimensional liquid cooling area and each three-dimensional layered liquid cooling area have a spatial boundary. By setting the plane boundary, the phenomenon of sub-planes overlapping during grid division can be prevented.

[0066] Furthermore, when a two-dimensional plane is established by simulating geometric data, the length and width of the two-dimensional plane are consistent with the length and width of the rectangular cube area. When performing grid division, the two-dimensional plane is divided into X sub-planes of equal area according to a grid such as one row and one column, two rows and two columns...X rows and X columns, and each sub-plane contains a corresponding simulation control module. After the sub-plane is divided, each sub-plane is numbered and marked to facilitate the subsequent execution of the simulation allocation strategy, the establishment of the liquid cooling database, and the establishment of the liquid cooling linear regression model.

[0067] For example, suppose the user places 9 analog control modules in a rectangular cube area. The 2D plane is divided into 3 rows and 3 columns of sub-planes, that is, 9 sub-planes. The calculation formula for each sub-plane is:

[0068]

[0069] Wherein, ν represents the length of the rectangular cube area, o represents the width of the rectangular cube area, c represents the number of analog control modules, and x represents the area of ​​each sub-plane of equal area.

[0070] S103: Perform layered processing on each three-dimensional liquid cooling area to obtain multiple three-dimensional layered liquid cooling areas, each sub-plane contains three corresponding three-dimensional layered liquid cooling areas, number each three-dimensional layered liquid cooling area, and obtain real-time temperature data of each three-dimensional layered liquid cooling area.

[0071] Specifically, the layered processing is to divide each three-dimensional liquid cooling area into an upper layer, a middle layer and a bottom layer in a vertical direction.

[0072] It can be understood that by layering each three-dimensional liquid cooling area, three-dimensional layered liquid cooling areas of different levels can be obtained, and each three-dimensional liquid cooling area is formed by each sub-plane extending vertically downward to the bottom of the rectangular cube area. Therefore, the three different levels of three-dimensional layered liquid cooling areas divided by each three-dimensional liquid cooling area are contained in a corresponding sub-plane. Based on the three-dimensional layered liquid cooling area, the temperature data of different areas and different heights in the rectangular cube area can be obtained, avoiding the problem of only focusing on the overall average temperature of the rectangular cube area and ignoring the local problem. During the operation of the simulation control module, the heating temperature of each simulation control module is inconsistent, so the temperature data of the area where each simulation control module is located is also inconsistent. Through this division method, the liquid cooling system with regional control can be simulated during simulation.

[0073] Furthermore, when acquiring the real-time temperature data of each three-dimensional layered liquid cooling area, the real-time temperature data is acquired every 0.05 minutes, and each three-dimensional liquid cooling area has a spatial boundary with each three-dimensional layered liquid cooling area to prevent spatial overlap between the three-dimensional liquid cooling areas.

[0074] S104: Execute a simulation allocation strategy based on real-time temperature data, obtain multiple real-time allocation data through the simulation allocation strategy, obtain a set of real-time allocation data every 0.05 minutes, the multiple real-time allocation data correspond to multiple simulation control modules respectively, and import the multiple real-time allocation data into multiple corresponding simulation control modules respectively. The multiple simulation control modules adjust the liquid cooling liquid flow in different three-dimensional liquid cooling areas in real time according to the corresponding real-time allocation data, and record the first simulation data and the second simulation data.

[0075] Specifically, the real-time allocation data represents relevant data that the analog control module needs to perform real-time allocation and adjustment of the liquid cooling liquid flow. The real-time allocation data may include data such as working electrical parameters for real-time adjustment of the liquid cooling distribution unit and speed parameters for real-time adjustment of the cooling water pump. After multiple analog control modules receive multiple real-time allocation data respectively, they perform real-time adjustment of the liquid cooling liquid flow in different three-dimensional liquid cooling areas based on the real-time allocation data. For example, when the analog control module receives the working electrical parameters for real-time adjustment of the liquid cooling distribution unit or the speed parameters for real-time adjustment of the cooling water pump, the analog control module performs real-time adjustment of the liquid cooling distribution unit or the cooling water pump based on these parameters, thereby realizing real-time adjustment of the liquid cooling liquid flow.

