Electrostatic protection circuit, display panel and electrostatic protection method

Abstract

The invention relates to the field of electronic circuit protection, and discloses an electrostatic protection circuit, a display panel and an electrostatic protection method, and the electrostatic protection circuit comprises a sensing module which is used for monitoring multi-dimensional physical parameters in real time and extracting features; the processing unit is used for constructing a vector containing a system state and selecting an optimal protection action according to the current state; the memory is used for storing a program instruction operated by the processing unit, a risk assessment model parameter and a reinforcement learning model; and the protection execution module is used for driving a corresponding physical execution element to implement a specific electrostatic protection measure according to the action instruction. According to the invention, the electrostatic protection system can adaptively adjust and optimize the protection strategy according to the actually encountered ESD threat intensity, type and generation stage, the working state of the device and the external environment condition, thereby improving the precision, effectiveness and response speed of electrostatic protection.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic circuit protection, and in particular to an electrostatic protection circuit, a display panel and an electrostatic protection method. Background Art

[0002] With the rapid development of electronic technology, the integration density and operating frequency of integrated circuits continue to increase, while feature sizes continue to shrink. This has led to an increasing sensitivity of electronic devices to electrostatic discharge (ESD). ESD is one of the main causes of failure and even permanent damage to electronic devices and integrated circuits. This is especially true during manufacturing, transportation, assembly, and daily use. When static charge accumulates to a certain level in the human body, equipment, or the environment, it rapidly discharges through device pins or device interfaces, generating transient overvoltage and overcurrent, posing a serious threat to the delicate internal circuits.

[0003] To address the challenges posed by ESD, various electrostatic protection solutions have been developed. These solutions typically rely on integrating passive or active protection components into the input / output (I / O) ports, power lines, or critical signal lines of sensitive circuits. For example, widely adopted technologies include the use of shunt elements such as transient voltage suppressors (TVS), metal oxide varistors (MOVs), and gas discharge tubes (GDTs). When the monitored voltage exceeds their preset breakdown threshold, these components quickly turn on, directing the ESD current to ground, thereby clamping the voltage across the sensitive circuit. In addition, some protection circuits based on silicon-controlled rectifier (SCR) structures utilize their avalanche triggering and low conduction voltage characteristics to provide ESD protection. At the circuit design level, emphasis is also placed on reducing ESD risks through optimized PCB layout, grounding design, and shielding.

[0004] However, existing ESD protection measures are gradually revealing their inherent limitations in addressing increasingly complex electromagnetic environments and diverse device application scenarios. A significant issue is that the responses of these protection mechanisms are typically pre-set and relatively fixed, primarily triggering protection based on a single voltage or current threshold. This static protection strategy struggles to dynamically adapt to the diverse nature of ESD events—for example, the energy level, waveform characteristics (such as rise time and pulse width), occurrence frequency, and variations in the device's vulnerability under different operating modes. Traditional protection solutions often lack the ability to detect subtle environmental changes and early warning signs before an ESD event occurs, nor can they comprehensively assess the severity of potential risks based on historical experience or current system status. As a result, they may be unable to accurately distinguish between mild, tolerable ESD disturbances and highly destructive ESD strikes, resulting in either insufficient protection that effectively prevents damage; overreaction, with unnecessary protection actions that can disrupt normal device operation; or the long-term use of conservative, fixed protection parameters that compromise system performance. In addition, these solutions usually do not have the ability to learn from past ESD events and optimize their own protection strategies, and their protection effects are difficult to continuously improve as the operating time increases. Summary of the Invention

[0005] The present invention aims to overcome the defects of the existing electrostatic protection strategy, which is fixed, poorly adaptable, and difficult to cope with complex and changeable electrostatic discharge (ESD) events and equipment working conditions, and to provide an electrostatic protection solution that can dynamically perceive, make intelligent decisions, and continuously optimize.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A first aspect of the present invention provides an electrostatic protection circuit.

[0007] The electrostatic protection circuit includes a sensing module, a memory, a protection execution module, and a processing unit. The sensing module is configured to monitor at least two different types of physical parameters related to potential electrostatic discharge (ESD) events to obtain sensor data. These physical parameters can include at least two of the following: ground potential, power bus current gradient, transient interference on critical signal lines, local electric field strength, and ambient temperature and humidity, to achieve multi-dimensional perception of ESD event precursors and occurrence processes.

[0008] The memory is used to store instructions and learning models executed by the processing unit. The protection execution module is used to execute the protection action selected by the processing unit.

[0009] The processing unit is in communication with the sensing module, the memory and the protection execution module. The processing unit implements the following functions by executing the instructions in the memory: First, based on the sensor data obtained by the sensing module and in combination with the current working mode and / or environmental parameters of the device, the current system state is comprehensively determined. The current system state can be accurately represented by a state vector, which may specifically include a feature vector extracted from the sensor data, a risk score calculated based on the feature vector, a time rate of change of the risk score, a code representing the working mode of the device, and related environmental parameters. The construction of this multi-dimensional state enables the system to have a more comprehensive understanding of the current electrostatic risk environment and the sensitivity of the device itself.

[0010] Next, based on the determined current system state and a pre-stored learning model, the system selects the most appropriate protective action from a predefined set of protective actions with varying intensities and types. This learning model, such as a reinforcement learning model, uses historical experience to learn which protective actions achieve the best long-term results under different system states.

[0011] Thirdly, the protection execution module is controlled to execute the selected protection action to suppress or discharge the potential or ongoing ESD event.

