FT test time prediction method and device and electronic equipment

Abstract

The invention relates to an FT test time prediction method and apparatus, and an electronic device. The method comprises the steps of collecting historical test data including a test vector number, a test machine specification, a process node, a packaging type, a test temperature and a working voltage; extracting a plurality of test features of the historical test data and constructing a feature set, the test features being constructed based on the historical test data and reflecting a coupling relationship between test conditions; training a machine learning model by adopting the feature set, wherein the training process comprises the step of adjusting and optimizing model parameters based on historical prediction errors; and utilizing the trained machine learning model to predict the FT test time, and outputting a test time prediction value. According to the method, isolated test condition parameters are converted into input characteristics with relevance, so that the machine learning model can learn multi-dimensional coupling rules such as interaction between the process node and the test temperature, matching characteristics between the packaging type and the working voltage and the like, and the test time prediction precision is improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor testing technology, and in particular to a method, apparatus, and electronic device for predicting FT test time. Background Technology

[0002] In integrated circuit manufacturing, final testing (FT) is crucial for ensuring chip yield and reliability. With the continuous expansion of testing scale and the increasing complexity of test vectors, test time has become one of the main bottlenecks in the production line. Currently, FT test time assessment typically relies on empirical parameters or fixed rules, such as linear estimation methods based on the number of test vectors and test equipment specifications. However, actual testing conditions involve multiple factors, such as process node, package type, test temperature, and operating voltage, and these factors exhibit complex nonlinear coupling relationships. Traditional linear estimation methods cannot accurately reflect these complex relationships, resulting in low test time prediction accuracy. Low prediction accuracy not only affects the accuracy of test planning but also restricts the efficient scheduling of test resources and the improvement of overall production line utilization. Therefore, improving the accuracy of FT test time prediction has become an urgent technical problem to be solved. Summary of the Invention

[0003] Therefore, it is necessary to provide an FT test time prediction method, apparatus, and electronic device to address the aforementioned technical problem of low FT test time prediction accuracy.

[0004] This invention provides a method for predicting the FT test time, the method comprising: Collect historical test data including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage; Multiple test features are extracted from the historical test data and a feature set is constructed. The test features are constructed based on the historical test data and reflect the coupling relationship between test conditions. The machine learning model is trained using the feature set, wherein the training process includes tuning the model parameters based on historical prediction errors to optimize the model's prediction accuracy. The trained machine learning model is used to predict the FT test time, and the predicted test time value is output.

[0005] In one embodiment, the extraction of multiple test features from the historical test data includes: Construct at least one of the following: a combination variable of process node and test temperature, and a combination variable of package type and operating voltage, to strengthen the coupling relationship between the test conditions.

[0006] In one embodiment, the machine learning model is one of the XGBoost algorithm, the random forest algorithm, or the neural network algorithm.

[0007] In one embodiment, the method further includes generating a test scheduling scheme based on the predicted test time value, including optimizing test scheduling using heuristic algorithms.

[0008] In one embodiment, the generated test scheduling scheme further includes dynamically adjusting the allocation of test resources based on the real-time machine status.

[0009] In one embodiment, the generated test scheduling scheme further includes a multi-objective optimization framework, which includes defining multiple optimization objectives and using a genetic algorithm to generate a Pareto optimal solution set. The optimization objectives include minimizing the total test duration, maximizing machine utilization, and balancing the test load.

[0010] In one embodiment, the multi-objective optimization framework further includes a dynamic scheduling adjustment mechanism, which includes real-time monitoring of machine status changes and triggering dynamic rescheduling. The dynamic rescheduling uses a reinforcement learning algorithm to adjust the test sequence and resource allocation based on real-time feedback to minimize the impact of scheduling interruptions.

[0011] In one embodiment, the tuning of model parameters based on historical prediction errors includes: The deviation between historical prediction errors and a preset threshold is calculated, and when the deviation exceeds the threshold, the model parameters are recalibrated. The recalibration includes adjusting the hyperparameters of the machine learning model to minimize the cumulative effect of prediction errors. Newly generated test data is continuously collected and the model is incrementally trained. The incremental training uses a sliding window method to update the feature set, retaining only the data within the most recent preset time period to reduce computational complexity.

[0012] The present invention also provides an FT test time prediction device, the device comprising: The test data acquisition module is used to collect historical test data, including the number of test vectors, test equipment specifications, process nodes, package types, test temperatures, and operating voltages. The test feature extraction module is used to extract multiple test features from the historical test data and construct a feature set. The test features are constructed based on the historical test data and reflect the coupling relationship between test conditions. The model training module is used to train the machine learning model using the feature set. The training process includes tuning the model parameters based on historical prediction errors to optimize the model prediction accuracy. The prediction output module is used to predict the FT test time using the trained machine learning model and output the predicted test time value.

