Integrated circuit high-temperature aging platform test box thermal runaway self-healing control method and device

Abstract

The embodiment of the application provides a kind of integrated circuit high temperature aging station test box thermal runaway self-healing control method and device, its method includes: using the health state parameter of the key component obtained by history temperature sequence and a plurality, obtain the degradation prediction sequence information of high temperature aging station test box health index in future period and temperature drift amount prediction value;The actual temperature of high temperature aging station test box is acquired, and by comparing the actual temperature with the set temperature set by user, real-time error is obtained;By inputting the real-time error and the degradation prediction sequence information into the model predictive control (MPC) controller based on the first-order inertia-hysteresis transfer function, the anti-drift optimization control amount for actively inhibiting the temperature drift amount about to occur is obtained;According to the anti-drift optimization control amount, the integrated circuit high temperature aging station test box thermal runaway is self-healing controlled.

Description

Technical Field

[0001] This application relates to the field of integrated circuit aging bench technology, and in particular to a method and device for thermal runaway self-healing control of an integrated circuit high-temperature aging bench test chamber. Background Technology

[0002] Aging test benches simulate various real-world usage environments and operating conditions, subjecting products to changes in condition similar to those after prolonged use in a short period, thereby assessing product reliability and durability. Simultaneously, aging testing can also help products quickly overcome the early failure stages of the bathtub curve, reducing the product's failure rate. A key component of the aging test bench, the high-temperature test chamber, provides a constant or adjustable high-temperature environment for accelerated aging testing, evaluating the stability and reliability of electronic products and materials under extreme temperatures, and simulating the impact of long-term high-temperature operation on performance. A malfunction in the high-temperature test chamber can lead not only to temperature runaway, test interruptions, and data distortion, but also potentially to overheating and damaging samples or posing safety risks. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method for controlling thermal runaway self-healing in an integrated circuit high-temperature aging test chamber. One or more embodiments of this application also relate to a device for controlling thermal runaway self-healing in an integrated circuit high-temperature aging test chamber, a computing device, a computer-readable storage medium, and a computer program, to address the technical deficiencies existing in the prior art.

[0004] According to a first aspect of the embodiments of this application, a method for controlling thermal runaway self-healing in an integrated circuit high-temperature aging test chamber is provided, comprising: Acquire the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of several key components used to express the performance degradation information of the high-temperature aging test chamber; Using the historical temperature sequence and the health status parameters of the multiple key components, the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift are obtained in the future. The actual temperature of the high-temperature aging test chamber is obtained, and the real-time error is obtained by comparing the actual temperature with the set temperature set by the user. By inputting the real-time error and the degradation prediction sequence information into the model predictive control (MPC) controller based on the first-order inertial-hysteresis transfer function, an anti-drift optimized control quantity is obtained to actively suppress the temperature drift that is about to occur. The thermal runaway of the high-temperature aging test chamber for integrated circuits is self-healed based on the anti-drift optimization control quantity.

[0005] Preferably, using the historical temperature sequence and the health status parameters of the multiple key components, the degradation prediction sequence information and temperature drift prediction values ​​of the high-temperature aging test chamber's health indicators over a future period are obtained, including: The Long Short-Term Memory (LSTM) network is used to extract the dynamic change patterns of the historical temperature sequence to obtain temperature sequence feature information. The health status parameters of the multiple key components are extracted using a multilayer perceptron (MLP) network to obtain the health status feature information of each key component. The temperature sequence feature information and the health feature information of each key component are fused by an attention mechanism to obtain a fused feature that reflects the overall health status of the high-temperature aging bench test chamber. An LSTM model is constructed and trained, and the fused features are input into the trained LSTM model to obtain the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift over a future period.

[0006] Preferably, the temperature sequence feature information and the health feature information of each key component are fused using an attention mechanism to obtain fused features reflecting the overall health status of the high-temperature aging bench test chamber, including: Using the temperature sequence feature information, the attention weight of temperature is calculated. At the same time, using the health feature information of each key component, the attention weight of the component health status of different key components under different operating conditions and different failure levels on the impact of future temperature drift is calculated. Based on the attention weight of the temperature and the temperature sequence feature information, a weighted temperature feature is calculated, and the overall health status of the component is obtained based on the health feature information of each key component. Based on the weighted temperature characteristics, the attention weight of temperature, the attention weight of component health status, and the overall health status of the components, a fusion characteristic reflecting the overall health status of the high-temperature aging bench test chamber is obtained.

[0007] Preferably, the attention weight for temperature is calculated using the temperature sequence feature information, and the attention weight for component health status is calculated using the health feature information of each key component, including: Using the temperature sequence feature information, the attention score for temperature is calculated, and using the health feature information of each key component, the attention score for the health status of the component is calculated. The attention weight of temperature is calculated using the attention score of temperature, and the attention weight of component health status is calculated using the attention score of component health status.

[0008] Preferably, based on the weighted temperature characteristics, the attention weight of temperature, the attention weight of component health status, and the overall health status of the components, the fused characteristics used to reflect the overall health status of the high-temperature aging bench test chamber include:

[0009] in, Features of fusion; Attention weight for temperature; Weighted temperature characteristics; Attention weights for component health status; This refers to the overall health status of the components.

[0010] Preferably, the LSTM model includes an input layer, three hidden layers, and an output layer. The hidden layers fit a nonlinear relationship using the ReLU activation function, and the output layer outputs a health score in the [0,1] interval using the Sigmoid function.

[0011] Preferably, by inputting the real-time error and the degradation prediction sequence information into a model predictive control (MPC) controller based on a first-order inertial-hysteresis transfer function, the anti-drift optimized control quantity obtained for actively suppressing the impending temperature drift includes: The Model Predictive Control (MPC) controller based on the first-order inertial-hysteresis transfer function obtains an anti-drift optimized control quantity for actively suppressing the impending temperature drift by using online rolling optimization, based on the real-time error and the degradation prediction sequence information, combined with an objective function that includes an added temperature drift penalty term.

[0012] According to a second aspect of the embodiments of this application, a thermal runaway self-healing control device for an integrated circuit high-temperature aging test chamber is provided, comprising: The parameter degradation prediction module is configured to acquire the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of multiple key components used to express the performance degradation information of the high-temperature aging test chamber; using the historical temperature sequence and the health status parameters of the multiple key components, the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift are obtained in the future. The model predictive control module is configured to acquire the actual temperature of the high-temperature aging test chamber and obtain the real-time error by comparing the actual temperature with the set temperature set by the user; by inputting the real-time error and the degradation prediction sequence information into the model predictive control (MPC) controller based on the first-order inertial-hysteresis transfer function, an anti-drift optimized control quantity is obtained to actively suppress the temperature drift that is about to occur. The self-healing control module is configured to perform self-healing control on the thermal runaway of the integrated circuit high-temperature aging test chamber based on the anti-drift optimized control quantity.

[0013] According to a third aspect of the embodiments of this application, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement any of the steps of the thermal runaway self-healing control method for the integrated circuit high-temperature aging test chamber.

[0014] According to a fourth aspect of the present application, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of any one of the integrated circuit high-temperature aging bench test chamber thermal runaway self-healing control methods.

[0015] According to a fifth aspect of the present application, a computer program is provided, wherein when the computer program is executed in a computer, the computer is instructed to perform the steps of the above-described integrated circuit high-temperature aging bench test chamber thermal runaway self-healing control method.

[0016] The thermal runaway self-healing control scheme for an integrated circuit high-temperature aging test chamber provided in this application embodiment first obtains the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of multiple key components used to express the performance degradation information of the high-temperature aging test chamber. Using the historical temperature sequence and the health status parameters of the multiple key components, a degradation prediction sequence information and a predicted temperature drift value for the health indicators of the high-temperature aging test chamber over a future period are obtained. Then, the actual temperature of the high-temperature aging test chamber is obtained, and the real-time error is obtained by comparing the actual temperature with the user-set temperature. The real-time error and the degradation prediction sequence information are input into a Model Predictive Control (MPC) controller based on a first-order inertial-hysteresis transfer function to obtain an anti-drift optimization control quantity for actively suppressing the impending temperature drift. Finally, self-healing control of the thermal runaway of the integrated circuit high-temperature aging test chamber is performed based on the anti-drift optimization control quantity. That is, by quantifying the temperature drift risk in advance through degradation prediction and embedding drift penalties in MPC rolling optimization, active compensation and high-precision self-healing control before temperature deviation are achieved. Attached Figure Description

[0017] Figure 1 This is a flowchart of a thermal runaway self-healing control method for an integrated circuit high-temperature aging test chamber provided in one embodiment of this application; Figure 2 This is a general block diagram of the thermal runaway self-healing control system of an integrated circuit high-temperature aging test chamber provided in one embodiment of this application; Figure 3 This is a flowchart of a parameter degradation prediction method based on multi-component health status assessment provided in one embodiment of this application; Figure 4 This is a schematic diagram of the Model Predictive Control (MPC) principle provided in one embodiment of this application; Figure 5 This is a multilayer perceptron (MLP) network diagram provided in one embodiment of this application; Figure 6 This is a schematic diagram of a thermal runaway self-healing control device for an integrated circuit high-temperature aging test chamber provided in one embodiment of this application. Detailed Implementation

[0018] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0019] This application provides a method for controlling thermal runaway self-healing in an integrated circuit high-temperature aging test chamber. This application also relates to a device for controlling thermal runaway self-healing in an integrated circuit high-temperature aging test chamber, a computing device, a computer-readable storage medium, and a computer program, which will be described in detail in the following embodiments.