[0012] Finally, and also an important innovation of the present invention is that the processing unit will generate a feedback signal based on the system response such as the change in system sensor data after the selected protection action is executed. The feedback signal quantifies the effectiveness of the protection action (for example, whether ESD is successfully suppressed, whether unnecessary interference is caused to the equipment, etc.). The processing unit then uses this feedback signal to update the learning model. For example, if a reinforcement learning model is adopted, the parameters in the model (such as the Q value in the Q table) will be adjusted according to the feedback signal and the preset update rules of the reinforcement learning model (such as the update rules of Q learning). Through this continuous "action-feedback-learning" cycle, the electrostatic protection circuit can continuously optimize its decision-making strategy, dynamically adapt to different ESD threats and equipment states, and achieve adaptive improvement of protection performance.

[0013] A second aspect of the present invention provides a display panel that integrates the electrostatic protection circuit described in any of the above embodiments. By integrating this intelligent electrostatic protection circuit, the display panel can significantly improve its ESD resistance and long-term operating reliability in complex electromagnetic environments.

[0014] A third aspect of the present invention provides an electrostatic protection method.

[0015] The method first monitors at least two different types of physical parameters related to potential electrostatic discharge (ESD) events through a sensing module to obtain sensor data.

[0016] Subsequently, the current system status is determined based on the acquired sensor data and in combination with the device's operating mode and / or environmental parameters.

[0017] In a specific embodiment, before determining the current system state, the process may further include extracting a feature vector from the sensor data; and calculating a current risk score based on the feature vector and a pre-set risk assessment model. In this case, the determination of the current system state further references the current risk score. The determined current system state may also be represented by a state vector comprising the feature vector, the risk score, the time rate of change of the risk score, a device operating mode code, and environmental parameters.

[0018] Next, based on the determined current system state and the pre-stored learning model, a protection action is selected from a predefined set of protection actions that include multiple intensities and types (for example, at least two of monitoring and recording, activating an early warning signal, connecting a discharge path through a controllable impedance, connecting to the ground wire through a low-impedance switch, etc.).

[0019] Then, the selected protection action is executed.

[0020] Finally, a feedback signal is generated based on the system response (e.g., changes in sensor data) after the selected protection action is executed, and this feedback signal is used to update the learning model. For example, if the learning model is a Q-learning model, its Q-table stores the Q-values ​​of state-action pairs. The learning model update process can specifically include adjusting the Q-values ​​corresponding to the current system state and the selected protection action in the Q-table based on the feedback signal, the next state the system enters after the protection action is executed, and the update rules of the Q-learning model.

[0021] An exemplary update rule could be: Among them, s t is the current system status, act t is the selected protection action, r t is the return signal, s t+1 is the next state, α lr is the learning rate, γ df is the discount factor, is the predefined protection action set, action ′ To enter the next state s t+1 After that, the set of all possible actions Any action in .

[0022] In summary, the present invention includes at least one of the following beneficial technical effects: 1. This invention monitors at least two different types of physical parameters, including ground potential, power bus current gradient, and transient interference on critical signal lines, and combines them with device operating modes and environmental parameters to comprehensively determine the current system status. This allows the system to more comprehensively and dynamically assess the true extent of ESD risk, rather than relying on a single, fixed threshold. This enables more precise selection of protective actions, effectively avoiding false triggering or insufficient protection caused by insufficient information or simplistic judgments under traditional strategies, and ensuring that the most appropriate protective measures are implemented at the most appropriate time.

[0023] 2. This invention introduces a learning model that continuously updates and optimizes its decision-making logic based on the actual system response after the protection action is executed. This means that the ESD protection strategy is no longer static and preset. Instead, it can dynamically adjust its optimal action selection for a specific system state based on historical experience, different device operating modes, and changing environmental conditions. This allows it to better adapt to various complex and ever-changing ESD threat scenarios.

[0024] 3. The system of the present invention can select from a set of protection actions with multiple levels of intensity, and its learning goal is to maximize long-term cumulative rewards. Therefore, the learning model gradually learns to select the protection action that effectively suppresses the electrostatic risk while minimizing interference with normal equipment operation under the current state. This avoids the unnecessary system disturbances caused by overly strong protection or the protection failure caused by too weak protection, thereby ensuring the protection effect while minimizing the potential negative impact of the protection action on equipment performance.

[0025] 4. This invention provides a more comprehensive and effective ESD protection barrier for devices by more accurately identifying potential ESD threats, more intelligently selecting and executing protective actions, and continuously optimizing protection strategies. This protection mechanism, based on multi-dimensional perception and self-learning optimization, significantly reduces the likelihood of transient device failures, performance degradation, or permanent damage due to ESD, thereby extending the device's service life and improving its stable operation and overall reliability in various application environments.

[0026] 5. When determining system status, the present invention not only considers current sensor data and risk assessment results, but also dynamic characteristics such as the time rate of change of the risk score. This enables the learning model to make decisions based not just on static snapshots, but also to perceive whether the risk is rapidly accumulating, slowly increasing, or leveling off. This understanding of risk evolution trends enables the protection system to adjust its response strategy more proactively, for example, taking more proactive intervention in the early stages of a rapid risk escalation, thereby improving the proactiveness and effectiveness of responding to sudden and rapidly changing electrostatic risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the electrostatic protection circuit of the present invention; Figure 2 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0028] The following is combined with Figure 1 -Attached Figure 2 , the present invention is described in further detail.

[0029] like Figure 1 FIG2 is a block diagram of an embodiment of an electrostatic protection circuit of the present invention. The present invention provides an electrostatic protection circuit, the core concept of which is to deal with electrostatic discharge (ESD) events through multi-dimensional perception, intelligent evaluation, and learning-based adaptive decision-making.

[0030] The ESD protection circuit mainly includes a sensor module, a processing unit, a memory, and a protection execution module. These modules work together to form a protection system that can dynamically respond and continuously optimize.

[0031] The sensing module is responsible for real-time monitoring of various physical parameters related to potential ESD events.

