A high-reliability high-precision test system and method for a sensor chip for extreme environments

Abstract

This invention discloses a high-reliability and high-precision testing system and method for sensor chips in extreme environments, belonging to the field of sensor testing technology. The testing method includes steps such as multi-physics coupled environment loading, automotive operating condition excitation injection, anti-interference synchronous acquisition, anti-drift real-time error compensation, dynamic performance degradation analysis, and drift immunity test report generation. Anti-drift real-time error compensation uses a distributed auxiliary sensor array to monitor temperature gradient kurtosis, vibration spectrum entropy, and electromagnetic shielding effectiveness in real time. After triggering a drift flag, the weights of the CNN convolutional layers are frozen, and incremental federated compensation is performed only on fully connected layers. Furthermore, physical constraint correction compensation output based on FEM simulation is introduced. This invention can effectively suppress compensation coefficient mismatch caused by environmental drift, improve the long-term stability and accuracy of sensor chip testing in extreme environments, and meet the AEC-Q100 full lifecycle reliability verification requirements.

Description

Technical Field

[0001] This invention belongs to the field of sensor testing technology, specifically relating to a high-reliability and high-precision testing system and method for sensor chips designed for extreme environments. Background Technology

[0002] With the rapid development of the automotive industry, especially the rise of new energy vehicles and intelligent connected vehicles, extremely high demands are being placed on the performance and reliability of sensor chips. During vehicle operation, sensor chips face various extreme environments, such as high temperatures in the engine compartment, severe chassis vibrations, low temperatures in frigid regions, and high humidity in humid and hot areas. These extreme environments significantly impact the performance of sensor chips, leading to decreased sensitivity, zero-point drift, and increased full-scale error, ultimately affecting the safety and reliability of the vehicle.

[0003] Existing extreme testing methods for sensor chips typically involve: first, constructing a multi-physics coupled extreme operating condition using a programmable environment simulation chamber to simulate real-world application scenarios such as high engine compartment temperatures and chassis vibrations; then, injecting high-fidelity excitation signals and simultaneously acquiring multiple outputs, internal node signals, and environmental feedback signals at a sampling rate greater than 1 MSPS; finally, through dynamic performance degradation assessment, fault mode diagnosis, and test result reliability verification, a reliability report conforming to the AEC-Q100 standard is generated. This method, through end-to-end collaborative design, solves the problems of incomplete extreme environment simulation and insufficient error compensation capabilities in traditional testing.

[0004] One step involves real-time compensation based on a multidimensional error elimination model. This model uses a pre-trained adaptive error elimination model to accurately suppress test errors. The model first integrates real-time data from environmental sensors—cavity temperature gradient, vibration spectrum, and electromagnetic interference spectrum—with simulation data from the chip package's thermo-mechanical-electrical coupling. It then uses a deep learning algorithm to construct a transfer function for "environmental stress-package deformation-electrical parameter offset." The training process includes: acquiring the test system's background noise and the chip's ideal output in a standard, interference-free environment as a benchmark dataset; obtaining a system error feature library introduced by pure environmental stress under extreme environmental no-load conditions; and training a convolutional neural network (CNN) to identify error propagation patterns using simulation data from the chip's three-dimensional finite element model (FEM). During online operation, the model indexes the error feature library based on real-time environmental parameters, dynamically corrects the compensation coefficients, and separates inherent errors of the test system, such as wire temperature drift and connector fretting potential, as well as parasitic effects caused by chip package deformation. This achieves sub-microvolt residual error suppression, providing crucial technical support for high-precision testing of sensor chips under extreme environments.

[0005] However, the aforementioned real-time compensation technique based on a multidimensional error elimination model suffers from a mismatch in CNN compensation coefficients caused by environmental parameter drift coupling during long-term continuous testing. Firstly, there is temperature field gradient distortion; aging of heating / cooling elements leads to nonlinear shifts in the cavity temperature distribution, such as a higher heating rate at the edges than at the center. This causes changes in the temperature gradient distribution relied upon by the adaptive error elimination model, and the pre-trained environmental error feature library in the model fails to update in time, resulting in the CNN calling incorrect compensation coefficients and thus amplifying the residual error.