[0013] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the FT test time prediction method as described above.

[0014] The aforementioned Fourier Transform (FT) test time prediction method, apparatus, and electronic equipment, by collecting historical test data including the number of test vectors, test equipment specifications, process nodes, package types, test temperatures, and operating voltages, constructs a feature set reflecting the coupling relationships between test conditions. It employs a machine learning model training method based on parameter tuning of historical prediction errors, effectively overcoming the technical limitations of traditional linear estimation methods in handling complex nonlinear couplings of multiple factors. Simultaneously, the feature set construction process transforms previously isolated test condition parameters into correlated input features, enabling the machine learning model to learn multi-dimensional coupling patterns such as the interaction between process nodes and test temperatures, and the matching characteristics between package types and operating voltages, thereby improving test time prediction accuracy. The parameter tuning mechanism during model training further enhances the model's adaptability to changes in test conditions. By continuously optimizing model parameters, the prediction error converges, ultimately achieving high-precision FT test time prediction output. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 Here is a flowchart of an embodiment of the FT test time prediction method; Figure 2 Here is a flowchart of an FT test time prediction method according to another embodiment; Figure 3 This is a schematic diagram of an FT test time prediction device according to one embodiment; Figure 4 This is an internal structural diagram of an electronic device according to one embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The following is combined with Figures 1-4 The present invention describes the FT test time prediction method, apparatus, and electronic device.

[0019] like Figure 1 As shown, in one embodiment, an FT test time prediction method includes the following steps: Step S110: Collect historical test data including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage.

[0020] The system automatically acquires historical test data, including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage. The number of test vectors reflects test complexity; test equipment specifications include processor performance and memory capacity parameters; process nodes cover the 28nm to 5nm range; package types include common forms such as BGA and QFN; test temperatures cover the industry standard range of -40℃ to 125℃; and operating voltage includes multiple power supply domains for core and I / O voltages. The acquisition process also includes obtaining the equipment ID and its corresponding actual test time, forming a complete data record with spatiotemporal correlation to establish a precise mapping relationship between test conditions and test results. Test data traceability is achieved through the equipment ID, and the actual test time is used as label data for supervised learning. By constructing a complete dataset covering multi-dimensional test conditions, comprehensive data support is provided for subsequent feature engineering, enabling machine learning models to learn the time patterns under different test configurations, ultimately reducing test time prediction errors.

[0021] Step S120: Extract multiple test features from historical test data and construct a feature set. The test features are constructed based on historical test data and reflect the coupling relationship between test conditions.

[0022] The collected historical test data is preprocessed, including missing value imputation, outlier removal, and data standardization. Then, basic and combined features are constructed based on test condition parameters. Further, an interaction variable between process node and test temperature is constructed by encoding discrete process nodes as categorical variables and combining them with continuous test temperature values. Optionally, the numerical combination uses a product interaction method: the categorical variable encoded with the process node (e.g., a vector encoded using one-thermal multiplication) is multiplied element-wise with the test temperature value to generate a process node-test temperature interaction feature vector. Similarly, the interaction variable between package type and operating voltage is obtained by multiplying the one-thermal encoded package type vector with the operating voltage value. For example, for a process node encoded as [0,1,0] (corresponding to a 28nm node) and a test temperature of 25℃, the interaction feature is calculated as [0,1,0]*25 = [0,25,0]. Simultaneously, an interaction variable between package type and operating voltage is constructed by performing feature cross-validation between the one-thermal encoded package type and operating voltage. Optionally, the feature cross-validation uses an outer product operation to generate a second-order interaction term. Specifically, the outer product of the one-hot encoded package type vector (dimension m) and the operating voltage vector (dimension n) is taken to obtain an m×n matrix, which is then flattened into feature vectors. For example, if the package type is encoded as [1,0] (corresponding to BGA) and the operating voltage is 1.2V, the outer product result is [1.2, 0; 0, 0], which flattens to [1.2, 0, 0, 0]. This feature construction process explicitly characterizes the coupling effect between different test condition parameters. The interaction variable between process node and test temperature can capture the nonlinear influence of advanced process technology on temperature sensitivity, while the interaction variable between package type and operating voltage can reflect the dependence of different package structures on voltage stability. By constructing such a feature set of combined features, the machine learning model's ability to understand the complex interactions of test conditions can be enhanced, enabling the prediction model to more accurately fit the nonlinear relationships in the actual test environment. This reduces the prediction error of the XGBoost model and improves prediction accuracy compared to traditional single-feature methods.