[0076] The simulation allocation strategy includes the following steps:

[0077] S1041: Time recording of real-time temperature data, which can be expressed as T i,j,k , where i represents the number mark of each three-dimensional layered liquid cooling area, k represents the time point of time recording, the time point interval is 0.05min, and j represents the layer where the three-dimensional layered liquid cooling area is located.

[0078] It can be understood that by recording the time of real-time temperature data, when executing the simulation allocation strategy, the real-time temperature data at different time points can be called in sequence according to the order of time records, thereby preventing parameter errors from occurring during the calculation processing of the simulation allocation strategy.

[0079] S1042: Obtaining a real-time temperature gradient of adjacent layers of each three-dimensional layered liquid cooling region based on the real-time temperature data, and obtaining a real-time heat load within each three-dimensional layered liquid cooling region based on the real-time temperature data.

[0080] Furthermore, the real-time temperature gradient calculation formula is:

[0081]

[0082] in, It is represented by the real-time temperature gradient in the three-dimensional layered liquid cooling area numbered i, T (i,j-1,k) Represented as the real-time temperature data of the adjacent upper layer, T (i,j,k) Represented as real-time temperature data of the layer.

[0083] It can be understood that when obtaining the real-time temperature gradient, it is obtained in the order of upper layer-middle layer, and middle layer-lower layer.

[0084] When obtaining the real-time temperature gradient of the upper layer to the middle layer, the calculation formula of the real-time temperature gradient is:

[0085]

[0086] When obtaining the real-time temperature gradient from the middle layer to the lower layer, the calculation formula of the real-time temperature gradient is:

[0087]

[0088] Among them, j represents the layer where the three-dimensional layered liquid cooling area is located, j=1 represents the upper layer, j=2 represents the middle layer, and j=3 represents the lower layer.

[0089] When obtaining the real-time heat load in each three-dimensional layered liquid cooling area, it can be calculated and obtained based on parameters such as the real-time temperature data of the three-dimensional layered liquid cooling area and the specific heat capacity data of the liquid cooling liquid.

[0090] S1043: Acquire multiple real-time distribution data based on the real-time temperature gradient and the real-time heat load.

[0091] Specifically, the real-time allocation data calculation formula is:

[0092]

[0093] Where n represents the number of the sub-plane, It represents the real-time allocation data corresponding to the sub-plane numbered n. It is expressed as the upper limit of the flow rate that the rectangular cube area can withstand, l is the reduction proportional constant, Q n(i,j,k) The real-time heat load of the three 3D layered liquid cooling areas contained in the sub-plane numbered n is represented as follows: It is represented by the real-time temperature gradient in the three 3D layered liquid cooling regions contained in the sub-plane numbered n, η nIt represents the flow weight of the sub-plane numbered n, j represents the level at which the three-dimensional layered liquid cooling area is located, j=1 represents the upper level, j=2 represents the middle level, and j=3 represents the lower level.

[0094] Furthermore, the sum of all real-time distribution data is no greater than the upper limit of the flow that the rectangular cube area can bear. The reduction ratio l is related to the specific heat capacity data of the liquid coolant and the real-time heat load. The flow weight of each sub-plane numbered n is less than 1, and the sum of the flow weights of all sub-planes is 1. The flow weight is related to the real-time heat load and real-time temperature gradient of the three three-dimensional layered liquid cooling areas contained in the sub-plane.

[0095] S105: Obtain a liquid cooling linear regression model, stop using the simulation allocation strategy to obtain real-time allocation data, import the real-time temperature data into the liquid cooling linear regression model, obtain multiple predicted allocation data based on the liquid cooling linear regression model, the multiple predicted allocation data respectively correspond to multiple simulation control modules, import the multiple predicted allocation data respectively into multiple corresponding simulation control modules, and record the third simulation data and the fourth simulation data.

[0096] Specifically, the multiple simulation control modules pre-adjust the liquid cooling liquid flow rates in different three-dimensional liquid cooling areas according to the corresponding predicted allocation data.