[0006] Meanwhile, fatigue in piezoelectric ceramic actuators causes attenuation of high-frequency harmonic components, such as a 10% decrease in amplitude for components >1kHz. Since the adaptive error elimination model relies on vibration spectrum characteristics to identify and compensate for errors, a shift in vibration spectrum characteristics can prevent the model from accurately determining the source of the error, leading to incorrect application of compensation coefficients and a decrease in test accuracy.

[0007] Furthermore, oxidation of the cavity metal shielding layer reduces the shielding effectiveness in specific frequency bands (such as the 2MHz-5MHz band for CANFD communication). This increases electromagnetic interference in that band. When the adaptive error cancellation model processes this interference, due to the mismatch between the environmental feature library and real-time monitoring data, it may mistakenly identify noise caused by electromagnetic interference as other errors, incorrectly applying compensation coefficients, ultimately causing the residual error to rise from the sub-microvolt level to the millivolt level.

[0008] Therefore, developing a system capable of performing highly reliable and accurate testing of sensor chips under extreme environments is of great significance for ensuring the quality and performance of automobiles. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a high-reliability and high-precision testing system and method for sensor chips in extreme environments, which can effectively solve the problems in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A high-reliability and high-precision testing method for sensor chips designed for extreme environments includes the following steps:

[0012] Step S1, Multi-physics coupling environment loading: Apply temperature field, mechanical vibration field, electromagnetic field and humidity field to the sensor chip under test through a programmable environment simulation cavity;

[0013] Step S2, Automotive operating condition excitation injection: Inject sensor excitation signals and power supply disturbance signals into the chip input terminal;

[0014] Step S3, Synchronous Signal Acquisition: Synchronously sample and capture the chip's output response signal and the cavity's environmental reference signal;

[0015] Step S4, Anti-drift real-time error compensation: Input the signal captured in S3 into the adaptive error elimination model to perform the core operation and then output it;

[0016] Step S5, Dynamic performance degradation analysis: Perform wavelet packet transform on the signal output from S4, extract the signal-to-noise ratio and zero-point drift within the -3dB bandwidth, and establish a degradation mapping function with accumulated environmental stress.

[0017] Step S6, Drift Immunity Test Report Generation: When the number of federated model updates is ≥5, an AEC-Q100 certification report containing long-term stability confidence intervals is automatically generated.

[0018] Furthermore, in step S1, the temperature field has a temperature of -40℃ to +150℃, the mechanical vibration field has a frequency of 5Hz to 2000Hz, the electromagnetic field strength is DC-6GHz, and the humidity field has a humidity of 5% to 98%RH.

[0019] Furthermore, in step S2, the sensor excitation signal is matched with the WLTP driving cycle; the power supply disturbance standard is ISO16750-2.

[0020] Furthermore, in step S3, the sampling protocol is a 24-bit Σ-Δ ADC array based on the JESD204B protocol, with a sampling rate of ≥1MSPS.

[0021] Furthermore, step S4 includes the following steps:

[0022] S4.1 Environmental Drift Detection: Real-time monitoring of cavity temperature gradient kurtosis via a distributed auxiliary sensor array. Vibrational spectrum entropy and electromagnetic shielding effectiveness Calculate kurtosis separately The relative rate of change of vibrational spectrum entropy and the attenuation of electromagnetic shielding effectiveness, if kurtosis If either the relative rate of change of vibration spectrum entropy or the attenuation of electromagnetic shielding effectiveness satisfies the drift criterion, incremental federal compensation and physical constraint correction are performed; otherwise, the current compensation output is maintained.

[0023] S4.2 Incremental Federated Compensation: After the drift criterion response, the weights of the CNN convolutional layers are frozen, the fully connected layers are trained in a lightweight manner to generate local compensation coefficients to update parameters, and the parameters are uploaded to the federated server through differential privacy encryption.

[0024] S4.3 Physical Constraint Correction: Receives global model parameters aggregated from the federated server, combines them with the chip FEM thermal-mechanical-electrical coupling transfer function matrix to construct a regularization term, and outputs the result while meeting accuracy requirements under physical consistency constraints.