[0020] (1) This application is based on parameter degradation prediction of multi-component health status assessment: This system not only utilizes the strong time-series signal of historical temperature sequence, but also takes the health status of key execution and structural components such as heating wire, cooling fan, and housing seal as important inputs to the prediction model.

[0021] Specifically, for temperature sequences with strong temporal correlation, a Long Short-Term Memory (LSTM) network is used for modeling to capture their long-term dependencies. This involves using historical temperature data to build a mathematical model that quantitatively estimates the degradation trends and failure thresholds of certain key parameters of the test chamber over time at a future point in time. By comparing and analyzing predicted temperature, target temperature, and real-time temperature, a temperature degradation prediction is output, allowing for optimization of maintenance cycles or design improvements. For component health indicators with weak temporal correlation or event-triggered characteristics (such as slight changes in heating wire resistance, fan vibration characteristics, and seal aging cycles), a Multilayer Perceptron (MLP) is used for feature extraction. Building upon this, an attention mechanism is innovatively introduced to dynamically calculate the weights of different components' impact on future temperature drift under varying degrees of failure during the information fusion stage. This enables the prediction model to more accurately quantify the temperature drift risk caused by the degradation of specific components, providing a basis for targeted feedforward compensation.

[0022] (2) This application further integrates degradation prediction information under the model predictive control (MPC) framework: the online estimated drift amount is injected into the current initial temperature value in real time to form the corrected state; then the discrete first-order hysteresis temperature chamber model is used to recursively deduce the future multi-step temperature trajectory, and a drift penalty term is added to the cost function of rolling optimization, so that the control sequence obtained by the solution actively suppresses the temperature drift that is about to occur while tracking the set value, thereby realizing the advance adjustment of the actuator and the online self-healing of temperature drift.

[0023] (3) This application takes the parameter degradation prediction results as the guide, first performs drift correction on the current temperature within the model predictive control (MPC) framework, then uses the discrete first-order hysteresis temperature chamber model to generate the future time domain temperature trajectory, and embeds a drift penalty term in the cost function so that the rolling optimization process considers the future deviation simultaneously, thereby completing adaptive compensation before the temperature drift becomes apparent, and realizing online self-healing and high-precision tracking of temperature drift in the high-temperature test chamber.

[0024] Figure 1 This application provides a flowchart of a method for controlling thermal runaway and self-healing in an integrated circuit high-temperature aging test chamber, which includes the following steps: Step S101: Obtain the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of several key components used to express the performance degradation information of the high-temperature aging test chamber; Step S102: Using the historical temperature sequence and the health status parameters of the multiple key components, obtain the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift in the future period. In one embodiment of this application, using the historical temperature sequence and the health status parameters of the multiple key components, the degradation prediction sequence information and temperature drift prediction values ​​of the health indicators of the high-temperature aging bench test chamber over a future period are obtained, including: The Long Short-Term Memory (LSTM) network is used to extract the dynamic change patterns of the historical temperature sequence to obtain temperature sequence feature information. The health status parameters of the multiple key components are extracted using a multilayer perceptron (MLP) network to obtain the health status feature information of each key component. The temperature sequence feature information and the health feature information of each key component are fused by an attention mechanism to obtain a fused feature that reflects the overall health status of the high-temperature aging bench test chamber. An LSTM model is constructed and trained, and the fused features are input into the trained LSTM model to obtain the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift over a future period.

[0025] In one embodiment of this application, the temperature sequence feature information and the health feature information of each key component are fused using an attention mechanism to obtain fused features reflecting the overall health status of the high-temperature aging bench test chamber, including: Using the temperature sequence feature information, the attention weight of temperature is calculated. At the same time, using the health feature information of each key component, the attention weight of the component health status of different key components under different operating conditions and different failure levels on the impact of future temperature drift is calculated. Based on the attention weight of the temperature and the temperature sequence feature information, a weighted temperature feature is calculated, and the overall health status of the component is obtained based on the health feature information of each key component. Based on the weighted temperature characteristics, the attention weight of temperature, the attention weight of component health status, and the overall health status of the components, a fusion characteristic reflecting the overall health status of the high-temperature aging bench test chamber is obtained.

[0026] In one embodiment of this application, the attention weight for temperature is calculated using the temperature sequence feature information, and the attention weight for the health status of each key component is calculated using the health feature information of each key component, including: Using the temperature sequence feature information, the attention score for temperature is calculated, and using the health feature information of each key component, the attention score for the health status of the component is calculated. The attention weight of temperature is calculated using the attention score of temperature, and the attention weight of component health status is calculated using the attention score of component health status.

[0027] In one embodiment of this application, based on the weighted temperature characteristics, the attention weight of temperature, the attention weight of component health status, and the overall health status of the components, a fused feature reflecting the overall health status of the high-temperature aging bench test chamber is obtained, including:

[0028] in, Features of fusion; Attention weight for temperature; Weighted temperature characteristics; Attention weights for component health status; This refers to the overall health status of the components.

[0029] In one embodiment of this application, the LSTM model includes an input layer, three hidden layers and an output layer, wherein the hidden layers fit a nonlinear relationship through the ReLU activation function, and the output layer outputs a health score in the [0,1] interval through the Sigmoid function.

[0030] Step S103: Obtain the actual temperature of the high-temperature aging test chamber, and obtain the real-time error by comparing the actual temperature with the set temperature set by the user; Step S104: By inputting the real-time error and the degradation prediction sequence information into the model predictive control (MPC) controller based on the first-order inertia-hysteresis transfer function, an anti-drift optimized control quantity is obtained to actively suppress the temperature drift that is about to occur. Step S105: Perform self-healing control on the thermal runaway of the integrated circuit high-temperature aging test chamber according to the anti-drift optimization control quantity.

[0031] In one embodiment of this application, by inputting the real-time error and the degradation prediction sequence information into a model predictive control (MPC) controller based on a first-order inertial-hysteresis transfer function, an anti-drift optimized control quantity for actively suppressing the impending temperature drift is obtained, including: The Model Predictive Control (MPC) controller based on the first-order inertial-hysteresis transfer function obtains an anti-drift optimized control quantity for actively suppressing the impending temperature drift by using online rolling optimization, based on the real-time error and the degradation prediction sequence information, combined with an objective function that includes an added temperature drift penalty term.

[0032] Figure 2 This is a block diagram of the thermal runaway self-healing control system for an integrated circuit high-temperature aging test chamber provided in one embodiment of this application, as shown below. Figure 2 As shown, the overall system block diagram is mainly divided into two layers from top to bottom, forming a composite control architecture that combines feedforward compensation and feedback regulation, aiming to achieve self-healing control of temperature runaway in the high-temperature test chamber.

[0033] The upper layer is a temperature degradation feedforward compensation channel, responsible for proactively compensating for system performance degradation. The feedforward function cancels out potential future drift before the temperature deviates significantly from the set value, thereby significantly shortening the system's response time and achieving self-healing preparedness. First, the temperature sensor is the first step in ensuring system reliability. Its core function is to operate stably and accurately in extreme high-temperature environments for a long time, collecting the most realistic temperature data in the test chamber in real time, providing a highly reliable data foundation for the entire prediction and control loop. Second, the degradation prediction module receives continuous historical temperature sequences from the temperature sensor in real time and simultaneously collects health status parameters of key components such as heating wire resistance and fan vibration. By integrating algorithms of Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP) networks and introducing an attention mechanism for dynamic weighting, this module can accurately quantify the temperature drift risk at a specific future moment and output a predicted temperature drift value. This predicted value is directly superimposed on the output of the lower-level MPC controller after feedforward gain, constituting advance compensation for inertial hysteresis.

[0034] The lower layer is the Model Predictive Control (MPC) closed loop, such as... Figure 4 As shown, the system is responsible for precise tracking and robust adjustment, ensuring the real-time performance and adaptability of the control. The actual temperature of the "high-temperature test chamber" is measured and fed back in real time by the "temperature sensor," and compared with the "set temperature" set by the user at a comparison point to obtain the real-time error e(t). This error signal, along with degradation prediction information from the upper-level channel, is sent to the "MPC controller" module. This module optimizes the control strategy online based on the current temperature error, the rate of change of error, and the degradation prediction results. The controller performs rolling optimization based on the established discrete first-order hysteresis model of the high-temperature test chamber: it not only considers the tracking deviation of the current and future setpoints, but also embeds a "drift penalty term" in the objective function to actively suppress the predicted temperature drift. By solving this optimization problem online, the MPC controller dynamically generates the optimal feedback control sequence and outputs the control quantity, so that the control system can always maintain fast, accurate, and stable adjustment performance even when the thermodynamic characteristics of the temperature chamber drift with time or operating conditions.