[0097] Specifically, the predicted allocation data represents relevant data that the analog control module needs to perform pre-allocation adjustment of the liquid cooling liquid flow. The predicted allocation data may include data such as working electrical parameters for pre-adjusting the liquid cooling distribution unit, speed parameters for pre-adjusting the cooling water pump, etc. After multiple analog control modules receive multiple predicted allocation data respectively, they pre-adjust the liquid cooling liquid flow in different three-dimensional liquid cooling areas based on the predicted allocation data. For example, when the analog control module receives the working electrical parameters for pre-adjusting the liquid cooling distribution unit or the speed parameters for pre-adjusting the cooling water pump, the analog control module pre-adjusts the liquid cooling distribution unit or the cooling water pump in advance based on these parameters. The advance time can be set by the user, thereby achieving pre-adjustment of the liquid cooling liquid flow.

[0098] It can be understood that when performing the simulation allocation strategy, multiple real-time allocation data are obtained, and at the same time, the multiple real-time allocation data are respectively imported into multiple corresponding simulation control modules, and the first simulation data and the second simulation data are recorded. At this time, the first simulation data and the second simulation data have been recorded, and the real-time adjusted liquid cooling system has been simulated, so the simulation allocation strategy can be stopped and the real-time allocation data will no longer be obtained.

[0099] Specifically, a liquid cooling data set is obtained, and a liquid cooling simulation database is established based on the liquid cooling data set.

[0100] Furthermore, the liquid-cooled database is built based on InfluxDB, and when relevant data is stored in the liquid-cooled database, the timestamp is matched with the data.

[0101] The liquid cooling data set includes the real-time temperature data of each three-dimensional layered liquid cooling area, real-time allocation data, the real-time temperature gradient of adjacent layers of each three-dimensional layered liquid cooling area obtained when executing the simulated allocation strategy, and the real-time heat load in each three-dimensional layered liquid cooling area. The liquid cooling data set is updated every 0.05 minutes, and the updated liquid cooling data set is continuously imported into the liquid cooling database until the linear regression model derives the liquid cooling linear regression model.

[0102] A linear regression model is established, and the liquid-cooled dataset contained in the liquid-cooled database is imported into the linear regression model as a training set. The linear regression model performs model iteration. The linear regression model is trained based on the loss function and the regularization function, and optimized based on the gradient descent method. When the maximum number of model iterations is reached, the linear regression model stops iterating, obtains the liquid-cooled linear regression model, and exports the liquid-cooled linear regression model.

[0103] Specifically, during the generation process of the linear regression model, relevant data from the liquid cooling database is continuously obtained in sequence according to the timestamp of the liquid cooling data set. The relevant data contains the real-time temperature data of each three-dimensional layered liquid cooling area, real-time allocation data, the real-time temperature gradient of adjacent layers of each three-dimensional layered liquid cooling area obtained when executing the simulated allocation strategy, and the real-time heat load in each three-dimensional layered liquid cooling area. The corresponding relevant data are combined into a feature vector, and these feature vectors are aligned according to the sampling time.

[0104] Furthermore, assuming that the feature vector is m = [q, w, e, r], where q, w, e, r represent four different related data, and assuming that the linear regression model is G(m, θ), where θ is the regression coefficient of the linear regression model, the linear regression model can be implemented by the mean square error loss function and the regularization function.

[0105] Furthermore, during training, optimization can be performed based on the gradient descent method. When the maximum number of iterations is reached, the model iteration of the linear regression model is stopped, and the liquid-cooled linear regression model is derived. This is because when the maximum number of iterations is reached, the model can already accurately predict the data. Continuing to train the model may lead to problems such as local optimal solutions and overfitting.

[0106] S106: Establish a simulation information storage architecture, store the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometry data and the parameters of the liquid cooling liquid in the simulation information storage architecture, and the user can establish the liquid cooling system and hardware of the actual semiconductor device based on the relevant simulation data in the simulation information storage architecture.

