Multi-probe collaborative contact control and adjustment method and system for wafer-level testing

Abstract

This invention provides a multi-probe coordinated contact control and adjustment method and system for wafer-level testing. This system involves generating a trajectory planning scheme by acquiring initial and target probe position information, using a multi-dimensional sensor array to collect probe data and establish an evaluation model, constructing an optimization objective function, and using a hierarchical progressive predictive control algorithm to optimize the trajectory to obtain subgroup control parameters, achieve precise probe positioning, and trigger a parameter optimization cycle based on the evaluation indicators. This invention effectively improves the precision of multi-probe coordinated motion and contact reliability.

Description

Technical Field

[0001] The present invention relates to wafer-level testing technology, and in particular to a multi-probe coordinated contact control and adjustment method and system for wafer-level testing. Background Art

[0002] Wafer-level testing typically utilizes a multi-probe test system, where multiple probes simultaneously contact test points on the wafer to perform electrical performance testing. During testing, precise positioning and coordinated contact of the probes are critical factors for ensuring test accuracy and reliability. Traditional multi-probe test systems primarily utilize a single control scheme, controlling the movement of each probe individually. However, with the continuous reduction in integrated circuit feature sizes and the increase in test density, higher requirements are placed on the accuracy and consistency of the coordinated contact of the probe groups.

[0003] The existing multi-probe testing technology has the following major defects: First, there is a lack of an effective mechanism for evaluating coupling interference between probes. When multiple probes move at the same time, the deformation and vibration of the mechanical structure will cause mutual interference between the probes, affecting the accuracy and consistency of the probe contact, and thus leading to deviations in the test results. Secondly, the existing technology generally adopts a static preset scheme for probe trajectory planning, which cannot be dynamically adjusted according to the real-time test environment and probe status. When the test environment changes, it is difficult to ensure the stability and reliability of the test. Finally, there is a lack of a real-time monitoring and evaluation mechanism for the probe contact status during the test process, and contact anomalies cannot be discovered and corrected in a timely manner, resulting in low test efficiency and easy damage to the wafer or probe. Summary of the Invention

[0004] The embodiments of the present invention provide a multi-probe coordinated contact control and adjustment method and system for wafer-level testing, which can solve the problems in the prior art.

[0005] A first aspect of an embodiment of the present invention provides a multi-probe coordinated contact control and adjustment method for wafer-level testing, comprising:

[0006] Obtain the initial position information of each probe in the multi-probe test system and the target test point position information, and generate the initial probe displacement trajectory planning scheme;

[0007] A multi-dimensional heterogeneous sensor array is used to collect probe motion characteristic data in real time, and a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector is established based on the probe motion characteristic data;

[0008] Generate a probe group motion optimization objective function based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model and the initial probe displacement trajectory planning scheme;

[0009] Based on the probe group motion optimization objective function, a hierarchical progressive predictive control algorithm is used to optimize the trajectory and obtain subgroup control parameters; according to the subgroup control parameters, single probe precise positioning control is achieved at the micro level, and the prediction model of each layer is dynamically optimized through a sliding time domain window mechanism;

[0010] The evaluation index of the current contact state is calculated according to the probe group motion state evaluation model, and the evaluation index is compared with the preset evaluation threshold. When it exceeds the preset evaluation threshold range, the evaluation result is fed back to the probe group motion state evaluation model to trigger the parameter optimization cycle.

[0011] Based on the probe motion characteristic data, establishing a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector includes:

[0012] Based on the probe motion characteristic data, the inter-probe force is obtained by calculating a weighted sum of a ratio of a displacement difference to a square of a distance multiplied by a first coupling coefficient and a ratio of a relative speed to a distance multiplied by a second coupling coefficient;

[0013] Constructing a coupling interference matrix based on the inter-probe force, wherein the diagonal elements of the coupling interference matrix represent the characteristics of the probes themselves, and the non-diagonal elements represent the coupling interference coefficients between the probes;

[0014] Obtain motion phase data of each probe, subtract the average phase of the probe group from the motion phase data to obtain a phase deviation, perform complex exponential mapping on the phase deviation and sum and average it within a time window to obtain a phase synchronization evaluation index;

[0015] Converting the phase synchronization evaluation index into a synchronization deviation vector through nonlinear mapping, wherein each component of the synchronization deviation vector represents the degree of synchronization of the corresponding probe;

[0016] The eigenvalues ​​of the coupling interference matrix and the synchronization deviation vector are fused using a weighted average method to generate a fusion feature that characterizes the overall motion characteristics of the probe group;

[0017] Based on the fusion features, a coupling evaluation component characterizing the coupling strength, a synchronization evaluation component characterizing the degree of synchronization, and an accuracy evaluation component characterizing the position accuracy are calculated respectively. The coupling evaluation component, the synchronization evaluation component, and the accuracy evaluation component are weighted by an adjustable weight coefficient to obtain a comprehensive evaluation index. According to the comprehensive evaluation index, a probe group motion state evaluation model is constructed.

[0018] Based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model, combined with the initial probe displacement trajectory planning scheme, the probe group motion optimization objective function is generated, including:

[0019] According to the coupling interference matrix and the synchronization deviation vector output by the probe group motion state evaluation model, a coupling optimization term and a synchronization optimization term are calculated respectively, wherein the coupling optimization term is obtained by setting a weight coefficient on the coupling interference matrix and summing the results, and the synchronization optimization term is obtained by performing a square operation on the synchronization deviation vector and weighting the result;

[0020] Calculating trajectory curvature continuity based on the initial probe displacement trajectory planning scheme, multiplying the trajectory curvature continuity with the trajectory smoothness weight coefficient and integrating the result in the motion time domain to obtain a trajectory optimization term;

[0021] The coupling optimization item, the synchronization optimization item and the trajectory optimization item are linearly combined to generate a probe group motion optimization objective function.