[0032] In a specific embodiment, the sensing module can integrate multiple types of sensors, such as a ground potential sensor for monitoring abnormal fluctuations in the ground potential of the device, a power bus current gradient sensor for capturing rapid changes in the power bus current, and a signal line transient interference sensor for detecting transient noise or interference on key signal lines (such as data lines or control lines of a display panel).

[0033] In addition, in order to more comprehensively understand the external conditions under which ESD occurs, the sensing module can also selectively integrate an electric field strength sensor array for detecting the local electric field strength around the device, as well as an environmental parameter sensor for obtaining the current ambient temperature and humidity.

[0034] Through the combination of these sensors, the system can obtain rich data about the ESD event precursors, occurrence process and environmental background from different perspectives.

[0035] The processing unit is the control and decision-making core of the entire ESD protection circuit. It can be implemented by a high-performance microcontroller (MCU), field programmable gate array (FPGA) or application-specific integrated circuit (ASIC).

[0036] The processing unit is tightly coupled with the sensor module, receiving the raw data or preliminarily processed feature data from the sensor module. It is also connected to the memory and the protection execution module.

[0037] Memory plays multiple roles in circuits. It not only stores the program instructions (firmware) required for processing unit operation, but also stores characteristic parameters extracted from sensor data, intermediate calculation results, system status information, and crucial learning model parameters. For example, for reinforcement learning solutions, Q tables or neural network model weights are stored here. Memory can be a combination of volatile memory (such as RAM) and non-volatile memory (such as Flash).

[0038] The protection execution module is the unit that actually implements the electrostatic protection measures. It drives the corresponding physical execution components to act according to the instructions issued by the processing unit.

[0039] These actuators can include various types, such as low-impedance switches (such as MOSFETs) for directing excessive voltage or current to ground, controllable impedance elements (such as digital potentiometers or adjustable resistor networks) for providing adjustable discharge paths at different risk levels, and interface circuits for triggering alarm signals or adjusting system operating conditions.

[0040] When working, the system's information flow and control flow are roughly as follows: The sensor module continuously collects data and transmits it to the processing unit; The processing unit analyzes and processes the data, extracts features, assesses risks, and determines the current state of the system based on the current device operating mode and environmental parameters; Subsequently, the processing unit selects an optimal protection action for the current state based on the learning model stored in the memory; the action instruction is sent to the protection execution module, which is implemented by the latter; After the action is executed, the processing unit evaluates the effect of the action (usually by analyzing subsequent changes in sensor data) and generates a reward signal, which is used to feed back to the learning model to update and optimize the parameters of the learning model so that it performs better in future decisions.

[0041] The above process is a closed-loop, continuous learning and adaptation process.

[0042] When the electrostatic protection circuit of the present invention is applied to electronic devices such as display panels (e.g., TFT-LCD, OLED display panels), it can be integrated into the driver circuit board of the display panel, or exist as an independent protection module adjacent to sensitive circuits.

[0043] The sensing element can be placed near key display panel interfaces (such as FPC connectors), power inputs, and primary signal transmission paths to accurately and promptly capture ESD events. The protection execution module's discharge path is connected to the device ground or a dedicated ESD discharge network. This integrated approach provides the display panel with a dynamic, intelligent, and adaptively adaptable ESD protection layer to the environment and usage conditions.

[0044] The sensing module in the present invention is used to capture physical signals related to ESD events from multiple dimensions, and through feature extraction, converts the original, high-dimensional sensor data into low-dimensional feature vectors that are more meaningful for subsequent risk assessment and decision-making.

[0045] Specifically, the sensing module may include a ground potential monitoring unit, a power bus current gradient monitoring unit, a key signal line transient interference monitoring unit, and may selectively integrate a local electric field strength detection unit and an environmental parameter perception unit according to application requirements.

[0046] In one specific embodiment, a ground potential monitoring unit within the sensor module monitors potential fluctuations of the device or circuit board ground (GND) relative to a reference point (e.g., chassis ground or earth). This unit can use a high-input impedance voltage probe or differential amplifier to accurately measure transient voltages on the ground line.

[0047] The acquired analog voltage signal first passes through a signal conditioning circuit, which typically includes a bandpass filter to filter out power-frequency interference or high-frequency noise unrelated to ESD, and an amplifier to amplify the weak signal to a range suitable for the analog-to-digital converter (ADC). The ADC then converts the conditioned signal into a digital signal.

[0048] The processing unit then extracts key characteristic parameters from the digital voltage sequence, such as the ground voltage peak value V detected within a preset time window. gnd_pk , the approximate rise time of the ESD pulse t gnd_rise (It can be estimated by analyzing the time required for the voltage to rise from a lower threshold to a higher threshold), as well as pulse width, etc. These characteristics can reflect the intensity and waveform characteristics of the ESD event.

[0049] The power bus current gradient monitoring unit focuses on monitoring the rapid changes in current flowing through the device's main power lines (such as VCC and VDD), because ESD events often cause drastic fluctuations in power current.

[0050] The unit can indirectly obtain current information by connecting a precise small-resistance sampling resistor in series in the power path and measuring the voltage difference between its two ends; alternatively, a non-contact Hall effect current sensor can be used.

[0051] The acquired current signal or its equivalent voltage signal is fed into a differential circuit, or a differential operation is performed by a processing unit in the digital domain to obtain the current change rate dI PWR / dt. The gradient signal is also filtered and amplified.

[0052] The characteristic parameters extracted from the gradient signal by the processing unit may include the peak value dI of the current change rate pwr_pk / dt, and the integral of the absolute value of the current change rate within a specific time window ∫|dI pwr / dt|dt, the latter can indirectly reflect the energy of ESD injection.