[0107] Among them, establishing a simulation information storage architecture includes:

[0108] Establish a data lake layer, data processing layer, time series data layer, and table output layer, and connect the data lake layer, data processing layer, time series data layer, and table output layer based on links. The output end of the data lake layer is connected to the input end of the data processing layer, the output end of the data processing layer is connected to the input end of the time series data layer, and the output end of the time series data layer is connected to the input end of the table output layer.

[0109] The data lake layer is used to store the source data of the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometry data, and the parameters of the liquid cooling liquid. The data lake layer is established based on the distributed system HDFS, and the relevant data is stored in the JSON file format.

[0110] Specifically, all source data is stored in the data lake layer to maintain data diversity and integrity and support flexible access. The data lake layer uses the distributed system HDFS to support large-scale data storage. Data is stored in JSON file format to facilitate subsequent data processing and query. In actual applications, users can also choose Parquet or Avro file formats for storage.

[0111] The data processing layer is used to extract relevant data stored in the data lake layer and perform data cleaning, data conversion, and data standardization operations on the relevant data based on the stream processing platform. After the data processing is completed, the relevant data after data processing is sent to the time series data layer.

[0112] Specifically, data is extracted from the data lake layer and processed in real time through stream processing platforms, including Apache Kafka or Apache Flink. This layer is mainly responsible for cleaning, converting and standardizing source data to ensure data consistency. During the conversion, the data is converted into structured data that is more suitable for analysis and can be compressed as needed to save space.

[0113] The time series data layer is used to align the timestamps of the relevant data after data processing to ensure the correct timing of the relevant data, and then send the processed relevant data to the table output layer.

[0114] The table output layer generates the processed relevant data into an interactive simulation data table based on the Apache Zeppel in report generation engine and exports it to a file.

[0115] The interactive simulation data table is a visual table that supports multi-dimensional cross-queries.

[0116] Specifically, the relevant simulation data in the simulation information storage architecture includes:

[0117] The first simulation data represents the relevant working parameters of the simulation control module that is allocated in real time based on the simulation allocation strategy; the second simulation data represents the relevant system parameters of the simulation system that is allocated in real time based on the simulation allocation strategy; the third simulation data represents the relevant working parameters of the simulation control module that is predictively allocated based on multiple predictive allocation data; the fourth simulation data represents the relevant system parameters of the simulation system that is predictively allocated based on multiple predictive allocation data.

[0118] The user can establish a liquid cooling system for an actual semiconductor device based on the second simulation data and the fourth simulation data, select the control module actually used based on the first simulation data and the third simulation data, establish an actual rectangular cube area of ​​equal proportion and adjust the depth of the liquid coolant used based on the simulation geometry data, and select the liquid coolant actually used based on the parameters of the liquid coolant.

[0119] A semiconductor device simulation system based on machine learning, comprising:

[0120] The regional simulation module is used to simulate and generate a rectangular cube area and simulate the parameters and depth of the liquid coolant in the rectangular cube area. Users can adjust the parameters of the rectangular cube area and related parameters of the liquid coolant through the regional module.

[0121] The placement simulation module is used to simulate the placement of multiple groups of semiconductor devices in a rectangular cube area, and immerse the multiple groups of semiconductor devices in a simulated liquid cooling liquid.

[0122] The startup simulation module is used to simulate and start multiple groups of semiconductor devices placed in simulation and multiple groups of simulation control modules placed in simulation.

[0123] The simulation control module is used to simulate the real-time distribution or real-time pre-distribution of the liquid cooling liquid flow, and the simulation control modules are arranged in multiple groups, with a 3*3 grid-like arrangement of the simulation control modules at the bottom of the rectangular cube area;

[0124] The processing module is used to execute the simulation allocation strategy to obtain multiple real-time allocation data, and is used to establish a linear regression model, derive a liquid-cooled linear regression model, and obtain multiple predicted allocation data based on the liquid-cooled linear regression model.

[0125] The capture module is used to establish a two-dimensional plane, mesh the two-dimensional plane, generate multiple sub-planes of equal area, extend each sub-plane to generate multiple three-dimensional liquid cooling areas, and perform layered processing based on each three-dimensional liquid cooling area to generate multiple three-dimensional layered liquid cooling areas.