[0022] Based on the probe group motion optimization objective function, a hierarchical progressive predictive control algorithm is used for trajectory optimization, and the subgroup control parameters obtained include:

[0023] Dividing the control time domain into multiple time domain levels, generating corresponding optimization objectives and constraints according to the control requirements of each time domain level, and forming a multi-level control objective set;

[0024] constructing a probe group state vector based on the multi-level control target set, and designing a control instruction vector according to the probe group state vector;

[0025] Constructing a comprehensive reward function based on the coupling relationship between the probe group state vector and the control instruction vector, wherein the comprehensive reward function is obtained by weighted combination of optimization target reward, constraint satisfaction reward and synchronization performance reward;

[0026] A state prediction model is constructed using a deep neural network, the probe group state vector and the control instruction vector are input into the state prediction model to obtain a predicted state vector, and the control instruction vector is updated using a rolling optimization method with an adaptive step size based on the predicted state vector and the comprehensive reward function;

[0027] The probe group is divided into multiple subgroups according to the control instruction vector, and each subgroup is optimized in parallel using a distributed computing architecture. The subgroup control parameters are updated in an asynchronous iterative manner based on the evaluation results of the comprehensive reward function.

[0028] According to the subgroup control parameters, single probe precise positioning control is achieved at the micro level, and the prediction models at each level are dynamically optimized through the sliding time domain window mechanism, including:

[0029] Acquiring an actual motion trajectory of the probe within a preset time window, calculating a predicted motion trajectory of the probe based on the subgroup control parameters, comparing the actual motion trajectory with the predicted motion trajectory, and calculating a trajectory prediction error;

[0030] Dynamically adjusting the time window length based on the trajectory prediction error, resampling the probe motion data according to the adjusted time window length to generate a new sampling data sequence; using the new sampling data sequence to correct the subgroup control parameters, inputting the corrected subgroup control parameters into the single probe control unit to generate the probe target position;

[0031] The current position of the probe is collected, the current position of the probe is compared with the target position of the probe, a position deviation value is calculated, and a position compensation instruction is generated according to the position deviation value; the current position of the probe is adjusted in real time based on the position compensation instruction to generate an optimized prediction model.

[0032] An evaluation index of the current contact state is calculated according to the probe group motion state evaluation model, and the evaluation index is compared with a preset evaluation threshold. When the preset evaluation threshold range is exceeded, the evaluation result is fed back to the probe group motion state evaluation model. The triggering parameter optimization cycle includes:

[0033] Processing the motion data of the probe group using the probe group motion state evaluation model to obtain the position deviation value, contact force deviation value, and synchronization deviation value of the probe group in the current contact state, and generating a contact state evaluation index;

[0034] The contact state evaluation index is compared with a preset evaluation threshold. When the contact state evaluation index exceeds the range of the preset evaluation threshold, evaluation deviation data is generated; the evaluation deviation data is fed back to the probe group motion state evaluation model to trigger a parameter optimization cycle.

[0035] A second aspect of an embodiment of the present invention provides a multi-probe coordinated contact control and adjustment system for wafer-level testing, comprising:

[0036] The first unit is used to obtain the initial position information of each probe and the target test point position information of the multi-probe test system, and generate an initial probe displacement trajectory planning scheme;

[0037] The second unit is used to collect probe motion feature data in real time using a multi-dimensional heterogeneous sensor array, and establish a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector based on the probe motion feature data;

[0038] The third unit is used to generate a probe group motion optimization objective function based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model and the initial probe displacement trajectory planning scheme;

[0039] The fourth unit is configured to optimize the trajectory of the probe group using a hierarchical progressive predictive control algorithm based on the probe group motion optimization objective function to obtain subgroup control parameters; implement single probe precise positioning control at the micro level based on the subgroup control parameters, and dynamically optimize the prediction models at each level through a sliding time domain window mechanism;

[0040] The fifth unit is used to calculate the evaluation index of the current contact state according to the probe group motion state evaluation model, compare the evaluation index with the preset evaluation threshold, and when it exceeds the preset evaluation threshold range, feed back the evaluation result to the probe group motion state evaluation model to trigger the parameter optimization cycle.

[0041] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0042] processor;

[0043] a memory for storing processor-executable instructions;

[0044] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0045] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0046] The beneficial effects of this application are as follows:

[0047] The multi-probe collaborative contact control and adjustment method for wafer-level testing provided by the present invention can achieve efficient collaborative movement of probe groups in a multi-probe testing system, effectively solving the shortcomings of traditional testing methods in probe collaborative control.

[0048] The present invention adopts a multi-dimensional heterogeneous sensor array to monitor the motion state of the probe in real time, establishes an evaluation model including a coupling interference matrix and a synchronization deviation vector, and combines it with a hierarchical progressive predictive control algorithm. It can accurately capture the mutual influence between probes, greatly improve the positioning accuracy and testing efficiency of the probe group, and meet the semiconductor industry's demand for high-precision wafer-level testing.

[0049] The present invention dynamically optimizes the prediction model through a sliding time domain window mechanism and designs a complete evaluation feedback mechanism, which enables the system to adaptively adjust control parameters, improves the robustness and reliability of the multi-probe test system under complex working conditions, and provides an effective solution for precise contact control during wafer testing. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1Schematic diagram of the flow of a multi-probe coordinated contact control and adjustment method for wafer-level testing according to an embodiment of the present invention;

[0051] Figure 2 This is a flow chart of the probe group coupling characteristics and synchronization evaluation according to an embodiment of the present invention;

[0052] Figure 3 A flow chart for constructing an objective function for optimizing the motion of a probe group according to an embodiment of the present invention;

[0053] Figure 4 This is a flow chart of the motion state and contact evaluation of a probe group according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0055] The technical solution of the present invention is described in detail below with reference to specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0056] Figure 1 FIG. 1 is a flow chart of a multi-probe coordinated contact control and adjustment method for wafer-level testing according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0057] Obtain the initial position information of each probe in the multi-probe test system and the target test point position information, and generate the initial probe displacement trajectory planning scheme;

[0058] A multi-dimensional heterogeneous sensor array is used to collect probe motion characteristic data in real time, and a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector is established based on the probe motion characteristic data;

[0059] Generate a probe group motion optimization objective function based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model and the initial probe displacement trajectory planning scheme;

[0060] Based on the probe group motion optimization objective function, a hierarchical progressive predictive control algorithm is used to optimize the trajectory and obtain subgroup control parameters; according to the subgroup control parameters, single probe precise positioning control is achieved at the micro level, and the prediction model of each layer is dynamically optimized through a sliding time domain window mechanism;

[0061] The evaluation index of the current contact state is calculated according to the probe group motion state evaluation model, and the evaluation index is compared with the preset evaluation threshold. When it exceeds the preset evaluation threshold range, the evaluation result is fed back to the probe group motion state evaluation model to trigger the parameter optimization cycle.

