Chip liquid cooling heat dissipation channel path planning method based on temperature score

Abstract

The invention discloses a chip liquid cooling heat dissipation channel path planning method based on temperature scoring. A complete heat dissipation channel design framework is constructed, PCB temperature field distribution data are acquired in real time through a thermal imaging technology, a'temperature gradient driven 'path planning strategy is creatively developed based on the second law of thermodynamics and the countercurrent flow heat exchange principle, and the heat dissipation performance of the PCB is improved. Due to the self-adaptive characteristic, the device is particularly suitable for application scenes with strict heat dissipation requirements and limited cooling flow; accurately positioning a temperature peak region as a main runner end point; generating a main runner path by adopting a reverse derivation algorithm; a random backtracking mechanism is introduced to solve a local optimum problem, when space constraint is encountered, backtracking is automatically performed to a nearest feasible node, a secondary heat source is selected to continue planning, and a global optimal solution is obtained through iterative optimization. The heat dissipation efficiency and reliability of the high-power electronic equipment can be remarkably improved, meanwhile, the manufacturing cost is reduced, and the high-power electronic equipment has important engineering application value.

Description

Technical Field

[0001] This invention relates to thermal management technology for electronic devices, and more specifically to a chip liquid cooling heat dissipation channel path planning method based on temperature scoring. Background Technology

[0002] In the field of new energy technology, the explosive growth of high-efficiency data processing and storage devices has placed stringent demands on the stability, hardware operating efficiency, and lifespan of high-power computing equipment. Simultaneously, deep learning and large-scale data processing are imposing increasingly stringent requirements on device stability, computing speed, and hardware security, necessitating improvements in the computing performance and processing efficiency of AI computing chips such as GPUs and TPUs, as well as servers. These technological advancements all rely on the thermal management capabilities of high-power electronic devices, but traditional heat dissipation solutions are no longer sufficient, becoming a key bottleneck restricting industry development and providing important practical evidence for the research and development of intelligent heat dissipation channel technology.

[0003] Currently, liquid cooling for printed circuit boards (PCBs) primarily relies on placing a liquid cooling plate on the surface of the PCB, with refrigerant channels inside the plate. The refrigerant flows through these channels to cool the PCB. These channels typically employ regular shapes such as parallel direct current channels or S-shaped channels, or variations thereof. However, these channels cannot optimize the flow path based on the specific object being cooled, resulting in insufficient cooling efficiency. Furthermore, while computer graphics algorithms such as topology optimization can generate optimized channels, their computational time is long, the channel shapes are complex, and they are limited by processing precision and material properties, making them less feasible for large-scale and micro-scale manufacturing and difficult to apply in practice.

[0004] Current heat dissipation designs lack the ability to dynamically respond to temperature fields and cannot optimize flow channel layout in real time. While thermal imaging-based temperature field analysis technology can provide accurate temperature distribution data, it has not yet been effectively combined with intelligent path planning algorithms and cannot automatically generate optimal heat dissipation channels. Therefore, there is an urgent need for a heat dissipation design method that can integrate thermal imaging technology and intelligent optimization algorithms to improve heat dissipation efficiency and reduce manufacturing costs.

[0005] Traditional path planning algorithms are prone to getting trapped in local optima when optimizing heat flow paths and do not fully consider thermodynamic equilibrium characteristics, resulting in flow channels that do not conform to actual heat diffusion patterns. Furthermore, they lack adaptability to dynamic changes in temperature gradients, making it difficult to achieve global optimization in complex thermal fields.

[0006] The following problems exist in existing technologies and research: Flow channel designs limited to regular layout paths such as S-shaped, spiral, or parallel DC channels, or variations thereof, fail to be specifically planned for the actual heat dissipation object, making it difficult to effectively meet the cooling requirements of high-temperature critical components. While traditional optimization algorithms and topology optimization methods are theoretically feasible, they suffer from practical problems such as excessive computation time and high equipment costs. Furthermore, the complex flow channels they generate often exceed the capabilities of current manufacturing processes, resulting in severe manufacturing scale limitations. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention proposes a chip liquid cooling heat dissipation channel path planning method based on temperature scoring. Through innovative path generation algorithms that extract hot zone features and simulate heat conduction, high-precision heat flow path optimization is achieved. This invention not only solves the problem of low heat dissipation efficiency of traditional regular flow channels, but also overcomes the defect of difficult manufacturing of topology-optimized flow channels, providing an efficient and feasible solution for the thermal management of high-power electronic devices.

