An AI automatic mass production preparation method and production line for diamond copper heat dissipation cold plate

Abstract

This invention discloses an AI-automated mass production method and production line for diamond copper heat dissipation plates, relating to the field of intelligent manufacturing technology for thermally conductive composite materials. The method acquires batch data of diamond particles with a tungsten-containing interface layer on their surface, generates powder gradation and process formulation by an edge AI server, and forms sheet-like preforms through mixing, powder spreading, compaction, and pressing. The sheet-like preforms are then fed into a melt pressure infiltration furnace or a vacuum hot-pressing sintering furnace for high-density forming, followed by annealing to obtain a diamond copper core plate. Further post-processing forms the finished product. The production line employs layered collaborative control between an edge AI server and a PLC controller. The AI ​​outputs process suggestions, and the PLC executes within the hardware safety interlock range. This invention is applicable to the mass production of liquid-cooled heat dissipation plates for ultra-high-power GPUs, AI servers, and other high heat flux density devices.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology for high thermal conductivity composite materials, specifically to an AI-automated mass production method and production line for diamond copper heat dissipation cold plates, and particularly to the application of melt pressure infiltration, vacuum hot pressing sintering and edge AI collaborative control technology in the manufacturing of liquid-cooled cold plates. Background Technology

[0002] Ultra-high power GPUs, AI accelerators, and high heat flux density chip packaging require cold plate substrates with higher heat dissipation capabilities. Diamond-copper composite materials combine the high thermal conductivity of diamond with the machinability of copper, making them an important candidate material for GPU liquid cooling cold plate substrates.

[0003] In diamond-copper composites, the interfacial wettability and interfacial thermal resistance between diamond and copper are key factors determining the final thermal conductivity and reliability. The applicant has filed a prior application related to the upstream raw material interfacial layer, entitled "A Method for Preparing a Nano-Tungsten Interfacial Layer on the Surface of Diamond Micropowder and a Dynamic Coating Apparatus," application number 2026110636942, filed on July 16, 2026. This prior application, through dynamic coating, powder state acquisition, multi-point QCM film thickness monitoring, and equivalent film thickness closed-loop control, enables the formation of a 40–100 nm tungsten interfacial layer on the surface of diamond micropowder. This layer can then form a tungsten-containing carbide transition layer of WC / W2C or a combination thereof during subsequent heat treatment, providing a stable upstream raw material basis for diamond-copper composites.

[0004] However, the prior application mainly addresses the issues of preparing the nano-tungsten interface layer on the surface of diamond micropowder, controlling the film thickness, and dynamic coating equipment. It does not involve further mass-producing the powder into diamond copper core boards for GPU cold plates in a highly densified manner, nor does it involve subsequent copper plating, edge banding, microchannel processing, and AI automated mass production systems.

[0005] Based on this, the existing technology still has the following shortcomings: First, the melt pressure infiltration process relies on a fixed formula and human experience, making it difficult to judge the infiltration completion of different layers and flakes in real time, which can easily lead to local incomplete infiltration, insufficient shrinkage, and cooling warping.

[0006] Second, the vacuum hot pressing sintering process usually uses a fixed time for holding pressure, which does not make full use of signals such as pressure head displacement, displacement change rate, plate parallelism, mold temperature difference and acoustic emission to determine the densification endpoint.

[0007] Third, existing manufacturing systems are typically configured around a single forming device, lacking a control mechanism that can cover both melt pressure infiltration and vacuum hot pressing sintering routes on the same data platform, and also lacking a decision-making model that automatically selects the route based on product size, batch size, and target performance.

[0008] Fourth, existing processes generally lack a full-process AI multimodal closed-loop architecture, have not established a hierarchical safety collaborative control between edge AI servers and PLC controllers, and have not established a system for porosity prediction, infiltration digital twin, hot-pressing shrinkage stage identification, acoustic emission defect identification, and quality traceability.

[0009] To address the aforementioned issues, proposing an AI-automated mass production method and production line for diamond copper heat dissipation plates has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] In view of this, the present invention provides an AI-automated mass production method and production line for diamond copper heat dissipation cold plates. It covers two high-density process routes, melt pressure infiltration and vacuum hot pressing sintering, within the same upper-level technology framework. It opens up the data link between upstream diamond interface layer preparation data and downstream forming process, and realizes full-process AI closed-loop control and automatic endpoint determination.

[0011] To achieve the above objectives, the present invention adopts the following technical solution: An AI-automated mass production method for diamond copper heat sinks includes the following steps: S1: Obtain batch characteristic data of diamond particles with a tungsten-containing interface layer on the surface; S2: Based on batch feature data and target product parameters, an edge AI server generates a preparation scheme that includes at least powder gradation and process formulation; S3: According to the preparation plan, the diamond particles are prepared into sheet-like preforms; S4: The sheet preform is fed into the high-density forming unit for high-density forming to form a diamond copper core plate; S5: Post-process the diamond copper core board to form the finished diamond copper heat dissipation cold plate; The edge AI server collects sensor data from at least one step in the preparation method and outputs process control instructions through a multimodal process model; the PLC controller with hardware safety interlock receives the control instructions and executes them within a preset safety boundary to control at least some steps of the preparation method to run automatically.

[0012] Optionally, in the above method, the tungsten-containing interface layer in S1 is a tungsten, tungsten / tungsten carbide, WC / W2C or a combination thereof interface layer of 40-100 nm. Batch characteristic data should include at least equivalent film thickness, particle size distribution, powder loading mass, interface layer batch number, and film thickness sampling results.

[0013] Optionally, the target product parameters in S2 of the above method include target thermal conductivity, board thickness, copper cladding structure, product size, and production line cycle time. The process formulations generated by the edge AI server include powder gradation, preform forming formulation, high densification forming formulation, annealing stress relief formulation, and post-treatment formulation.

[0014] The above method, optionally, includes the following specific content for S3: According to the preparation plan, sheet-like preforms are formed by powder gradation and mixing, automatic powder spreading and compaction, and cold pressing. The sheet-type preform is either a diamond framework preform or a diamond-copper powder mixed preform.

[0015] Optionally, the above method involves automatic powder compaction and preform forming by acquiring powder surface images through a machine vision camera and identifying powder shortages, agglomerations, local accumulation, scraper marks, edge damage, and abnormal powder height. When the percentage of agglomerated pixels, powder shortage areas, local accumulation areas, or thickness deviations exceed a threshold, the edge AI server outputs suggestions to the PLC controller for re-compacting, powder replenishment, reducing the thickness of a single layer of powder, rework, or rejection.

[0016] The above method, optionally, includes the following specific content for S4: The sheet-like preform is fed into a high-density forming unit for high-density forming. The high-density forming unit is either a melt pressure infiltration furnace or a vacuum hot pressing sintering furnace. If a melt pressure infiltration furnace is used, diamond copper core plates are formed through preheating degassing, copper liquid pressure infiltration, pressure holding solidification, and slow cooling annealing. If a vacuum hot-pressing sintering furnace is used, diamond copper core plates are formed under vacuum or inert atmosphere through heating, pressurization, pressure holding, pressure release and slow cooling.