[0126] The recording module is used to record the parameters of the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, and the simulated geometric data in the liquid cooling liquid, and store the parameters of the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, and the simulated geometric data in the liquid cooling liquid based on the simulation information storage architecture or export them in the form of an interactive table.

[0127] The specific usage and function of this embodiment 1 are as follows:

[0128] First, a rectangular cube area is simulated and generated, and liquid cooling liquid is simulated and injected into the rectangular cube area, and the parameters of the liquid cooling liquid are recorded, the working parameters of the semiconductor device are set, and multiple groups of semiconductor devices are simulated and placed, and multiple groups of semiconductor devices are immersed in the simulated liquid cooling liquid. At the same time, multiple groups of simulation control modules are added, and multiple groups of simulation control modules and multiple groups of semiconductor devices are simulated and started, so as to simulate the operating environment of the liquid cooling system of the actual semiconductor device. Afterwards, the simulation geometric data of the rectangular cube area is obtained, the simulation geometric data is recorded, a two-dimensional plane is established, and the two-dimensional plane is gridded into sub-planes of equal area. Based on the simulation geometric data, each sub-plane is extended to the bottom of the rectangular cube area to form multiple three-dimensional liquid cooling areas, and then layered processing is performed to obtain multiple three-dimensional layered liquid cooling areas, and the real-time temperature data of each three-dimensional layered liquid cooling area is obtained. Then, the simulation allocation strategy is executed to obtain multiple real-time allocation data, and the data are imported into multiple corresponding simulation control modules, and the first simulation data and the second simulation data are recorded. Simulate data, then obtain the liquid cooling linear regression model, stop the simulation allocation strategy, import the real-time temperature data into the liquid cooling linear regression model, obtain multiple predicted allocation data, and import them into multiple corresponding simulation control modules, record the third simulation data and the fourth simulation data, and finally, establish a simulation information storage architecture, store the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometry data and the parameters of the liquid cooling liquid into the simulation information storage architecture, and the user can establish the liquid cooling system and hardware of the actual semiconductor device based on the relevant simulation data in the simulation information storage architecture. This method can establish sub-regions by grid division and layered processing of the two-dimensional plane. Not only can it simulate the liquid cooling system of the sub-region in real time based on the simulation classification strategy, but it can also simulate the liquid cooling system of the sub-region in pre-adjustment based on the liquid cooling linear regression model. Through this simulation method based on machine learning, users can evaluate and optimize different design schemes more quickly and accurately, shorten the R&D cycle, and reduce R&D costs.

[0129] An electronic device, comprising:

[0130] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method proposed in the first embodiment of the present invention.

[0131] The following is a detailed introduction to the various components of electronic equipment:

[0132] The processor is the control center of an electronic device and can be a single processor or a collective term for multiple processing elements. For example, the processor can be one or more central processing units (CPUs), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the first embodiment of the present invention, such as one or more digital signal processors (DSPs) or one or more field programmable gate arrays (FPGAs).

[0133] The processor can execute various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0134] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0135] The memory may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only (CD-ROM), or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor or exist independently and be coupled to the processor through an interface circuit of the electronic device, and the embodiments of the present invention do not specifically limit this.