[0053] The key signal line transient interference monitoring unit is used to detect transient noise or interference caused by ESD induction or conduction on sensitive signal lines (such as LVDS data lines of display panels, SPI control lines, touch screen signal lines, etc.).

[0054] Monitoring can be done by directly contacting the signal line with a high-impedance probe to sample the voltage, or indirectly by sensing electromagnetic field changes on the signal line through a small coupling loop. Because ESD interference typically manifests as high-frequency, narrow pulses, signal conditioning circuits may include high-pass filters to focus on these high-frequency components.

[0055] The processing unit analyzes the interference waveforms on these signal lines, and the extracted characteristic parameters may include the peak overshoot amplitude A of the noise signal. os_pk and undershoot amplitude A us_pk , the number of times overshoot / undershoot occurs within a certain period of time N os_us , and possible ringing frequency, etc. These characteristics help determine the extent of signal integrity impairment.

[0056] In applications with higher ESD protection requirements, the sensing module may also include a local electric field strength detection unit. This unit may consist of an array of one or more small electric field sensors (such as micro electric field probes based on MEMS technology) and be placed near key areas of the device that are susceptible to ESD.

[0057] These sensors can sense the changes in local electric field strength caused by the accumulation of static charge or the formation of discharge paths in the space. The processing unit can extract the maximum electric field strength value E from the readings of the sensor array. field_max , the differential field strength ΔV between different probes efield_max or field intensity gradients, which can serve as early warning signals that an ESD event is about to occur or is occurring.

[0058] In addition, the environmental parameter sensing unit can be integrated with a temperature sensor and a humidity sensor to obtain the current ambient temperature T of the device. env and relative humidity H env These environmental parameters have a significant impact on the generation and accumulation of static electricity (for example, a low humidity environment is more likely to generate and accumulate static electricity), so using them as input features helps the system assess ESD risks more accurately.

[0059] To ensure that information from different sensors can be effectively integrated and analyzed synchronously, the sensing module or processing unit implements a data synchronization mechanism. This can be achieved by adding a unified timestamp to the sampled data of all sensors, or by synchronously triggering the conversion of all ADCs to ensure the time alignment of the data. After extracting the individual feature parameters, the processing unit combines these parameters to form a unified multi-dimensional feature vector: x(t)=[V gnd_pk (t),dI pwr_pk (t) / dt,N os_us (t),E field_max (t),T env (t),H env (t),…] T ; Before being fed into the subsequent risk assessment model, the components of this feature vector are typically normalized (for example, through min-max normalization or Z-score normalization) to eliminate the effects of differences in the dimensions and value ranges of the different feature parameters, making model training and inference more stable and efficient. This normalized feature vector x(t) serves as input to the next stage of the dynamic risk assessment module.

[0060] After multimodal sensing and feature extraction, the sensor module provides a real-time updated, processed, and normalized multidimensional feature vector x(t) to the processing unit. To accurately and dynamically quantify the electrostatic discharge (ESD) risk faced by the current system, the processing unit integrates dynamic risk assessment functions or executes specific instructions in memory.

[0061] The core task of this function is to receive and process the feature vector x(t) from the sensing module and convert it into a scalar value that can intuitively reflect the current ESD risk level, namely the instantaneous risk score S(t), through a built-in or configurable risk assessment model.

[0062] In a specific embodiment, the risk assessment model executed by the processing unit can be based on a parameterized mathematical function. For example, the processing unit can use a structure that combines weighted linear combination and nonlinear activation function mapping to calculate the risk score.

[0063] Specifically, during the system design or initialization phase, the storage area of ​​the processing unit is configured with a weight matrix W and a bias vector b.

[0064] The dimensions of the weight matrix W match those of the characteristic vector x(t) output by the sensor module. Each element in the matrix represents the relative importance of the corresponding characteristic parameter in the comprehensive assessment of the overall ESD risk. For example, if an abnormal ground potential peak has been shown to be a strong indicator of an ESD event, its corresponding weight coefficient in the weight matrix will be relatively large. The bias vector b provides a basic adjustment for risk calculations.

[0065] These model parameters (W and b) can be pre-set based on design experience, or obtained through offline training of a historical data set containing known ESD event samples and their corresponding risk levels, and the learning results are solidified into the storage of the processing unit.

[0066] Based on the selected model structure and configured parameters, the process of the processing unit calculating the instantaneous risk score S(t) can be specifically described by the following mathematical expression: S(t)=σ(W T x(t)+b) Where x(t) is the feature vector provided by the sensor module at the current moment. T x(t) represents the weighted sum of the components in the feature vector by the processing unit according to the weight matrix. This weighted sum is added to the bias term b, and the result is input into a nonlinear activation function σ(·). A commonly used and effective activation function is the Sigmoid function, which is defined as σ(z) = 1 / (1+e -z The advantage of using the Sigmoid function is that it can smoothly map any real-valued output of the internal linear combination to the open interval (0, 1). This allows the risk score S(t) output by the processing unit to have a probability-like meaning (for example, indicating the probability of a potentially harmful ESD event), or at least a normalized risk level measure that is easy to understand and compare.

[0067] Of course, the processing unit can also be programmed to use other types of activation functions, such as the hyperbolic tangent function (tanh) or piecewise functions customized according to specific application requirements to achieve optimal risk characterization effects.

[0068] The dynamically updated instantaneous risk score S(t) calculated by the processing unit provides a clear, quantitative indicator of the current ESD threat level for the entire ESD protection circuit. For example, when the S(t) value calculated by the processing unit is low (close to 0), it indicates that the current system-assessed ESD risk is at a low level; conversely, when the S(t) value is high (close to 1), it indicates that the current ESD risk is very significant.