[0017] Optionally, when high-density forming is achieved using melt pressure infiltration, the melt pressure infiltration furnace is configured to load 100 to 300 pieces per furnace; the cycle time per furnace is 5.0 to 8.0 hours. Under the condition of an overall equipment efficiency of 75% to 85%, a single production line can achieve a designed capacity of 18,000 to 20,000 pieces per month, and a total capacity of 100,000 pieces per month can be achieved by connecting multiple single lines in parallel. When the melt pressure infiltration furnace adopts a multi-layer graphite infiltration module, the multi-layer graphite infiltration module includes two 100-piece modules, one 200-piece module, or multiple equivalent plate position modules; each layer includes a graphite tray layer, a sheet diamond preform receiving cavity, a copper liquid diversion channel, an exhaust channel, an overflow channel, and a feeding channel, and has independent or semi-independent diversion, exhaust, overflow, and feeding paths.

[0018] Optionally, when the high-density forming is performed by melt pressure infiltration, the copper liquid temperature is 1080–1280℃, the infiltration pressure is 0.1–30MPa, and the holding time is 0.5–120 minutes. The edge AI server recommends the heating curve, pressure curve, and holding time based on the preform porosity, infiltration resistance index, mold layer, copper liquid temperature, and interface layer data. When high-density forming is vacuum hot-pressing sintering, diamond particles with a tungsten-containing interface layer on the surface are mixed with copper powder to form a sheet-like diamond-copper powder mixed preform, which is then hot-pressed and sintered under vacuum or inert atmosphere; the sintering temperature is 850-1050℃, the sintering pressure is 20-100MPa, and the holding time is 10-90 minutes.

[0019] Optionally, the edge AI server reads the equivalent film thickness, QCM integration record, powder visual state, particle size distribution, and film thickness sampling data output by the upstream diamond dynamic coating equipment or process database; when the equivalent film thickness is 45-60nm, it matches forming parameters with lower interface reaction risk; when the equivalent film thickness is 70-100nm, it adjusts the preheating, forming, and annealing process windows. Sensor data includes at least one of machine vision, temperature, pressure, vacuum, flow rate, liquid level, acoustic emission, displacement, weighing, platen parallelism, and quality inspection data; The multimodal process model includes at least the following sub-models: preform porosity estimation, infiltration resistance prediction, copper liquid filling state identification, hot-pressing shrinkage stage identification, solidification feeding prediction, warpage risk prediction, post-processing quality prediction, and production line cycle time scheduling. The multimodal process model outputs process routes, formula suggestions, target curves, alarm levels, wafer location quality risks, and scheduling strategies. The PLC controller operates within the hardware safety range of temperature, pressure, vacuum, pressure head displacement, cooling water, furnace door, valve group, and emergency stop interlock. The edge AI server generates an impregnation completion score based on the pressure decay rate, gas supply flow rate, valve position change, copper liquid level drop, mold temperature gradient, exhaust channel temperature rise, and acoustic emission events. When the pressure decay rate, gas supply volume per unit time, mold temperature difference, liquid level change rate, and acoustic emission events meet the preset endpoint conditions, it outputs an impregnation completion label and performs temperature compensation on the pressure and gas supply volume based on the effective gas volume of the furnace cavity, gas temperature, and gas state equation. The edge AI server identifies the densification stage and densification endpoint based on the pressure head, displacement, displacement change rate, sintering shrinkage curve, temperature difference between the mold center and edge, vacuum degree, platen parallelism, and acoustic emission events. It then outputs suggestions for heating, pressurizing, holding, depressurizing, and slow cooling curves to the PLC controller. When the displacement change rate is below a preset threshold, the pressure remains stable, the acoustic emission abnormality events are below a preset threshold, and the mold temperature difference and parallelism meet the requirements, a sintering completion label is output. After high-density forming, staged pressure relief and controlled annealing and slow cooling are performed; in the melt pressure infiltration route, after pressure holding and solidification, stress relief and heat preservation steps are performed at 300-700℃ for 5-120 minutes; in the vacuum hot pressing sintering route, after sintering and pressure holding, pressure is released according to a preset gradient, and annealing and slow cooling are performed in a vacuum or inert atmosphere.

[0020] An AI-automated mass production line for implementing the AI-automated mass production method for a diamond copper heat sink as described in any of the above claims, comprising: Powder gradation and mixing unit, automatic powder spreading and vibration and preform forming unit, high densification forming unit, annealing slow cooling furnace, double-sided grinding and inspection unit, upper and lower copper plating unit, four-sided copper edging unit, shovel tooth or microchannel processing unit, final inspection and sorting unit, edge AI server, PLC controller, MES or process database connected to automatic loading and unloading or transfer mechanism. The high-density forming unit includes a melt pressure infiltration furnace, a vacuum hot pressing sintering furnace or a combination thereof; the melt pressure infiltration furnace is used in conjunction with a preheating degassing furnace and a multi-layer graphite infiltration module, and the vacuum hot pressing sintering furnace is used in conjunction with an upper pressure head, a lower pressure head, a graphite hot pressing mold, a thickness limiting structure, a displacement sensor and a pressure plate parallelism detection unit. The edge AI server includes a GPU or AI accelerator card, an industrial data acquisition interface, an image inference module, a timing signal inference module, a route recognition module, a process database, a model version management module, and an industrial Ethernet interface for communicating with the PLC controller. The PLC controller connects to the heating power supply, vacuum valve group, inert gas valve group, proportional pressure valve, pressure head loading mechanism, displacement limit mechanism, gas replenishment valve, cooling medium valve, furnace door interlock, automatic transfer mechanism, alarm, and emergency stop circuit. The hardware safety interlock priority of the PLC controller is higher than the AI ​​recommended parameters.

[0021] As can be seen from the above technical solution, compared with the prior art, the present invention provides an automated mass production method and production line for diamond copper heat dissipation cold plates, which has the following beneficial effects: (1) Forming a high-density process protection system covering two routes: melt pressure infiltration and vacuum hot pressing sintering, which can be flexibly selected according to production volume, size, consistency and verification requirements.

[0022] (2) Determine the endpoint for two routes. The pressure impregnation route uses pressure decay, gas replenishment, liquid level, temperature difference, temperature rise of exhaust channel and acoustic emission to determine the degree of impregnation completion; the hot pressing sintering route uses pressure, displacement, displacement change rate, shrinkage curve, parallelism, vacuum degree and acoustic emission to determine the densification endpoint.

[0023] (3) Achieve data exchange between upstream and downstream. The equivalent film thickness t_eq, particle size distribution and powder state of the upstream tungsten powder are used as inputs to two high densification models, so that the temperature, pressure, holding pressure and annealing parameters can be adaptively adjusted according to the state of the interface layer.

[0024] (4) The melt pressure infiltration route can form a multi-layer array graphite infiltration module with 200 pieces / furnace level. Each layer is equipped with independent diversion, venting, overflow and feeding structures, and the layer position, row and column and piece position data are bound to the quality results.

[0025] (5) A layered safety closed loop of AI+PLC is adopted. AI is responsible for route identification, model reasoning, quality prediction, parameter suggestion and production scheduling, while PLC controller is responsible for deterministic execution and hardware safety interlock.