[0136] The above embodiments can be implemented in whole or in part through software, hardware (e.g., circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wireless method (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0137] It should be understood that the term "and / or" as used herein simply describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent the existence of A alone, the existence of both A and B, or the existence of B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the related objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0138] It should be understood that in the embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0139] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. 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. 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 various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A semiconductor device simulation method based on machine learning, characterized in that: The following steps are included: Simulating the generation of a rectangular cube region, injecting a liquid coolant into the region, placing multiple semiconductor devices in the region and setting operating parameters so that the multiple semiconductor devices are immersed in the simulated liquid coolant, adding multiple simulated control modules, and simulating the activation of the multiple simulated control modules and the multiple semiconductor devices. Acquiring simulation geometry data of a rectangular cube region, establishing a two-dimensional plane based on the simulation geometry data, meshing the two-dimensional plane into sub-planes of equal area, numbering each sub-plane, and extending each sub-plane to the bottom of the rectangular cube region based on the simulation geometry data to form a plurality of three-dimensional liquid cooling regions; Performing layered processing on each 3D liquid cooling area to obtain multiple 3D layered liquid cooling areas, where each sub-plane contains three corresponding 3D layered liquid cooling areas. Each 3D layered liquid cooling area is numbered and labeled, and real-time temperature data of each 3D layered liquid cooling area is obtained. Executing a simulation allocation strategy based on the real-time temperature data, obtaining a plurality of real-time allocation data, importing the plurality of real-time allocation data into a plurality of corresponding simulation control modules, and recording the first simulation data and the second simulation data; Obtaining a liquid cooling linear regression model, stopping the use of the simulated allocation strategy to obtain real-time allocation data, importing the real-time temperature data into the liquid cooling linear regression model to obtain a plurality of predicted allocation data, importing the plurality of predicted allocation data into a plurality of corresponding simulation control modules, and recording the third simulation data and the fourth simulation data; A simulation information storage architecture is established, and each set of simulation data and liquid cooling liquid parameters are stored in the simulation information storage architecture. Users can establish the liquid cooling system and hardware of the actual semiconductor device based on the data in the simulation information storage architecture.

2. The method for simulating a semiconductor device based on machine learning according to claim 1, wherein: The simulation allocation strategy includes: The real-time temperature data is recorded in time, and the real-time temperature data can be expressed as T i,j,k , where i represents the number mark of each three-dimensional layered liquid cooling area, k represents the time point of time recording, the time point interval is 0.05min, and j represents the layer where the three-dimensional layered liquid cooling area is located; Based on the real-time temperature data, the real-time temperature gradient of adjacent layers in each three-dimensional layered liquid cooling area is obtained, and the real-time heat load in each three-dimensional layered liquid cooling area is obtained based on the real-time temperature data; Acquire multiple real-time distribution data based on real-time temperature gradient and real-time heat load.

3. The method for simulating a semiconductor device based on machine learning according to claim 2, wherein: The calculation formula of the real-time distribution data is: Where n represents the number of the sub-plane, It represents the real-time allocation data corresponding to the sub-plane numbered n. It is expressed as the upper limit of the flow rate that the rectangular cube area can withstand, l is the reduction proportional constant, Q n(i,j,k) The real-time heat load of the three 3D layered liquid cooling areas contained in the sub-plane numbered n is represented as follows: It is represented by the real-time temperature gradient in the three 3D layered liquid cooling regions contained in the sub-plane numbered n, η n It represents the flow weight of the sub-plane numbered n, j represents the level at which the three-dimensional layered liquid cooling area is located, j=1 represents the upper level, j=2 represents the middle level, and j=3 represents the lower level.

4. The method for simulating a semiconductor device based on machine learning according to claim 2, wherein: The real-time temperature gradient calculation formula is: in, It is represented by the real-time temperature gradient in the three-dimensional layered liquid cooling area numbered i, T (i,j-1,k) Represented as the real-time temperature data of the adjacent upper layer, T (i,j,k) Represented as real-time temperature data of the layer.

5. The method for simulating a semiconductor device based on machine learning according to claim 1, wherein: The establishment of the simulation information storage architecture includes: The data lake layer is used to store the source data of the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometry data, and the parameters of the liquid cooling liquid. The data lake layer is established based on the distributed system HDFS; The data processing layer is used to extract relevant data stored in the data lake layer and perform data cleaning, data conversion, and data standardization operations on the relevant data based on the stream processing platform. After data processing is completed, the processed data is sent to the time series data layer; The time series data layer is used to align the timestamps of the processed data to ensure the correct timing of the data, and then send the processed data to the table output layer; The table output layer generates the processed relevant data into an interactive simulation data table based on the report generation engine and exports the file; The interactive simulation data table is a visual table that supports multi-dimensional cross-queries.