[0008] The chip liquid cooling heat dissipation channel path planning method based on temperature scoring of the present invention includes the following steps:

[0009] 1) Acquire thermal imaging images:

[0010] A high-resolution infrared thermal imager acquires red, green, and blue (RGB) thermal images representing the temperature field distribution on the surface of a printed circuit board (PCB). The thermal images contain complete information about the PCB surface temperature field.

[0011] 2) Extracting thermal features:

[0012] A grid-based partitioning strategy is adopted to divide the RGB thermal imaging image into multiple rectangular unit regions of equal size;

[0013] Independent spectral analysis is performed on each unit region, and a weighting strategy that is more conducive to temperature difference analysis is adopted: by fusing the intensity information of the red, green and blue channels, the red channel is enhanced and the blue channel term is suppressed, and a temperature characteristic value reflecting the thermodynamic properties is assigned to each unit region.

[0014] 3) Construct the matrix:

[0015] a) Constructing a temperature feature matrix: Generate a two-dimensional temperature feature matrix from the temperature feature values ​​of all unit regions. The elements of the temperature feature matrix are positively correlated with the thermal load intensity of the corresponding unit region, and retain the spatial correlation characteristics of the original temperature field distribution. That is, the temperature feature matrix corresponds completely to the thermal imaging image, and the position points of the elements of the temperature feature matrix correspond to the coordinate positions of the unit regions in the thermal imaging image.

[0016] b) Storage Matrix: The temperature feature matrix is ​​converted into a standardized data format, and a matrix table storage mode is used to persist the temperature feature matrix data to ensure the convenience and integrity of subsequent modules.

[0017] c) Standardized temperature feature matrix: The temperature feature matrix after persistence is linearly transformed using a normalization method to map all feature values ​​to a uniform dimension range, resulting in a standardized feature matrix. The elements in the standardized feature matrix are standardized feature values.

[0018] 4) Locate the main heat source:

[0019] A global extremum search is performed on the standardized feature matrix. Points whose standardized feature values ​​exceed the temperature feature threshold are selected as candidate core heat sources. The candidate core heat source closest to the geometric center of the temperature field is selected as the main heat source based on the location of the point, thereby determining the location of the main heat source in the temperature field.

[0020] 5) Optimize flow path:

[0021] a) Set the main path:

[0022] The path generation algorithm based on heat conduction simulation starts from the main heat source and takes the main heat source as the endpoint of the main channel path. It establishes a propagation model driven by the temperature gradient of the four neighborhoods, reversely derives the appropriate main channel path, and updates the standardized feature matrix. This is called the path finding process.

[0023] b) Introduce a backtracking mechanism:

[0024] If spatial constraints are encountered during the pathfinding process, and the entire effective neighborhood of the current position point has been set as the main path and the next step cannot be performed, while there are still position points in the entire normalized feature matrix that are not completely covered by the main path, the path will backtrack to the most recent step number. A heat diffusion path backtracking mechanism is introduced, so that the neighborhood of the backtracked main path node position has at least one unvisited position point, which is called the backtracking point. The pathfinding process is re-executed, while the main path of the original pathfinding process is retained as an optional main path. If the main path cannot completely cover all position points in the entire normalized feature matrix, this process needs to be performed multiple times.

[0025] c) Adaptive termination determination:

[0026] Set an adaptive termination condition. When the adaptive termination condition is met, the pathfinding process terminates, the result is output, and the optimized flow path is obtained based on the output result.

[0027] d) Optimize the flow path to generate a visual output with thermal rendering, and the flow path optimization ends.

[0028] In step 1), the thermal imaging image uses color to represent temperature information; the input is a thermal imaging image that has been standardized in length and width.