[0026] (6) A segmented mass production chain is adopted, which includes core board forming, copper plating on the top and bottom, edge wrapping on all four sides, and microchannel processing with copper plating on the top. This avoids complex processing on the diamond-copper composite core material directly and allows for detection and sorting before each high-value post-processing step. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0028] Figure 1 This is a flowchart of an AI-automated mass production method for diamond copper heat dissipation cold plates disclosed in this invention. Figure 2 This is a schematic diagram of the overall layout of a single AI automated production line disclosed in this invention; Figure 3 This is a schematic diagram of the internal structure of the multilayer graphite infiltration module disclosed in this invention; Figure 4 This is a schematic diagram of the diamond copper heat sink structure disclosed in this invention; Figure 5 This is a schematic diagram of the collaborative control architecture of edge AI server, PLC and MES disclosed in this invention; Figure 6This is a schematic diagram of the AI ​​quality write-back and model iteration closed loop disclosed in this invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0031] The purpose of this invention is to provide an AI-automated mass production method and production line for diamond copper heat dissipation cold plates. Within the same upper-level technology framework, it covers two high-density process routes: melt pressure infiltration and vacuum hot pressing sintering. It establishes a data link between upstream diamond interface layer preparation data and downstream forming process, realizing full-process AI closed-loop control and automatic endpoint determination.

[0032] The high-density forming unit referred to in this invention is a core forming device used to transform sheet preforms into diamond copper core plates. It can be a melt pressure infiltration furnace or a vacuum hot-pressing sintering furnace. The melt pressure infiltration furnace is suitable for array-type mass production routes; the vacuum hot-pressing sintering furnace is suitable for pilot-scale verification, small-batch high-consistency preparation, and products with special dimensions. Both routes utilize an edge AI server to read powder batch data, equivalent interfacial layer thickness, preform visual data, temperature, pressure, vacuum, acoustic emission data, and quality inspection data, which are then executed by a PLC controller within hardware safety boundaries.

[0033] This invention employs a control architecture that coordinates an edge AI server and a PLC controller. The edge AI server runs multimodal process models, route recognition models, infiltration digital twin models, hot-pressing shrinkage stage models, visual defect recognition models, quality prediction models, anomaly detection models, and Bayesian process optimization models; the PLC controller executes heating, vacuuming, valve grouping, pressure, pressure head loading, displacement limiting, gas replenishment, cooling, transfer, alarm, and safety interlocking.

[0034] The AI ​​automated production line of the present invention includes a powder gradation and mixing unit, an automatic powder spreading and compaction and preform forming unit, a high-density forming unit, an annealing slow cooling furnace, a double-sided grinding and inspection unit, an upper and lower copper plating unit, a four-sided copper edging unit, a toothed or microchannel processing unit, a final inspection and sorting unit, an edge AI server, a PLC controller and an MES or process database.

[0035] When the melt pressure infiltration route is selected, a sheet-like diamond skeleton preform is inserted into a multi-layer graphite infiltration module, followed by preheating and degassing, copper melt pressure infiltration, pressure holding and solidification, and slow annealing and cooling to form a diamond copper core plate. When the vacuum hot pressing sintering route is selected, diamond particles and copper powder form a sheet-like mixed preform, which is inserted into an array-type graphite hot pressing mold with thickness limits. Under vacuum or inert atmosphere, heating, pressurizing, holding, depressurizing, and slow cooling are performed to form a diamond copper core plate. See also Figure 3 This is a schematic diagram of the internal structure of the multilayer graphite impregnation module disclosed in this invention, including: sheet preform 3, graphite tray layer 23, copper liquid diversion channel 24, exhaust channel 25, overflow channel 26, and shrinkage channel 27.

[0036] This invention employs a segmented approach: first, a diamond copper core board is prepared; then, a second copper plating and edge banding are performed. The upper thick copper layer and the lower thin copper layer are composited by vacuum brazing or diffusion welding; the copper banding around the edges is achieved by chemical copper plating followed by thickening electroplating or copper frame brazing; finally, serrations or microchannels are machined within the upper copper plating layer.

[0037] The specific technical solution is as follows: See Figure 1 As shown, this invention discloses an AI-automated mass production method for diamond copper heat dissipation plates, comprising the following steps: S1: Obtain batch characteristic data of diamond particles with a tungsten-containing interface layer on the surface; S2: Based on batch feature data and target product parameters, an edge AI server generates a preparation scheme that includes at least powder gradation and process formulation; S3: According to the preparation plan, the diamond particles are prepared into sheet-like preforms; S4: The sheet preform is fed into the high-density forming unit for high-density forming to form a diamond copper core plate; S5: Post-process the diamond copper core board to form the finished diamond copper heat dissipation cold plate; The edge AI server collects sensor data from at least one step in the preparation method and outputs process control instructions through a multimodal process model; the PLC controller with hardware safety interlock receives the control instructions and executes them within a preset safety boundary to control at least some steps of the preparation method to run automatically.

[0038] Furthermore, the tungsten-containing interface layer in S1 is a tungsten, tungsten / tungsten carbide, WC / W2C or a combination thereof interface layer of 40-100 nm. Batch characteristic data should include at least equivalent film thickness, particle size distribution, powder loading mass, interface layer batch number, and film thickness sampling results.

[0039] Furthermore, the target product parameters in S2 include target thermal conductivity, board thickness, copper cladding structure, product size, and production line cycle time. The process formulations generated by the edge AI server include powder gradation, preform forming formulation, high densification forming formulation, annealing stress relief formulation, and post-treatment formulation.

[0040] Furthermore, the specific content of S3 is as follows: According to the preparation plan, sheet-like preforms are formed by powder gradation and mixing, automatic powder spreading and compaction, and cold pressing. The sheet-type preform is either a diamond framework preform or a diamond-copper powder mixed preform.

[0041] Furthermore, the automatic powder spreading and compaction and preform forming process uses a machine vision camera to collect powder surface images and identify powder shortages, agglomerations, local accumulation, scraper marks, edge damage, and abnormal powder spreading height. When the percentage of agglomerated pixels, powder shortage areas, local accumulation areas, or thickness deviations exceed the threshold, the edge AI server outputs suggestions to the PLC controller for re-compacting, powder replenishment, reducing the thickness of a single layer of powder spreading, rework, or rejection.

[0042] Furthermore, the specific content of S4 is as follows: The sheet-like preform is fed into a high-density forming unit for high-density forming. The high-density forming unit is either a melt pressure infiltration furnace or a vacuum hot pressing sintering furnace. If a melt pressure infiltration furnace is used, diamond copper core plates are formed through preheating degassing, copper liquid pressure infiltration, pressure holding solidification, and slow cooling annealing. If a vacuum hot-pressing sintering furnace is used, diamond copper core plates are formed under vacuum or inert atmosphere through heating, pressurization, pressure holding, pressure release and slow cooling.