[0025] Furthermore, in step S4.1,

[0026] cliff Defined as:

[0027] ;

[0028] Where T is the temperature difference vector, μ is the mean, and σ is the standard deviation. This represents the expectation operation;

[0029] Spectral entropy Defined as:

[0030] ;

[0031] in The energy percentage of the i-th frequency band in the normalized vibrational power spectral density. Number of frequency bands;

[0032] Electromagnetic shielding effectiveness Defined as:

[0033] ;

[0034] in Electromagnetic field strength under reference conditions The average field strength;

[0035] The relative rate of change of vibrational spectrum entropy is defined as:

[0036] ;

[0037] in The current frame spectral entropy, Baseline spectral entropy;

[0038] Electromagnetic shielding effectiveness attenuation is defined as follows:

[0039] ;

[0040] in This represents the output field strength at a frequency of 5MHz. Output field strength values ​​for historical static baselines.

[0041] Furthermore, in step S4.1, the temperature gradient kurtosis >4.0 satisfies the drift criterion; >0.05 relative change rate of vibration spectrum entropy satisfies the drift criterion; >20dB electromagnetic shielding effectiveness attenuation satisfies the drift criterion.

[0042] Furthermore, in step S4.2, the compensation coefficient is defined as:

[0043] ;

[0044] in It is a multidimensional physical field drift characteristic vector, composed of temperature gradient kurtosis, vibration spectrum entropy, and electromagnetic shielding attenuation factor; This is the nonlinear compensation function obtained by mapping from the CNN output layer; Update parameters for weighted average aggregation;

[0045] Defined as:

[0046] ;

[0047] in This represents the number of incremental samples for the i-th device; This represents the change in the fully connected layer weights calculated by the i-th device.

[0048] Defined as:

[0049] ;

[0050] in This represents the weights of the fully connected layer after training on the i-th device. This represents the weights of the original fully connected layer.

[0051] Furthermore, in step S4.3, the thermo-mechanical-electric coupling transfer function matrix includes a frequency domain transfer function. The parasitic capacitance change caused by thermal deformation and its transmission through the transfer function are characterized. The change in carrier mobility caused by vibrational stress;

[0052] The model's predicted output is aligned with the theoretical output calculated based on the physical field input. Real-time correction is achieved by constructing a physical loss function, which is defined as follows:

[0053] ;

[0054] in =0.7、 =0.3 is the weighting coefficient for the loss term. This represents the current temperature gradient tensor. Represents the current vibrational spectrum tensor. This is the model's predicted output;

[0055] The accuracy requirement is defined as:

[0056] ;

[0057] in This represents the FEM simulation response under thermo-mechanical-electrical coupling. For multiphysics input vectors, This is the model's predicted output.

[0058] Furthermore, in step S5, the degradation mapping function of accumulated environmental stress is defined as:

[0059] ;

[0060] in The signal-to-noise ratio at the current time t. This is the zero-point drift amount. The cumulative environmental stress vector is defined as:

[0061]

[0062] in , , These represent the cumulative thermal stress, vibration stress, and humidity stress that the chip has endured.

[0063] Furthermore, in step S6, the confidence interval is defined as:

[0064] ;

[0065] in This represents the mean of the performance parameters. Let n be the standard deviation and n be the sample size. For degrees of freedom The critical value of the t-distribution with a confidence level of 95%.

[0066] This invention also provides a high-reliability and high-precision testing system for sensor chips in extreme environments, including a programmable environment simulation chamber, an isolated low-noise signal generation module, and a distributed high-speed data acquisition network;

[0067] The programmable environment simulation cavity is used to provide temperature, mechanical vibration, electromagnetic, and humidity fields to be applied to the sensor chip under test.

[0068] The isolated low-noise signal generation module is used to output excitation signals and power disturbance signals to the sensor chip under test;

[0069] The distributed high-speed data acquisition network is used to acquire and preprocess the captured signal data.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. In its testing method, this invention effectively solves the problem of compensation coefficient mismatch through an incremental federated compensation mechanism. This mechanism freezes the weights of the CNN convolutional layers after drift is triggered, performs lightweight incremental training only on the fully connected layers, and uploads the parameters to the federated server for aggregation using differential privacy encryption. This preserves the original feature extraction capability while dynamically updating the compensation coefficients based on real-time data. In contrast, existing pre-trained models cannot continuously optimize based on environmental parameter drift during long-term testing, leading to a mismatch between the compensation coefficients and the actual error, thus easily causing the compensation coefficient mismatch problem.