[0069] In actual deployment, different thresholds can be set based on this risk score, allowing the processing unit to map S(t) to predefined risk levels, such as "safe," "concern," "warning," or "danger." This risk score S(t), generated in real time by the processing unit, can not only be used directly to drive initial responses such as alarms or logging, but more importantly, it will serve as a key component in building a more comprehensive representation of the system state in the next phase, providing the core basis for subsequent learning-based adaptive decision-making.

[0070] Through this dynamic risk assessment mechanism led by the processing unit, the electrostatic protection circuit can more intelligently perceive and respond to changing risk environments.

[0071] In order for the subsequent learning-based adaptive decision module to make the best protection action selection, relying solely on the instantaneous risk score may not be enough to fully reflect the complex situation the system is in. Therefore, the processing unit needs to further integrate multiple aspects of information to construct a more comprehensive and rich system state representation, namely the state vector s t This state vector s t It will serve as the main basis for decision making by the learning model.

[0072] System state vector s t The construction of is completed by the processing unit using the current and historical information. In a typical embodiment, the state vector s t The dimensional composition of can be detailed including the following key components: First, construct the state vector s t The basis of the eigenvector x(t) is the processed and normalized eigenvector x(t) from the sensor module. This eigenvector x(t) directly reflects the raw environmental and device electrical parameter information collected from various physical sensors (such as the ground potential sensor and the power bus current gradient sensor). It is the system's most direct perception input of the external physical world and provides basic observation data for subsequent decision-making.

[0073] Secondly, the instantaneous risk score S(t) calculated by the processing unit in the previous stage is also the state vector s tThis score S(t) is a direct indication of the current ESD threat level obtained by abstracting and generalizing the feature vector x(t), which enables the state vector to contain a clear risk assessment result.

[0074] In order to capture the dynamic evolution trend of ESD risk rather than just the static current value, the processing unit also calculates and introduces the time change rate or change trend of the risk score ΔS(t) as the state vector s t an important dimension of .

[0075] For example, ΔS(t) can be simply calculated as the difference between the current risk score and the risk score at the previous moment, that is, S(t)-S(t-Δt hist ), where Δt hist is a very short time interval. More complex calculations can also include a sliding average or fitted slope of several recent risk score values. Introducing S(t) allows the learning model to distinguish whether the risk is rapidly accumulating, slowly increasing, remaining stable, or receding, which is crucial for predicting future risks and selecting more proactive protective actions.

[0076] In addition, since the sensitivity and vulnerability of electronic equipment to ESD may vary significantly in different working modes, the current working mode M of the device op (t) is also encoded by the processing unit and integrated into the state vector s t For example, a display panel may have several operating modes, including standby mode, normal display mode, high refresh rate mode, and test mode. The processing unit can obtain the current operating mode information from the device's main controller and convert it into an encoding format suitable for learning model input, such as using one-hot encoding to assign a unique binary vector to each mode, or using simple digital encoding.

[0077] If the sensing module includes an environmental parameter sensing unit (such as temperature and humidity sensors), then these related environmental parameter features E env_features (t) will also be included in the state vector s by the processing unit t These environmental parameters, such as lower humidity, may significantly increase the probability and intensity of static electricity generation, so including them as part of the state helps the learning model better understand the overall static electricity risk context of the current environment.

[0078] In this state vector s t Before providing it to the learning-based decision module, the processing unit may also need to appropriately process the state space according to the specific learning algorithm selected (such as Q learning, deep Q network, etc.).

[0079] If a traditional reinforcement learning method based on Q-table is used, and some continuous-valued components in the state vector (such as risk scores and certain parameters in the feature vector) make the state space too large or infinite, then the processing unit may need to discretize these continuous components, that is, divide their value range into several finite intervals, each interval representing a discrete state.

[0080] If a reinforcement learning method based on function approximation (such as neural network) is used, the state vector s t It can usually be used directly as the input of the neural network without or with only minimal discretization processing.

[0081] Ultimately, this comprehensive and rich system state vector s constructed by the processing units t , which will serve as the core input information for intelligent decision-making in the next stage, enabling it to select the most appropriate electrostatic protection action based on the complete situation of the current system.

[0082] Furthermore, the processing unit will use the state vector s according to its internally implemented adaptive decision logic based on learning (this logic can be regarded as a functional module) t To select and output an optimal protection action act t One of the core innovations of this invention is that the decision logic is not fixed, but can be learned and optimized through continuous interaction and feedback with the environment.

[0083] In a preferred embodiment, the adaptive decision-making function implemented by the processing unit adopts the framework of reinforcement learning (RL), specifically the classic Q-learning algorithm or its variants (such as the deep Q network (DQN). Q-learning is a model-free, value-iterative reinforcement learning algorithm whose goal is to learn an optimal action-value function (Q function), where Q(s, act) represents the expected cumulative reward that can be obtained in the future after performing action act in state s.

[0084] In order to apply the Q-learning algorithm, the processing unit first needs to maintain a predefined, limited set of protection actions This set includes various protection measures of different strengths and types that the electrostatic protection circuit can perform. The design of these actions needs to fully consider the effectiveness of the protection and possible side effects. For example, the protection action set Action 0: Monitoring and logging. This is the lowest level of response, applicable when the system is assessed as extremely low risk or in a safe state. The processing unit only records current sensor data and system status and does not perform any active physical intervention.

[0085] Action 1: Activate an early warning signal. When the system assesses that the risk begins to rise but has not yet reached a dangerous level, the processing unit can activate a visual or auditory warning signal to prompt the user to pay attention or conduct manual inspection.

[0086] Action 2: Through the adjustable impedance R ctrl Connecting sensitive circuits to ground. Suitable for medium risk levels. The processing unit controls the adjustable resistor element (such as a digital potentiometer) in the protection execution module to connect the monitored potential overvoltage circuit (such as a signal line or part of a power line) to ground with a medium impedance. This slowly discharges static charge and prevents excessive impact on normal signals.