[0062] In an optional embodiment, establishing a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector based on the probe motion characteristic data includes:

[0063] Based on the probe motion characteristic data, the inter-probe force is obtained by calculating a weighted sum of a ratio of a displacement difference to a square of a distance multiplied by a first coupling coefficient and a ratio of a relative speed to a distance multiplied by a second coupling coefficient;

[0064] Constructing a coupling interference matrix based on the inter-probe force, wherein the diagonal elements of the coupling interference matrix represent the characteristics of the probes themselves, and the non-diagonal elements represent the coupling interference coefficients between the probes;

[0065] Obtain motion phase data of each probe, subtract the average phase of the probe group from the motion phase data to obtain a phase deviation, perform complex exponential mapping on the phase deviation and sum and average it within a time window to obtain a phase synchronization evaluation index;

[0066] Converting the phase synchronization evaluation index into a synchronization deviation vector through nonlinear mapping, wherein each component of the synchronization deviation vector represents the degree of synchronization of the corresponding probe;

[0067] The eigenvalues ​​of the coupling interference matrix and the synchronization deviation vector are fused using a weighted average method to generate a fusion feature that characterizes the overall motion characteristics of the probe group;

[0068] Based on the fusion features, a coupling evaluation component characterizing the coupling strength, a synchronization evaluation component characterizing the degree of synchronization, and an accuracy evaluation component characterizing the position accuracy are calculated respectively. The coupling evaluation component, the synchronization evaluation component, and the accuracy evaluation component are weighted by an adjustable weight coefficient to obtain a comprehensive evaluation index. According to the comprehensive evaluation index, a probe group motion state evaluation model is constructed.

[0069] like Figure 2 As shown, the method includes:

[0070] Obtain the position coordinates and velocity vector data of each probe. This data can be collected in real time through the sensor network. For example, in a system with five probes, the position coordinates of each probe can be represented as a point (x, y, z) in three-dimensional space, and the velocity vector is represented as (vx, vy, vz).

[0071] The calculation of the interprobe force uses an interaction model from physics. For any two probes i and j, the system calculates the displacement difference vector between them—the position difference of probe j relative to probe i. Assuming probe i is at (2, 3, 1) and probe j is at (4, 5, 2), the displacement difference vector is (2, 2, 1). The Euclidean distance between the two probes is then calculated, which in this example is 3. The ratio of the displacement difference to the square of the distance is multiplied by the first coupling coefficient α, which can be set to 0.5.

[0072] The relative velocity is also calculated. If the velocity of probe i is (1,1,0) and the velocity of probe j is (2,2,1), the relative velocity is (1,1,1). The ratio of the relative velocity to the distance is multiplied by the second coupling coefficient β, which can be set to 0.3. The weighted sum of these two components is the interprobe force, which in this example is approximately 0.28.

[0073] Based on the forces calculated above for all probe pairs, a 5×5 coupling interference matrix M is constructed. The diagonal elements Mii of the matrix represent the intrinsic properties of probe i, such as mass or inertia, and can be set as the probe's own parameters. The off-diagonal elements Mij represent the coupling interference coefficients between probes i and j, that is, the interprobe forces calculated above. For example, the diagonal elements can be set to 1.0 to represent the normalized intrinsic properties, and the off-diagonal elements are filled in based on the calculated results to complete the coupling interference matrix.

[0074] Obtain motion phase data for each probe, with phase values ​​ranging from 0 to 2π. Assume that the phase values ​​for the five probes are 1.2, 1.5, 1.3, 1.8, and 1.4 radians, respectively. Calculate the average phase of the probe group to be 1.44 radians. Subtract the average phase from the phase of each probe to obtain the phase deviations: -0.24, 0.06, -0.14, 0.36, and -0.04 radians.

[0075] A complex exponential mapping is performed on the phase deviations, converting each deviation value δ into the complex number exp(iδ). Within a 10-second window, samples are taken every 0.1 seconds, for a total of 100 sampling points. These complex values ​​are summed and divided by the number of sampling points to obtain the average. The closer the absolute value of the resulting complex number is to 1, the better the synchronization. In this example, the calculated result is 0.92, indicating that the probe group has high phase synchronization.

[0076] The phase synchronization evaluation index is converted into a synchronization deviation vector S through nonlinear mapping. Using an exponential function mapping, for each probe's phase deviation δi, exp(-|δi| / σ) is calculated, where σ is a tuning parameter set to 0.5. The resulting five-dimensional vector S is the synchronization deviation vector. The closer the component values ​​are to 1, the better the synchronization of the probes. In this example, the five components of the synchronization deviation vector are approximately 0.62, 0.89, 0.76, 0.49, and 0.92, respectively.

[0077] Perform eigenvalue decomposition on the coupling interference matrix M to obtain the main eigenvalues. Assume that the calculated eigenvalues ​​are {2.1, 1.3, 0.8, 0.5, 0.3}. Use the weighted averaging method to fuse these eigenvalues ​​with the synchronization deviation vector S, with weights set to 0.6 and 0.4. Averaging the eigenvalues ​​yields an eigenvalue component of 1.0, and averaging the synchronization deviation vectors yields a synchronization component of 0.736. The weighted combination of these two yields the fused feature F = 0.6 × 1.0 + 0.4 × 0.736 = 0.894.

[0078] Based on the fused feature F, the system calculates three evaluation components: the coupling evaluation component EC uses the coefficient of variation of the eigenvalues, which is 0.68 in this example; the synchronization evaluation component ES is the average value of the synchronization deviation vector, 0.736; and the precision evaluation component EP is calculated based on the probe position deviation. Assuming an average deviation of 0.15 meters, EP = 1 - 0.15 = 0.85. The weights of the three components are set to w1 = 0.4, w2 = 0.4, and w3 = 0.2, respectively, resulting in a comprehensive evaluation index I = 0.4 × 0.68 + 0.4 × 0.736 + 0.2 × 0.85 = 0.738.

[0079] The resulting probe swarm motion state assessment model maps the input probe motion feature data to a comprehensive evaluation index, I, and distinguishes the probe swarm's motion state based on the value of I. For example, when I > 0.8, it is considered "excellent," 0.6 ≤ I ≤ 0.8 is "good," 0.4 ≤ I < 0.6 is "fair," and I < 0.4 is "poor." In this example, I = 0.738 is considered "good," indicating that the overall motion characteristics of the probe swarm meet system requirements, but still have room for improvement. This assessment model can be applied to scenarios such as multi-probe collaborative detection and swarm robot control, enabling effective assessment and optimization of the probe swarm's motion state.