[0072] 2. This invention introduces physical consistency constraints into the CNN output layer and constructs a regularization term by combining the chip FEM thermal-mechanical-electrical coupling transfer function matrix. A physical loss function is used to align the model predictions with the FEM simulation results, limiting the error to within 0.01V and ensuring the physical reliability of the compensation model. Existing error elimination models do not integrate physical field simulation data, making it difficult to guarantee the physical rationality of compensation results under complex coupling environments.

[0073] 3. This invention generates a test report containing confidence intervals after multiple updates of the federated model, evaluates the chip performance fluctuation range based on statistical methods, and stops training when the confidence interval width meets a threshold, achieving a balance between test accuracy and resource consumption, and improving the long-term stability of the test system. Existing technologies lack quantitative evaluation and adaptive convergence mechanisms for long-term test stability, making it difficult to meet the requirements of full lifecycle reliability verification.

[0074] 4. The testing system of this invention employs a distributed auxiliary sensor array, including an infrared thermometer, a laser vibrometer, and a near-field probe, and constructs a multi-dimensional drift detection mechanism: monitoring temperature gradient kurtosis, vibration spectrum entropy, and electromagnetic shielding effectiveness. It can capture non-stationary disturbances such as temperature field distortion and vibration spectrum shift in real time. When continuous sampling data deviates from the baseline by more than 3σ, a drift flag is quickly triggered, enabling more accurate detection of environmental drift. Existing technologies lack such comprehensive real-time monitoring and multi-dimensional joint criteria, making it difficult to detect subtle changes in environmental parameters in a timely manner. Attached Figure Description

[0075] Figure 1 This is a flowchart illustrating the high-reliability and high-precision testing method for sensor chips in extreme environments as described in this invention.

[0076] Figure 2 This is a flowchart illustrating the real-time error compensation step for anti-drift in this invention. Detailed Implementation

[0077] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0078] This invention provides a high-reliability and high-precision testing system for sensor chips designed for extreme environments. The system includes a programmable environment simulation chamber, an isolated low-noise signal generation module, and a distributed high-speed data acquisition network. The programmable environment simulation chamber provides temperature, mechanical vibration, electromagnetic, and humidity fields to be applied to the sensor chip under test. The isolated low-noise signal generation module outputs excitation and power disturbance signals to the sensor chip under test. The distributed high-speed data acquisition network acquires and preprocesses the captured signal data.

[0079] The programmable environment simulation chamber utilizes equipment with high-precision control and excellent environmental simulation capabilities. Test fixtures hold the sensor chip under test within the chamber's internal space. The chamber boasts excellent sealing performance to ensure stability and accuracy when simulating various environmental conditions. The chamber also includes a temperature field control unit, a rapid temperature change unit, and an infrared imaging unit. The temperature field control unit employs a high-precision platinum resistance temperature sensor (PT1000 model), combined with a PID algorithm to control the power of the semiconductor cooler array, achieving temperature stability of ±0.1℃ in the core area of ​​the chamber. The infrared imaging unit, located on the outer ring of the chamber, is equipped with a high-resolution infrared thermal imager to monitor the temperature distribution on the chip package surface in real time, dynamically adjusting the power of each zone of the TEC array to ensure that the temperature gradient in the sensitive areas of the chip is ≤2℃ / cm. The rapid temperature change unit uses a rapid switching device between liquid nitrogen injection and high-temperature nitrogen turbulence to achieve a temperature change rate ≥30℃ / min from -40℃ to +150℃, accurately simulating the thermal shock of a car during cold starts and rapid acceleration.

[0080] The isolated low-noise signal generation module features multiple signal output functions, capable of generating analog / digital excitation signals matched to actual automotive application scenarios. Internally, it employs a high-precision clock source and low-noise signal processing circuitry, while utilizing opto-isolation and other technologies to effectively isolate input and output signals, reducing signal interference and ensuring signal purity and stability. The module also includes an automotive operating condition scenario library, pre-stored with sensor excitation waveforms for typical operating conditions. Through high-speed storage chips and optimized algorithms, the required excitation waveforms can be quickly and accurately recalled. Furthermore, the module includes a fault injection unit implemented using a programmable logic device (PLD), capable of flexibly inserting power drop (ISO16750-2), bus error frame (ISO11898), and sensor open / short circuit fault modes to meet diverse testing requirements.