[0023] Step S130: The machine learning model is trained using a feature set. The training process includes tuning the model parameters based on historical prediction errors to optimize the model's prediction accuracy.

[0024] During machine learning model training, the constructed feature set is first divided into a training set and a validation set, and the model performance is evaluated using a 10-fold cross-validation method. The machine learning model is one of the following algorithms: XGBoost, Random Forest, or Neural Network. If XGBoost is chosen to build the prediction model, this algorithm iteratively generates a combination of decision trees using a gradient boosting framework. In each iteration, the model parameters are updated based on the negative gradient direction, and the initial hyperparameter values ​​are configured as follows: maximum tree depth of 6, learning rate of 0.1, and subsampling ratio of 0.8. During training, parameter tuning is performed based on historical prediction errors. Specifically, this includes calculating the mean absolute error and root mean square error on the validation set. When the error decreases less than a certain value over multiple iterations, the number of trees and the learning rate are automatically adjusted, and a grid search method is used to find the optimal hyperparameter combination within a preset parameter space. The preset parameter space for XGBoost includes: maximum tree depth range [3, 10], learning rate range [0.01, 0.3], and subsampling ratio range [0.6, 1.0]. The grid search step size is set to 0.05. The tuning trigger condition for historical prediction errors is: when the mean absolute error on the validation set decreases by less than 1 second for five consecutive iterations, grid search recalibration is automatically triggered. Dynamic parameter optimization allows the model to continuously adapt to changes in the testing environment, and the XGBoost algorithm's ability to handle nonlinear relationships between features improves prediction accuracy.

[0025] Step S140: Use the trained machine learning model to predict the FT test time and output the predicted test time value. In FT test time prediction, the number of test vectors, test equipment specifications, process nodes, package types, test temperatures, and operating voltages collected in real time are input into a trained machine learning model. The model outputs the corresponding predicted test time values ​​through forward computation. Subsequently, a test scheduling scheme is generated based on the predicted test time values, including optimizing test scheduling using heuristic algorithms. The predicted values ​​are input into the scheduling optimization module, and a genetic algorithm is used as a heuristic algorithm for test scheduling optimization. Specifically, this includes: establishing a task allocation model based on the predicted time and equipment capacity; iteratively generating a scheduling scheme through selection, crossover, and mutation operations. The parameters of the genetic algorithm are set as follows: population size 100, crossover rate 0.8, mutation rate 0.1, and maximum number of iterations 500. The selection operation uses roulette wheel selection, the crossover operation uses single-point crossover, and the mutation operation uses random exchange mutation.

[0026] The test scheduling scheme also includes dynamically adjusting the allocation of test resources based on the real-time status of the test equipment. The scheduling process acquires the working status data of each test equipment in real time, and dynamically reallocates the test tasks when a test equipment failure or task backlog is detected.

[0027] The test scheduling scheme also includes a multi-objective optimization framework. This framework defines multiple optimization objectives and uses a genetic algorithm to generate a Pareto optimal solution set. The optimization objectives include minimizing the total test duration, maximizing machine utilization, and balancing the test load. A multi-objective optimization framework is constructed that includes minimizing the total test duration, maximizing machine utilization, and balancing the test load. A fast non-dominated sorting genetic algorithm with an elitist retention strategy is used to generate the Pareto optimal solution set, and a compromise solution is selected from the solution set through a membership function. The multi-objective optimization framework also includes a dynamic scheduling adjustment mechanism. This mechanism involves real-time monitoring of machine status changes and triggering dynamic rescheduling. Dynamic rescheduling uses a reinforcement learning algorithm to adjust the test sequence and resource allocation based on real-time feedback to minimize the impact of scheduling interruptions. The reinforcement learning algorithm uses Q-learning. The state space is defined as machine state (idle, busy, faulty) and task queue length (discreteized into 10 levels). The action space is the task reassignment instructions (such as moving a task from a faulty machine to an idle machine). The reward function is designed as: Reward = Number of completed tasks × 10 - Machine idle time × 0.1 - Number of scheduling interruptions × 5. The learning rate is set to 0.1, the discount factor to 0.9, and the initial exploration rate to 0.5, which decays over time. A reinforcement learning dynamic scheduling mechanism based on Q-learning is established. This mechanism defines the state space as machine status and task queues, the action space as task reassignment instructions, and the reward function as the reduction in production capacity loss. Online rescheduling is triggered when machine anomalies or emergency orders are detected. Prediction results and scheduling schemes are transmitted to the manufacturing execution system via a RESTful API interface. Data is encapsulated in JSON format and bidirectional communication is achieved via HTTP. This prediction and scheduling process forms a closed-loop control from time prediction to resource allocation. It balances production efficiency and resource consumption through multi-objective optimization, enhances the system's anti-interference capability through dynamic scheduling, and automates production line-level decision-making through system integration. This results in a reduction in total testing time, improved machine utilization, and ultimately, improved testing efficiency through the synergistic effect of prediction and scheduling.

