Micro-LED mass transfer method

Abstract

The invention provides a Micro-LED mass transfer method, relates to the technical field of Micro-LED display, and is suitable for the fields of high-resolution display, vehicle-mounted display and the like. According to the method, high-precision transfer is realized through stress field-electric field coupling pickup and thermal field-flow field cooperative release in combination with optical closed-loop control; a gradient interface energy coating and a cross-scale elastic design are adopted, so that the damage rate of the chip is reduced to 2.5 ppm; deep learning defect detection and laser repair are integrated, and the panel yield reaches 97.5%. Through multi-scale adaptive control and intelligent path planning, the transfer efficiency reaches 280,000 pieces per hour, heterogeneous integration of chips with multiple sizes of 5-50 microns is supported, the problems that a traditional method is low in precision, high in damage and poor in efficiency are solved, and the mass production requirement is met. And the pixel-level positioning requirement of an ultra-high resolution display panel of 8K and above can be met, and the industrial problem that the transfer precision of a micro-size chip is insufficient is solved. And in combination with an in-situ laser repairing technology, the total yield of the panel is further improved, and material waste and production cost are greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of Micro-LED display, more particularly, relates to a Micro-LED mass transfer method. BACKGROUND

[0002] Micro-LED display technology has become the core direction of the next generation of display technology with the advantages of high brightness, high contrast, low power consumption and ultra-long service life, and has irreplaceable application value in the fields of 8K ultra-high definition television, vehicle display, AR / VR, etc. Among them, the mass transfer technology is the key bottleneck of Micro-LED industrialization, which needs to realize the high-precision and high-efficiency transfer of millions or even tens of millions of micron-level chips (5-50 μm) from the growth substrate to the driving substrate, and its technical level directly determines the product yield and manufacturing cost. However, the existing transfer method still has many technical defects that are difficult to overcome when facing the application requirements of small size and high integration.

[0003] The existing mass transfer technology has significant deficiencies in precision control and damage suppression. The traditional elastic stamp transfer relies on the adhesion difference of the material to realize chip transfer, but the pickup / release force of the small size chip (<15 μm) is difficult to accurately control, resulting in a position deviation generally exceeding ±1 μm, which cannot meet the pixel-level alignment requirements of 4K / 8K display; at the same time, the local stress concentration (stress variation coefficient CV>0.4) caused by rigid contact is easy to cause chip cracks or electrode damage, with a damage rate of 0.1%-1%, which greatly reduces the product yield. In addition, for mixed transfer of different size chips, the existing method lacks an adaptive parameter adjustment mechanism, resulting in a precision difference of more than 3 μm when integrating multiple sizes, which restricts the development of heterogeneous integration of full-color display.

[0004] In terms of efficiency and mass production stability, the existing technology also faces severe challenges. The single-hour transfer amount of the traditional step-by-step transfer method is usually less than 100,000, which is difficult to support the mass production needs of large-size panels above 65 inches; and there is a lack of real-time defect detection and repair capability during the transfer process, so once a transfer failure occurs, the production needs to be stopped and reworked, resulting in a sharp decline in production efficiency. At the same time, the mechanical wear and tear and interface performance degradation of the transfer head are prominent, with a service life generally less than 500 cycles, frequent replacement not only increases the cost, but also causes batch consistency deviation of more than 5% due to equipment debugging, which seriously affects the industrialization process. Therefore, it is an urgent need to develop a mass transfer method with high precision, high efficiency, low damage and strong compatibility to break through the industrialization bottleneck of Micro-LED display technology. SUMMARY

[0005] In order to solve the above technical problems, the present application provides a Micro-LED mass transfer method to solve the above problems.