[0035] Ultimately, the control quantity u(t) = u_{MPC} output by the MPC controller serves as a signal, forming a control command to drive the actuator, namely the combination of heating power and fan speed. This combination acts simultaneously on the heating wire and the fan actuator, achieving a triple synergy of "prediction (degradation feedforward) - compensation (feedforward superposition) - adaptation (MPC optimization)". Through the coordinated control of heating and air cooling, the system can more precisely and energy-efficiently suppress temperature drift, enabling the high-temperature test chamber to complete self-healing correction before temperature drift occurs, balancing speed, robustness, and energy efficiency.

[0036] The thermal runaway self-healing control scheme for the integrated circuit high-temperature aging test chamber in this application specifically includes two parts: Part 1: Prediction of parameter degradation.

[0037] Parameter degradation prediction anticipates the risk of temperature runaway by setting alarm thresholds to achieve lifespan prediction. For example... Figure 3As shown, firstly, feature information expressing the performance degradation information of the high-temperature test chamber is extracted (here, temperature and health status parameters of other key components, such as resistance value, vibration signal, and sealing index). Secondly, the main degradation trends and fuzzy degradation boundaries are extracted through trend filtering and fuzzy information granulation. An LSTM network is used to process the strong temporal data of temperature, and an MLP network is used to process the weak temporal component parameters to extract their features. Then, an attention mechanism is introduced to dynamically fuse the above heterogeneous features, accurately quantifying the impact weight of different component degradation on the system. Finally, based on the fused features, trend extraction and LSTM prediction are performed to output the future degradation trend and range. Finally, a dynamic LSTM model is used for comprehensive performance degradation prediction, giving the main degradation trend and fuzzy degradation range of the equipment at future moments. The specific process includes: The first step is data acquisition. Historical temperature time series are obtained through seamless monitoring. Simultaneously, health status parameters of key components such as heating wires, cooling fans, and enclosure seals are collected. Specifically, the data collected includes: minute changes in the resistance of the heating wires (reflecting oxidation and aging), vibration signal characteristics of the cooling fan (such as amplitude and frequency, reflecting bearing wear and dynamic balance), and sealing performance indicators of the enclosure seals (such as leakage rate measured by pressure decay method, reflecting aging and cracking). At the same time, the system also records the optimal control sequence u(k) previously output by the MPC controller (such as heating / cooling power and fan speed commands) as a historical control command sequence for subsequent model updates and degradation correlation analysis. This constructs a raw data stream covering the entire lifecycle of the equipment and the status of multiple components, providing a complete information foundation for subsequent feature extraction.

[0038] The second step is feature extraction. The purpose of feature extraction is to transform the previously extracted raw signals into single, monotonic, dimensionless quantitative health indicators. These features are used to quantify the real-time health status of each component and to provide structured input for subsequent attention fusion and degradation prediction. Here, the spectral distance index method is used to compress the sequence into health indicators; the closer the value is to 1, the healthier it is, and the closer it is to 0, the more severe the degradation.

[0039] Let the power spectral density of the reference vibration signal of a certain device in a healthy state and the power spectral density of the vibration signal monitored at the current moment be respectively... and The power spectral density here is obtained by performing a Fast Fourier Transform (FFT) on the original vibration signal. Its symmetric distance is then:

[0040] This formula ensures that the distance metric is consistent regardless of which signal is used as the reference.

[0041] Here, d is a fundamental asymmetric distance function that quantifies the degree of unidirectional difference between the power spectral densities of two signals through integration. The parameter q is the order of the distance norm, which determines how the distance in two directions is expressed. and It is merged into a single, symmetric scalar value. and Represents two asymmetric distance metrics, respectively indicating from... and from The difference is that the two are usually not equal.

[0042] When q = 1, and hour, J divergence:

[0043] In the formula, It is a symmetric, non-negative scalar that reflects the overall difference between two power spectral densities. The larger the value, the further the health status deviates from the baseline. N represents the Nth group of signals. The subscript IS represents the Itakura-Saito distance function, which is asymmetric, but becomes a symmetric distance metric after being processed by a symmetric formula (such as the J divergence used).

[0044] Here we introduce the J divergence (denoted as J0). The goal is to merge two asymmetric Itakura-Saito distances into a single symmetric comprehensive difference metric, thereby providing a stable and interpretable distance input for the subsequent unified mapping to health scores. Then, the spectral distance function is used to... Health score mapped to the [0,1] interval: Spectral distance function:

[0045] in, Let J be the divergence value of the Nth group of signals. This represents the J-divergence between two signals; The final output health indicator, the closer the value is to 1, the healthier it is; The sensitivity coefficient, determined by the equipment performance degradation trend, serves as one of the inputs for the subsequent "multi-component health feature extraction and fusion." It is input along with temperature sequence features and component health parameters to the attention fusion module to generate fused features reflecting the overall health status of the system. This, in turn, drives the LSTM model to perform degradation prediction.

[0046] Simultaneously, multi-component health feature extraction and fusion are performed. To achieve quantitative assessment of the health status of key components in the high-temperature test chamber, weakly time-series or quasi-static health status parameters such as heating wires (resistance value), fans (vibration signal), and seals (leakage rate) are abstracted using Multi-Level Processing (MLP) and mapped into high-dimensional feature vectors to quantify the real-time health level of each component. The specific steps of MLP are as follows: a. Data preprocessing: Min-Max normalization is performed on heterogeneous parameters such as resistance value and vibration amplitude to eliminate dimensional differences and map them to the [0,1] interval; b. Network architecture: A three-layer fully connected structure is adopted, such as... Figure 5 As shown, the input layer dimension matches the number of parameters, the hidden layer fits the nonlinear relationship through the ReLU activation function, and the output layer outputs a health score in the [0,1] interval through the Sigmoid function (here z is the input value of the activation function, representing the output value of all neurons in the previous layer).

[0047]

[0048] c. Model training: Using mean squared error as the loss function, the weights and biases are iteratively updated through the Adam optimizer to ensure that the predicted values ​​closely match the actual state of the components; d. Real-time inference: Input the preprocessed real-time parameters into the trained model to quickly output quantified health features, providing a foundation for subsequent fusion.

[0049] The third step is attention mechanism fusion, where an attention mechanism is introduced for feature fusion. This mechanism dynamically calculates the correlation weights between temperature sequence features and the health features of various components, automatically identifying which component degradation has a more critical impact on the current system performance under different operating conditions. By incorporating the health of other system components during feature extraction, it improves upon the inaccuracy of LSTM predictions that only consider the temperature sequence. For example, in the high-temperature steady-state phase, the health status of the heating element may dominate; while in the rapid cooling phase, the performance of the fan becomes the main influencing factor. Through weighted fusion, a comprehensive feature vector reflecting the overall health status of the system is generated. The mechanism fusion specifically includes: a. Use This represents the hidden state sequence input after LSTM processes historical temperature data, where T is the number of time steps. It is the dimension of the LSTM hidden state; Similarly, using The feature vector data representing the health status output of the MLP processing components (such as the health baseline offset rate of heating wires, fans, etc.) b. Calculate the attention scores for temperature and component health status separately:

[0050]

[0051] in, , , , , k is a learnable parameter and a hyperparameter.

[0052] c. Using the softmax operation, calculate the attention weights for temperature and component health status, respectively: For temperature time series features, calculate the weight of each time step:

[0053] For the component health status feature, which is itself a comprehensive vector (non-sequence), the global weight is calculated as follows:

[0054] d. Calculate the weighted fusion features: Weighted representation of temperature characteristics:

[0055] Component health status characteristics This is a composite vector extracted from the aforementioned Multilayer Perceptron (MLP). It does not require further weighting and can be directly used with the weight scalars. Scaling is applied for blending.

[0056] Among them, the weight scalar It is a value calculated by the attention mechanism, representing the overall health status of the components under the current system state. This indicates the degree of importance for predicting temperature drift.

[0057] The final fusion characteristics are:

[0058] in, This is used to balance the overall contribution of temperature features in the fusion process. Here, a comprehensive feature vector that fully reflects the overall health of the system is obtained, which will serve as the direct input to the subsequent LSTM prediction model.

[0059] The fourth step is dynamic LSTM modeling. A single-layer Long Short-Term Memory (LSTM) network is constructed, which, after being fused with the attention mechanism from the third step, forms a comprehensive feature vector that fully reflects the temporal characteristics of the temperature in the high-temperature test chamber and the health status of key components (heating wire, fan, and seals). As input, the LSTM model is used for comprehensive performance degradation prediction. The output of the LSTM model is a predicted sequence of health indicators (or temperature drift) of the high-temperature test chamber over a future period, which is denoted as the comprehensive health indicator sequence. A single LSTM block consists of three gate structures: a forget gate, an input gate, and an output gate, along with a cell state. The cells at each time step are updated according to the following equation: a) The Gate of Oblivion:

[0060] in , Here are the weight matrix and bias terms for the forget gate. This is the sigmoid function. This is the output of the previous layer. This is the data for this moment.

[0061] This layer performs data elimination and selection, partially reducing the input data.

[0062] b) Input gate:

[0063] in , Here are the weight matrix and bias terms of the input gate.