[0043] Furthermore, when high-density forming is achieved using melt pressure infiltration, the melt pressure infiltration furnace is configured to load 100 to 300 pieces per furnace; the cycle time per furnace is 5.0 to 8.0 hours. Under the condition of an overall equipment efficiency of 75% to 85%, a single production line can achieve a designed capacity of 18,000 to 20,000 pieces per month, and a total capacity of 100,000 pieces per month can be achieved by connecting multiple single lines in parallel. When the melt pressure infiltration furnace adopts a multi-layer graphite infiltration module, the multi-layer graphite infiltration module includes two 100-piece modules, one 200-piece module, or multiple equivalent plate position modules; each layer includes a graphite tray layer, a sheet diamond preform receiving cavity, a copper liquid diversion channel, an exhaust channel, an overflow channel, and a feeding channel, and has independent or semi-independent diversion, exhaust, overflow, and feeding paths.

[0044] Furthermore, when high-density forming is performed by melt pressure infiltration, the copper liquid temperature is 1080–1280℃, the infiltration pressure is 0.1–30MPa, and the holding time is 0.5–120 minutes; the edge AI server recommends the heating curve, pressure curve, and holding time based on the preform porosity, infiltration resistance index, mold layer, copper liquid temperature, and interface layer data. When high-density forming is vacuum hot-pressing sintering, diamond particles with a tungsten-containing interface layer on the surface are mixed with copper powder to form a sheet-like diamond-copper powder mixed preform, which is then hot-pressed and sintered under vacuum or inert atmosphere; the sintering temperature is 850-1050℃, the sintering pressure is 20-100MPa, and the holding time is 10-90 minutes.

[0045] Furthermore, the specific details of S5 are as follows: The diamond copper core board undergoes double-sided grinding inspection, top and bottom copper plating, copper edging around the edges, toothing or microchannel processing, and final inspection and sorting to form the finished diamond copper heat dissipation cold plate.

[0046] Specifically, double-sided grinding inspection includes at least four of the following: double-sided grinding, flatness inspection, ultrasonic C-scan, X-ray inspection, CT inspection, density inspection, thermal conductivity sampling inspection, and visual defect inspection; the inspection results are written back to the MES or process database and used to correct the process parameters for the next batch.

[0047] Specifically, the top and bottom copper cladding includes placing a 1.0-5.0 mm thick oxygen-free copper plate on the upper side and a 0.2-1.0 mm thick oxygen-free copper plate on the lower side of the diamond copper core board, forming the top and bottom copper cladding layers through vacuum brazing, thin-layer brazing, thermo-pressure diffusion welding, or a combination thereof; the surrounding copper cladding includes forming a 0.2-1.0 mm thick surrounding copper cladding layer through edge roughening, activation, chemical copper plating as a base layer, and thick copper electroplating, or forming the surrounding copper cladding layer through a combination of copper frame brazing and electroplating repair; the chiseling or microchannel machining is performed within the top copper cladding layer.

[0048] See Figure 4 The diagram shows a diamond copper heat sink structure disclosed in this invention, including: an upper copper layer 29, a diamond copper core layer 28, a lower copper layer 30, and a surrounding copper cladding layer 31. A toothed or microchannel layer 32 is provided on the upper copper layer 29.

[0049] Furthermore, the edge AI server reads the equivalent film thickness, QCM integration records, powder visual status, particle size distribution, and film thickness sampling data output from the upstream diamond dynamic coating equipment or process database; when the equivalent film thickness is 45-60nm, it matches forming parameters with a lower risk of interface reaction; when the equivalent film thickness is 70-100nm, it adjusts the preheating, forming, and annealing process windows. Sensor data includes at least one of machine vision, temperature, pressure, vacuum, flow rate, liquid level, acoustic emission, displacement, weighing, platen parallelism, and quality inspection data; The multimodal process model includes at least the following sub-models: preform porosity estimation, infiltration resistance prediction, copper liquid filling state identification, hot-pressing shrinkage stage identification, solidification feeding prediction, warpage risk prediction, post-processing quality prediction, and production line cycle time scheduling. The multimodal process model outputs process routes, formula suggestions, target curves, alarm levels, wafer location quality risks, and scheduling strategies. The PLC controller operates within the hardware safety range of temperature, pressure, vacuum, pressure head displacement, cooling water, furnace door, valve group, and emergency stop interlock. The edge AI server generates an impregnation completion score based on the pressure decay rate, gas supply flow rate, valve position change, copper liquid level drop, mold temperature gradient, exhaust channel temperature rise, and acoustic emission events. When the pressure decay rate, gas supply volume per unit time, mold temperature difference, liquid level change rate, and acoustic emission events meet the preset endpoint conditions, it outputs an impregnation completion label and performs temperature compensation on the pressure and gas supply volume based on the effective gas volume of the furnace cavity, gas temperature, and gas state equation. The edge AI server identifies the densification stage and densification endpoint based on the pressure head, displacement, displacement change rate, sintering shrinkage curve, temperature difference between the mold center and edge, vacuum degree, platen parallelism, and acoustic emission events. It then outputs suggestions for heating, pressurizing, holding, depressurizing, and slow cooling curves to the PLC controller. When the displacement change rate is below a preset threshold, the pressure remains stable, the acoustic emission abnormality events are below a preset threshold, and the mold temperature difference and parallelism meet the requirements, a sintering completion label is output. After high-density forming, staged pressure relief and controlled annealing and slow cooling are performed; in the melt pressure infiltration route, after pressure holding and solidification, stress relief and heat preservation steps are performed at 300-700℃ for 5-120 minutes; in the vacuum hot pressing sintering route, after sintering and pressure holding, pressure is released according to a preset gradient, and annealing and slow cooling are performed in a vacuum or inert atmosphere.

[0050] This invention also discloses an AI-automated mass production line for implementing the AI-automated mass production method for diamond copper heat sinks described in any of the above claims, comprising: Powder gradation and mixing unit, automatic powder spreading and vibration and preform forming unit, high densification forming unit, annealing slow cooling furnace, double-sided grinding and inspection unit, upper and lower copper plating unit, four-sided copper edging unit, shovel tooth or microchannel processing unit, final inspection and sorting unit, edge AI server, PLC controller, MES or process database connected to automatic loading and unloading or transfer mechanism. The high-density forming unit includes a melt pressure infiltration furnace, a vacuum hot pressing sintering furnace or a combination thereof; the melt pressure infiltration furnace is used in conjunction with a preheating degassing furnace and a multi-layer graphite infiltration module, and the vacuum hot pressing sintering furnace is used in conjunction with an upper pressure head, a lower pressure head, a graphite hot pressing mold, a thickness limiting structure, a displacement sensor and a pressure plate parallelism detection unit. The edge AI server includes a GPU or AI accelerator card, an industrial data acquisition interface, an image inference module, a timing signal inference module, a route recognition module, a process database, a model version management module, and an industrial Ethernet interface for communicating with the PLC controller. The PLC controller connects to the heating power supply, vacuum valve group, inert gas valve group, proportional pressure valve, pressure head loading mechanism, displacement limit mechanism, gas replenishment valve, cooling medium valve, furnace door interlock, automatic transfer mechanism, alarm, and emergency stop circuit. The hardware safety interlock priority of the PLC controller is higher than the AI ​​recommended parameters.

[0051] The present invention also discloses a diamond copper heat dissipation plate, comprising a diamond copper core layer, an upper copper cladding layer located above the diamond copper core layer, a lower copper cladding layer located below the diamond copper core layer, and a surrounding copper cladding layer located at the four edges; the diamond particles in the diamond copper core layer have a 40-100 nm tungsten, tungsten / tungsten carbide, WC / W2C or a combination thereof interface layer on their surface, and the upper copper cladding layer forms a toothed or microchannel layer.