[0081] The distributed high-speed data acquisition network utilizes an ADC array based on the JESD204B protocol, employing a 24-bit Σ-Δ ADC chip with an independent ADC for each channel to ensure high accuracy and synchronization in signal acquisition. Data is aggregated to the FPGA preprocessing unit via a high-speed SerDes link, leveraging its parallel processing capabilities to perform real-time preprocessing on the acquired data, including data buffering, format conversion, and preliminary error correction. A differential current sensing amplifier with a gain error of <0.01% and a digital domain adaptive notch filter are embedded in the signal transmission path to perform spectral cancellation of switching power supply noise and RF interference, improving the accuracy and anti-interference capability of signal acquisition.

[0082] This invention also provides a high-reliability, high-precision testing method for sensor chips in extreme environments, basically as follows: Figure 1 As shown, it includes the following steps:

[0083] Step S1, Multi-physics coupling environment loading: Apply temperature field, mechanical vibration field, electromagnetic field and humidity field to the sensor chip under test; wherein the temperature field is -40℃~+150℃, the mechanical vibration field frequency is 5Hz~2000Hz, the electromagnetic field strength is DC-6GHz, and the humidity field humidity is 5%~98%RH.

[0084] Step S2, Automotive operating condition excitation injection: Inject sensor excitation signals and power supply disturbance signals into the chip input terminal;

[0085] Specifically, the sensor chip is excited by precisely simulating actual automotive operating conditions to realistically reproduce the dynamic signal characteristics of the WLTP driving cycle and the impact of power supply disturbances on chip performance. First, a timing sensor excitation signal is generated based on the WLTP standard operating condition curve. This signal covers typical operating conditions such as acceleration, cruising, and deceleration, with a dynamic frequency range from low frequency 0.1Hz to high frequency hundreds of Hz. The amplitude reflects the actual vehicle speed and acceleration changes, ensuring that the changes in physical quantities received at the chip input accurately reflect the changes in road conditions. The excitation signal is converted into voltage or current form by a high-precision signal generator and injected into the chip's sensitive area, ensuring that both linear and nonlinear characteristics of the excitation amplitude and frequency are fully captured. Simultaneously, a power supply disturbance signal is designed according to the ISO16750-2 standard, including various electrical anomalies such as transient voltage spikes, drops, voltage fluctuations, and electromagnetic interference. The disturbance signal is synchronously superimposed on the excitation signal to simulate the complex electrical stresses experienced by the chip under actual automotive power supply conditions. Real-time acquisition of the injected signal and the chip response signal ensures signal timing synchronization and guarantees a high correlation between the dynamic characteristics of the stimulus injection and the chip output, meeting the input requirements for subsequent anti-interference synchronous acquisition and drift compensation analysis. Through this step, the chip exposes its dynamic performance limits and disturbance rejection capabilities in a dual-coupled environment of operating condition stimulus and power supply disturbance, laying a solid foundation for reliability assessment and lifetime prediction.

[0086] Step S3, Synchronous Signal Acquisition: Synchronously sample and capture the chip's output response signal and the cavity's environmental reference signal;

[0087] Specifically, a 24-bit Σ-Δ analog-to-digital converter array based on the JESD204B high-speed serial interface protocol is employed to achieve high-precision synchronous acquisition of the chip output response signal and the cavity environment reference signal. This Σ-Δ ADC array possesses excellent noise shaping characteristics and high-resolution quantization capabilities, effectively suppressing quantization noise and electromagnetic interference during the conversion process, ensuring signal sampling accuracy at the microvolt level. The sampling rate is set to no less than 1 MSPS to meet the dynamic response frequency requirements of automotive sensors and the detailed capture of operating condition signals, achieving accurate recording of high-frequency vibrations and transient temperature fluctuations. Utilizing the multi-channel synchronization mechanism and clock data recovery technology of the JESD204B protocol, the clock phase of all sampling channels is ensured to be consistent, achieving time alignment and synchronous acquisition of the signal, avoiding signal distortion and subsequent processing errors caused by sampling time deviations. The acquired data is transmitted to the digital signal processing unit in real time via a link, supporting subsequent error compensation and dynamic performance analysis input. The synchronous acquisition includes both chip output signals and cavity environment reference signals, such as temperature, vibration, and electromagnetic intensity, to ensure that the subsequent adaptive error elimination model can accurately correlate chip output drift with environmental changes, improve the accuracy and adaptability of the compensation model, and achieve an overall improvement in the anti-interference performance of the system and high reliability of the test results.