[0028] The FT test time prediction method in this embodiment collects historical test data including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage to construct a feature set reflecting the coupling relationship between test conditions. It employs a machine learning model training method that optimizes parameters based on historical prediction errors, effectively overcoming the limitations of traditional linear estimation methods in handling complex nonlinear couplings of multiple factors. Simultaneously, the feature set construction transforms previously isolated test condition parameters into correlated input features, enabling the machine learning model to learn multi-dimensional coupling patterns such as the interaction between process node and test temperature, and the matching characteristics between package type and operating voltage, thereby improving test time prediction accuracy. The parameter optimization mechanism during model training further enhances the model's adaptability to changes in test conditions. By continuously optimizing model parameters, the prediction error converges, ultimately achieving high-precision FT test time prediction output.

[0029] like Figure 2 As shown, in one embodiment, the model parameters are tuned based on historical prediction errors, including the following steps: Step S210: Calculate the deviation between the historical prediction error and the preset threshold, and trigger the recalibration of the model parameters when the deviation exceeds the threshold. The recalibration includes adjusting the hyperparameters of the machine learning model to minimize the cumulative effect of the prediction error.

[0030] By calculating the deviation between the historical prediction error generated on the validation set and the preset threshold, when the mean absolute error exceeds 8 seconds for three consecutive training cycles or the root mean square error is consistently higher than 9 seconds, the model recalibration mechanism is automatically triggered. The recalibration process specifically includes: using a grid search method to systematically adjust the maximum depth of the XGBoost model's trees, the learning rate, and the subsampling ratio within the preset parameter space; evaluating the prediction performance of each parameter combination through ten-fold cross-validation; and selecting the optimal parameter combination that minimizes the validation set error.

[0031] Step S220: Continuously collect newly generated test data and incrementally train the model. The incremental training uses the sliding window method to update the feature set, retaining only the data within the most recent preset time period to reduce computational complexity.

[0032] Newly generated test data is continuously collected based on an online learning mechanism. A sliding window technique is used to retain the test records of the most recent 30 days to form an incremental training sample set. The sliding window size is dynamically adjusted according to the frequency of test data: for production lines with more than 1000 test data points per day, the window size is set to 7 days; for production lines with less than 100 data points per day, the window size is set to 90 days. Incremental training uses an online gradient descent algorithm. Each update calculates the gradient only based on the new data within the window, with a learning rate set to 0.01. An incremental update is automatically triggered every 24 hours. During each incremental training, the gradient is calculated and the model weights are updated only based on the newly added data. A model performance monitoring mechanism is set up to trigger a complete retraining process when the prediction error continuously exceeds a warning value.

[0033] The parameter tuning scheme in this embodiment effectively addresses the performance degradation of the model caused by changes in test conditions through dynamic error monitoring and parameter adjustment. The incremental training controlled by sliding window ensures the timeliness of the model while controlling the consumption of computing resources, so that the model can still maintain a stable mean absolute error within a small range after changes in the production line environment. Moreover, the computation time required for the recalibration process is significantly reduced compared to complete retraining, ultimately achieving a synergistic improvement in the model's prediction accuracy and adaptive capability.

[0034] The FT test time prediction device provided by the present invention is described below. The FT test time prediction device described below can be referred to in correspondence with the FT test time prediction method described above.

[0035] like Figure 3 As shown, in one embodiment, an FT test time prediction device includes a test data acquisition module 310, a test feature extraction module 320, a model training module 330, and a prediction value output module 340.

[0036] The test data acquisition module 310 is used to collect historical test data including the number of test vectors, test equipment specifications, process nodes, package types, test temperatures, and operating voltages.

[0037] The test feature extraction module 320 is used to extract multiple test features from historical test data and construct a feature set. The test features are constructed based on historical test data and reflect the coupling relationship between test conditions.

[0038] The model training module 330 is used to train a machine learning model using a feature set. The training process includes tuning the model parameters based on historical prediction errors to optimize the model's prediction accuracy.

[0039] The prediction output module 340 is used to predict the FT test time using the trained machine learning model and output the predicted test time value.