[0064] According to this formula, some memories are discarded, while others are preserved and updated.

[0065]

[0066] in , The weight matrix and bias terms are used when updating information; It is the hyperbolic tangent function.

[0067] The formula is used to calculate the alternative content to be updated.

[0068] Information from the previous moment This process is passed through this structure and ultimately output to the next time step. The state is then updated according to the following formula, allowing long-term memory and current memory to combine and form a new unit state: .

[0069] c) Output gate:

[0070] in , Here are the weight matrix and bias terms for the output gate. The above formula determines which states are output, and the new unit states are then calculated using the following formula:

[0071] in, Output the new cell state.

[0072] For example, if the current state The code encodes two messages: "high temperature" and "insufficient fan speed," while the output gate... If the current decision relies more on the "fan speed" information, it will assign a higher weight to that information, thus affecting the final output. It mainly reflects the impact of fan status on temperature.

[0073] Subsequently, a digital set is constructed using a "multi-step input - single-step output" approach, and a rolling retraining strategy is employed to reduce errors. The spectral distance values ​​of the previous n time steps are taken as the input vector, and the predicted value of the next time step is output. The prediction results of each round are incorporated into the training set, and the network parameters are immediately retrained to suppress error accumulation. Thus, while maintaining the nonlinear modeling capability of LSTM, it dynamically adapts to degradation evolution and significantly reduces medium- and long-term prediction drift.

[0074] Next, the comprehensive health indicator sequence is generated from the final output of the LSTM. The process separates a definite main degradation trend from uncertain fluctuation residuals. This is primarily achieved by first performing trend filtering, followed by fuzzy information granulation of the filtered residuals. Furthermore, to extract a smooth main degradation trend, the obtained index sequence... The Λ-trend filter is applied to solve the convex optimization problem. The specific formula is as follows:

[0075] in, This is the observation sequence (i.e., the comprehensive health index sequence). The regularization parameter λ for the smooth trend sequence (i.e., the main degenerate trend) is a non-negative parameter used to control the smoothness of x and the size of the balance remainder. The weighted objective function is a strictly convex function with respect to x, so it has only one minimum value, and the obtained optimal solution... This refers to the dominant degradation trend, where n is the total length of the observation sequence (i.e., n+1 times from t=0 to t=n), and t is the time step number (t=0,1,…,n), representing the discrete time position index in the sequence.

[0076] Furthermore, to quantify and characterize the uncertainty inherent in the residual sequence after trend filtering, the residual sequence is segmented according to a window length w=10, and a triangular fuzzy particle (a,m,b) is constructed for each segment, where a is the minimum value, m is the mean value, and b is the maximum value; the membership functions are as follows:

[0077] The minimum and maximum values ​​after fuzzification represent the fuzzy boundaries of the degraded information. This separates the "main trend" from the "fluctuation range," characterizing the overall degradation path and quantifying the uncertainty of prediction.

[0078] Step 5: Output the results.

[0079] Output the main degradation trend sequence, and output the fuzzy upper and lower boundary sequences to form a confidence interval. Simultaneously, calculate the root mean square according to the following formula. The root mean square (RMSE) is used to evaluate the degradation effect; the smaller the RMSE value, the better the prediction effect.

[0080]

[0081] in: The value at time i is the predicted sequence of health indicators (or temperature drift) of the high-temperature test chamber over a future period of time output by the LSTM model. is the actual observed value at the corresponding time (such as measured health indicators or actual temperature drift); n is the number of samples within the evaluation period.

[0082] This approach not only provides a dual basis for decision-making in health management—point estimation and interval estimation—but also offers a reliable error metric.

[0083] Part 2: Temperature prediction model based on first-order inertial-hysteresis transfer function.

[0084] To enable rapid, accurate, and online prediction of the future temperature evolution of a high-temperature test chamber within a model predictive control framework, a continuous transfer function with "first-order inertia + pure time delay" is used to model the chamber's thermal dynamics. Furthermore, a recursive multi-step prediction equation is established through discretization using the forward Euler method. The specific steps are as follows: Step 1: Continuous domain model structure; The temperature chamber is considered a single-input, single-output thermal system, with its input being the percentage of rated heating / cooling power u(t) (%) and its output being the average operating temperature T(t) (°C). Based on the step response experiment, this process exhibits typical first-order inertial plus pure time-delay characteristics; therefore, the transfer function... Written as

[0085] In the formula: K—static gain (℃ / %), T—time constant (seconds), τ—pure time delay (seconds), and s is a complex variable in the Laplace transform, representing the system's frequency response. The three parameters can be identified using the two-point method through a single offline step response test, without the need for complex modeling.

[0086] Step 2, Discretization Derivation; Let the sampling period be (seconds), the transfer function of the continuous model Discretize using the forward Euler method and calculate the pure time delay. Integer number of beats This yields the embedded programmable difference equation:

[0087] The coefficients are only related to the model parameters: Rounding down. The above formula is the "temperature update law" in discrete state space.

[0088] Step 3: Multi-step prediction and drift penalty term generation; The measured temperature is obtained at time k. Then, using the above formula as the core, control the sequence for the next N steps. By performing recursion, we obtain the temperature prediction sequence based on the first-order inertial-hysteresis model:

[0089]

[0090] in, , These are the discretized model parameters, reflecting the state decay coefficient (related to the inertial time constant), control gain (related to the static gain), and hysteresis steps (the number of hysteresis steps after discretization, representing the delay of the control action, which takes effect after d sampling periods). For control quantity (heating / cooling power); This represents the predicted temperature at time k.

[0091] , The specific formula is:

[0092]

[0093] ) in, The time constant (in seconds) for a continuous system. Sampling period (seconds); Static gain (°C / %) The pure time delay (in seconds) for a continuous system.

[0094] Simultaneously, a comprehensive health indicator prediction sequence with values ​​ranging from (0, 1) is obtained from the LSTM model output obtained from the "parameter degradation prediction". This reflects the impact of the system's health status on future temperature drift.

[0095] Here, the health indicators output by the LSTM model are converted into temperature drift, making them comparable to those predicted by the physical model. To perform weighted fusion under the same unit (°C), first define the mapping function:

[0096] in >0 is the drift ratio coefficient, which represents the maximum temperature drift (unit: °C) that may be caused when the health level drops to 0. It can be obtained by calibration through historical data.

[0097] Subsequently, a weighted fusion strategy was employed to generate a comprehensive temperature prediction sequence combining the output of the LSTM model.

[0098] in: and For adjustable fusion weights, and satisfying =1; The specific values ​​for these weights can be adjusted based on the system's historical performance, model confidence, or user preferences. When the system is running stably and the model confidence is high, the weights should be increased. When a significant decline in component health or an increase in historical drift error is detected, improve... .

[0099] To further suppress temperature drift, a "drift penalty term" is added to the classic MPC objective function, which is then applied to the fused prediction sequences. Substitute into the MPC objective function:

[0100] The three terms on the right side of the above equation are as follows: a. Setpoint deviation penalty – ensures the predicted trajectory tracks the setpoint temperature; The target temperature (°C) is set. The above weighted and fused temperature prediction results (°C) are shown. The tracking error weighting coefficient (positive number) indicates that the larger the value, the higher the required tracking accuracy at that moment. To predict the time domain length (steps); b. Control incremental penalties – avoid frequent actuator actions; Changes in adjacent control quantities; To control the incremental weighting coefficient (positive number), the larger the value, the stronger the suppression of actuator action; To predict the time domain length (steps); c. Temperature drift penalty — To predict the interstep temperature change, Its weighting coefficient (positive number) is larger, and the more sensitive it is to future drift, the more reverse correction can be applied before the drift becomes apparent.

[0101] Q, R, and P are coefficient matrices used to balance control performance and control quantity. By adjusting the relative values ​​of Q, R, and P, a flexible trade-off can be made between tracking performance, actuator lifespan, and drift suppression. An MPC objective function with a drift penalty term is added. It is a quadratic cost function that can be solved analytically or quickly calculated using the standard QP library, making it suitable for embedded implementations.

[0102] Step 4: Feedback and correction; To overcome model mismatch and external disturbances, a "one-step error compensation" strategy is introduced:

[0103] This error will be added to all future forecasts:

[0104] Corrected As an output constraint benchmark for subsequent rolling optimization problems, it can significantly improve long-term prediction accuracy.

[0105] Step 5: Online update mechanism.

[0106] Calculate the cumulative average prediction error: ; When the cumulative average prediction error exceeds a set threshold, model re-identification is triggered: repeat the small-amplitude step experiment and recursively update. This ensures that the prediction model always matches the actual thermal properties.

[0107] Through the above steps, a temperature prediction scheme with a complete process of "identification-discretion-recursion-correction-update" was established. It has the advantages of simple structure, small amount of calculation and easy programming implementation. It can directly serve the subsequent MPC rolling optimization module and achieve early suppression and self-healing of temperature drift caused by inertia and hysteresis in high temperature test chambers.