[0088] When collecting temperature field data, four infrared thermometers, each placed at one of the four corners of the chip's sensitive area, are positioned inside the cavity to obtain temperature measurements. This directly constitutes the temperature difference vector. This is used for calculating the kurtosis K in step S4.1.

[0089] When collecting vibration field data, a piezoelectric ceramic vibration sensor is used to detect the micro-strain caused by mechanical excitation at the bottom of the chip in real time. The vibration signals collected by the system are used to construct a micro-strain spectrum. And further extract the power spectral entropy To quantify the complexity of the vibration environment:

[0090] ;

[0091] in The normalized vibration power spectral density represents the energy proportion of the i-th frequency band, and n is the number of frequency bands. A higher entropy value indicates a more dispersed vibration energy distribution, suggesting whether the system has atypical excitations that exceed the characteristic range of the SAEJ211 waveform.

[0092] When acquiring electromagnetic field data, eight near-field electromagnetic probes are arranged in a ring scanning array. Each probe has a sensitivity threshold of -120 dBm and is used to scan the spatial distribution of electromagnetic disturbances. The response of the probe array within a 5 MHz bandwidth is analyzed. Normalization is performed before calculating the shielding effectiveness. :

[0093] ;

[0094] in Electromagnetic field strength under reference conditions The average field strength collected by the array is used to determine whether there are electromagnetic leakage paths or insufficient attenuation issues in the cavity at various frequency bands. The entire auxiliary sensor system works in conjunction with the main control system to provide high-precision, low-latency environmental data support for subsequent drift detection, compensation correction, and dynamic analysis.

[0095] Step S4, Anti-drift real-time error compensation: Input the signal captured in S3 into the adaptive error elimination model to perform the core operation and then output it;

[0096] Specifically, basically as Figure 2 As shown, a multi-dimensional drift detection mechanism is set up for four key interference factors: temperature, vibration, electromagnetic interference, and humidity, to achieve rapid sensing and quantification of non-stationary disturbances in the sensor's operating environment. In the temperature channel, a spatial difference vector is constructed from the collected temperature values, and its kurtosis is calculated as a criterion for the stability of the temperature field. The kurtosis K is defined as:

[0097] ;

[0098] Where T is the temperature difference vector, μ is the mean, and σ is the standard deviation. This indicates the expected operation. When K > 4.0, it indicates a sharp local abrupt change in the temperature field, which is identified as thermal drift distortion, and the system marks the temperature channel drift for that period as valid.

[0099] The vibration channel uses spectral entropy as the monitoring indicator. The system divides the 0.5 to 10 kHz frequency band into 100 equal sub-bands and calculates the energy-normalized probability distribution for each sub-band. The spectral entropy is:

[0100] ;

[0101] The system continuously tracks the change of spectral entropy over time. If the relative rate of change of the difference between the current frame's spectral entropy and the baseline value exceeds 5%, then... If this occurs, it is determined that a structural change has occurred in the distribution of vibration energy, triggering the vibration drift indicator.

[0102] For the electromagnetic field channel, the system scans the output field strength values ​​of the 8-channel near-field probe at a frequency of 5MHz every 10 milliseconds. and compared with historical static baseline For comparison, if all three consecutive samples satisfy: If the value is greater than 20, it indicates that the shielding effectiveness at that frequency point has decreased by more than 20dB, the electromagnetic shielding of the cavity is deemed to be ineffective, and the electromagnetic channel drift flag is activated.

[0103] The humidity channel uses the rate of change of relative humidity as the monitoring indicator. A capacitive humidity sensor is placed inside the cavity, sampling at a frequency of 1Hz to record the relative humidity value in real time. Define the humidity baseline. This is the arithmetic mean of the humidity over the first 10 minutes after the test begins. Calculate the current relative rate of change of humidity. If the value is greater than 0.1 and the duration exceeds 30 seconds, humidity drift is determined to be triggered, and the humidity channel drift flag is activated.

[0104] The drift judgment results of the four sub-channels are merged in the system to form the final multiphysics drift flag. Once any channel meets the drift criterion, the current model output is immediately frozen, and the incremental federated compensation and physical constraint correction process is initiated to ensure that the system response is not affected by implicit environmental changes and thus fails.