[0029] In step 2), the side length of the unit region is arbitrary; the number of unit regions is ≥3×3.

[0030] In conventional methods, the weights of the three color channels are summed to 1, meaning each channel falls within the range of 0 to 1. However, this method specifically enhances the red spectral response and suppresses blue component interference, taking into account the RGB three-channel characteristics of thermal imaging. A temperature feature value is assigned to each unit region, involving the following steps: A unit region of an RGB thermal imaging image includes the intensity values ​​of the red, green, and blue channels. The intensity values ​​of the three channels are weighted, increasing the weight of the red channel and decreasing the weight of the blue channel to obtain the feature value for this unit region. The weight of the red channel is increased by a factor of x (0.4 ≤ x ≤ 1.6), and the weight of the blue channel is decreased by a factor of y (0.4 ≤ y ≤ 1.6), allowing the blue channel value to be reduced to 0 or negative. The optimal values ​​are 0.5 ≤ x ≤ 1.2 and 0.4 ≤ y ≤ 0.96, meaning a larger weight for the red channel and a smaller weight for the blue channel to eliminate the influence of temperature scale settings. This combination improves the sensitivity to high and low temperature regions and is more effective in addressing color and temperature mismatch issues caused by temperature scale settings. The red, green, and blue channels of the unit region are a, b, and c, respectively, and the original weights are l, m, and n. The weight of the red channel is increased by x times and the weight of the blue channel is decreased by y times. Then the temperature characteristic value T = (a×l + b×m + c×n) + a×l×xc×n×y.

[0031] In step 3), the extreme value normalization method is used to maintain the relative relationship of the original temperature field distribution, eliminate the dimensional influence caused by equipment differences, and eliminate the influence of the ambient reference temperature.

[0032] In step 4), firstly, a global extremum search is used to standardize the eigenvalues, identifying locations exceeding the temperature characteristic threshold as candidate core heat sources. Then, a geometric center evaluation algorithm based on Euclidean distance is employed to select the point closest to the PCB geometric center from among the candidate core heat sources as the primary heat source. Other candidate core heat sources are designated as secondary heat sources. Therefore, in step 5), for secondary heat sources, a priority queue is established based on a trade-off between temperature gradient and spatial distribution density. This involves sorting the standardized eigenvalues ​​of each secondary heat source, prioritizing those with larger standardized eigenvalues ​​as path nodes to ensure the systematic nature of the heat dissipation path planning. The temperature characteristic threshold is 0.8–0.99.

[0033] In step 5)a), a propagation model driven by the temperature gradient of four neighborhoods is established. The reverse derivation process includes: defining the four points above, below, left, and right of a position in the normalized feature matrix as a neighborhood; a position in the neighborhood is called a neighbor; if a neighbor has already been set as the main channel path, it is not considered; the difference between the normalized feature value corresponding to the current position and the normalized feature value corresponding to the neighborhood is obtained; the neighbor corresponding to the smallest difference is the next step path of the main channel. Specifically, starting from the main heat source, the position of each step of the main channel path in the updated normalized feature matrix is ​​assigned an integer value of the current step number, indicating the path order of the main channel path; in each step, starting from the current position, a valid neighbor is selected, and the absolute value of the difference between the normalized feature value of the neighbor and the normalized feature value of the current position should be as small as possible; next, this neighbor will be designated as the next node position of the main channel path; this operation continues until as many positions as possible are assigned a value greater than or equal to 1.

[0034] Within each computational step, the temperature distribution of the surrounding neighborhood is analyzed, and the direction with the lower temperature difference is selected as the main channel path extension direction. To enhance physical plausibility, the algorithm introduces a heat diffusion path backtracking mechanism to prevent the main channel path calculation from getting stuck.

[0035] In step 5)b), when starting the pathfinding attempt, the step number of the flow path between the location point encountering spatial constraints and the backtracking point is retained, and the diffusion continues from the newly selected node next to the backtracking point. The entire process involves multiple calculations. Due to complex geometric constraints, in many cases, it is impossible for the main flow path to completely cover the entire normalized feature matrix. Therefore, after meeting the stopping condition, the output result with the longest main flow path is finally selected as the final main flow path. A backtracking mechanism is introduced to solve the local optimum problem. When encountering spatial constraints, it automatically backtracks to the nearest feasible node and selects a secondary heat source to continue planning, obtaining the global optimum through iterative optimization.