[0052] The overall technical approach of this invention is described in detail below: The diamond particles used in this invention have a tungsten-containing interface layer on their surface. The tungsten-containing interface layer is preferably a tungsten, tungsten / tungsten carbide, WC / W2C, or a combination thereof, with a wavelength of 40–100 nm. This powder can be prepared using the method described in the applicant's prior application, entitled "A Method for Preparing a Nano-Tungsten Interface Layer on the Surface of Diamond Micropowder and a Dynamic Coating Device," application number 2026110636942. This invention reads batch data of the powder, including equivalent film thickness t_eq, particle size distribution D10 / D50 / D90, interface layer batch number, and film thickness sampling results, and uses this data as input parameters for the high-density forming process.

[0053] The overall technical route is as follows: upstream diamond dynamic tungsten-plated powder batch data reading → interface layer diamond particle batch identification → AI powder gradation and mixing → machine vision powder spreading → vibration compaction and cold pressing → sheet preform modeling → high densification process route selection → high densification forming → controlled annealing and slow cooling → double-sided grinding inspection → top and bottom copper plating → copper edging around the perimeter → tooth scraping or microchannel processing → final inspection, sorting and data write-back.

[0054] For the melt pressure infiltration route, the diamond skeleton preform is loaded into a multi-layer graphite infiltration module, and vacuum degassing, copper melt melting and homogenization, pressure infiltration, pressure holding solidification, staged pressure relief and slow cooling annealing are performed in sequence.

[0055] For the vacuum hot pressing sintering route, diamond particles and copper powder form a mixed preform, which is then loaded into an array-type graphite hot pressing mold. The process involves sequentially performing vacuuming or inert gas replacement, heating, pressurization, pressure holding, endpoint determination, staged pressure release, and slow cooling.

[0056] The edge AI server generates route tags and corresponding process formulas based on order quantity, product size, target thermal conductivity, powder batch, interface layer status, equipment availability, and delivery requirements. A shared set of pre-treatment and post-treatment equipment can be set up in the same factory, and at least one of the following can be configured: pressure impregnation station and vacuum hot pressing station.

[0057] like Figure 2 As shown, a single production line connects shared stations for powder gradation, preform forming, annealing, grinding, copper plating, edge banding, tooth removal or microchannel processing, and final inspection to at least one melt pressure infiltration station or vacuum hot pressing sintering station; edge AI servers, PLC controllers, and MES or process databases interact with each station through an industrial network.

[0058] This invention implements the diamond copper core board forming, top and bottom copper coating, surrounding copper edging and tooth or microchannel processing in segments. Each segment is equipped with independent quality inspection points and AI vision inspection points, and defective products are sorted out before subsequent high-value processes.

[0059] The AI ​​vision and multi-sensor data system of this invention is as follows: The AI ​​vision system is deployed at various stages, including powder application, pre-pressing, mold assembly, demolding, post-grinding, post-copper plating, post-electroplated edge wrapping, and post-tooth removal. Powder application vision identifies powder shortages, agglomerations, and powder surface ripples; demolding vision identifies cracks, missing corners, and surface copper nodules; post-copper plating vision identifies bulges, edge solder overflow, and surface contamination; and post-tooth removal vision detects tooth height, tooth pitch, burrs, and tooth collapse.

[0060] In the melt pressure infiltration route, temperature sensors collect copper liquid temperature, furnace temperature, mold center temperature, mold edge temperature, and interlayer temperature; pressure, vacuum, flow, gas replenishment, and liquid level sensors are used to determine degassing, copper liquid filling, gas leakage, and blockage; acoustic emission sensors are coupled to the mold or mold support through a high-temperature acoustic waveguide rod to assist in determining preform cracking, copper liquid impact, or mold loosening.

[0061] In the vacuum hot pressing sintering process, pressure sensors collect the loading force of the pressure head, displacement sensors collect the displacement and displacement change rate of the upper or lower pressure head, and the plate parallelism detection unit collects the relative tilt of the two pressure plates or the displacement difference at multiple points; temperature, vacuum and acoustic emission signals are used to identify the exhaust, particle rearrangement, plastic deformation, sintering shrinkage, abnormal cracking and pressure holding stabilization stages.

[0062] All data is entered into the database according to powder batch, process route, furnace number, mold number, layer, sheet position, and model version. Each product has a complete data history from upstream powder coating to final inspection.

[0063] This invention relates to a multimodal process model and an edge AI server: The multimodal process model referred to in this invention is a manufacturing process model that takes diamond powder batch, equivalent film thickness t_eq, particle size distribution, visual images, temperature time series, pressure time series, flow time series, liquid level change, acoustic emission events, pressure head displacement, pressure plate parallelism, preform forming curve and quality inspection results as inputs, and outputs route selection, formula recommendation, stage identification, anomaly alarm, quality prediction, Bayesian optimization and production line scheduling.

[0064] Among them, particle size distributions D10, D50, and D90 represent the particle size values ​​corresponding to cumulative distribution percentages of 10%, 50%, and 90%, respectively; QCM stands for Quartz Crystal Microbalance, used for in-situ monitoring of coating thickness.

[0065] The process model is composed of a physical constraint model, a machine vision model, a time-series feature model, a digital twin agent model, a quality prediction model, a Bayesian optimization model, and a production scheduling model. Its output must be confirmed by PLC safety limiting and process authority.

[0066] During the preform stage, the model outputs porosity, connectivity index, infiltration resistance index, or expected hot-pressing shrinkage value based on particle size distribution, weighing, powder thickness, compaction curve, pressing curve, and visual images. For substandard preforms, the system rejects or reworks them before they enter the high-densification equipment.

[0067] During the melt pressure infiltration stage, the model judges the infiltration status based on pressure decay, gas supply, valve position change, copper liquid level drop, temperature gradient, exhaust channel temperature rise and acoustic emission events, and outputs labels such as infiltration not started, rapid infiltration, slow infiltration, near completion, completion, suspected blockage, suspected gas leakage or suspected cracking.

[0068] During the vacuum hot pressing sintering stage, the model judges the state of exhaust, particle rearrangement, initial compaction, rapid shrinkage, slow densification, final stabilization, off-center loading or suspected cracking based on pressure, displacement, displacement change rate, displacement-time curve, mold temperature difference, vacuum degree, platen parallelism and acoustic emission events.

[0069] During the production line scheduling phase, the model generates a production scheduling strategy based on the remaining time of the two high-density equipment, the remaining time of the annealing furnace, the grinding queue length, the copper plating furnace load, the electroplating tank status, the gear scraper load, and the delivery priority.

[0070] like Figure 5 and Figure 6 As shown, the edge AI server is responsible for model inference and parameter suggestions, the PLC controller is responsible for deterministic execution and hardware safety interlocking, and the MES or process database is responsible for batch, furnace, layer, wafer position, test results and model version traceability; after the quality results are verified, they are written back to the model training and process optimization stages.