[0087] Action 3: Quickly connect the main power or ground line to the main ground using a low-impedance switch (such as a MOSFET). This is the highest level of emergency response, suitable for situations where the system assesses the risk to be extremely high, or a destructive ESD event is imminent or occurring. The processing unit quickly turns on a low-on-resistance power switch to provide a low-impedance discharge path for the ESD current, maximizing protection for downstream circuits. These actions can be added, deleted, or refined based on actual application scenarios.

[0088] When the processing unit obtains the current state vector s t After that, it needs to choose an action according to the learned Q function t In Q learning, the commonly used action selection strategy is ∈ exp -greedy strategy. This strategy aims to balance "exploration" (trying new, under-evaluated actions) and "exploitation" (selecting the best action currently known). Specifically, the processing unit will use ∈ exp The probability of randomly selecting an action from the set Choose an action to explore, and take 1-∈ exp The probability of selecting the action with the largest current Q value is parameter∈ exp It usually decreases as learning progresses, allowing the system to explore more in the early stages and exploit more in the later stages.

[0089] Execute the selected action t Afterwards, the system will transition to a new state s t+1(Determined by sensor data at the next moment, etc.), and the processing unit calculates a reward signal r based on the actual effect of this action. t = R(s t , act t , s t+1 ). The design of the reward function R is crucial for the success of reinforcement learning, as it directly guides the learning direction. The design principles of the reward function should include: Effectiveness reward: If the selected action successfully suppresses the ESD event or reduces the risk score (e.g., S(t + 1) < S(t) and no overshoot occurs), a positive reward is given.

[0090] Failure penalty: If the ESD event still occurs or the risk deteriorates further, a negative reward is given.

[0091] Side effect or overprotection penalty: If the selected protection action is too strong, resulting in unnecessary system interference (e.g., normal signal transmission interruption, significant power fluctuations but no serious ESD threat in reality), even if the risk is reduced, a certain negative reward or a reduced positive reward should be given. A penalty term C(act i ) related to its "invasiveness" or "cost" can be set for each action act i . An exemplary reward value setting can be: A large positive reward R is given for successfully suppressing a high-risk ESD. success_high A medium positive reward R is given for successfully suppressing a medium-risk ESD. suacess_mid A negative reward R is given for failing to suppress the ESD or causing problems. failure And a certain negative reward R is given for triggering high-intensity protection but with a very low actual risk. overprotect .

[0092] After obtaining the reward r t and the next state s t+1 , the processing unit can update the Q value corresponding to the state-action pair (s t , act t ) stored in its internal Q table according to the classic Q-learning update rule: In the formula: α lr is the learning rate, which controls the step size of each update and is usually a decimal between 0 and 1.

[0093] γ df is the discount factor, which also lies between 0 and 1 and represents the importance of future rewards relative to immediate rewards. γdf The closer it is to 1, the more the model values ​​long-term cumulative returns.

[0094] is the next state s t+1 The maximum estimate of the Q-value of all possible actions represents the best future reward that can be obtained starting from the next state.

[0095] A Q-table is typically a two-dimensional array or hash table, where rows correspond to all possible (discretized) states, columns correspond to all possible actions, and each cell in the table stores the Q value for the corresponding state-action pair. In the early stages of system operation, the Q-table can be initialized to zero or a small random value. By continuously interacting with the environment (perceiving state - selecting action - executing action - obtaining reward - updating Q value), the processing unit can gradually learn what kind of protection action in various states can bring the greatest long-term cumulative reward, thereby enabling the electrostatic protection strategy to be adaptively optimized.

[0096] In order to improve learning efficiency and stability, especially when the state space is large or a deep neural network (such as DQN) is used to approximate the Q function, the processing unit can also implement the Experience Replay mechanism. This mechanism will replay the transformation tuples (s t ,act t ,r t ,s t+1 ) is stored in an experience replay buffer. When updating the Q value, instead of just using the current transition tuple, a batch of historical experience tuples are randomly sampled from the buffer for learning. This breaks the correlation between the data, making the learning process more stable and making more effective use of historical experience.

[0097] Through the above-mentioned learning-based adaptive decision-making process, the electrostatic protection circuit of the present invention no longer relies on fixed, manually set rules, but can, like an intelligent entity, continuously improve its protection capabilities and decision-making level under complex and changing ESD threats through interaction with the environment and trial and error.

[0098] In the learning-based adaptive decision-making process, the processing unit constructs the system state vector s according to the current t And the action-value function optimized by Q learning (or other reinforcement learning algorithms), and finally select an optimal protection action act t To address the assessed electrostatic discharge (ESD) risk.

[0099] Once a decision is made, the processing unit sends the corresponding control instructions to the protection action execution module. This module is the part of the ESD protection circuit that directly interacts with the protected object (such as the sensitive circuits of a display panel or the power bus). It is responsible for converting the processing unit's decision into actual physical protection measures.

[0100] The protection action execution module integrates a series of controllable physical execution elements, whose types and configurations depend on the predefined protection action set. Different protection actions t Different hardware units in the module will be activated accordingly or these units will be configured with different parameters.

[0101] For example, when the processing unit decides to perform a lower level protection action, such as by adjusting the impedance R ctrl Connecting a specific sensitive line to ground (corresponding to action 2 in the previous example) causes the protection action execution module to drive a corresponding control signal to adjust one or more controllable impedance elements. These controllable impedance elements can be digital potentiometers. The processing unit sends data to the digital potentiometers via a serial interface such as SPI or I2C to precisely set the resistance between the sliding terminal and two fixed terminals, thereby controlling the discharge current.