6. The method for simulating a semiconductor device based on machine learning according to claim 1, wherein: The user can establish a liquid cooling system and hardware for an actual semiconductor device based on the data in the simulated information storage architecture, including: The user can establish a liquid cooling system for an actual semiconductor device based on the second simulation data and the fourth simulation data, select the control module actually used based on the first simulation data and the third simulation data, establish an actual rectangular cube area of ​​equal proportion and adjust the depth of the liquid coolant used based on the simulation geometry data, and select the liquid coolant actually used based on the parameters of the liquid coolant.

7. The method for simulating a semiconductor device based on machine learning according to claim 1, wherein: The obtaining of the liquid-cooled linear regression model comprises: Acquire a liquid cooling data set, and establish a liquid cooling simulation database based on the liquid cooling data set; A linear regression model is established, and the liquid-cooled dataset contained in the liquid-cooled database is imported into the linear regression model as a training set. The linear regression model performs model iteration. The linear regression model is trained based on the loss function and the regularization function, and optimized based on the gradient descent method. When the maximum number of model iterations is reached, the linear regression model stops iterating, obtains the liquid-cooled linear regression model, and exports the liquid-cooled linear regression model.

8. The method for simulating a semiconductor device based on machine learning according to claim 7, wherein: The obtaining of the liquid cooling data set and establishing a liquid cooling simulation database based on the liquid cooling data set includes: Liquid-cooled database, when relevant data is stored in the liquid-cooled database, the timestamp is matched with the data; The liquid cooling data set includes the real-time temperature data of each three-dimensional layered liquid cooling area, real-time allocation data, the real-time temperature gradient of adjacent layers of each three-dimensional layered liquid cooling area obtained when executing the simulated allocation strategy, and the real-time heat load in each three-dimensional layered liquid cooling area. The liquid cooling data set is updated every 0.05 minutes, and the updated liquid cooling data set is continuously imported into the liquid cooling database until the linear regression model derives the liquid cooling linear regression model.

9. The method for simulating a semiconductor device based on machine learning according to claim 1, wherein: The method further comprises: The simulation geometric data includes the length, width and depth of the rectangular cube area and the depth of the liquid coolant; Establishing a two-dimensional plane based on the length and width in the simulated geometric data, the two-dimensional plane being at the highest point of the depth of the liquid coolant, and extending each sub-plane in a vertical direction to the bottom of the rectangular cube region based on the depth of the liquid coolant in the simulated geometric data; The plurality of sub-planes respectively correspond to the plurality of simulation control modules, each sub-plane has a plane boundary, and each three-dimensional liquid cooling region and each three-dimensional layered liquid cooling region have a spatial boundary; The stratification process is specifically to divide each three-dimensional liquid cooling area into an upper layer, a middle layer and a bottom layer in the vertical direction.

10. A semiconductor device simulation system based on machine learning, characterized in that: include: Region simulation module, used to simulate and generate rectangular cube regions; a placement simulation module, for simulating the placement of multiple groups of semiconductor devices in a rectangular cube area, and immersing the multiple groups of semiconductor devices in a simulated liquid cooling liquid; A start-up simulation module is used to simulate and start multiple groups of semiconductor devices placed in simulation and multiple groups of simulation control modules placed in simulation; A simulation control module is used to simulate the real-time distribution or real-time pre-distribution of the liquid cooling liquid flow, and the simulation control modules are arranged in multiple groups, with a 3×3 grid arrangement of simulation control modules provided at the bottom of the rectangular cube area; a processing module, configured to execute a simulated allocation strategy to obtain a plurality of real-time allocation data, and to establish a linear regression model, derive a liquid-cooled linear regression model, and obtain a plurality of predicted allocation data based on the liquid-cooled linear regression model; A capture module is used to establish a two-dimensional plane, mesh the two-dimensional plane into sub-planes of equal area, and extend each sub-plane to generate multiple three-dimensional liquid cooling areas; The recording module is used to record the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometric data and the parameters of the liquid coolant, and store the first simulation data, the second simulation data, the third simulation data, the fourth simulation data, the simulation geometric data and the parameters of the liquid coolant based on the simulation information storage architecture.

Images