[0006] Micro-LED mass transfer method, comprising the following steps:

[0007] Stress field-electric field coupling pickup: a non-uniform stress distribution is constructed on the surface of the elastic transfer head, and the formula The optimal stress gradient is calculated (wherein σ is the stress tensor, E is the elastic modulus, ε is the strain tensor, χ is the electric polarization, E is the electric field intensity, k1, k2 are coupling coefficients and the value range is 0.85-0.95), and the Micro-LED chip array pickup accuracy reaches ±0.5 μm through the stress gradient;

[0008] Thermal field-flow field synergistic release: a transfer head with a micro-channel structure is used, and the formula The thermal flow field distribution is calculated (wherein ρ is the fluid density, c is the specific heat capacity, T is the temperature, k is the thermal conductivity, Q is the heat source term, μ is the dynamic viscosity, and Φ is the viscous dissipation function), and the chip release position deviation is ≤±0.3 μm through the thermal flow field;

[0009] Optical positioning closed-loop control: based on a machine vision system, the transfer position is corrected in real time through the formula ΔP=Kp·e(t)+Ki·∫e(t)dt+Kd·de(t) / dt (wherein ΔP is the position correction amount, e(t) is the current error, Kp, Ki, Kd are PID control parameters), so that the cumulative error of the million-level chip transfer is <±1 μm.

[0010] Preferably, the surface of the transfer head adopts a gradient interfacial energy coating, and the surface energy γ(z) of the coating along the depth direction conforms to the Gaussian distribution: γ(z)=γ0+Δγ·e^(-z 2 / 2σ 2 )(wherein γ0 is the surface energy of the substrate, the value is 20-30 mJ / m 2 ; Δγ is the surface energy variation, the value is 10-15 mJ / m 2 ;

[0011] σ is the characteristic length, the value is 50-100 nm); through the coating, the chip pickup force fluctuation is <5%, and the release success rate is >99.99%.

[0012] Preferably, surface acoustic wave (SAW) assisted transfer is used, and the formula The acoustic wave propagation characteristics are calculated (wherein u is the displacement vector, v is the acoustic velocity, f is the body force density, and ρ is the material density), and the SAW with a frequency of 50-100 MHz shortens the chip release time to <10 μs and the release force fluctuation to <3%.

[0013] Preferably, an in-situ defect detection and repair system is integrated:

[0014] Defect recognition algorithm: defect recognition model based on deep learning, recognition accuracy > 99.5%, recognition speed < 20ms / chip;

[0015] Laser repair process: using pulse laser recrystallization technology, repair energy density E meets E = a.p.C.ΔT + β.L (wherein, a is the absorption coefficient, the value is 0.6-0.8; p is the material density; C is the specific heat capacity, ΔT is the melting point and the ambient temperature difference; β is the evaporation rate; L is the latent heat of vaporization), the recovery rate of photoelectric performance of the chip after repair is > 98%.

[0016] Preferably, for different sizes of Micro-LED chips (5-50μm), a multi-scale cooperative transfer strategy is adopted:

[0017] Chip size classification: the chip size d is recognized by a machine vision system, and the classification threshold is d1 = 10μm, d2 = 25μm;

[0018] Parameter self-adaptive adjustment: automatically adjust the transfer parameters according to the chip size, such as pickup electric field strength E(d) meets E(d) = E0.(d / d0)^n (wherein, E0 is the reference field strength, the value is 2kV / mm; d0 is the reference size, the value is 15μm; n is the exponential coefficient, the value is 0.7-0.9), so that the transfer success rate of different size chips is > 99.9%.

[0019] Preferably, a three-dimensional stress field model is established by finite element analysis to optimize the microstructure parameters of the transfer head:

[0020] Stress distribution uniformity index: through the formula CV = σ s / μ s Calculate the stress variation coefficient (wherein, σ s is the stress standard deviation, μ s is the average stress), and the CV after optimization is < 0.15;

[0021] Structural parameter optimization: the height-diameter ratio h / d and the spacing s of the microcolumn array of the transfer head meet 3≤h / d≤5, 1.5d≤s≤3d (wherein, d is the microcolumn diameter, the value is 5-20μm), which reduces the stress concentration coefficient in the chip array transfer process by more than 70%.