[0108] In summary, the temperature prediction model and its rolling optimization mechanism established in this application are the core execution links for realizing the self-healing control of temperature runaway in high-temperature test chambers. Its output directly reflects the self-healing function in the following two aspects: First, the forward-looking prediction trajectory serves as the basis for self-healing decisions: the model, through recursive calculation, obtains a multi-step temperature prediction sequence, which can anticipate temperature drift trends that may be caused by system inertia, hysteresis, and component degradation, enabling the control system to "prevent problems before they occur." Second, the optimized control quantity embedded with drift penalty serves as the execution action for self-healing: a "drift penalty term" is added to the objective function. The optimization solver obtains the optimal control sequence by minimizing this objective function J. It no longer simply tracks the setpoint, but includes feedforward compensation commands to actively counteract predicted drift. This control quantity acts on the actuator (heating wire, fan), which can apply reverse adjustment before drift actually occurs, thereby stabilizing the temperature near the setpoint. This is the essence of "online self-healing".

[0109] In summary, the self-healing control process of this application presents a closed loop: the model-based prediction results (output) are used for optimization calculations, the optimized control actions (results) then act back on the controlled object, and finally, a closed loop is formed through sensor feedback. Therefore, the high-precision prediction and anti-drift optimization control output of this model together constitute the direct technical carrier of temperature runaway self-healing control.

[0110] It should be noted that the thermal runaway self-healing control method for an integrated circuit high-temperature aging test chamber provided in this application embodiment includes, before obtaining the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of multiple key components used to express the performance degradation information of the high-temperature aging test chamber, a health assessment of the integrated circuit high-temperature aging test chamber, which specifically includes: Step 101: Determine the health value of each module in the test chamber, and calculate the degradation rate of each module based on the health value.

[0111] In one optional embodiment, the method for determining the health value of each module in the test chamber and calculating the corresponding degradation rate of each module based on the health value can be as follows: for each module in the test chamber, determine the difference between the current health value of the module and the health value collected in the previous time window; and determine the degradation rate of the module by the ratio of the difference to the duration of the collection time window.

[0112] In actual implementation, the quantitative health index, i.e. the health value, at a certain moment can be obtained from the independent health assessment algorithm of each module. Assuming that at time t, the normalized health index is obtained from the independent health status algorithm of each module, with a range of [0,1], where 1 indicates that the module health status is optimal.

[0113] Taking the test chamber, which includes a heating module, a fan module, and a sealing module, as an example, the health values ​​of the three modules are obtained as follows. The health indicators of the three modules are represented as follows, and the values ​​corresponding to the three health indicators are the health values.

[0114] Heating module health status:

[0115] Fan module health status:

[0116] Sealed module health status:

[0117] To extract more information from the limited data, a degradation gradient is introduced into the input data. A time window Δt is defined, and the degradation rate of each module is calculated. :

[0118] like A value greater than 0 indicates that health is declining.

[0119] In this step, the above formula is used to calculate the degradation rate of each module based on the health value of each module.

[0120] Step 102: Based on the number of modules contained in the test chamber and the influence coupling coefficient between each module, construct the coupling influence relationship matrix.

[0121] The constructed coupling influence matrix is ​​the coupling influence matrix C based on the operating mechanism.

[0122] In one optional embodiment, constructing the coupling influence matrix based on the number of modules contained in the test chamber and the influence coupling coefficient between each module may include the following sub-steps: Sub-step 1: Determine the size of the coupling influence matrix based on the number of modules contained in the test chamber; Wherein, the coupling influence relationship matrix is ​​an N×N matrix, where N is the number of modules; Sub-step 2: Combine the influence coupling coefficients between modules with the correlation relationships between modules in pairs, and use them as elements in the coupling influence relationship matrix.

[0123] In this embodiment, considering that the weight vector is not static but dynamically adjusted according to the real-time health status of each module, a coupling influence matrix C is introduced. Matrix C defines the impact of the performance degradation of module j on the weight of module i. Since the test chamber includes three modules, and the health values ​​of the three modules are inputs, a 3×3 matrix C is constructed, where... The coupling coefficient representing the influence of module j on module i:

[0124] For example This represents the coupling coefficient between the sealing module and the heating module. Due to the strong coupling between the two, it can be assigned a value of 0.8. This is because the failure of the sealing module will lead to an increase in the load on the heating module and a sharp increase in the risk of overload. Therefore, the weight of the heating module must be given special attention.

[0125] This represents the coupling coefficient between the fan module and the heating module, which can be assigned a value of 0.6, because reduced air circulation inside the test chamber can lead to localized overheating.

[0126] , Each can be assigned a value of 0.1, because the failure of the heating module has a relatively small impact on the fan module and the sealing module. , The corresponding influence coupling is weak coupling.

[0127] Step 103: Based on the health value, degradation rate, predefined baseline weight vector, and coupling influence matrix of each module, generate the dynamic weight vector corresponding to the test chamber.

[0128] Most variable weighting methods are simple linear additions. This application uses an exponential state-variable weighting vector algorithm to generate a dynamic weight vector corresponding to the test chamber, so as to achieve an amplification effect for more severe faults.

[0129] Define the baseline weight vector The importance of each module to overall performance under ideal conditions can be determined by using the analytic hierarchy process or by inviting domain experts to score the data, thus obtaining a set of benchmark weight vectors. ={ , , The baseline weights represent the inherent importance of each module. Based on the predefined baseline weight vector, the health values ​​and degradation rates of each module, the dynamic weight vector corresponding to the test chamber is generated.

[0130] In one optional embodiment, the method for generating the dynamic weight vector corresponding to the test chamber based on the health value, degradation rate, predefined baseline weight vector, and coupling influence matrix of each module may include the following steps: Step 1: For each module, determine the corresponding penalty factor based on the module's health value, degradation rate, and coupling influence matrix.

[0131] In practical implementation, a feasible way to determine the penalty factor for each module based on its health value, degradation rate, and coupling influence matrix may include the following sub-steps: Sub-step 11: For each module, calculate its own severity based on the module's health value and the preset amplitude sensitivity coefficient; Sub-step 12: Calculate the module's own deterioration rate based on the module's degradation rate and the preset trend sensitivity coefficient; Sub-step 13: Based on the health values ​​and coupling influence matrix of other modules in the test chamber besides the module, calculate the coupling load pressure caused by other modules to the module; Sub-step 14: Determine the penalty factor corresponding to the module based on its own severity, its own deterioration rate, and the coupled load pressure.

[0132] Penalty factor for each module It consists of two parts: its own degradation level + the coupled load pressure caused by other modules. The relationship is expressed as follows:

[0133] in, The amplitude sensitivity coefficient represents the degree of influence of the health status itself. This is a trend sensitivity coefficient, used to provide early warning of rapidly deteriorating situations.

[0134] Step 2: Calculate the dynamic weight of the module based on the penalty factor, the module's baseline weight value in the predefined baseline weight vector, and the adjustment constant.

[0135] The baseline weight vector is ={ , , The penalty factor for each module is: The generated dynamic weights can be represented as: Specifically, the dynamic weights corresponding to the calculation module can be calculated using an exponential scaling function:

[0136] Where k is an adjustment constant, the specific value of which can be flexibly adjusted according to the accuracy requirements of the experiment.

[0137] The use of an exponential state-variable weighting function is intended to introduce a strong penalty mechanism. This overcomes the "state amortization effect" of the traditional constant weighting model, ensuring that when a single critical module experiences a serious failure or is at high coupling risk, its weight can increase rapidly and non-linearly, thereby dominating the overall health assessment results and reflecting the "weakest link effect" of the system.

[0138] Step 3: Normalize the dynamic weights corresponding to each module in the test chamber to generate the dynamic weight vector corresponding to the test chamber.

[0139] Finally, normalization is performed to obtain the final dynamic weight vector. , The dynamic weights in It can be represented as:

[0140] Step 104: Construct a fuzzy relation matrix for health levels.

[0141] The fuzzy relation matrix of health levels includes the membership function of each module under the evaluation level of each health state.

[0142] In one optional embodiment, constructing the health level fuzzy relation matrix includes the following sub-steps: Sub-step 1: Treat each module in the test chamber as an evaluation object and construct an evaluation object set; wherein the set includes N evaluation objects; This sub-step involves constructing a factor set U, also known as the set of evaluation objects, i.e., U = {heating module, fan module, sealing module}.

[0143] Sub-step 2: Define the evaluation levels for Y health states; This sub-step involves constructing an evaluation set V: defining the levels of equipment health status, i.e., V = {Excellent, Good, Average, Poor, Terrible} evaluation levels.

[0144] Sub-step 3: Based on the membership function of each module under the evaluation level of each health status, construct an N×Y health level fuzzy relation matrix.

[0145] This sub-step constructs a membership matrix, namely the health level fuzzy relation matrix R. Specifically, for each level (Excellent, Good, Average, Poor, Poor) in the evaluation set V, each module is assigned a membership function, such as a trapezoidal membership function, to convert the precise input health value into a degree of membership to each level. Constructing membership functions separately based on the importance of different modules allows for precise capture of important information. Finally, these are combined to form the health level fuzzy relation matrix: the membership vectors of the three modules are arranged in rows to form a 3x5 health level fuzzy relation matrix R.