[0105] In incremental federated compensation, local training data is first constructed on each device. When the environmental drift detection module triggers a drift flag, the system automatically selects high-precision environmental sensing data collected over the next 30 minutes and the corresponding sensor chip output response to form the incremental dataset for that device. It is used for local model fine-tuning.

[0106] Building upon this, the transfer learning mechanism comes into play. Each device loads a pre-trained convolutional neural network model from the central server. The convolutional layers of this model are used to extract multiphysics coupling features and keep the parameters frozen, only unfreezing the weight matrices of the fully connected layers at the ends. Lightweight training is performed to achieve adaptive fine-tuning of the local compensation model without compromising the original feature extraction capabilities. After training, each device calculates the change in its fully connected layer weights. The data is encrypted using a differential privacy mechanism and then uploaded to the federated server. The server receives compensation parameter updates from k devices. Perform the following weighted average aggregation:

[0107] ;

[0108] in, This represents the number of incremental samples from the i-th device. The aggregation process weights each device by its data size, ensuring that devices with sufficient samples contribute more to the global model. The updated global compensation weights are mapped to the feature library using the following formula to maintain the environment drift vector. With compensation coefficient Matching relationships:

[0109] ;

[0110] in It is a multidimensional physical field drift eigenvector, composed of temperature gradient kurtosis, vibrational spectral entropy, and electromagnetic shielding attenuation factor. The nonlinear compensation function is obtained by mapping from the CNN output layer, which ultimately enables the updated system to be immune to different types and degrees of drift and to have continuous learning capabilities.

[0111] In the physical constraint correction, the thermal, mechanical, and electrical response relationships of the sensor chip under extreme environments are first simulated using finite element modeling to extract the coupling transfer function between key physical fields. The parasitic capacitance change caused by thermal deformation is then analyzed through the frequency domain transfer function. Characterization shows that the effect of vibrational stress on carrier mobility is achieved through the transfer function. The description states that these functions are all established through FEM simulation and mapped to response surfaces within the test frequency range. Based on this, the system introduces physical consistency constraints into the CNN output layer to adjust the model's predicted output. Aligning with the theoretical output calculated based on the physical field input, real-time correction is achieved by constructing a physical loss function:

[0112] ;

[0113] in =0.7、 =0.3 is the weighting coefficient for the loss term. This represents the current temperature gradient tensor. This represents the current vibration spectrum tensor, which is acquired in real time by a distributed auxiliary sensor array and input into the neural network. This physical loss, together with the main loss function, forms the overall objective function of the CNN, driving the model output to maintain adherence to physical laws while satisfying data fitting accuracy. After training, the system maps the temperature gradient compensation output of the CNN into regional power control commands for the thermoelectric cooler array, keeping the sensitive areas of the chip in a physically controllable thermal distribution state. Simultaneously, the vibration compensation is converted into real-time waveform control signals for a six-degree-of-freedom actuator, and through inverse demapping, interference suppression actions conforming to FEM simulation characteristics are reconstructed. Thus, adaptive stability compensation under environmental drift is achieved through the combined action of the algorithm and hardware. The ultimate goal is to ensure that the prediction error meets the following accuracy requirements under physical consistency constraints:

[0114] ;

[0115] in This represents the FEM simulation response under thermo-mechanical-electrical coupling. As a multi-physics input vector, the error constraint ensures the physical reliability and engineering usability of the chip response prediction results in complex coupled environments.

[0116] Step S5, Dynamic performance degradation analysis: Perform wavelet packet transform on the signal output from S4, extract the signal-to-noise ratio and zero-point drift within the -3dB bandwidth, and establish a degradation mapping function with accumulated environmental stress.

[0117] Specifically, wavelet packet transform is performed on the signal output from S4. A multi-scale decomposition tree is constructed using Daubechies wavelets to extract local energy distribution features within the complete frequency domain, reconstructing the dominant frequency component within a -3dB bandwidth. Two key metrics are extracted from this bandwidth: signal-to-noise ratio (SNR), defined as the ratio of dominant frequency component energy to background noise energy, used to measure the effectiveness of the chip signal; and zero-point drift, referring to the accumulated error caused by signal baseline offset to sensor accuracy, measured in millivolts. The SNR at the current time point t is used as an example. and zero-point drift With the corresponding cumulative environmental stress vector Construct a performance degradation mapping function:

[0118] ;