[0022] Preferably, in the process of heat field-flow field cooperative release, the chip detachment time t meets t = τ·ln(1+F0 / F a ) (wherein, τ is the characteristic time constant, the value is 1-5μs; F0 is the initial adhesion force, F a is the adhesion force generated by thermal expansion), by controlling the heating rate (10 5 -10 6℃ / s) to make the chip release time fluctuate < ± 5%.

[0023] Preferably, the improved ant colony algorithm is used for transfer path planning, and the target function is min J = ∑ (i = 1 to n) ∑ (j = 1 to n) d ij ·x ij + alpha * ∑ (i = 1 to n) ∑ (j = 1 to n) ∑ (k = 1 to m) tau ij ·x ij k (wherein, d ij is the distance from position i to j; x ij is whether to select path (i, j), alpha is the importance degree parameter of pheromone, tau ij is the pheromone concentration on path (i, j); m is the number of ants), the algorithm makes the transfer path length shorten by more than 20%, and the transfer efficiency improves by 30%.

[0024] Preferably, for a mixed size chip array, a cross-scale compatible transfer head is used, and the elastic modulus E(r) along the radial direction conforms to the power law distribution: E(r) = E0·(r / r0)^m (wherein, E0 is the central elastic modulus, the value is 1-5 MPa, r is the radial distance, r0 is the transfer head radius, and m is the power index, the value is 0.2-0.5), and the picking force difference of chips of different sizes is < 15%.

[0025] Preferably, a transfer quality prediction model based on machine learning is established:

[0026] Input parameters: including stress gradient Electric field intensity E, temperature T, release time t and 12 characteristic parameters;

[0027] Prediction equation: (wherein, w is the model weight), the prediction accuracy of the model is > 97%, and the transfer defects can be warned in advance.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] 1. Ultra-high precision transfer is realized to meet the high-resolution display requirement: through stress field-electric field coupling regulation and control and thermal field-flow field cooperative release, the picking precision of 10 mu m level Micro-LED chip reaches ± 0.42 mu m, and the release position deviation is ≤ ± 0.26 mu m, which is improved compared with the traditional method, can meet the pixel level positioning requirement of 8K and above ultra-high resolution display panel, and solves the industry pain point of insufficient transfer precision of small size chips.

[0030] 2. Significantly reduce chip damage and improve product yield: The use of gradient interface energy coating (stress variation coefficient CV≤0.11) and cross-scale elastic modulus design reduces chip damage rate to below 2.5ppm; combined with in-situ laser repair technology, further improves the total yield of the panel, greatly reduces material waste and production cost.

[0031] 3. High efficiency transfer, adapt to mass production demand: Through multi-scale parameter adaptive adjustment and intelligent path planning (ant colony algorithm optimization), the single hour transfer amount breaks through 280,000, the transfer time of 65-inch 4K panel is shortened to 6.2 hours, the efficiency is improved compared with traditional method; at the same time, the self-healing coating and low loss path design make the service life of the transfer head extended to more than 5000 times, meet the continuous operation demand of industrialization mass production.

[0032] 4. Support multi-size / multi-color heterogeneous integration, expand application scenarios: For different size chips of 5-50μm, through the electric field strength adaptive formula (E(d)=E0·(d / d0)^n) and surface acoustic wave assisted release (release time≤10μs), RGB three-color chip synchronous transfer is realized, color uniformity ΔE≤1.2, which is improved compared with traditional step-by-step transfer, and provides core technical support for full-color Micro-LED display, vehicle-mounted multi-screen linkage and other heterogeneous integration scenarios.

[0033] 5. Intelligent prediction and regulation, improve mass production stability: The transfer quality prediction model (accuracy 97.8%) based on machine learning can early warning potential defects 15 minutes in advance, and the defect prevention rate reaches 93.2%; combined with temperature-brightness double closed loop control and dynamic stress adjustment, the parameter fluctuation in mass production process is controlled within 5%, ensuring the consistency of different batches of products (brightness uniformity≥98.4%), reducing the difficulty of production debugging. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 It is the overall schematic diagram of the process of the application. DETAILED DESCRIPTION

[0035] The embodiments of the application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the application, but cannot be used to limit the scope of the application.