[0146] Step 105: Generate a comprehensive health evaluation vector based on the dynamic weight vector and the fuzzy relation matrix of health levels.

[0147] Among them, the comprehensive health evaluation vector represents the degree of certainty of the overall health status of the test chamber belonging to each evaluation level.

[0148] This step is the fuzzy synthesis step, which involves dynamic weight vectors. Performing fuzzy matrix multiplication with the health level fuzzy relation matrix R yields the final comprehensive evaluation vector B. ×R.

[0149] B={ , , , , }

[0150] Among them, the comprehensive evaluation vector B can also be called the comprehensive health evaluation vector, and each element in vector B... This represents the certainty that the overall health status of the test chamber belongs to the j-th evaluation level.

[0151] Step 106: Determine the target health value of the test chamber based on the comprehensive health evaluation vector and the health score of the test chamber at each evaluation level.

[0152] Once the target health value of the test chamber is determined, it can be directly output as the overall quantitative health score of the test chamber. Alternatively, the corresponding health level can be determined based on the target health value, and the overall health level of the test chamber can be output as: Excellent, Good, etc. Of course, both can be output so that technicians can understand the health status of the test chamber in a timely manner.

[0153] This step is the defuzzification process. The comprehensive health evaluation vector B only describes the fuzzy distribution of the state. To make monitoring and analysis more intuitive, the fuzzy vector is transformed into a precise health value, and a weighted average method is used to calculate the final health value, i.e., the target health value. An example formula for calculating the target health value is as follows:

[0154] In a preferred embodiment, the health assessment method for the high-temperature test chamber of the integrated circuit aging bench provided in this application may further include the calculation process of the weight drift amount corresponding to each module, as follows: For each module, the weight drift amount corresponding to the module is calculated based on the dynamic weight corresponding to the module and the benchmark weight value of the module in the predefined benchmark weight vector.

[0155] In actual implementation, the weight shift of the module can be calculated using the following formula. :

[0156] The calculated weight drift of each module can be used as the basis for subsequent test chamber fault type judgment.

[0157] In addition to outputting the overall health level of the integrated circuit high-temperature aging test chamber (e.g., "Good") and / or the overall quantitative health score (e.g., 82.5), the health assessment method can also provide more assessment and diagnostic reference information. Specifically, when the output overall health level is lower than a preset standard (e.g., "Good") or the quantitative score is lower than a specific threshold (e.g., 75), the diagnostic analysis trigger condition is met, and the system will automatically trigger the diagnostic analysis program and output maintenance recommendation information.

[0158] In an optional embodiment, after determining the target health value of the test chamber based on the comprehensive health evaluation vector and the health score of the test chamber at each evaluation level, the health assessment method for the integrated circuit high-temperature aging bench test chamber provided in this application embodiment further includes: a fault type judgment and maintenance suggestion generation process, as follows: If the target health value and / or target health evaluation level of the test chamber meet the diagnostic analysis trigger conditions, the fault type of the test chamber is determined based on the health value and weight drift of each module; maintenance suggestions are generated based on the fault type.

[0159] The fault types include: internally-driven faults and coupled-affected faults. For each module, whether it has failed can be determined by whether its corresponding health value is within the normal range; if it is outside the normal range, the module is considered to have failed; if it is within the normal range, the module is considered not to have failed. The normal range can be a range preset by those skilled in the art, and the specific value can be flexibly set by those skilled in the art; this application embodiment does not impose specific limitations on this. For modules that have not failed, whether the weight drift corresponding to the module has increased abnormally can be used to determine whether other modules coupled with it have failed. Specifically, if the weight drift corresponding to the module increases abnormally, it indicates that other modules coupled with it have failed; conversely, if the weight drift corresponding to the module does not increase abnormally, it indicates that other modules coupled with it have not failed. For example, when the weight drift corresponding to the module is greater than a preset weight drift threshold, it indicates that the weight drift has increased abnormally.

[0160] Specifically, for internally-driven faults, the corresponding test chamber characteristics are: module i itself malfunctions; the output maintenance suggestion can be: replace or repair module i.

[0161] For coupled-affected faults, the corresponding test chamber characteristics are: the health status HI of module i is within the normal range, but its An abnormally high level of activity indicates that another module j has failed, and the importance of module i has been elevated through the coupling matrix. Module i is currently operating under high load. The recommended maintenance output is: focus on checking the source of the coupling, module j, and perform preventative maintenance on module i.

[0162] It should be noted that the performance recovery of the integrated circuit high temperature aging test chamber includes pre-event self-healing and post-event self-healing. Post-event self-healing includes performance recovery from heater failure and performance recovery from sensor failure. That is, the thermal runaway self-healing control method and temperature drift self-healing control method of the integrated circuit high temperature aging test chamber provided in this application embodiment can realize the performance recovery of the integrated circuit high temperature aging test chamber.

[0163] Furthermore, the temperature drift self-healing control method for the high-temperature aging stage of integrated circuits provided in this application embodiment addresses the stringent requirements for precise temperature control in high-temperature test chambers used for electronic device testing. Traditional single temperature sensors are susceptible to environmental interference and their own faults (such as deviations and drift), leading to inaccurate temperature measurements and consequently causing output deviations in the temperature control system, affecting experimental results and product quality. Employing redundant sensor configuration and rationally fusing their data, while introducing self-healing control to promptly address sensor faults, is crucial for improving the temperature control performance of the chamber. Current sensor fusion methods largely rely on simple averaging and other coarse approaches, lacking sufficient identification and adaptability for faulty sensors; and they lack a self-healing control coordination mechanism specific to the temperature chamber scenario, making it difficult to quickly ensure the stable operation of the temperature control system in the event of sensor failure.

[0164] The integrated circuit high-temperature aging bench temperature drift self-healing control method provided in this application includes: Step 102: Obtain the first output results of at least two temperature sensors of the aging bench high temperature test chamber at the current moment and the second output results of multiple historical moments.

[0165] Specifically, the aging bench, also known as an integrated circuit high-temperature aging test bench, accelerates various physical and chemical reactions within components by continuously applying electrical stress over a prolonged period. This causes potential faults within the components to surface early, eliminating early-failure products and ensuring that electronic components enter a period of low failure rate and relative stability from the outset. In practical use, the aging bench high-temperature test chamber controls the internal temperature to test the performance and reliability of electronic devices and materials under high-temperature conditions. Temperature sensors installed within the aging bench high-temperature test chamber are used to detect the internal temperature. This embodiment of the application includes at least two temperature sensors within the high-temperature test chamber to ensure the accuracy of the internal temperature detection results by fusing the outputs of multiple temperature sensors.

[0166] In practical applications, for temperature monitoring inside a high-temperature test chamber, the temperature variable to be monitored is defined as follows: ,for Configuration One redundant temperature sensor ( , It is a positive integer, and (This can be preset), and each temperature sensor can be used to detect the output results of the aging bench high-temperature test chamber at various times; among them, it can be... Indicates the first The first output result of each temperature sensor at the current moment, through Indicates the first The second output of a temperature sensor at a historical moment. .

[0167] Step 104: Determine the distance between each temperature sensor at the current moment based on the first output result and the second output result.

[0168] Specifically, first define the distance between temperature sensors. (Measure the first) The temperature sensor and the first (Differences in the output results of individual temperature sensors)

[0169] in, , These parameters were designed for temperature monitoring in a high-temperature test chamber, and are used to balance instantaneous and long-term errors. As a forgetting factor, distance affects the degree to which historical data is "remembered".

[0170] Furthermore, distance After the corresponding formula is constructed, the first step in the preceding steps will be... The temperature sensor and the first By inputting the first and second output results of the first temperature sensor into the formula, the value of the first temperature sensor can be calculated. The temperature sensor and the first The distance between the temperature sensors at the current moment.

[0171] Step 106: Determine the weight of each temperature sensor based on the distance.

[0172] In one optional implementation, determining the weight corresponding to each temperature sensor based on the distance includes: determining a first function value of the distance function corresponding to each temperature sensor at the current moment based on the distance; determining a second function value of the distance function corresponding to each temperature sensor at multiple historical moments based on the second output result; and determining the weight corresponding to each temperature sensor at the current moment based on the first function value and the second function value.

[0173] Furthermore, determining the first function value of the distance function corresponding to each temperature sensor at the current moment based on the distance includes: determining the distance parameter corresponding to the first temperature sensor based on the target distance between the first temperature sensor and each of the second temperature sensors at the current moment, wherein the first temperature sensor is each of the at least two temperature sensors, and the second temperature sensor is each of the at least two temperature sensors other than the first temperature sensor; and determining the first function value of the distance function corresponding to the first temperature sensor at the current moment based on the distance parameter and the target distance.

[0174] The step of determining the distance parameter corresponding to the first temperature sensor based on the target distance between the first temperature sensor and each second temperature sensor at the current moment includes: comparing the target distance between the first temperature sensor and each second temperature sensor at the current moment with a preset distance threshold, and determining the number of target distances greater than the preset distance threshold; if the number is greater than a preset number threshold, then the value of the distance parameter corresponding to the first temperature sensor is determined to be 1; if the number is less than the preset number threshold, then the value of the distance parameter corresponding to the first temperature sensor is determined to be 0.