[0040] Figure 4 This example illustrates a schematic diagram of the physical structure of an electronic device, which can be a smart terminal. Its internal structure diagram can be as follows: Figure 4 As shown. The electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an FT test time prediction method, which includes: Collect historical test data including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage; Extract multiple test features from historical test data and construct a feature set. The test features are constructed based on historical test data and reflect the coupling relationship between test conditions. The machine learning model is trained using a feature set. The training process includes tuning the model parameters based on historical prediction errors to optimize the model's prediction accuracy. The trained machine learning model is used to predict the FT test time, and the predicted test time value is output.

[0041] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the electronic device to which the present invention is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0042] On the other hand, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements a Fourier Transition (FT) test time prediction method, the method comprising: Collect historical test data including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage; Extract multiple test features from historical test data and construct a feature set. The test features are constructed based on historical test data and reflect the coupling relationship between test conditions. The machine learning model is trained using a feature set. The training process includes tuning the model parameters based on historical prediction errors to optimize the model's prediction accuracy. The trained machine learning model is used to predict the FT test time, and the predicted test time value is output.

[0043] In another aspect, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, it implements an FT test time prediction method, the method comprising: Collect historical test data including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage; Extract multiple test features from historical test data and construct a feature set. The test features are constructed based on historical test data and reflect the coupling relationship between test conditions. The machine learning model is trained using a feature set. The training process includes tuning the model parameters based on historical prediction errors to optimize the model's prediction accuracy. The trained machine learning model is used to predict the FT test time, and the predicted test time value is output.

[0044] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0045] By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0046] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0047] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method of FT test time prediction, characterized by, The method includes: Collect historical test data including the number of test vectors, test equipment specifications, process node, package type, test temperature, and operating voltage; Multiple test features are extracted from the historical test data and a feature set is constructed. The test features are constructed based on the historical test data and reflect the coupling relationship between test conditions. The machine learning model is trained using the feature set, wherein the training process includes tuning the model parameters based on historical prediction errors to optimize the model's prediction accuracy. The trained machine learning model is used to predict the FT test time, and the predicted test time value is output.

2. The FT test time prediction method of claim 1, wherein The extraction of multiple test features from the historical test data includes: Construct at least one of the following: a combination variable of process node and test temperature, and a combination variable of package type and operating voltage, to strengthen the coupling relationship between the test conditions.

3. The FT test time prediction method according to claim 1, characterized in that, The machine learning model is one of the following: XGBoost algorithm, random forest algorithm, or neural network algorithm.

4. The FT test time prediction method according to claim 1, characterized in that, The method also includes generating a test scheduling scheme based on the predicted test time value, including optimizing test scheduling using heuristic algorithms.

5. The FT test time prediction method according to claim 4, characterized in that, The generated test scheduling scheme also includes dynamically adjusting the allocation of test resources based on the real-time status of the equipment.

6. The FT test time prediction method according to claim 4, characterized in that, The generated test scheduling scheme also includes a multi-objective optimization framework, which includes defining multiple optimization objectives and using a genetic algorithm to generate a Pareto optimal solution set. The optimization objectives include minimizing the total test duration, maximizing machine utilization, and balancing the test load.

7. The FT test time prediction method according to claim 6, characterized in that, The multi-objective optimization framework also includes a dynamic scheduling adjustment mechanism, which includes real-time monitoring of machine status changes and triggering dynamic rescheduling. The dynamic rescheduling uses a reinforcement learning algorithm to adjust the test sequence and resource allocation based on real-time feedback to minimize the impact of scheduling interruptions.

8. The FT test time prediction method according to any one of claims 1 to 7, characterized in that, The optimization of model parameters based on historical prediction errors includes: The deviation between historical prediction errors and a preset threshold is calculated, and when the deviation exceeds the threshold, the model parameters are recalibrated. The recalibration includes adjusting the hyperparameters of the machine learning model to minimize the cumulative effect of prediction errors. Newly generated test data is continuously collected and the model is incrementally trained. The incremental training uses a sliding window method to update the feature set, retaining only the data within the most recent preset time period to reduce computational complexity.

9. An FT test time prediction device, characterized in that, The device includes: The test data acquisition module is used to collect historical test data, including the number of test vectors, test equipment specifications, process nodes, package types, test temperatures, and operating voltages. The test feature extraction module is used to extract multiple test features from the historical test data and construct a feature set. The test features are constructed based on the historical test data and reflect the coupling relationship between test conditions. The model training module is used to train the machine learning model using the feature set. The training process includes tuning the model parameters based on historical prediction errors to optimize the model prediction accuracy. The prediction output module is used to predict the FT test time using the trained machine learning model and output the predicted test time value.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the FT test time prediction method according to any one of claims 1 to 8.

Images