[0036] In step 5)c), three types of adaptive termination conditions are set:

[0037] (1) The preset maximum number of attempts has been reached;

[0038] (2) The main road path covers the entire standardized feature matrix;

[0039] (3) Tracing back to the main heat source, no new mainstream path was found;

[0040] If any of the above conditions are met, the entire optimized flow path ends, and the result is output. The maximum number of attempts is 0.4 to 0.8 times the number of location points in the standardized feature matrix, which is the number of unit regions divided in the thermal imaging image.

[0041] For the output results, if the main channel path can completely cover all positions of the entire normalized feature matrix (i.e., there are no integers greater than 1 and no numbers less than 1 in the matrix), then the optimized channel path is obtained. If there are multiple selectable main channel paths and none of them can completely cover all positions of the entire normalized feature matrix (i.e., there are integers greater than 1 or numbers less than 1 in the normalized feature matrix), then a secondary channel is opened for each selectable main channel path to cover all positions of the normalized feature matrix except for the part covered by the main channel, thus obtaining the candidate optimized channel path.

[0042] For the candidate optimization path, select the main flow path scheme with fewer generated secondary flow paths as much as possible. Under this premise, select the longest main flow path as the final result. At the same time, observe whether the number of secondary flow paths is reduced. If the number of secondary flow paths can be reduced, discard the output result with the longest main flow path and select the second longest main flow path as the optimization path.

[0043] One or more secondary channels are opened at locations not covered by the main channel. Each secondary channel is a complete S-shaped channel. The start and end points of each secondary channel are connected to the main channel and are as close as possible to the end point of the main channel path, i.e., the main heat source. The location with the smaller step number is used as the starting point of the secondary channel to obtain an optimized channel path that includes the main channel and the secondary channels. Each location point is required to be traversed by the channel path only once.

[0044] In step 5), d), a 3D model for manufacturing is drawn based on the optimized flow path data. The reverse order of the step numbers is used as the flow order of the main flow path.

[0045] Experimental tests showed that under typical low flow rate conditions of 0.001 kg / s to 0.010 kg / s, the flow channel designed in this invention exhibited significant advantages, with its heat dissipation efficiency increasing by an average of over 40% compared to parallel direct-flow channels and over 16% compared to S-shaped flow channels. Due to the flow-heat coupling effect, the heat dissipation capacity at high flow rates gradually approaches that of S-shaped and parallel direct-flow channels. The fewer the number of secondary flow channels generated and the larger the proportion of the main flow path, the better the heat dissipation performance.

[0046] The flow path and structure generated by this invention can be planned considering the surface heat distribution of the object being cooled. The flow path shape, channel corners, and cross-section all adopt regular geometric configurations, which not only ensures fluid dynamic performance but also fully meets the requirements of conventional processing technology. It fully considers the characteristics of modern manufacturing technology: for micron-level precision scenarios, micro-milling or laser processing is used; for millimeter-level flow paths, conventional computer numerical control CNC milling or precision casting is suitable; and for larger-scale meter-level flow path systems, a modular assembly scheme of segmented processing followed by welding / bonding is adopted. This multi-scale adaptability allows the same design concept to be applied to various scenarios from chip-level heat dissipation to rack-level cooling, overcoming the difficult processing problems caused by the complex geometric configuration flow paths of other design methods.

[0047] Advantages of the present invention:

[0048] This invention constructs a complete heat dissipation channel design framework. It acquires real-time PCB temperature field distribution data using thermal imaging technology and creatively develops a "temperature gradient-driven" path planning strategy based on the second law of thermodynamics and the principle of counter-current heat transfer. It accurately locates the temperature peak region as the endpoint of the main channel; uses a reverse derivation algorithm to generate the main channel path; and introduces a random backtracking mechanism to solve local optima problems. When encountering spatial constraints, it automatically backtracks to the nearest feasible node and selects a secondary heat source to continue planning. Through iterative optimization, it obtains the globally optimal solution. This invention exhibits excellent heat dissipation performance under low-flow conditions, significantly improving the heat dissipation efficiency and reliability of high-power electronic devices while reducing manufacturing costs, thus possessing significant engineering application value.