[0080] In an optional embodiment, generating a probe group motion optimization objective function based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model and the initial probe displacement trajectory planning scheme includes:

[0081] According to the coupling interference matrix and the synchronization deviation vector output by the probe group motion state evaluation model, a coupling optimization term and a synchronization optimization term are calculated respectively, wherein the coupling optimization term is obtained by setting a weight coefficient on the coupling interference matrix and summing the results, and the synchronization optimization term is obtained by performing a square operation on the synchronization deviation vector and weighting the result;

[0082] Calculating trajectory curvature continuity based on the initial probe displacement trajectory planning scheme, multiplying the trajectory curvature continuity with the trajectory smoothness weight coefficient and integrating the result in the motion time domain to obtain a trajectory optimization term;

[0083] The coupling optimization item, the synchronization optimization item and the trajectory optimization item are linearly combined to generate a probe group motion optimization objective function.

[0084] like Figure 3 As shown, the method includes:

[0085] The probe swarm motion state assessment model continuously monitors and analyzes the real-time position, velocity, and acceleration data of multiple probes. By processing this data, the assessment model generates two key outputs: a coupling interference matrix and a synchronization deviation vector. The coupling interference matrix represents the degree of mutual influence between probes and is an N×N matrix (N is the number of probes). Matrix elements Cij represent the interference strength of probe i on probe j.

[0086] In a four-probe system, if probe 1's interference on probe 2 is 0.35, on probe 3 is 0.28, and on probe 4 is 0.22, then the first row of the coupling interference matrix is ​​[0, 0.35, 0.28, 0.22]. The synchronization offset vector records the time deviation of each probe relative to the planned trajectory. For example, the synchronization offset vector for a four-probe system is [0.05, -0.03, 0.02, -0.04] seconds, indicating the lead or lag of each probe.

[0087] The coupling optimization term is calculated using a weighted allocation strategy. The system assigns a weight coefficient to each element of the coupling interference matrix based on the importance of the probe and the mission requirements. The weight coefficient matrix W is element-wise multiplied by the coupling interference matrix C to produce the weighted coupling interference matrix. All elements of the weighted matrix are then summed to obtain the coupling optimization term Jcoupling.

[0088] Set the diagonal elements of the weight coefficient matrix to 0 (ignoring self-interference) and the off-diagonal elements to values ​​ranging from 0.2 to 0.8 based on the relative importance of the probes. For the four-probe example above, if the first row of the weight matrix is ​​[0, 0.6, 0.5, 0.4], the coupling optimization value contributed by this row is 0.35 × 0.6 + 0.28 × 0.5 + 0.22 × 0.4 = 0.379.

[0089] The calculation of the synchronization optimization term involves squaring and weighting the synchronization deviation vector. Each element in the synchronization deviation vector is squared to ensure that both positive and negative deviations are considered performance degradations. A weight coefficient α is then assigned based on the task priority of each probe and multiplied by the squared deviation value. The sum of all weighted squared deviations constitutes the synchronization optimization term, Jsync.

[0090] The weight coefficients of the four probes can be set to [0.25, 0.25, 0.25, 0.25] (indicating equal importance) or [0.4, 0.3, 0.2, 0.1] (indicating decreasing importance). When using equal weights, the synchronization optimization term of the above synchronization deviation vector is calculated as 0.25×(0.05²) + 0.25×(-0.03²) + 0.25×(0.02²) + 0.25×(-0.04²) = 0.00185.

[0091] The trajectory optimization term is calculated based on the initial probe displacement trajectory plan, focusing on the curvature continuity of the trajectory. The system first calculates the rate of change of curvature of each probe trajectory at each time point. A smaller rate of change indicates a smoother trajectory. For each probe i, the square of the rate of change of curvature over the entire time domain is calculated and multiplied by the trajectory smoothness weight β to obtain the trajectory optimization term for probe i.

[0092] The sum of all probe trajectory optimization sub-items constitutes the trajectory optimization term J-trajectory. In typical applications, the curvature change rate can be calculated using the second-order derivative change rate of the trajectory points. For example, if the squared integral of the curvature change rate of four probes in a certain time period is [0.15, 0.22, 0.18, 0.25], and the smoothness weight coefficient is 0.5, then the trajectory optimization term is 0.5 × (0.15 + 0.22 + 0.18 + 0.25) = 0.4.

[0093] The probe swarm motion optimization objective function is generated by linearly combining the three optimization terms described above. Based on the specific application scenario and task requirements, the system assigns overall weight coefficients λ1, λ2, and λ3 to the coupling, synchronization, and trajectory optimization terms. The optimization objective function is expressed as Jtotal = λ1 × Jcoupling + λ2 × Jsynchronization + λ3 × Jtrajectory.

[0094] In practical applications, if the task focuses more on minimizing interference between probes, the weights can be set to [0.5, 0.3, 0.2]; if the focus is on synchronization accuracy, the weights can be set to [0.3, 0.5, 0.2]; if the focus is on trajectory smoothness, the weights can be set to [0.2, 0.3, 0.5]. For the above calculated J coupling = 0.379, J synchronization = 0.00185, and J trajectory = 0.4, if the weights [0.4, 0.4, 0.2] are used, the total objective function value is 0.4 × 0.379 + 0.4 × 0.00185 + 0.2 × 0.4 = 0.15226.

[0095] During actual system operation, the controller continuously evaluates the current state of the probe group, calculates the objective function value, and adjusts the motion parameters of the probes through an optimization algorithm to minimize the objective function value. For example, if the system detects excessive interference between two probes, it automatically adjusts their relative position or velocity profile. If it finds that the synchronization deviation is increasing, it adjusts the speed of each probe accordingly to restore synchronization. If the trajectory has an uneven section, it replans the section to improve smoothness. In this way, the probe group can achieve efficient coordination, precise synchronization, and smooth motion.