[0175] In addition, determining the first function value of the distance function corresponding to the first temperature sensor at the current moment based on the distance parameter and the target distance includes: summing the target distance and determining the product of the summation result and the distance parameter as the first function value of the distance function corresponding to the first temperature sensor at the current moment.

[0176] Specifically, first, define the distance function. For the first A temperature sensor, defined as follows:

[0177] in, The function (distance parameter) is set based on the distance between the j-th temperature sensor and other temperature sensors, and there are two cases: (a): If the first A temperature sensor and at least other The distance between the temperature sensors meets the requirements ,but ; (b): If the first One temperature sensor and up to [number] other [other] The distance between the temperature sensors meets the requirements ,but .

[0178] in, This is the maximum permissible deviation between different monitoring points measured through standard testing procedures when the high-temperature test chamber leaves the factory. This parameter is determined by the factory calibration specification of the high-temperature test chamber.

[0179] at this time, Presentation: If (a) is true, If (b) is true, .

[0180] In practical applications, the j-th temperature sensor can be the first temperature sensor, and the k-th temperature sensor can be the second temperature sensor. This refers to the target distance between the j-th temperature sensor and the k-th temperature sensor at the current moment. If the target distance satisfies condition (a), then the distance parameter corresponding to the j-th temperature sensor is 1; if the target distance satisfies condition (b), then the distance parameter corresponding to the j-th temperature sensor is 0. After determining the target distance and the corresponding distance parameter, substitute them into the aforementioned distance function, that is, sum the target distances, and use the product of the summation result and the distance parameter as the first function value of the distance function corresponding to the j-th temperature sensor at the current moment.

[0181] Similarly, the calculation process for the second function value is similar to that for the first function value, and will not be repeated here.

[0182] Furthermore, after calculating and obtaining the first function value and the second function value, the weight of each temperature sensor at the current moment can be determined based on the first function value and the second function value.

[0183] That is, the distance function Once defined, helper functions can be constructed. With final weight .

[0184] Specifically, to make the weight function smooth, we construct:

[0185] in, ,and and These are all design parameters used to adjust the... The use of historical data included in it.

[0186] Further definition:

[0187] Finally, the first The weighting function for each temperature sensor is:

[0188] And satisfy .

[0189] Based on this, after calculating the first function value and the second function value corresponding to the j-th temperature sensor, the first function value and the second function value are input into the auxiliary function to calculate the result. The value, then based on The relationship between the weighting function and the weighting function can be used to calculate the weight of the j-th temperature sensor at the current moment.

[0190] Step 108: If it is determined that the target temperature sensor is faulty according to the weight, determine the fault type of the target temperature sensor based on the first output result and the second output result.

[0191] Specifically, after calculating the weight of each temperature sensor at the current moment, each weight can be compared with a preset weight threshold. If the weight of one or more temperature sensors is less than the preset weight threshold, then these one or more temperature sensors are identified as target temperature sensors and determined to be faulty. In this case, the fault type of the target temperature sensor can be determined based on the first output result and the second output result.

[0192] The fault types in the embodiments of this application include, but are not limited to, deviation faults and drift faults, which can be identified by combining the operating environment of the high temperature test chamber and the historical fault database of the temperature sensor.

[0193] In one optional implementation, determining the fault type of the target temperature sensor based on the first output result and the second output result includes: fusing the first output result based on the weight to obtain the first actual temperature of the aging bench high-temperature test chamber at the current moment; comparing the first actual temperature with the first output result of the target temperature sensor to generate a first comparison result; comparing the second actual temperature of the aging bench high-temperature test chamber at each historical moment with the second output result of the target temperature sensor at the corresponding historical moment to generate a second comparison result; and determining the fault type of the target temperature sensor based on the changing trend between the first comparison result and the second comparison result.

[0194] Specifically, a fusion function can be constructed based on the output results of each temperature sensor. To obtain more reliable temperature measurements:

[0195] After the fusion function is constructed, the weight of each temperature sensor and the first output result of each temperature sensor at the current moment are input into the fusion function, and the first actual temperature of the aging bench high temperature test chamber at the current moment can be calculated.

[0196] The calculation method for the second actual temperature of the high-temperature test chamber at each historical moment is similar to that for the first actual temperature, and will not be repeated here.

[0197] After calculating and obtaining the first actual temperature and the second actual temperature, the first output result of the target temperature sensor can be compared with the first actual temperature, and the second output result can be compared with the second actual temperature to determine the trend of change between the two comparison results.

[0198] For example, if the first output of the target temperature sensor is compared with the first actual temperature, the first comparison result is that the two differ by 2℃; if the second actual temperature of the aging bench high-temperature test chamber at time tn (n is a positive integer less than t) is compared with the second output of the target temperature sensor at time tn, the second comparison result is that the two differ by 2℃. That is, at any time, the difference between the output of the target temperature sensor and the actual temperature at that time is a constant value, then the fault type of the target temperature sensor can be determined to be a deviation fault; if the difference between the output of the target temperature sensor at different times and the actual temperature at the corresponding time changes with time, then the fault type of the target temperature sensor can be determined to be a drift fault.

[0199] This application embodiment determines whether the temperature sensor has changed from having guaranteed accuracy (AG) to having no guaranteed accuracy (NAG) through weighting. If so, it is determined that the temperature sensor is faulty. In this case, the temperature data output by the aforementioned sensor fusion algorithm can be used. By comparing historical temperature data, the fault type of the faulty sensor can be accurately identified.

[0200] Step 110: Correct the fault in the target temperature sensor based on the fault type.

[0201] Specifically, the fault types of the target temperature sensor in the embodiments of this application include, but are not limited to, deviation faults and drift faults.

[0202] Taking the aforementioned fault type as a deviation fault as an example, fault correction is performed on the target temperature sensor based on the fault type. Specifically, temperature compensation can be performed based on the output result of the target temperature sensor at the next moment, and the compensation value is the deviation value between its output result and the actual temperature.

[0203] Taking drift fault as an example, since the difference between the output of the target temperature sensor at different times and the actual temperature at the corresponding time changes over time, fault correction of the target temperature sensor can be performed based on the fault type. Specifically, the relationship between the difference and time can be fitted to obtain a fitting function. For example, if the relationship between the difference and time conforms to the difference... Then, k and b can be directly calculated from the difference between time 0 and time t. Then, the difference between the output of the target temperature sensor and the actual temperature at the next time can be calculated based on the fitting function. Based on this difference, temperature compensation can be performed on the output of the target temperature sensor at the next time.

[0204] This application embodiment uses a redundant temperature sensor fusion algorithm to comprehensively consider instantaneous and long-term errors, dynamically adjust sensor weights, effectively suppress the influence of single sensor failures and environmental interference on temperature measurement, and make the temperature monitoring value in the temperature chamber closer to the true value, providing a reliable basis for precise temperature control.

[0205] In an optional embodiment, the temperature drift self-healing control method for the high-temperature aging chamber of the integrated circuit further includes: when it is determined that the target temperature sensor is faulty according to the weight, determining the fusion convergence degree based on the target weight greater than a preset weight threshold and the number of the target weight; and adjusting the relevant parameters of the temperature control system of the high-temperature test chamber of the aging chamber according to the fusion convergence degree.

[0206] Specifically, the fusion convergence is calculated based on the weights of the temperature sensors, including: Filtering effective weight sets ,in The preset weight threshold.

[0207] calculate ,in , It represents the number of effective weights.

[0208] The fusion convergence is ,in The design parameters are positive.

[0209] The relevant parameters of the temperature control system, including the proportional coefficient, are dynamically adjusted based on the fusion convergence. Integral coefficient and differential coefficients Specifically: ( When the value is large, although the fused value is approaching the accurate value, there may still be lag or bias (for example, a faulty temperature sensor has just had its weight reduced, and the information from a normal temperature sensor has not yet fully dominated the fusion result). Increase the impact of the current deviation on the control quantity to accelerate the correction process.

[0210] ( When the integral action is large, reducing the integral action can decrease the accumulation of "spurious biases during the transition phase" and avoid interference from integral saturation to the system.

[0211] ( When the temperature is high, it enhances the ability to predict and suppress fluctuations, using a larger differential action to offset temperature change trends in advance and reduce the risk of oscillations during the transition phase.

[0212] in These are parameters under normal operating conditions. This represents the increment of the controller parameters.

[0213] In this embodiment of the application, when a temperature sensor malfunction causes abnormal fusion temperature data, resulting in the actual temperature of the high-temperature test chamber deviating from the set value, the temperature control system determines the fusion temperature based on the temperature sensor malfunction. The deviation from the set value is analyzed, and based on the above parameter adjustment strategy, the output of the heating, cooling, and ventilation actuators is adjusted. Through a fusion algorithm and self-healing control mechanism, the output can be adjusted according to the actual needs of the high-temperature test chamber (such as different volumes and different temperature control accuracy requirements). , , It adapts to various parameters and can be extended to other similar environmental control equipment requiring precise sensing and self-healing control (such as humidity chambers and temperature and humidity integrated chambers), demonstrating good versatility and scalability. This processing method has the ability to detect, identify, and self-heal temperature sensor faults. When temperature sensors experience deviations or drift, it can automatically adjust the fusion strategy, trigger self-calibration, or regulate the temperature control actuator without manual intervention, ensuring the continuous and stable operation of the temperature control system and reducing the risk of temperature control failure due to temperature sensor malfunctions. It is suitable for scenarios with high requirements for temperature stability.