[0049] (1) This invention pioneered the introduction of the concept of intelligent temperature field allocation into the field of chip heat dissipation design. Through the deep integration of thermal imaging and algorithms, it realizes the adaptive planning design of heat dissipation channels. Compared with the traditional heat dissipation scheme that relies on empirical rules, this method can accurately match the actual temperature distribution characteristics, providing a brand-new technical route for the heat dissipation design of high-power electronic devices. (2) The innovative dynamic path planning algorithm with autonomous backtracking capability effectively solves the problem that traditional methods are prone to getting trapped in local optima. This algorithm monitors the temperature gradient changes in real time, adjusts the search strategy, and automatically backtracks and selects the suboptimal path when encountering local path blockages, which significantly improves the global optimization performance of the flow channel design. This adaptive characteristic makes it particularly suitable for application scenarios with strict heat dissipation requirements but limited cooling flow.

[0050] (3) The present invention establishes a complete heat dissipation performance evaluation and verification system, and the superiority of the method is confirmed by comparative experiments; the test data shows that, under the same mass flow rate conditions, the heat dissipation capacity is significantly improved compared with the traditional parallel flow channel and S-shaped flow channel.

[0051] (4) The flow channel structure designed in this invention has significant engineering advantages; taking into account the surface heat distribution of the object being cooled, all flow channel shapes and cross-sections adopt regular geometric configurations, which not only ensures fluid dynamic performance but also fully meets the requirements of conventional processing technology; this design feature enables it to adapt to the manufacturing needs of various scales (from micrometer to meter) and various materials (metal, plastic and ceramic, etc.), which greatly reduces the threshold for industrialization. Attached Figure Description

[0052] Figure 1 This is a flowchart of the chip liquid cooling heat dissipation channel path planning method based on temperature scoring according to the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0054] like Figure 1 As shown, the chip liquid cooling heat dissipation channel path planning method based on temperature scoring in this embodiment includes the following steps:

[0055] 1) Acquire thermal imaging images:

[0056] A high-resolution infrared thermal imager is used to acquire red, green, and blue (RGB) thermal images representing the temperature field distribution on the PCB surface. The thermal images contain complete PCB surface temperature field information, and the temperature information is presented using color. Input is a thermal image that has been standardized in length and width.

[0057] 2) Extracting thermal features:

[0058] A gridded partitioning strategy is adopted to divide the RGB thermal imaging image into 5×5 rectangular unit regions of equal size. Independent spectral analysis is performed on each unit region, employing a weighting strategy more conducive to temperature difference analysis: by fusing the intensity information of the red, green, and blue channels, the red channel is enhanced while the blue channel is suppressed. Each unit region is assigned a temperature characteristic value reflecting its thermodynamic properties. This includes the following steps: A unit region of the RGB thermal imaging image includes the intensity values ​​of the red, green, and blue channels. The intensity values ​​of the three channels are weighted, increasing the weight of the red channel and decreasing the weight of the blue channel to obtain the characteristic value of this unit region. The weight of the red channel is increased by x times, and the weight of the blue channel is decreased by y times, i.e., the weight of the red channel is larger and the weight of the blue channel is smaller, to eliminate the influence of temperature scale settings. This combination improves the sensitivity to high and low temperature regions, making it easier to address color and temperature mismatch issues caused by temperature scale settings. In this embodiment, the original weights of red, green, and blue are 0.3, 0.4, and 0.3, respectively, with x = 1 and y = 1. The assigned weights are 0.3 × (1 + x) = 0.6, 0.4, and 0.3 × (1 - y) = 0, respectively. That is, in this embodiment, the value of the blue channel is reduced to 0.