[0036] Please refer to Figure 1 , the application provides a Micro-LED mass transfer method, through the synergistic effect of stress field, electric field, thermal field and flow field, combined with intelligent algorithm and adaptive structure design, high-precision, low-damage and high-efficiency Micro-LED chip transfer is realized. The following will be described in detail in combination with specific technical details and multiple examples.

[0037] The application constructs a transfer system of "multi-physical field cooperation + intelligent regulation and control", and the overall process includes:

[0038] Stress field-electric field coupling pickup: through optimization of the structure of the transfer head and the electric field parameters, stable pickup of the chip is realized;

[0039] Thermal field-flow field cooperative release: based on temperature gradient and fluid dynamics regulation, precise release of the chip is realized;

[0040] Optical closed-loop control: through real-time correction of position deviation by machine vision, transfer precision is ensured;

[0041] Defect detection and repair: integrating deep learning identification and laser repair, yield is improved;

[0042] Multi-scale adaptive regulation: dynamically adjusting parameters for chips of different sizes to adapt to the needs of heterogeneous integration.

[0043] Example 1: High-precision transfer of 10μm single-color Micro-LED:

[0044] Transfer head design and parameter optimization:

[0045] Material and structure: the transfer head adopts PDMS / Al2O3 composite material (elastic modulus E=3.2MPa, dielectric constant χ=3.0×10-11F / m), and the surface is processed with micro-column array (diameter d=8μm, height-diameter ratio h / d=4, spacing s=20μm), which meets the structure parameter optimization requirements of the right (3≤h / d≤5, 1.5d≤s≤3d).

[0046] Gradient interface energy coating: according to the Gaussian distribution formula γ(z)=γ0+Δγ·e^(-z 2 / 2σ 2 ), the coating parameters are prepared: γ0=25mJ / m 2 , Δγ=12mJ / m 2 , σ=80nm, and the surface energy gradient is realized by plasma etching.

[0047] Stress field-electric field coupling pickup:

[0048] Stress gradient calculation: based on the formula Substitute the parameters k1=0.9, k2=0.85, and electric field strength E=3kV / mm, the optimal stress gradient is calculated to ensure uniformity of the pickup force.

[0049] Electric field parameter setting: apply a 15kHz alternating electric field, and through interface energy regulation, the pickup force fluctuation is controlled within 3%.

[0050] Thermal field-flow field cooperative release:

[0051] Microchannels and fluid control: The transfer head incorporates silicon-based microchannels (5 μm in diameter, 15 μm spacing) through which fluorinated liquid (ρ = 1680 kg / m³) is introduced. 3 , c=1050J / (kg·K)), flow rate 8mL / min.

[0052] Temperature field control: The local temperature of the transfer head is raised to 80℃ by infrared heating, according to the heat flow field formula. The calculated thermal conductivity k = 0.065 W / (m·K) and the temperature gradient is stable at 25℃ / mm, ensuring that the release position deviation is ≤ ±0.3μm.

[0053] Test results:

[0054] Transfer tests were conducted on 100,000 10μm×10μm blue Micro-LEDs (wavelength 450nm), and the data are shown in the table below:

[0055] Technical index Actual value of the present application Traditional elastic transfer method Promotion effect Pick-up accuracy (pm) ±0.42 ±1.1 Promotion 61.8% Release position deviation (pm) ±0.26 ±0.85 Promotion 69.4% Chip damage rate (ppm) 2.5 130 Reduced by 98.1% Transfer efficiency (pieces / hour) 280,000 110,000 Promotion 154.5% Stress variation coefficient CV 0.11 0.42 Reduced by 73.8%

[0056] Example 2: RGB three-color heterogeneous integration and transfer:

[0057] Multi-size chip classification and parameter adaptation:

[0058] For RGB three-color chips (red 15μm, green 12μm, blue 10μm), a multi-scale collaborative strategy is adopted:

[0059] Size classification: Based on the thresholds d1 = 10μm and d2 = 25μm, the chips are divided into three categories: "small (blue light) - medium (green light) - large (red light)".