[0214] This application embodiment acquires the first output results of at least two temperature sensors of the aging bench high-temperature test chamber at the current moment and the second output results at multiple historical moments. Based on the first and second output results, the distance between each temperature sensor at the current moment is determined. A weight is then determined for each temperature sensor based on the distance. If a target temperature sensor is determined to be faulty based on the weight, the fault type of the target temperature sensor is determined based on the first and second output results. Fault correction is then performed on the target temperature sensor based on the fault type. This method can accurately identify and correct faulty temperature sensors, thereby helping to ensure the stable operation of the aging bench high-temperature test chamber.

[0215] Figure 6 This illustration shows a structural schematic diagram of a thermal runaway self-healing control device for an integrated circuit high-temperature aging bench test chamber according to an embodiment of this application. Figure 6 As shown, it includes: a parameter degradation prediction module, configured to acquire the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of multiple key components used to express the performance degradation information of the high-temperature aging test chamber; using the historical temperature sequence and the health status parameters of the multiple key components, to obtain the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift over a future period of time; a model predictive control module, configured to acquire the actual temperature of the high-temperature aging test chamber and obtain the real-time error by comparing the actual temperature with the set temperature set by the user; by inputting the real-time error and the degradation prediction sequence information into a model predictive control (MPC) controller based on a first-order inertial-hysteresis transfer function, to obtain an anti-drift optimization control quantity for actively suppressing the temperature drift that is about to occur; and a self-healing control module, configured to perform self-healing control on the thermal runaway of the integrated circuit high-temperature aging test chamber according to the anti-drift optimization control quantity.

[0216] The above is a schematic diagram of a thermal runaway self-healing control device for an integrated circuit high-temperature aging test chamber according to this embodiment. It should be noted that the technical solution of this integrated circuit high-temperature aging test chamber thermal runaway self-healing control device belongs to the same concept as the technical solution of the aforementioned integrated circuit high-temperature aging test chamber thermal runaway self-healing control method. Details not described in detail in the technical solution of the integrated circuit high-temperature aging test chamber thermal runaway self-healing control device can be found in the description of the aforementioned integrated circuit high-temperature aging test chamber thermal runaway self-healing control method.

[0217] A computing device is provided according to one embodiment of this application. The components of the computing device include, but are not limited to, a memory and a processor. The processor and the memory are connected via a bus, and a database is used to store data.

[0218] The computing device also includes access devices that enable the computing device to communicate via one or more networks. Examples of such networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. Access devices may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0219] The computing device can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device can also be a mobile or stationary server.

[0220] The processor is used to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned integrated circuit high-temperature aging test chamber thermal runaway self-healing control method.

[0221] An embodiment of this application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-described integrated circuit high-temperature aging test chamber thermal runaway self-healing control method.

[0222] An embodiment of this application also provides a computer program, wherein when the computer program is executed in a computer, the computer is instructed to perform the steps of the above-described integrated circuit high-temperature aging test chamber thermal runaway self-healing control method.

[0223] The above are illustrative schemes of a computing device, a computer-readable storage medium, and a computer program according to this embodiment. It should be noted that each technical solution belongs to the same concept as the above-described battery swapping method based on battery appearance identification and battery swapping authorization. Details not described in detail for each technical solution can be found in the description of the above-described integrated circuit high-temperature aging test chamber thermal runaway self-healing control method.

[0224] The preferred embodiments of this application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of this application shall be within the scope of the claims.

Claims

1. A method for controlling thermal runaway and self-healing in a high-temperature aging test chamber for integrated circuits, characterized in that, include: Acquire the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of several key components used to express the performance degradation information of the high-temperature aging test chamber; Using the historical temperature sequence and the health status parameters of the multiple key components, the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift are obtained in the future. The actual temperature of the high-temperature aging test chamber is obtained, and the real-time error is obtained by comparing the actual temperature with the set temperature set by the user. By inputting the real-time error and the degradation prediction sequence information into the model predictive control (MPC) controller based on the first-order inertial-hysteresis transfer function, an anti-drift optimized control quantity is obtained to actively suppress the temperature drift that is about to occur. The thermal runaway of the high-temperature aging test chamber for integrated circuits is self-healed based on the anti-drift optimization control quantity.

2. The method according to claim 1, characterized in that, Using the historical temperature sequence and the health status parameters of the multiple key components, the degradation prediction sequence information and temperature drift prediction values ​​of the high-temperature aging test chamber's health indicators for a future period are obtained, including: The Long Short-Term Memory (LSTM) network is used to extract the dynamic change patterns of the historical temperature sequence to obtain temperature sequence feature information. The health status parameters of the multiple key components are extracted using a multilayer perceptron (MLP) network to obtain the health status feature information of each key component. The temperature sequence feature information and the health feature information of each key component are fused by an attention mechanism to obtain a fused feature that reflects the overall health status of the high-temperature aging bench test chamber. An LSTM model is constructed and trained, and the fused features are input into the trained LSTM model to obtain the degradation prediction sequence information of the health indicators of the high-temperature aging bench test chamber and the predicted value of temperature drift over a future period.

3. The method according to claim 2, characterized in that, By using an attention mechanism to fuse the temperature sequence feature information with the health feature information of each key component, the fused features reflecting the overall health status of the high-temperature aging bench test chamber are obtained, including: Using the temperature sequence feature information, the attention weight of temperature is calculated. At the same time, using the health feature information of each key component, the attention weight of the component health status of different key components under different operating conditions and different failure levels on the impact of future temperature drift is calculated. Based on the attention weight of the temperature and the temperature sequence feature information, a weighted temperature feature is calculated, and the overall health status of the component is obtained based on the health feature information of each key component. Based on the weighted temperature characteristics, the attention weight of temperature, the attention weight of component health status, and the overall health status of the components, a fusion characteristic reflecting the overall health status of the high-temperature aging bench test chamber is obtained.

4. The method according to claim 3, characterized in that, Using the temperature sequence feature information, the attention weight for temperature is calculated. Simultaneously, using the health status feature information of each key component, the attention weight for the component's health status is calculated, including: Using the temperature sequence feature information, the attention score for temperature is calculated, and using the health feature information of each key component, the attention score for the health status of the component is calculated. The attention weight of temperature is calculated using the attention score of temperature, and the attention weight of component health status is calculated using the attention score of component health status.

5. The method according to claim 3, characterized in that, Based on the weighted temperature characteristics, the attention weight of temperature, the attention weight of component health status, and the overall health status of the components, the fused characteristics used to reflect the overall health status of the high-temperature aging bench test chamber include: in, Features of fusion; Attention weight for temperature; Weighted temperature characteristics; Attention weights for component health status; This refers to the overall health status of the components.

6. The method according to claim 2, characterized in that, The LSTM model includes an input layer, three hidden layers, and an output layer. The hidden layers fit a non-linear relationship using the ReLU activation function, and the output layer outputs a health score in the [0,1] interval using the Sigmoid function.

7. The method according to claim 1, characterized in that, By inputting the real-time error and the degradation prediction sequence information into a Model Predictive Control (MPC) controller based on a first-order inertial-hysteresis transfer function, the anti-drift optimized control quantities for actively suppressing the impending temperature drift include: The Model Predictive Control (MPC) controller based on the first-order inertial-hysteresis transfer function obtains an anti-drift optimized control quantity for actively suppressing the impending temperature drift by using online rolling optimization, based on the real-time error and the degradation prediction sequence information, combined with an objective function that includes an added temperature drift penalty term.

8. A self-healing control device for thermal runaway in an integrated circuit high-temperature aging test chamber, characterized in that, include: The parameter degradation prediction module is configured to acquire the historical temperature sequence of the high-temperature aging test chamber and the health status parameters of multiple key components used to express the performance degradation information of the high-temperature aging test chamber; using the historical temperature sequence and the health status parameters of the multiple key components, the degradation prediction sequence information of the health indicators of the high-temperature aging test chamber and the predicted value of temperature drift are obtained in the future. The model predictive control module is configured to acquire the actual temperature of the high-temperature aging test chamber and obtain the real-time error by comparing the actual temperature with the set temperature set by the user; by inputting the real-time error and the degradation prediction sequence information into the model predictive control (MPC) controller based on the first-order inertial-hysteresis transfer function, an anti-drift optimized control quantity is obtained to actively suppress the temperature drift that is about to occur. The self-healing control module is configured to perform self-healing control on the thermal runaway of the integrated circuit high-temperature aging test chamber based on the anti-drift optimized control quantity.

9. A computing device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the thermal runaway self-healing control method for the integrated circuit high-temperature aging bench test chamber according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the thermal runaway self-healing control method for an integrated circuit high-temperature aging test chamber according to any one of claims 1 to 7.

Images