[0059] 3) Construct the matrix:

[0060] a) Constructing a temperature feature matrix: Generate a two-dimensional temperature feature matrix from the temperature feature values ​​of all unit regions. The elements of the temperature feature matrix are positively correlated with the thermal load intensity of the corresponding unit region, and retain the spatial correlation characteristics of the original temperature field distribution. That is, the temperature feature matrix corresponds completely to the thermal imaging image, and the position points of the elements of the temperature feature matrix correspond to the coordinate positions of the unit regions in the thermal imaging image.

[0061] b) Storage matrix: The temperature feature matrix is ​​converted into a standardized data format and persistently processed using a matrix table storage mode to ensure the convenience and integrity of subsequent modules reading the data;

[0062] c) Standardized temperature feature matrix: The temperature feature matrix after persistence is linearly transformed using a normalization method to map all feature values ​​to a unified dimensional range, resulting in a standardized feature matrix. The elements in the standardized feature matrix are standardized feature values. An extreme value normalization method is used to maintain the relative relationship of the original temperature field distribution, eliminate the dimensional influence caused by equipment differences, and eliminate the influence of the environmental reference temperature.

[0063] 4) Locate the main heat source:

[0064] First, the location points that exceed the temperature characteristic threshold are determined by global extreme value search and standardized feature values ​​as candidate core heat sources. Then, a geometric center evaluation algorithm based on Euclidean distance is used to select the point closest to the PCB geometric center from multiple similar candidate core heat sources as the main heat source. Candidate core heat sources other than the main heat source are secondary heat sources.

[0065] 5) Iterative optimization of flow path:

[0066] a) Set the main path:

[0067] A propagation model driven by the temperature gradient of four neighborhoods is established. The reverse derivation process includes: defining the four points above, below, left, and right of a given location as its neighborhood; a location within the neighborhood is called a neighbor; if a neighbor has already been set as the main path, it is not considered; the difference between the standardized eigenvalue of the current location and the standardized eigenvalue of the neighborhood is obtained; the neighbor corresponding to the smallest difference is the next path of the main path; the specific operation is as follows:

[0068] Starting from the main heat source, the position of each step in the main channel path is assigned an integer value in the updated normalized feature matrix, representing the path order of the main channel path. In each step, starting from the current position, a valid neighbor is selected, where the absolute value of the difference between the temperature feature value of the neighbor and the temperature feature value of the current position should be as small as possible. This neighbor will then be designated as the next node position in the main channel path. This process continues until as many positions as possible are assigned a value greater than or equal to 1.

[0069] b) Introduce a backtracking mechanism:

[0070] A backtracking mechanism for the thermal diffusion path is employed to ensure that the calculation of the main channel path does not get stuck. Specifically, the following steps are included: If spatial constraints are encountered during pathfinding, and the entire effective neighborhood of the current point consists of integers greater than or equal to 1, preventing further progress, but there are still floating-point numbers less than 1 within the entire normalized feature matrix, indicating that the entire normalized feature matrix has not been completely covered by the main channel path, then the main channel path will backtrack to the nearest step number t. The backtracking mechanism ensures that the neighboring points of the backtracked main channel path node have at least one unvisited index point, i.e., the neighbor of the next highest temperature. Then, the node with the highest normalized eigenvalue in the effective neighborhood is selected. The new index point is used for pathfinding attempts, and the pathfinding process is re-executed while retaining the main path of the original pathfinding process. It's important to note that if the standardized eigenvalues ​​of candidate neighborhoods are equal, a random selection is made from the neighborhoods with the same standardized eigenvalue. If, after backtracking, the main path still cannot completely cover all points in the standardized eigenvalue matrix, then backtracking is performed to the nearest step label t to select another untested index point with the second highest standardized eigenvalue. When all index points in step label t have undergone pathfinding attempts, then backtracking is performed to the step label t preceding step label t. -1 And so on, the global optimal solution is obtained through iterative optimization;

[0071] c) Adaptive termination determination:

[0072] Set three types of termination conditions:

[0073] (1) The maximum number of attempts has been reached. In this embodiment, the maximum number of attempts is 0.6 times the number of position points of the standardized feature matrix, that is, the number of unit regions divided by the thermal imaging image, i.e., 5×5×0.6=15 times;

[0074] (2) The main road path covers the entire standardized feature matrix;

[0075] (3) Tracing back to the main heat source, no new mainstream path was found;

[0076] If any of the above conditions are met, the entire pathfinding process terminates and the result is output.