[0096] In an optional embodiment, based on the probe group motion optimization objective function, a hierarchical progressive predictive control algorithm is used to perform trajectory optimization, and the subgroup control parameters obtained include:

[0097] Dividing the control time domain into multiple time domain levels, generating corresponding optimization objectives and constraints according to the control requirements of each time domain level, and forming a multi-level control objective set;

[0098] constructing a probe group state vector based on the multi-level control target set, and designing a control instruction vector according to the probe group state vector;

[0099] Constructing a comprehensive reward function based on the coupling relationship between the probe group state vector and the control instruction vector, wherein the comprehensive reward function is obtained by weighted combination of optimization target reward, constraint satisfaction reward and synchronization performance reward;

[0100] A state prediction model is constructed using a deep neural network, the probe group state vector and the control instruction vector are input into the state prediction model to obtain a predicted state vector, and the control instruction vector is updated using a rolling optimization method with an adaptive step size based on the predicted state vector and the comprehensive reward function;

[0101] The probe group is divided into multiple subgroups according to the control instruction vector, and each subgroup is optimized in parallel using a distributed computing architecture. The subgroup control parameters are updated in an asynchronous iterative manner based on the evaluation results of the comprehensive reward function.

[0102] Control time domain segmentation is a fundamental step in the optimization process. The system divides the entire control time domain into three levels: short-term, medium-term, and long-term. Each level corresponds to a different control accuracy and prediction range. The short-term level focuses on immediate response within 0-5 seconds, with a control sampling period of 0.1 seconds; the medium-term level focuses on strategy adjustments within 5-30 seconds, with a control sampling period of 0.5 seconds; and the long-term level focuses on global planning within 30-120 seconds, with a control sampling period of 2 seconds.

[0103] The short-term goal was to minimize the distance deviation between probes, with the constraint that the probe speed should not exceed 3 meters per second. The mid-term goal was to optimize the swarm's energy consumption, with the constraint that the probe acceleration should not exceed 0.5 meters per square second. The long-term goal was to cover the target area, with the constraint that the swarm maintain a specific geometric configuration. Through hierarchical design, a multi-level control objective set consisting of 9 control objectives and 12 constraints was formed.

[0104] To construct the probe swarm state vector, the system integrates each probe's position, velocity, acceleration, energy state, and mission completion into a single state vector. For a swarm of 100 probes, the state vector has a dimension of 700, including each probe's three-dimensional spatial coordinates, three-dimensional velocity, and a single-dimensional energy value.

[0105] The control command vector design includes two components: direction control and speed control, precisely regulating the motion of each probe. Direction control is represented using quaternions, avoiding the singularity of Euler angles. Speed ​​control uses a scalar value ranging from 0 to 3 meters per second. Discretization reduces computational complexity by mapping the continuous speed value to 20 discrete levels.

[0106] The construction of the comprehensive reward function incorporates multiple evaluation metrics. The optimization target reward uses an exponential function of target achievement, with a baseline value of 0.8 and a weighting factor of 0.5. The constraint satisfaction reward uses a soft constraint, converting the degree of constraint violation into a penalty value, with a weighting factor of 0.3. The synchronization performance reward measures the collaborative behavior of the probe group and is calculated by calculating the correlation between the position and velocity of the probes, with a weighting factor of 0.2. In practice, when the probe group maintains a predetermined formation and evenly covers the target area, and the speed difference between the probes does not exceed 0.5 meters per second, the comprehensive reward function reaches its maximum value of 0.95.

[0107] The state prediction model is constructed using a five-layer deep neural network, consisting of three fully connected layers and two LSTM layers. The input layer has 1100 nodes (a 700-dimensional state vector plus a 400-dimensional control instruction vector), the hidden layers have 512, 256, and 256 nodes, respectively, and the output layer has 700 nodes, corresponding to the predicted state vector.

[0108] The network was trained using 64,000 sets of historical data. The learning rate was initially set to 0.001 and decayed by 10% every 2,000 iterations. In the rolling optimization method, the adaptive step size was dynamically adjusted based on the gradient of the reward function, with an initial step size of 0.05, a minimum step size of 0.001, and a maximum step size of 0.2. During the actual optimization process, the system predicted the next 10 time steps and, based on the comprehensive reward function evaluation, selected the optimal control command vector for implementation at the current moment, then rolled over to the next moment for continued optimization.

[0109] Subgrouping and parallel optimization are key to improving system computational efficiency. The system uses the K-means clustering algorithm to partition 100 probes into five subgroups, each containing 20 probes, based on the similarity of their control instruction vectors. Similarity is calculated using cosine distance, and cluster centers are initialized using K-means++. The distributed computing architecture employs a master-slave structure, with the master node responsible for global coordination and the five slave nodes each optimizing a subgroup.

[0110] Each slave node is equipped with an 8-core CPU and 16GB of memory, with an optimization cycle of 0.2 seconds. During the asynchronous iteration process, subgroup optimization results are reported to the master node at a frequency of 5Hz. After comprehensive evaluation, the master node issues update instructions at a frequency of 2Hz. In actual testing, when a probe group needs to traverse a complex obstacle environment, the system divides probes in areas with more obstacles into the same subgroup and allocates more computing resources, shortening the subgroup optimization cycle to 0.1 seconds, improving response speed in complex environments.

[0111] Through the above technical implementation, the hierarchical progressive predictive control algorithm of the present invention can effectively optimize the motion trajectory of the probe group, while ensuring the coordination of the probe group, achieving adaptability to complex environments and efficient completion of task objectives. Practical application tests have shown that compared with traditional control methods, the control parameter optimization method of the present invention can shorten task completion time by 22.3% and reduce energy consumption by 17.8%, while ensuring a coordination index of more than 95% for the probe group.

[0112] In an optional embodiment, the single probe precise positioning control is implemented at the micro level according to the subgroup control parameters, and the prediction models at each level are dynamically optimized through a sliding time domain window mechanism, including:

[0113] Acquiring an actual motion trajectory of the probe within a preset time window, calculating a predicted motion trajectory of the probe based on the subgroup control parameters, comparing the actual motion trajectory with the predicted motion trajectory, and calculating a trajectory prediction error;

[0114] Dynamically adjusting the time window length based on the trajectory prediction error, resampling the probe motion data according to the adjusted time window length to generate a new sampling data sequence; using the new sampling data sequence to correct the subgroup control parameters, inputting the corrected subgroup control parameters into the single probe control unit to generate the probe target position;

[0115] The current position of the probe is collected, the current position of the probe is compared with the target position of the probe, a position deviation value is calculated, and a position compensation instruction is generated according to the position deviation value; the current position of the probe is adjusted in real time based on the position compensation instruction to generate an optimized prediction model.