[0071] This invention relates to PLC collaborative control and safety interlocking: The PLC controller is responsible for deterministic execution and safety interlocking. The AI ​​model outputs process routes, recommended temperature curves, pressure curves, pressure head displacement targets, holding time, cooling curves, and scheduling instructions, but the PLC controller decides whether to execute based on preset safety upper and lower limits, displacement limits, and interlocking conditions.

[0072] The PLC controller is directly connected to the heating power supply, vacuum valve group, inert gas valve group, proportional pressure valve, gas replenishment valve, hydraulic or servo pressure head, displacement limit mechanism, cooling medium valve, furnace door lock, automatic transfer mechanism, audible and visual alarm, and emergency stop circuit. When the furnace temperature, furnace pressure, pressure head load, displacement, cooling water, vacuum, or furnace door status is abnormal, the PLC controller will directly enter the safety handling procedure without waiting for AI judgment.

[0073] The edge AI server communicates with the PLC controller via OPC UA, Modbus TCP, Profinet, EtherCAT, or Industrial Ethernet. AI outputs include route labels, recipe numbers, stage switching suggestions, parameter correction amounts, alarm levels, and batch risk labels; the PLC controller returns actual execution curves, alarm status, equipment status, and human intervention records.

[0074] This invention relates to two high-density processing routes and production capacity configurations: Route A is a melt pressure infiltration mass production route. In a preferred embodiment, a single production line uses a melt pressure infiltration furnace with 200 wafers / furnace. The 200 wafers can be composed of two 100-wafer graphite modules, each 100-wafer module including 50 wafers / layer × 2 layers; or it can be composed of a 200-wafer graphite module with 50 wafers / layer × 4 layers.

[0075] Each graphite tray layer has independent copper liquid distribution channels, venting channels, overflow channels, and feeding channels. The edge AI server records the temperature, impregnation signals, and quality results for each layer, row, column, and sheet to identify system differences between the central area, edge area, corner area, and different layers.

[0076] The cycle time for a single furnace is 5.0 to 8.0 hours. When a single furnace is loaded with 200 wafers, the cycle time is 6 hours, there are 4 furnaces per day, 30 days per month, and the overall equipment efficiency is 75%, the monthly production capacity of a single production line is 200 × 4 × 30 × 0.75 = 18,000 wafers; after improving the overall equipment efficiency, it can reach approximately 19,000 to 20,000 wafers per month.

[0077] Route B is a vacuum hot pressing sintering route, suitable for pilot-scale verification, small-batch high-consistency preparation, special-size products, and process window verification before pressure impregnation mass production. This route can use single-cavity, multi-cavity, array-type graphite hot pressing molds or multiple hot pressing stations in parallel, and shares mixing, grinding, copper plating, edge banding, tooth scraping, inspection, and MES resources with Route A.

[0078] The array-type graphite hot pressing mold includes an upper graphite pressure plate, a sheet-type diamond-copper powder mixed preform, a lower graphite shallow cavity mold, and multiple array cavities. It works in conjunction with a thickness limiting structure, a displacement sensor, and a pressure plate parallelism detection unit to control the thickness consistency and off-center load risk during the multi-sheet synchronous densification process.

[0079] The two routes can be configured independently or in parallel on the same production line. The edge AI server selects the route based on order size, dimensions, material formula, delivery cycle, and equipment status. The quality results of different routes are entered into a unified database.

[0080] The segmented mass production process of this invention: The first stage involves the high-density forming of diamond copper core boards. Route A forms the core board through melt pressure infiltration and pressure-holding solidification; Route B forms the core board through vacuum hot pressing sintering. After the core boards are removed from the furnace, they undergo annealing, grinding, and inspection. Unqualified core boards are rejected before copper plating.

[0081] The second section consists of top and bottom copper cladding. A 3.0–3.5 mm thick oxygen-free copper plate is placed on the top side, and a 0.5–0.8 mm thick oxygen-free copper plate is placed on the bottom side. The top and bottom copper cladding layers are formed by vacuum brazing, thin-layer brazing, or thermo-pressure diffusion welding.

[0082] The third section consists of a copper cladding around the perimeter. This is achieved by roughening the edges, activating the metal, applying a base coat of electroless copper plating, and then thickening the cladding with electroplating to form a 0.2–1.0 mm copper cladding layer. The target cladding thickness is then obtained through edge grinding. Alternatively, a combination of copper frame brazing and electroplating repair can be used.

[0083] The fourth stage involves tooth-shaving or microchannel machining. This machining is performed only within the overlying copper layer, preventing the diamond copper core layer from directly bearing the complex micro-machining and reducing the risk of edge chipping and particle shedding.

[0084] The following are embodiments based on the technical solutions disclosed in this invention: Example 1: A single AI pressure impregnation production line with a capacity of 18,000 pieces / month Each production line is equipped with one melt pressure infiltration furnace, with a capacity of 200 wafers per furnace. Two 100-wafer graphite modules are loaded into the furnace side by side, with each module consisting of 50 wafers / layer × 2 layers. The preform forming line is configured with a gross capacity of no less than 300 wafers / hour, and the infiltration furnace operates on a 6-hour cycle.

[0085] The edge AI server reads the weight, thickness, visual image, and pressing curve of each preform and assigns a unique number to each piece. During the impregnation process, the copper melt temperature, mold center temperature, interlayer temperature, furnace pressure, gas supply flow rate, liquid level, and acoustic emission events are collected to determine the impregnation status in real time. Based on 200 pieces / furnace, 4 furnaces / day, 30 days / month, and 75% equipment efficiency, the monthly output of a single production line is 18,000 pieces.

[0086] Example 2: Five pressure impregnation lines connected in parallel form a total production capacity of 100,000 pieces / month.

[0087] The single production line described in Example 1 can be replicated into five single lines, or five core melt pressure infiltration stations can be configured, sharing resources for powder gradation, post-processing, electroplating, and final inspection. The MES reads the equipment status and work-in-process quantity of each line and dynamically allocates queues for preforms, annealing furnaces, copper cladding furnaces, and tooth-shaving machines.

[0088] Example 3: AI Visual Quality Closed Loop

[0089] The machine vision camera at the powder spreading station identifies powder agglomeration and powder shortage areas. If the percentage of agglomerated pixels exceeds a set threshold, the system automatically increases the compaction time or requires re-spreading of powder. After demolding, visual identification is used to identify cracks and missing corners; after grinding, visual identification is used to identify surface scratches and edge chipping; and after tooth removal, visual identification is used to identify tooth height and tooth pitch deviations. After correlating the visual defect level with the final C-scan, thermal conductivity, and thermal resistance test results, the model learns the relationship between early visual features and final quality risks.

[0090] Example 4: Mass production of copper plating and edge banding processes

[0091] After double-sided grinding, qualified diamond copper core boards are vacuum brazed or diffusion-welded to upper and lower oxygen-free copper plates. After copper plating, interface C-scanning or random sampling is performed. The edges are activated and then chemically and electroplated with copper to thicken them, before grinding to the target edge thickness. Test results are written back to the database. If the interface voids in a particular batch of copper plating are too high, the system traces back the core board roughness, cleaning records, brazing temperature, holding time, and fixture pressure.