[0077] For the output results, if the main channel path can completely cover all points of the entire normalized feature matrix (i.e., there are no integers greater than 1 and no numbers less than 1 in the matrix), then the optimized channel path is obtained. If there are multiple selectable main channel paths and none of them can completely cover all points of the entire normalized feature matrix (i.e., there are integers greater than 1 or numbers less than 1 in the normalized feature matrix), then a secondary channel is opened for each selectable main channel path to cover all points of the normalized feature matrix except for the part covered by the main channel, thus obtaining the candidate optimized channel path.

[0078] One or more secondary channels are opened at locations not covered by the main channel. Each secondary channel is a complete S-shaped channel. The start and end points of each secondary channel are connected to the main channel and are as close as possible to the end point of the main channel path, i.e. the main heat source, to obtain an optimized channel path including the main channel and secondary channels. Each location point is required to pass through the channel path only once. In some cases, the longest main channel generates two unconnected secondary channels, but the second longest path can connect these secondary channels into a single path. In this case, the second longest path is selected as the main channel. For candidate optimized flow paths, the main channel path with the fewest generated secondary channels should be chosen. Under this premise, the longest main channel is selected as the optimized flow path. At the same time, it is observed whether the number of secondary channels is reduced. If the number of secondary channels is reduced, the output result with the longest main channel is discarded, and the second longest main channel is selected as the optimized flow path. In the following case, the longest main channel generates two unconnected secondary channels, but the second longest main channel path can connect these secondary channels into a single path. In this case, the second longest main channel path is selected as the main channel for the optimized flow path.

[0079] d) Generate a visualization output with thermal rendering from the optimized flow path. The optimization of the flow path ends. Draw a 3D model for manufacturing based on the optimized flow path data. The flow path of the main flow channel is in reverse order of the step numbers. The width of the flow channel is smaller than the size of the unit area.

[0080] Following an optimized flow path, refrigerant channels are configured within the liquid cooling plate, with the main flow path sequence following the reverse order of the step labels assigned in the final standardized feature matrix, for liquid cooling of the printed circuit board (PCB). Experimental testing demonstrates significant advantages of the flow path designed in this invention, with heat dissipation efficiency increasing by an average of over 40% compared to parallel DC channels and over 16% compared to S-shaped channels. Due to the flow-heat coupling effect, the heat dissipation capacity gradually approaches that of S-shaped and parallel DC channels at high flow rates. The larger the proportion of the generated main flow path, the better the heat dissipation performance.

[0081] Finally, it should be noted that the purpose of disclosing the embodiments is to help further understand the present invention. However, those skilled in the art will understand that various substitutions and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the embodiments, and the scope of protection of the present invention is defined by the claims.

Claims

1. A chip liquid cooling heat dissipation channel path planning method based on temperature scoring, characterized in that, The path planning method includes the following steps: 1) Acquire thermal imaging images: A high-resolution infrared thermal imager acquires red, green, and blue (RGB) thermal images representing the temperature field distribution on the surface of a printed circuit board (PCB). 2) Extracting thermal features: The RGB thermal imaging image is divided into multiple unit regions, and a weight allocation strategy that is more conducive to temperature difference analysis is adopted to assign temperature characteristic values ​​that reflect thermodynamic properties to each unit region. 3) Construct the matrix: a) Constructing a temperature feature matrix: Generate a two-dimensional temperature feature matrix from the temperature feature values, while preserving the spatial correlation characteristics of the original temperature field distribution; b) Storage matrix: Perform data persistence processing on the temperature feature matrix; c) Standardized temperature feature matrix: A linear transformation is performed using a normalization method to obtain the standardized feature matrix; 4) Locate the main heat source: A global extremum search is performed on the standardized feature matrix to determine the location of the main heat source in the temperature field; 5) Optimize flow path: a) Set the main path: The path generation algorithm based on heat conduction simulation is as follows: a propagation model driven by the temperature gradient of the four neighborhood is established, with the main heat source as the endpoint of the main channel path. Starting from the main heat source, the main channel path is derived in reverse and the standardized feature matrix is ​​updated. This is called the path finding process. b) Introduce a backtracking mechanism: If spatial constraints are encountered during the pathfinding process, a heat diffusion path backtracking mechanism is introduced to attempt a new path and re-execute the pathfinding process. c) Adaptive termination determination: Set an adaptive termination condition. When the adaptive termination condition is met, the optimized flow path ends, the result is output, and the optimized flow path is obtained based on the output result. d) Optimize the flow path to generate a visual output with thermal rendering, and the flow path optimization ends.