[0116] The probe's actual motion trajectory data is collected within a preset time window. This time window can be initially set to 500 milliseconds. During this time window, the system records the probe's coordinate position (x, y, z) in three-dimensional space at a sampling interval of 10 milliseconds, forming an actual trajectory sequence containing 50 data points.

[0117] The predicted trajectory is calculated based on pre-established subgroup control parameters. These subgroup control parameters include the mass coefficient, damping coefficient, and stiffness coefficient of the probe motion, with initial values ​​of m = 0.05 kg, c = 0.2 N·s / m, and k = 25 N / m, respectively. The system uses these parameters to predict the theoretical position of the probe within the same time window and generate a corresponding predicted trajectory sequence.

[0118] When comparing the actual trajectory with the predicted trajectory, the system calculates the root mean square error (RMSE) between the two trajectories. For example, if the coordinate difference between the actual and predicted trajectory points at a certain moment is △x = 0.02μm, △y = 0.03μm, and △z = 0.01μm, then the Euclidean distance error at that moment is 0.037μm. The RMSE is calculated for all data points within the entire time window. If the result exceeds the preset threshold of 0.05μm, a dynamic adjustment mechanism for the time window length is triggered.

[0119] The dynamic adjustment of the time window length is based on the error gradient principle. If the trajectory prediction error increases three times in a row, the system shortens the time window by 20%, from 500 milliseconds to 400 milliseconds. If the error decreases five times in a row, the window is increased by 15%, to 575 milliseconds. This adjustment affects the data sampling strategy. For example, if the window is shortened to 400 milliseconds, the system still maintains a 10 millisecond sampling interval, reducing the number of data points to 40.

[0120] Based on the adjusted time window, the system resamples the probe's historical motion data. Resampling uses linear interpolation to ensure that the newly sampled data points are evenly distributed along the time axis. For example, if the original data point timestamps are [0, 10, 20, ..., 500] milliseconds and the adjusted window is 400 milliseconds, the new sampling point timestamps become [0, 10, 20, ..., 400] milliseconds, and the system generates 40 new sampling data points.

[0121] Using the new sampled data sequence, the system corrects the subgroup control parameters. This correction process employs a recursive least squares approach, iteratively optimizing mass, damping, and stiffness parameters based on the deviation between the probe's actual motion and the theoretical model. For example, the original parameters m = 0.05 kg, c = 0.2 N·s / m, and k = 25 N / m are corrected to m = 0.048 kg, c = 0.22 N·s / m, and k = 24.8 N / m. The correction amplitude is generally kept within ±10% of the original values ​​to avoid system instability caused by sudden parameter changes.

[0122] The modified subgroup control parameters are fed into the single-probe control unit, which calculates the probe's target position based on the equations of motion. Assuming the current probe position is (10.25μm, 15.36μm, 5.42μm), the control unit predicts the target position at the next instant (10 milliseconds later) to be (10.28μm, 15.39μm, 5.44μm). This target position serves as the instantaneous reference point for the probe's motion.

[0123] The probe's current actual position is acquired using a high-precision position sensor. These sensors can be laser interferometers or capacitive displacement sensors, with nanometer-level accuracy. If the actual position acquired by the sensor is (10.27μm, 15.41μm, 5.43μm), the system compares it with the previously calculated target position (10.28μm, 15.39μm, 5.44μm) to calculate the position deviation. In this example, the deviation vector is (0.01μm, -0.02μm, 0.01μm).

[0124] A position compensation command is generated based on the calculated position deviation. This compensation command is generated using a proportional-integral-derivative (PID) control strategy with a proportional coefficient Kp = 0.8, an integral coefficient Ki = 0.05, and a differential coefficient Kd = 0.2. Given the above deviation vector, the system generates compensation commands that will move the probe 0.008 μm in the negative x-axis direction, 0.016 μm in the positive y-axis direction, and 0.008 μm in the negative z-axis direction.

[0125] The probe position is adjusted in real time by a piezoelectric actuator or linear motor based on position compensation commands. Upon receiving these commands, the actuator performs position correction with submicron accuracy. After correction, the probe's actual position is closer to the target position. For example, after adjustment, the actual position is (10.272μm, 15.394μm, 5.438μm), reducing the deviation from the target position to within 0.01μm.

[0126] Through this periodic adjustment and continuous optimization, the system continuously improves the accuracy of the prediction model. Every 10 control cycles (approximately 100 milliseconds), the system updates the statistical characteristics of the probe's motion, including parameters such as average velocity, acceleration distribution, and position fluctuation range. These statistical characteristics, combined with the revised subgroup control parameters, form the optimized prediction model. This model can adapt to subtle changes in the probe's motion environment, such as temperature drift and mechanical vibration, maintaining a positioning accuracy of less than 50 nanometers.

[0127] In an optional embodiment, an evaluation index of the current contact state is calculated according to the probe group motion state evaluation model, and the evaluation index is compared with a preset evaluation threshold. When the preset evaluation threshold range is exceeded, the evaluation result is fed back to the probe group motion state evaluation model, and the triggering parameter optimization cycle includes:

[0128] Processing the motion data of the probe group using the probe group motion state evaluation model to obtain the position deviation value, contact force deviation value, and synchronization deviation value of the probe group in the current contact state, and generating a contact state evaluation index;

[0129] The contact state evaluation index is compared with a preset evaluation threshold. When the contact state evaluation index exceeds the range of the preset evaluation threshold, evaluation deviation data is generated; the evaluation deviation data is fed back to the probe group motion state evaluation model to trigger a parameter optimization cycle.

[0130] like Figure 4 As shown, the method includes:

[0131] The motion assessment model monitors the motion data of the probe group in real time to ensure optimal contact between the probes and the object being measured. The system compares assessment indicators with preset thresholds and triggers a parameter optimization cycle when indicators fall outside of the range, thereby improving measurement accuracy and stability.

[0132] The probe swarm motion state assessment model uses a multi-layer perceptron architecture, consisting of an input layer, hidden layers, and an output layer. The input layer receives real-time motion data from the probe swarm, including position coordinates, contact force values, and timestamp information. The hidden layer contains three subnetworks, each containing 64 neurons, to process position information, force information, and timing information, respectively. The output layer generates three evaluation metrics: position deviation, contact force deviation, and synchronization deviation.