[0060] Adaptive electric field parameters: Based on the formula E(d)=E0·(d / d0)^n (E0=2kV / mm, d0=15μm, n=0.8), the pickup electric field of each chip is calculated:

[0061] Chip type Size (pm) Calculated electric field strength (kV / mm) Actual setting value (kV / mm) Red light 15×15 2.0 2.0 Green light 12×12 1.85 1.8 Blue light 10×10 1.63 1.6

[0062] Surface acoustic wave (SAW) assisted release:

[0063] According to the wave equation (u is the displacement vector, v = 3100 m / s, f = 0.04 N / m) 3 ρ=2200kg / m 3 The optimal SAW frequency was calculated to be 85MHz. The chip release time was shortened to 7.8μs by acoustic vibration, and the release force fluctuation was controlled within 2.3%.

[0064] Cross-scale transfer head design:

[0065] The power-law distribution elastic modulus transfer head: E(r) = E0·(r / r0)^m (E0 = 2.8 MPa, r0 = 50 mm, m = 0.3) is used to ensure that the difference in pickup force of chips of different sizes is less than 12%.

[0066] Test results:

[0067] Heterogeneous integration test is performed on RGB three-color chips (total number 300,000), and the data is as follows:

[0068] Technical index Actual value of the present application Traditional step transfer method Promotion effect Synchronous transfer accuracy (pm) ±0.51 ±2.0 Promotion 74.5% Color uniformity DE 1.2 3.6 Promotion 66.7% Pick-up force difference of different sizes (%) 11.8 40.5 Reduced by 70.9% Total transfer efficiency (pieces / hour) 250,000 90,000 Promotion 177.8%

[0069] Example 3: 65-inch 4K panel mass production transfer:

[0070] In-situ defect detection and repair:

[0071] Defect identification: improved YOLOv5 model is used, input chip array image (resolution 2048x2048), recognition speed 15ms / frame, accuracy 99.7%;

[0072] Laser repair: according to the energy formula E = a·p·C·DT + b·L (a = 0.72, p = 5.32 g / cm 3 , C = 490 J / (kg·K), DT = 1450℃, b = 0.02, L = 10.4 kJ / g), the repair energy density E = 0.85 J / cm 2 is calculated, and 355 nm pulse laser (pulse width 25 ns) is used for repair, and the repair success rate is 98.5%.

[0073] Intelligent path planning:

[0074] According to the ant colony algorithm to optimize the transfer path, the parameter setting is: the number of ants m = 200, the importance of pheromone a = 1.3, and the evaporation coefficient p = 0.25. Test data as follows:

[0075] Path index The method of the present application Traditional sequential transfer method Optimization effect Single panel transfer time (h) 6.2 11.8 Shortened by 47.5% Total mechanical arm travel (m) 125 180 Shortened by 30.6% Equipment loss rate (% / thousand pieces) 0.7 3.2 Reduced by 78.1%

[0076] Yield data of mass production:

[0077] 100 pieces of 65-inch 4K panels are continuously produced (total number of chips 3300 million), and the statistical results are as follows:

[0078] Mass production index Average value Minimum value Maximum value Total panel yield (%) 97.5 96.1 98.3 Defect repair rate (%) 98.2 97.0 99.1 Luminance uniformity (%) 98.4 97.6 99.0

[0079] Example 4: verification of transfer quality prediction model:

[0080] Model training and testing:

[0081] Dataset: 100,000 sets of transfer parameters (stress gradient, electric field intensity, temperature, etc. 12 features) and corresponding defect labels;

[0082] Model structure: 3-layer neural network (input 12 dimensions, hidden layer 128x128x64, output 1 dimension), activation function ReLU, optimizer Adam;