2. The path planning method as described in claim 1, characterized in that, In step 2), a temperature feature value is assigned to each unit region, including the following steps: A unit region of an RGB thermal imaging image includes the intensity values ​​of the red, green and blue channels. The intensity values ​​of the three channels are weighted, increasing the weight of the red channel and decreasing the weight of the blue channel to obtain the feature value of this unit region.

3. The path planning method as described in claim 1, characterized in that, In step 5)a), a propagation model driven by temperature gradient in four neighborhoods is established. The reverse derivation process includes: defining the four positions above, below, left, and right of a position point in the normalized feature matrix as a neighborhood, and a position point in the neighborhood is called a neighbor point. If a neighbor point has already been set as the main channel path, it is not considered. The difference between the normalized feature value corresponding to the current position point and the normalized feature value corresponding to the neighborhood is obtained. The neighbor point corresponding to the smallest difference is the next path of the main channel.

4. The path planning method as described in claim 3, characterized in that, Starting from the main heat source, the position of each step in the main channel path is assigned an integer value in the updated normalized feature matrix, representing the path order of the main channel path. In each step, starting from the current position, a valid neighbor is selected, and the absolute value of the difference between the normalized feature value of the neighbor and the normalized feature value of the current position should be as small as possible. Next, this neighbor will be designated as the next node position in the main channel path. This operation continues until as many positions as possible are assigned a value greater than or equal to 1.

5. The path planning method as described in claim 4, characterized in that, In step 5)b), if spatial constraints are encountered during the pathfinding process, and the entire effective neighborhood of the current position point has been set as the main path and the next step cannot be carried out, while there are still position points in the entire normalized feature matrix that are not completely covered by the main path, then the path will backtrack to the nearest step number. A heat diffusion path backtracking mechanism is introduced, so that the neighborhood of the backtracked main path node position has at least one unvisited position point, which is called the backtracking point. The pathfinding process is re-executed, and the main path of the original pathfinding process is retained as an optional main path. If the main path cannot completely cover all position points in the entire normalized feature matrix, this process needs to be performed multiple times.

6. The path planning method as described in claim 1, characterized in that, The adaptive termination condition is: (1) The preset maximum number of attempts has been reached; (2) The main road path covers the entire standardized feature matrix; (3) Tracing back to the main heat source, there is still no new mainstream path.

7. The path planning method as described in claim 1, characterized in that, In step 5)c), for the output result, if the main channel path can completely cover all the position points of the entire normalized feature matrix, then the optimized flow path is obtained; if there are multiple selectable main channel paths and none of them can completely cover all the position points of the entire normalized feature matrix, then a secondary flow path is opened for each selectable main channel path to cover all the position points of the entire normalized feature matrix except for the part covered by the main channel, thus obtaining the candidate optimized flow path.

8. The path planning method as described in claim 7, characterized in that, For the candidate optimization path, select the main flow path scheme with fewer generated secondary flow paths as much as possible. Under this premise, select the longest main flow path, and at the same time observe whether the number of secondary flow paths is reduced. If the number of secondary flow paths can be reduced, discard the output result with the longest main flow path and select the second longest main flow path as the optimization path.

9. The path planning method as described in claim 7, characterized in that, One or more secondary channels are opened at locations not covered by the main channel. Each secondary channel is a complete S-shaped channel. The start and end points of each secondary channel are connected to the main channel and are as close as possible to the end point of the main channel, i.e., the main heat source.

Images