[0133] The position deviation calculation process involves collecting the spatial coordinate data of each probe in microns, comparing the actual coordinates of each probe with the ideal coordinates, calculating the Euclidean distance as the position deviation of each probe, and finally summing the position deviations of all probes to obtain the overall position deviation value. For example, in a test with 8 probes, the ideal position is (100, 100, 0) microns, and the actual measured position is (102, 101, -1) microns, resulting in a position deviation of 3.16 microns.

[0134] The contact force deviation calculation step involves reading the pressure sensor data for each probe, comparing the actual contact force with the set contact force in millinewtons, and calculating the root mean square (RMS) of the contact force deviations for all probes to obtain the overall contact force deviation. For example, if the contact force is set to 50 millinewtons, the actual measured contact forces for the eight probes are 48.5, 51.2, 49.8, 50.3, 47.9, 52.1, 49.5, and 50.8 millinewtons, respectively. The calculated contact force deviation is 1.45 millinewtons.

[0135] The synchronization deviation calculation step involves recording the timestamp of each probe's contact with the target surface, calculating the absolute deviation between the timestamps of all probes and the average timestamp in microseconds, and taking the maximum value as the synchronization deviation. For example, if the contact timestamps of eight probes are 1000, 1003, 998, 1002, 997, 1004, 999, and 1001 microseconds, and the average timestamp is 1000.5 microseconds, the maximum deviation is 3.5 microseconds, indicating a synchronization deviation of 3.5 microseconds.

[0136] The evaluation metrics are compared against preset thresholds as follows: The system pre-sets position deviation thresholds of 5 microns, 2 millinewtons, and 10 microseconds for position deviation, force deviation, and synchronization deviation. The system compares the three deviation values ​​calculated in real time with the corresponding thresholds to determine whether they exceed the specified range. If any deviation value exceeds the corresponding threshold, the system generates evaluation deviation data, including the type of metric exceeding the threshold, the degree of excess, and the time of the excess.

[0137] When the evaluation metric exceeds a preset threshold, the evaluation deviation data is fed back to the probe group motion state evaluation model, triggering a parameter optimization loop. The parameter optimization process includes adjusting the model's weight parameters based on the feedback data, updating the probe control instructions, and optimizing the contact strategy. In the above example, if the contact force deviation of 1.45 millinewtons does not exceed the threshold of 2 millinewtons, optimization is not triggered. However, if the position deviation exceeds the 5 micron threshold, the system triggers a position calibration process.

[0138] The parameter optimization loop is implemented as follows: first, the deviation type is determined. If it is a position deviation, the stepper motor displacement is adjusted; if it is a force deviation, the pressure control unit setting is adjusted; if it is a synchronization deviation, the probe group trigger timing is adjusted. Parameter adjustments are made incrementally, with each adjustment proportional to the deviation. For example, if the position deviation is 7 microns, the system calculates an adjustment of (7-5) × 0.8 = 1.6 microns, adjusting the probe position in the direction of reducing the deviation.

[0139] Through a continuous monitoring and optimization cycle, the system maintains the probe cluster within the ideal contact state range. Experiments have shown that this approach reduces position deviation by an average of 62%, contact force deviation by an average of 57%, and synchronization deviation by an average of 73%, significantly improving measurement accuracy and stability. This method is particularly suitable for scenarios requiring strict control of contact states, such as high-precision microelectronics testing and biological tissue probing.

[0140] To further ensure system reliability, an abnormal situation handling mechanism has been implemented: when an indicator still exceeds the threshold after three consecutive optimizations, the system automatically enters deep diagnosis mode, analyzes the fault type and provides targeted solutions; when a certain indicator exceeds twice the threshold, the system enters emergency protection mode, suspends testing and issues an alarm; when three indicators exceed the threshold at the same time, the system determines it as a serious abnormality, immediately terminates the operation and saves the current status data for subsequent analysis.

[0141] A second aspect of an embodiment of the present invention provides a multi-probe coordinated contact control and adjustment system for wafer-level testing, comprising:

[0142] The first unit is used to obtain the initial position information of each probe and the target test point position information of the multi-probe test system, and generate an initial probe displacement trajectory planning scheme;

[0143] The second unit is used to collect probe motion feature data in real time using a multi-dimensional heterogeneous sensor array, and establish a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector based on the probe motion feature data;

[0144] The third unit is used to generate a probe group motion optimization objective function based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model and the initial probe displacement trajectory planning scheme;

[0145] The fourth unit is configured to optimize the trajectory of the probe group using a hierarchical progressive predictive control algorithm based on the probe group motion optimization objective function to obtain subgroup control parameters; implement single probe precise positioning control at the micro level based on the subgroup control parameters, and dynamically optimize the prediction models at each level through a sliding time domain window mechanism;

[0146] The fifth unit is used to calculate the evaluation index of the current contact state according to the probe group motion state evaluation model, compare the evaluation index with the preset evaluation threshold, and when it exceeds the preset evaluation threshold range, feed back the evaluation result to the probe group motion state evaluation model to trigger the parameter optimization cycle.

[0147] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0148] processor;

[0149] a memory for storing processor-executable instructions;

[0150] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0151] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0152] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-probe collaborative contact control and adjustment method for wafer-level testing, characterized in that: include: Obtain the initial position information of each probe in the multi-probe test system and the target test point position information, and generate the initial probe displacement trajectory planning scheme; A multi-dimensional heterogeneous sensor array is used to collect probe motion characteristic data in real time, and a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector is established based on the probe motion characteristic data; Generate a probe group motion optimization objective function based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model and the initial probe displacement trajectory planning scheme; Based on the probe group motion optimization objective function, a hierarchical progressive predictive control algorithm is used to optimize the trajectory and obtain subgroup control parameters; based on the subgroup control parameters, single probe precise positioning control is achieved at the micro level, and the prediction model of each layer is dynamically optimized through a sliding time domain window mechanism; The evaluation index of the current contact state is calculated according to the probe group motion state evaluation model, and the evaluation index is compared with the preset evaluation threshold. When it exceeds the preset evaluation threshold range, the evaluation result is fed back to the probe group motion state evaluation model to trigger the parameter optimization cycle.