[0102] Another implementation involves using a resistor network consisting of multiple fixed resistors of varying values ​​and analog switches (such as CMOS switches). The processing unit selects the equivalent resistance value of the connected circuit by controlling the on / off combination of the analog switches. This approach allows potential overvoltages to be slowly discharged to ground with a moderate impedance, making it suitable for scenarios where the risk level is low but preventive discharge is required.

[0103] When the processing unit determines that a high-intensity emergency protection action is necessary, such as quickly connecting the main power line or ground line to the main ground via a low-impedance switch (such as a MOSFET) (corresponding to action 3 in the aforementioned example), the protection action execution module rapidly activates one or more high-current, low-on-resistance power switching elements. These switching elements are typically power metal-oxide semiconductor field-effect transistors (MOSFETs) or insulated-gate bipolar transistors (IGBTs), capable of rapid turn-on in nanoseconds or microseconds, providing a very low-impedance discharge path for the high current generated by ESD to the system ground plane or a dedicated ESD protection ground. To ensure that these power switches operate quickly and reliably, the protection action execution module typically includes a dedicated gate drive circuit. This drive circuit provides sufficient drive current and a suitable voltage waveform to overcome the effects of the power MOSFET's gate capacitance, enabling rapid turn-on and turn-off. For example, a push-pull drive circuit or a dedicated gate drive IC may be used.

[0104] If the protection action set includes an activation warning signal (corresponding to the aforementioned example action act1), the protection action execution module will control the corresponding output interface, such as driving a light-emitting diode (LED) to flash, controlling a buzzer to sound, or sending an alarm message to the host computer or user interface through the communication interface.

[0105] For the monitoring and logging action (corresponding to the above example action act0), the protection action execution module may not perform any physical active intervention in this case, but the processing unit will take this decision and related data (such as the current state s t ) is recorded in the memory. Although this function is mainly completed by the processing unit and the memory, it can be regarded as a "do nothing" execution state of the protection action execution module.

[0106] When designing the protection action execution module, the response time and effectiveness of the protection action also need to be considered. The response time includes the time required from the issuance of the instruction by the processing unit to the complete action of the physical execution element, which is crucial for responding to rapidly changing ESD events. Therefore, it is necessary to select high-speed switching elements and optimize the drive circuit design. Effectiveness involves whether the rated parameters of the protection element can withstand the expected ESD impact (such as peak current and energy capacity), and whether the design of the discharge path can effectively guide the ESD energy to a safe ground without introducing secondary interference. For example, the wiring of the discharge path should be as short and wide as possible to reduce parasitic inductance.

[0107] Furthermore, the protection action execution module must possess a certain level of robustness and reliability. The actuator should be able to withstand multiple protection actions without experiencing performance degradation or damage. In some advanced designs, the module may also include a function to monitor the actuator's own status (for example, to detect MOSFET damage) and feed this information back to the processing unit to detect faults and adjust the protection strategy in a timely manner.

[0108] Through the hardware implementation of the above design and close cooperation with the processing unit, the protection action execution module can efficiently execute the protection decisions output by the learning model, converting the advantages of the intelligent algorithm into practical and effective electrostatic protection for electronic equipment.

[0109] like Figure 2 The figure shows a schematic flow chart of the electrostatic protection method of the present invention. The present invention also provides an electrostatic protection method. This method relies on the aforementioned electrostatic protection circuit hardware architecture and realizes dynamic perception, intelligent evaluation, adaptive decision-making and effective protection of electrostatic discharge (ESD) events through a series of orderly steps, and has the ability of continuous self-optimization. The specific implementation steps of this method are as follows: First, during the system initialization phase, when the electrostatic protection circuit of the present invention is powered on and started, the processing unit loads the necessary initial configuration parameters and learning models from its internal memory or external non-volatile memory. These parameters include, but are not limited to, the baseline calibration values ​​and sampling frequency settings of each sensor unit in the sensor module; the initial weights W and bias b required for the dynamic risk assessment model (if a parameterized model is used); and the initial Q table (for example, all Q values ​​are initialized to zero or a small random number) or the initial weights of the neural network model (if deep reinforcement learning is used) required by the learning-based adaptive decision module. In some embodiments, the initialization phase may also include a functional self-test program for each executive element in the protection action execution module to ensure that it is in normal working condition.

[0110] After initialization, the ESD protection circuit enters a real-time sensing and decision-making cycle. This cycle is the core of the method and continuously performs the following steps: 1. Multimodal Perception and Feature Extraction: Powered by the aforementioned sensing modules, the method begins by synchronously collecting raw data in real time from the ground potential monitoring unit, the power bus current gradient monitoring unit, the critical signal line transient interference monitoring unit, and the optional local electric field strength detection unit and environmental parameter sensing unit. The processing unit then conditions, filters, amplifies, and digitizes this raw data, extracting key characteristic parameters such as the ground voltage peak, the power current rate of change peak, and the number of signal line overshoots. These parameters are combined to form the feature vector x(t) at the current moment.

[0111] 2. Dynamic risk assessment: After receiving the feature vector x(t), the processing unit uses its internally implemented risk assessment model to calculate the current instantaneous ESD risk score S(t).

[0112] 3. System state construction: Subsequently, the processing unit integrates the current feature vector x(t), the instantaneous risk score S(t), the risk score change trend ΔS(t) (calculated by comparing the current and historical risk scores), the current device working mode M op (t) (obtained from the main system or internally determined) and environmental parameter characteristics E env_features (t), construct a state vector s that can fully describe the current situation of the system t .

[0113] 4. Adaptive decision-making and action selection: based on the current state vector s t The processing unit uses its internal storage and maintenance reinforcement learning strategy to select an optimal protection action from the predefined protection action set A. t This selection process aims to maximize expected long-term cumulative returns.