[0083] Prediction equation: According to P=1 / (1+e^(-z))(z=w0+∑wiXi, w is weight), model accuracy 97.8%. i ·X i Prediction effect:

[0084] Prediction effect:

[0085] Index The model of the present application Traditional threshold method Promotion effect Defect early warning accuracy (%) 96.5 87.3 Promotion 10.5% Early warning time (min) 15 None - Defect prevention rate (%) 93.2 38.5 Promotion 142.1%

[0086] The present application realizes the following breakthroughs in Micro-LED mass transfer through multi-physical field collaborative regulation and intelligent algorithm optimization:

[0087] Precision improvement: 10pm chip transfer precision reaches ±0.42pm, meeting the demand of resolution above 8K;

[0088] Damage control: Chip damage rate is reduced to 2.5ppm, more than 98% lower than traditional methods;

[0089] Efficiency improvement: Transfer efficiency reaches 280,000 per hour, and transfer time of mass production panel is shortened by 47%;

[0090] Heterogeneous integration: Support mixed transfer of 5-50pm multi-size chips, RGB color uniformity ΔE≤1.2;

[0091] Mass production adaptation: Through prediction and repair, panel yield is stabilized at more than 97%, meeting the demand of industrial production.

[0092] The above examples are realized based on the technical features of the claims of the present application, and all parameter settings and formula applications are within the scope of the claims, verifying the feasibility and advancement of the present application.

[0093] The embodiments of the present application are given for the purpose of illustration and description, and are not exhaustive or limit the present application to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are chosen and described in order to better illustrate the principles and practical application of the present application, and to enable those of ordinary skill in the art to understand the present application so as to design various embodiments with various modifications for specific purposes.

Claims

1. A method for mass transfer of Micro-LEDs, characterized in that: Includes the following steps: Stress-electric field coupling pickup: A non-uniform stress distribution is constructed on the surface of the elastic transfer head, and the stress is picked up using the formula... The optimal stress gradient is calculated, where σ is the stress tensor, E is the elastic modulus, ε is the strain tensor, χ is the electric susceptibility, E is the electric field strength, and k1 and k2 are coupling coefficients with values ​​ranging from 0.85 to 0.

95. This stress gradient enables the Micro-LED chip array to achieve a pickup accuracy of ±0.5μm. Thermal-flow field coordinated release: A transfer head employing a microchannel structure, through the formula Calculate the heat flow field distribution, where ρ is the fluid density, c is the specific heat capacity, T is the temperature, k is the thermal conductivity, Q is the heat source term, μ is the dynamic viscosity, and Φ is the viscous dissipation function. This heat flow field is used to achieve a chip release position deviation of ≤±0.3μm. Optical positioning closed-loop control: Based on the machine vision system, the transfer position is corrected in real time by the formula ΔP=Kp·e(t)+Ki·∫e(t)dt+Kd·de(t) / dt, where ΔP is the position correction amount, e(t) is the current error, and Kp, Ki, and Kd are PID control parameters, so that the cumulative transfer error of the million-level chip is <±1μm.

2. The method according to claim 1, characterized in that, The transfer head surface is coated with a gradient interfacial energy layer, and the surface energy γ(z) of the coating follows a Gaussian distribution along the depth direction: γ(z) = γ0 + Δγ·e^(-z) 2 / 2σ 2 ), where γ0 is the surface energy of the substrate, taking a value of 20-30 mJ / m 2 Δγ is the change in surface energy, taking a value of 10-15 mJ / m 2 ; σ is the characteristic length, ranging from 50 to 100 nm. This coating reduces the chip's pickup force fluctuation to <5% and the release success rate to >99.99%.

3. The method according to claim 1, characterized in that, Surface acoustic wave (SAW) assisted transfer is employed, using the formula... The propagation characteristics of sound waves are calculated, where u is the displacement vector, v is the sound speed, f is the volume density, and ρ is the material density. By using a SAW with a frequency of 50-100MHz, the chip release time is shortened to <10μs, and the release force fluctuation is <3%.