2. The method according to claim 1, characterized in that Based on the probe motion characteristic data, establishing a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector includes: Based on the probe motion characteristic data, the inter-probe force is obtained by calculating a weighted sum of a ratio of a displacement difference to a square of a distance multiplied by a first coupling coefficient and a ratio of a relative speed to a distance multiplied by a second coupling coefficient; Constructing a coupling interference matrix based on the inter-probe force, wherein the diagonal elements of the coupling interference matrix represent the characteristics of the probes themselves, and the non-diagonal elements represent the coupling interference coefficients between the probes; Obtain motion phase data of each probe, subtract the average phase of the probe group from the motion phase data to obtain a phase deviation, perform complex exponential mapping on the phase deviation and sum and average it within a time window to obtain a phase synchronization evaluation index; Converting the phase synchronization evaluation index into a synchronization deviation vector through nonlinear mapping, wherein each component of the synchronization deviation vector represents the degree of synchronization of the corresponding probe; The eigenvalues ​​of the coupling interference matrix and the synchronization deviation vector are fused using a weighted average method to generate a fusion feature that characterizes the overall motion characteristics of the probe group; Based on the fusion features, a coupling evaluation component characterizing the coupling strength, a synchronization evaluation component characterizing the degree of synchronization, and an accuracy evaluation component characterizing the position accuracy are calculated respectively. The coupling evaluation component, the synchronization evaluation component, and the accuracy evaluation component are weighted by an adjustable weight coefficient to obtain a comprehensive evaluation index. According to the comprehensive evaluation index, a probe group motion state evaluation model is constructed.

3. The method according to claim 1, characterized in that Based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model, combined with the initial probe displacement trajectory planning scheme, the probe group motion optimization objective function is generated, including: According to the coupling interference matrix and the synchronization deviation vector output by the probe group motion state evaluation model, a coupling optimization term and a synchronization optimization term are calculated respectively, wherein the coupling optimization term is obtained by setting a weight coefficient on the coupling interference matrix and summing the results, and the synchronization optimization term is obtained by performing a square operation on the synchronization deviation vector and weighting the result; Calculating trajectory curvature continuity based on the initial probe displacement trajectory planning scheme, multiplying the trajectory curvature continuity with the trajectory smoothness weight coefficient and integrating the result in the motion time domain to obtain a trajectory optimization term; The coupling optimization item, the synchronization optimization item and the trajectory optimization item are linearly combined to generate a probe group motion optimization objective function.

4. The method according to claim 1, wherein Based on the probe group motion optimization objective function, a hierarchical progressive predictive control algorithm is used for trajectory optimization, and the subgroup control parameters obtained include: Dividing the control time domain into multiple time domain levels, generating corresponding optimization objectives and constraints according to the control requirements of each time domain level, and forming a multi-level control objective set; constructing a probe group state vector based on the multi-level control target set, and designing a control instruction vector according to the probe group state vector; Constructing a comprehensive reward function based on the coupling relationship between the probe group state vector and the control instruction vector, wherein the comprehensive reward function is obtained by weighted combination of optimization target reward, constraint satisfaction reward and synchronization performance reward; A state prediction model is constructed using a deep neural network, the probe group state vector and the control instruction vector are input into the state prediction model to obtain a predicted state vector, and the control instruction vector is updated using a rolling optimization method with an adaptive step size based on the predicted state vector and the comprehensive reward function; The probe group is divided into multiple subgroups according to the control instruction vector, and each subgroup is optimized in parallel using a distributed computing architecture. The subgroup control parameters are updated in an asynchronous iterative manner based on the evaluation results of the comprehensive reward function.

5. The method according to claim 1, wherein According to the subgroup control parameters, single probe precise positioning control is achieved at the micro level, and the prediction models at each level are dynamically optimized through the sliding time domain window mechanism, including: Acquiring an actual motion trajectory of the probe within a preset time window, calculating a predicted motion trajectory of the probe based on the subgroup control parameters, comparing the actual motion trajectory with the predicted motion trajectory, and calculating a trajectory prediction error; Dynamically adjusting the time window length based on the trajectory prediction error, resampling the probe motion data according to the adjusted time window length to generate a new sampling data sequence; using the new sampling data sequence to correct the subgroup control parameters, inputting the corrected subgroup control parameters into the single probe control unit to generate the probe target position; The current position of the probe is collected, the current position of the probe is compared with the target position of the probe, a position deviation value is calculated, and a position compensation instruction is generated according to the position deviation value; the current position of the probe is adjusted in real time based on the position compensation instruction to generate an optimized prediction model.

6. The method according to claim 1, characterized in that An evaluation index of the current contact state is calculated according to the probe group motion state evaluation model, and the evaluation index is compared with a preset evaluation threshold. When the preset evaluation threshold range is exceeded, the evaluation result is fed back to the probe group motion state evaluation model. The triggering parameter optimization cycle includes: Processing the motion data of the probe group using the probe group motion state evaluation model to obtain the position deviation value, contact force deviation value, and synchronization deviation value of the probe group in the current contact state, and generating a contact state evaluation index; The contact state evaluation index is compared with a preset evaluation threshold. When the contact state evaluation index exceeds the range of the preset evaluation threshold, evaluation deviation data is generated; the evaluation deviation data is fed back to the probe group motion state evaluation model to trigger a parameter optimization cycle.

7. A multi-probe coordinated contact control and adjustment system for wafer-level testing, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to obtain the initial position information of each probe and the target test point position information of the multi-probe test system, and generate an initial probe displacement trajectory planning scheme; The second unit is used to collect probe motion feature data in real time using a multi-dimensional heterogeneous sensor array, and establish a probe group motion state assessment model including a coupling interference matrix and a synchronization deviation vector based on the probe motion feature data; The third unit is used to generate a probe group motion optimization objective function based on the coupling interference matrix and synchronization deviation vector output by the probe group motion state evaluation model and the initial probe displacement trajectory planning scheme; The fourth unit is configured to optimize the trajectory of the probe group using a hierarchical progressive predictive control algorithm based on the probe group motion optimization objective function to obtain subgroup control parameters; implement single probe precise positioning control at the micro level based on the subgroup control parameters, and dynamically optimize the prediction models at each level through a sliding time domain window mechanism; The fifth unit is used to calculate the evaluation index of the current contact state according to the probe group motion state evaluation model, compare the evaluation index with the preset evaluation threshold, and when it exceeds the preset evaluation threshold range, feed back the evaluation result to the probe group motion state evaluation model to trigger the parameter optimization cycle.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images