[0092] Example 5: AI Closed-Loop Vacuum Hot Press Sintering Route

[0093] Diamond particles with a tungsten-containing interface layer of 45–60 nm on their surface are mixed with copper powder at a target volume fraction. The mixture is then automatically spread, vibrated, and cold-pressed to form a sheet-like diamond-copper powder hybrid preform. One or more preforms are then loaded into an array-type graphite hot-pressing mold, which is equipped with a thickness limiting structure.

[0094] The vacuum hot-pressing sintering furnace is heated to 950–1000℃ under vacuum or high-purity argon protection, pressure of 40–70 MPa is applied, and the temperature is held for 20–60 minutes. An edge AI server identifies the densification stage based on pressure head displacement, displacement change rate, pressure holding curve, mold temperature difference, vacuum degree, platen parallelism, and acoustic emission events. When the displacement change rate is below a preset threshold, the pressure remains stable, the acoustic emission abnormality event is below the threshold, and the mold temperature difference meets the requirements, a sintering completion label is output, and the PLC enters the pressure relief and slow cooling stage. The resulting diamond copper core board then undergoes double-sided grinding, top and bottom copper plating, perimeter edging, and toothing or microchannel processing.

[0095] Example 6: AI-driven adaptive infiltration based on data linkage between upstream and downstream tungsten plating powders

[0096] The upstream dynamic coating equipment can use the method described in the prior application to output the equivalent film thickness t_eq=48nm, D50=150μm, D10 / D90 particle size range, QCM deposition record, powder dynamic image features, and FIB-SEM sampling results of the batch of diamond powder, and automatically synchronize them to the edge AI server. Based on the characteristic that the interface layer of this batch of powder is in the preferred window of 45-60nm, the edge AI server recommends a stress release annealing curve with a copper melt temperature of 1110℃, an impregnation pressure of 3.2MPa, a holding time of 12min, and a holding time of 500℃ for 60min.

[0097] Two hundred four-layer array graphite impregnation molds are loaded in layers, each with an independent flow channel and exhaust channel. During the impregnation stage, the pressure decay rate is 0.016 MPa / min, the air supply rate is 0.07 L / min, the temperature difference between the mold center and edge is 12℃, and the acoustic emission event is below the preset threshold. When these conditions are met, the AI ​​outputs an impregnation completion label, and the PLC controller switches to pressure holding solidification and stress relief annealing.

[0098] The counterexamples to the above embodiments are as follows: Comparative Example 1: Traditional single-piece fixed-duration air pressure infiltration It employs a single-cavity graphite mold, loading no more than 20 diamond preforms at a time, and lacks a layered, diverting venting structure. It also lacks industrial vision and acoustic emission sensors, relying on a fixed holding time to determine the impregnation endpoint. Furthermore, it does not integrate with diamond-coated tungsten powder data and lacks AI model control. During mass production, this can easily lead to issues such as inter-piece thermal conductivity dispersion, localized incomplete impregnation, and unstable interface layer reactions.

[0099] Comparative Example 2: Single-furnace multi-piece impregnation without AI

[0100] The system uses 100 single-layer molds, without multi-layer independent venting, and only collects furnace temperature and pressure signals, without visual, liquid level, or acoustic emission data acquisition; there is no process prediction model, and batch performance fluctuations cannot be automatically corrected. There is a significant difference in thermal conductivity between the center sheet and the corner sheets, and localized unimpregnated pore defects are prone to appear in the edge areas.

[0101] Comparative Example 3: Fixed-Time Vacuum Hot Press Sintering

[0102] Vacuum hot pressing sintering is performed at a fixed temperature, pressure, and holding time. Only the set pressure is recorded; indenter displacement, displacement change rate, plate parallelism, and acoustic emission events are not collected. This process makes it difficult to distinguish between particle rearrangement, rapid shrinkage, and the final stabilization stage, and is prone to insufficient or excessive holding pressure, localized uneven loading, or inconsistent thickness.

[0103] Comparative Example 4: A single high-density process simultaneously forms the core board, a thick top copper layer, a bottom copper layer, and copper cladding around the edges.

[0104] While this process is shorter, controlling the shrinkage of the thick copper layer at the top, the uniformity of the surrounding cladding, and the warping of the asymmetrical structure is challenging, and defects are difficult to repair once they occur. In contrast, this invention, through high-density core board formation, secondary copper plating, copper cladding around the edges, and segmented mass production with toothed edges, is more suitable for industrialization.

[0105] In summary, the weighing, mixing, powder spreading, vibration compaction, cold pressing, melt pressure impregnation, vacuum hot pressing sintering, vacuum brazing, electroplating, tooth scraping, machine vision, temperature, pressure and displacement sensing, PLC control, edge AI server and MES database involved in this invention can all be implemented by existing industrial equipment or customized equipment.

[0106] This invention allows for the initial construction of a vacuum hot-pressing sintering pilot plant to verify materials and processes, followed by the construction of a single pressure impregnation production line for verification at a rate of 18,000 to 20,000 pieces per month. Alternatively, both routes can be used in parallel for an extended period, depending on customer orders and product dimensions. Each process has a data interface and a quality write-back mechanism.

[0107] This invention is particularly suitable for ultra-high power GPU cold plates, AI server liquid cooling cold plates, power semiconductor heat sinks, laser heat sinks, and other diamond copper composite heat sinks that require high thermal conductivity, low thermal resistance, and high reliability for mass production.

[0108] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An AI-automated mass production method for diamond copper heat dissipation plates, characterized in that, Includes the following steps: S1: Obtain batch characteristic data of diamond particles with a tungsten-containing interface layer on the surface; S2: Based on batch feature data and target product parameters, an edge AI server generates a preparation scheme that includes at least powder gradation and process formulation; S3: According to the preparation plan, the diamond particles are prepared into sheet-like preforms; S4: The sheet preform is fed into the high-density forming unit for high-density forming to form a diamond copper core plate; S5: Post-process the diamond copper core board to form the finished diamond copper heat dissipation cold plate; The edge AI server collects sensor data from at least one step in the preparation method and outputs process control instructions through a multimodal process model; the PLC controller with hardware safety interlock receives the control instructions and executes them within a preset safety boundary to control at least some steps of the preparation method to run automatically.

2. The AI-automated mass production method for a diamond copper heat dissipation plate according to claim 1, characterized in that, S1 contains a tungsten-containing interface layer of 40–100 nm, which is a tungsten, tungsten / tungsten carbide, WC / W2C or a combination thereof interface layer. Batch characteristic data should include at least equivalent film thickness, particle size distribution, powder loading mass, interface layer batch number, and film thickness sampling results.

3. The AI-automated mass production method for diamond copper heat dissipation plates according to claim 1, characterized in that, The target product parameters in S2 include target thermal conductivity, board thickness, copper cladding structure, product size, and production line cycle time. The process formulations generated by the edge AI server include powder gradation, preform forming formulation, high densification forming formulation, annealing stress relief formulation, and post-treatment formulation.

4. The AI-automated mass production method for diamond copper heat dissipation plates according to claim 1, characterized in that, The specific content of S3 is as follows: According to the preparation plan, sheet-like preforms are formed by powder gradation and mixing, automatic powder spreading and compaction, and cold pressing. The sheet-type preform is either a diamond framework preform or a diamond-copper powder mixed preform.