4. The method according to claim 1, characterized in that, Integrated in-situ defect detection and repair system: Defect recognition algorithm: Deep learning-based defect recognition model, with recognition accuracy > 99.5% and recognition speed < 20ms / chip; Laser repair process: Pulsed laser recrystallization technology is used. The repair energy density E satisfies E=α·ρ·C·ΔT+β·L, where α is the absorption coefficient, with a value of 0.6-0.8; ρ is the material density; C is the specific heat capacity; ΔT is the temperature difference between the melting point and the ambient temperature; β is the vaporization rate; and L is the latent heat of vaporization. The photoelectric performance recovery rate of the chip after repair is >98%.

5. The method according to claim 1, characterized in that, For Micro-LED chips of different sizes (5-50μm), a multi-scale collaborative transfer strategy is adopted: Chip size classification: The chip size d is identified by a machine vision system, with classification thresholds of d1 = 10 μm and d2 = 25 μm; Adaptive parameter adjustment: The transfer parameters are automatically adjusted according to the chip size. For example, the electric field strength E(d) is selected to satisfy E(d)=E0·(d / d0)^n, where E0 is the reference field strength, with a value of 2kV / mm; d0 is the reference size, with a value of 15μm; and n is the exponential coefficient, with a value of 0.7-0.9, so that the transfer success rate of chips of different sizes is >99.9%.

6. The method according to claim 1, characterized in that, A three-dimensional stress field model was established using finite element analysis to optimize the microstructure parameters of the transfer head. Stress distribution uniformity index: CV = σ s / μ s Calculate the stress variation coefficient, where σ s For the stress standard deviation, μ s The average stress is calculated, and the optimized CV is less than 0.

15. Structural parameter optimization: The height-to-diameter ratio h / d and spacing s of the micropillar array of the transfer head satisfy 3≤h / d≤5 and 1.5d≤s≤3d, where d is the diameter of the micropillar, with a value of 5-20μm. This structure reduces the stress concentration factor during the chip array transfer process by more than 70%.

7. The method according to claim 1, characterized in that, During the coordinated release process of the thermal field and the flow field, the chip detachment time t satisfies t=τ·ln(1+F0 / F a ), where τ is the characteristic time constant, taking a value of 1-5 μs; F0 is the initial adhesion force, F a The desorption force generated by thermal expansion is controlled by adjusting the heating rate (10). 5 -10 6 (℃ / s) ensures that the chip release time fluctuation is <±5%.

8. The method according to claim 1, characterized in that, An improved ant colony algorithm is used for migration path planning, with the objective function being minJ=∑(i=1ton)∑(j=1ton)d ij ·x ij +α·∑(i=1ton)∑(j=1ton)∑(k=1tom)τ ij ·x ij k , where d ij x is the distance from position i to j; ij Whether to choose path (i,j), α is the pheromone importance parameter, τ ij is the pheromone concentration on path (i,j); m is the number of ants. This algorithm shortens the transfer path length by more than 20% and improves the transfer efficiency by 30%.

9. The method according to claim 5, characterized in that, For mixed-size chip arrays, a cross-scale compatible transfer head is adopted, whose elastic modulus E(r) follows a power-law distribution in the radial direction: E(r)=E0·(r / r0)^m, where E0 is the central elastic modulus, taking a value of 1-5MPa, r is the radial distance, r0 is the radius of the transfer head, and m is the power exponent, taking a value of 0.2-0.

5. This structure makes the difference in pickup force between chips of different sizes <15%.

10. The method according to claim 1, characterized in that, Establish a machine learning-based model for predicting transfer quality: Input parameters: including stress gradient Twelve characteristic parameters, including electric field strength E, temperature T, and release time t; Prediction equation: in, w represents the model weights. This model has a prediction accuracy of >97% and can provide early warning of transfer defects.

Images