[0114] 5. Protection action execution: The processing unit will select the action instruction act t The protection action execution module then drives the corresponding physical execution element (such as a controllable impedance, a low-impedance MOSFET switch, an alarm device, etc.) to implement the protection action.

[0115] 6. Result observation and reward calculation: In action act t For a short period of time after execution, the sensor module continues to collect data, and the processing unit evaluates the actual effect of the action based on this data, for example, observing whether the risk score S(t) decreases, whether the key signal returns to normal, whether there are any adverse side effects, etc. Based on these observations and the preset reward function, the processing unit calculates the instantaneous reward r corresponding to this state-action transition. t At the same time, the system also enters a new state s t+1 .

[0116] The above steps 2 to 6 constitute a complete decision-making and execution cycle.

[0117] Following this cyclical work phase is the core learning and model iterative optimization process of this method, which is closely integrated with real-time perception and decision-making: 7. The learning model update processing unit uses the newly acquired conversion experience tuple to update its internally stored Q table or neural network model parameters based on the selected reinforcement learning algorithm. If the experience replay mechanism is used, this experience tuple is stored in an experience pool, and a batch of historical experiences is randomly sampled from the pool for batch update. This update step allows the ESD protection circuit to learn from each interactive experience, gradually improving its decision-making strategy, so that it can make more optimal protection actions when encountering similar situations in the future.

[0118] By continuously repeating the above-mentioned closed-loop process of real-time perception, decision-making, execution, observation, feedback calculation and learning update, the electrostatic protection method provided by the present invention can enable its corresponding electrostatic protection circuit to exhibit strong adaptability and self-optimization capabilities.

[0119] For example, when the system is first deployed, its decisions may not be optimal. However, with continuous interaction with various ESD scenarios (including weak static electricity accumulation, sudden high-energy discharge, interference under different operating modes, etc.), the Q table or learning model will gradually converge, allowing the circuit to automatically select and implement the most appropriate protection level and protection method for ESD events of different intensities, different sources, and different stages of occurrence, as well as under different equipment operating modes and environmental conditions. This learning ability makes the protection effect no longer dependent on pre-set fixed thresholds and rigid rules, but can dynamically adapt to the complexity and variability of the specific application environment, thereby providing more efficient and more accurate ESD protection.

[0120] The entire method process embodies a complete closed loop from data collection, information processing, risk assessment, intelligent decision-making to feedback learning, ensuring that the electrostatic protection system can not only cope with current ESD threats, but also continuously improve its future protection performance.

[0121] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An electrostatic protection circuit, characterized in that: include: a sensing module configured to monitor at least two different types of physical parameters associated with a potential electrostatic discharge event to obtain sensor data; a memory configured to store instructions and a learning model; a protection execution module, configured to execute a protection action; and a processing unit, communicatively connected to the sensing module, the memory, and the protection execution module, the processing unit being configured to execute the instructions to: Determining a current system state based on the sensor data, device operating mode, and / or environmental parameters; selecting a protection action from a predefined protection action set based on the current system state and the learning model; Controlling the protection execution module to execute the selected protection action; as well as A feedback signal is generated based on the system response after the selected protection action is executed, and the learning model is updated using the feedback signal.

2. The electrostatic protection circuit according to claim 1, wherein: The at least two different types of physical parameters include at least two selected from ground potential, power bus current gradient, critical signal line transient interference, local electric field strength, and ambient temperature and humidity.

3. The electrostatic protection circuit according to claim 1, wherein: The current system state is represented by a state vector, which includes a feature vector extracted from the sensor data, a risk score calculated based on the feature vector, a time change rate of the risk score, an encoding of the device operating mode, and the environmental parameters.

4. The electrostatic protection circuit according to claim 1, wherein: The learning model is a reinforcement learning model, and the processing unit updating the learning model specifically includes: adjusting parameters in the model according to the reward signal and a preset update rule of the reinforcement learning model.

5. A display panel, characterized in that: The electrostatic protection circuit comprises the electrostatic protection circuit according to any one of claims 1 to 4.

6. An electrostatic protection method, characterized in that: The following steps are involved: monitoring at least two different types of physical parameters associated with potential electrostatic discharge events by a sensing module to obtain sensor data; Determining a current system state based on the sensor data, device operating mode, and / or environmental parameters; Selecting a protection action from a predefined protection action set based on the current system state and a pre-stored learning model; Execute the selected protection action; as well as A feedback signal is generated based on the system response after the selected protection action is executed, and the learning model is updated using the feedback signal.

7. The electrostatic protection method according to claim 6, characterized in that: Before determining the current system state, the method further includes: extracting a feature vector from the sensor data; Calculating a current risk score based on the feature vector and the risk assessment model; Wherein, the determination of the current system state is further based on the current risk score.

8. The electrostatic protection method according to claim 6, wherein: The current system state is represented by a state vector, which includes a feature vector extracted from the sensor data, the current risk score, the time change rate of the risk score, the encoding of the device operating mode, and the environmental parameters.

9. The electrostatic protection method according to claim 6, characterized in that: The learning model is a Q learning model, whose Q table stores the Q value of the state-action pair; Updating the learning model includes: The Q value corresponding to the current system state and the selected protection action in the Q table is adjusted according to the feedback signal, the next state entered by the system after the protection action is executed, and the update rule of the Q learning model. The update rule is: Among them, s t is the current system status, act t is the selected protection action, r t is the return signal, s t+1 is the next state, α lr is the learning rate, γ df is the discount factor, is the predefined protection action set, action ′ To enter the next state s t+1 After that, the set of all possible actions Any action in .

10. The electrostatic protection method according to claim 6, characterized in that: The predefined protection action set includes at least two protection actions of different strengths, and the protection action includes at least one of monitoring and recording, activating an early warning signal, connecting a discharge path through a controllable impedance, and connecting to a ground line through a low-impedance switch.

Images