5. The AI-automated mass production method for a diamond copper heat dissipation plate according to claim 1, characterized in that, Automatic powder spreading and compaction, as well as preform forming, use machine vision cameras to capture powder surface images and identify powder shortages, agglomerations, local accumulation, scraper marks, edge damage, and abnormal powder spreading height. When the percentage of agglomerated pixels, powder shortage areas, local accumulation areas, or thickness deviations exceed the threshold, the edge AI server outputs suggestions to the PLC controller for re-compacting, powder replenishment, reducing the thickness of a single layer of powder spreading, rework, or rejection.

6. The AI-automated mass production method for a diamond copper heat dissipation plate according to claim 1, characterized in that, The specific content of S4 is as follows: The sheet-like preform is fed into a high-density forming unit for high-density forming. The high-density forming unit is either a melt pressure infiltration furnace or a vacuum hot pressing sintering furnace. If a melt pressure infiltration furnace is used, diamond copper core plates are formed through preheating degassing, copper liquid pressure infiltration, pressure holding solidification, and slow cooling annealing. If a vacuum hot-pressing sintering furnace is used, diamond copper core plates are formed under vacuum or inert atmosphere through heating, pressurization, pressure holding, pressure release and slow cooling.

7. The AI-automated mass production method for a diamond copper heat dissipation plate according to claim 6, characterized in that, When high-density forming is achieved using melt pressure infiltration, the melt pressure infiltration furnace is configured to load 100 to 300 pieces per furnace; the cycle time per furnace is 5.0 to 8.0 hours. Under the condition of an overall equipment efficiency of 75% to 85%, a single production line can achieve a designed capacity of 18,000 to 20,000 pieces per month, and a total capacity of 100,000 pieces per month can be achieved by connecting multiple single lines in parallel. When the melt pressure infiltration furnace adopts a multi-layer graphite infiltration module, the multi-layer graphite infiltration module includes two 100-piece modules, one 200-piece module, or multiple equivalent plate position modules; each layer includes a graphite tray layer, a sheet diamond preform receiving cavity, a copper liquid diversion channel, an exhaust channel, an overflow channel, and a feeding channel, and has independent or semi-independent diversion, exhaust, overflow, and feeding paths.

8. The AI-automated mass production method for a diamond copper heat dissipation plate according to claim 7, characterized in that, When high-density forming is performed by melt pressure infiltration, the copper liquid temperature is 1080–1280℃, the infiltration pressure is 0.1–30MPa, and the holding time is 0.5–120 minutes. The edge AI server recommends the heating curve, pressure curve, and holding time based on the preform porosity, infiltration resistance index, mold layer, copper liquid temperature, and interface layer data. When high-density forming is vacuum hot-pressing sintering, diamond particles with a tungsten-containing interface layer on the surface are mixed with copper powder to form a sheet-like diamond-copper powder mixed preform, which is then hot-pressed and sintered under vacuum or inert atmosphere; the sintering temperature is 850-1050℃, the sintering pressure is 20-100MPa, and the holding time is 10-90 minutes.

9. The AI-automated mass production method for a diamond copper heat dissipation plate according to claim 1, characterized in that, The edge AI server reads the equivalent film thickness, QCM integration record, powder visual state, particle size distribution, and film thickness sampling data output from the upstream diamond dynamic coating equipment or process database; when the equivalent film thickness is 45-60nm, it matches the forming parameters with a lower risk of interface reaction; when the equivalent film thickness is 70-100nm, it adjusts the preheating, forming, and annealing process windows. Sensor data includes at least one of machine vision, temperature, pressure, vacuum, flow rate, liquid level, acoustic emission, displacement, weighing, platen parallelism, and quality inspection data; The multimodal process model includes at least the following sub-models: preform porosity estimation, infiltration resistance prediction, copper liquid filling state identification, hot-pressing shrinkage stage identification, solidification feeding prediction, warpage risk prediction, post-processing quality prediction, and production line cycle time scheduling. The multimodal process model outputs process routes, formula suggestions, target curves, alarm levels, wafer location quality risks, and scheduling strategies. The PLC controller operates within the hardware safety range of temperature, pressure, vacuum, pressure head displacement, cooling water, furnace door, valve group, and emergency stop interlock. The edge AI server generates an impregnation completion score based on the pressure decay rate, gas supply flow rate, valve position change, copper liquid level drop, mold temperature gradient, exhaust channel temperature rise, and acoustic emission events. When the pressure decay rate, gas supply volume per unit time, mold temperature difference, liquid level change rate, and acoustic emission events meet the preset endpoint conditions, it outputs an impregnation completion label and performs temperature compensation on the pressure and gas supply volume based on the effective gas volume of the furnace cavity, gas temperature, and gas state equation. The edge AI server identifies the densification stage and densification endpoint based on the pressure head, displacement, displacement change rate, sintering shrinkage curve, temperature difference between the mold center and edge, vacuum degree, platen parallelism, and acoustic emission events. It then outputs suggestions for heating, pressurizing, holding, depressurizing, and slow cooling curves to the PLC controller. When the displacement change rate is below a preset threshold, the pressure remains stable, the acoustic emission abnormality events are below a preset threshold, and the mold temperature difference and parallelism meet the requirements, a sintering completion label is output. After high-density forming, staged pressure relief and controlled annealing and slow cooling are performed; in the melt pressure infiltration route, after pressure holding and solidification, stress relief and heat preservation steps are performed at 300-700℃ for 5-120 minutes; in the vacuum hot pressing sintering route, after sintering and pressure holding, pressure is released according to a preset gradient, and annealing and slow cooling are performed in a vacuum or inert atmosphere.

10. An AI-automated mass production line, characterized in that, An AI-automated mass production method for manufacturing a diamond copper heat sink as described in any one of claims 1 to 9 includes: Powder gradation and mixing unit, automatic powder spreading and vibration and preform forming unit, high densification forming unit, annealing slow cooling furnace, double-sided grinding and inspection unit, upper and lower copper plating unit, four-sided copper edging unit, shovel tooth or microchannel processing unit, final inspection and sorting unit, edge AI server, PLC controller, MES or process database connected to automatic loading and unloading or transfer mechanism. The high-density forming unit includes a melt pressure infiltration furnace, a vacuum hot pressing sintering furnace or a combination thereof; the melt pressure infiltration furnace is used in conjunction with a preheating degassing furnace and a multi-layer graphite infiltration module, and the vacuum hot pressing sintering furnace is used in conjunction with an upper pressure head, a lower pressure head, a graphite hot pressing mold, a thickness limiting structure, a displacement sensor and a pressure plate parallelism detection unit. The edge AI server includes a GPU or AI accelerator card, an industrial data acquisition interface, an image inference module, a timing signal inference module, a route recognition module, a process database, a model version management module, and an industrial Ethernet interface for communicating with the PLC controller. The PLC controller connects to the heating power supply, vacuum valve group, inert gas valve group, proportional pressure valve, pressure head loading mechanism, displacement limit mechanism, gas replenishment valve, cooling medium valve, furnace door interlock, automatic transfer mechanism, alarm, and emergency stop circuit. The hardware safety interlock priority of the PLC controller is higher than the AI ​​recommended parameters.

Images