[0119] in , , The system accumulates and records the thermal stress, vibration stress, and humidity stress that the chip has endured, respectively, according to a standard stress integral model during processes S1-S3. This mapping function establishes a multivariate nonlinear relationship between performance and stress through historical model fitting. It is then used as a performance predictor on the system side to determine whether the chip has entered a critical degradation range. When the signal-to-noise ratio falls below a set threshold or the zero-point drift exceeds the compensation capability limit, the system marks the chip as "degradation risk" and cross-validates this with the confidence interval matching analysis results in S6, improving the closed-loop accuracy of the overall reliability analysis. The final output of this analysis module includes indicators such as degradation trend curves, key performance degradation rates, and corresponding environmental stress loading curves, providing quantitative support for chip reliability lifetime assessment and AEC-Q100 certification.

[0120] Step S6, Drift Immunity Test Report Generation: When the number of federated model updates is ≥5, an AEC-Q100 certification report containing long-term stability confidence intervals is automatically generated.

[0121] Specifically, after each round of federated updates, the corresponding global model version sequence {V1,V2,…,Vn} is recorded, and n repeated tests are performed using the same chip under standard environmental conditions to obtain the response values ​​of its key performance parameters, forming a set {P1,P2,…,Pn}. The system first calculates the sample mean of the performance parameters. and sample standard deviation σ P Then, confidence intervals are constructed based on the Studentt distribution theory:

[0122] ;

[0123] in This represents the mean of the performance parameters. Standard deviation, For the sample size, For degrees of freedom The t-distribution critical value with a 95% confidence level reflects the statistical reliability of the performance parameter fluctuation range of the model during long-term use. If the chip's performance changes consistently remain within this range, the system automatically marks the global model as "stable" and triggers the AEC-Q100 long-term stability certification module, generating a standardized report document containing information such as the model version number, test timestamp, environmental stress conditions, and performance range boundaries, forming a traceable and verifiable high-reliability test closed loop. Furthermore, the system performs a judgment on the width of the confidence interval; if... If ε is a set threshold, then it is considered to have reached long-term stable convergence, and further compensation training is stopped, thus saving system resources and shortening the testing cycle.

[0124] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A high-reliability and high-precision testing method for sensor chips designed for extreme environments, characterized in that: Includes the following steps: Step S1, Multi-physics coupling environment loading: Apply temperature field, mechanical vibration field, electromagnetic field and humidity field to the sensor chip under test; Step S2, Automotive operating condition excitation injection: Inject sensor excitation signals and power supply disturbance signals into the chip input terminal; Step S3, Synchronous Signal Acquisition: Synchronously sample and capture the chip's output response signal and the cavity's environmental reference signal; Step S4, Anti-drift real-time error compensation: Input the signal captured in S3 into the adaptive error elimination model to perform the core operation and then output it; Step S5, Dynamic performance degradation analysis: Perform wavelet packet transform on the signal output from S4, extract the signal-to-noise ratio and zero-point drift within the -3dB bandwidth, and establish a degradation mapping function with accumulated environmental stress. Step S6, Drift Immunity Test Report Generation: When the number of federated model updates is ≥5, an AEC-Q100 certification report containing long-term stability confidence intervals is automatically generated.

2. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 1, characterized in that: In step S1, the temperature field temperature is -40℃ to +150℃, the mechanical vibration field frequency is 5Hz to 2000Hz, the electromagnetic field strength is DC-6GHz, and the humidity field humidity is 5% to 98%RH. In step S2, the sensor excitation signal is matched with the WLTP driving cycle; the power supply disturbance standard is ISO16750-2. In step S3, the sampling protocol is a 24-bit Σ-Δ ADC array based on the JESD204B protocol, and the sampling rate is ≥1MSPS.

3. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 1, characterized in that: Step S4 includes the following steps: S4.1 Environmental Drift Detection: Real-time monitoring of cavity temperature gradient kurtosis via a distributed auxiliary sensor array. Vibrational spectrum entropy Electromagnetic shielding effectiveness and relative humidity value Calculate kurtosis separately The relative change rate of vibrational spectrum entropy, the attenuation of electromagnetic shielding effectiveness, and the change in relative humidity, if kurtosis If any one of the following indicators—the relative change rate of vibration spectrum entropy, the attenuation of electromagnetic shielding effectiveness, and the change in relative humidity—satisfies the drift criterion, incremental federal compensation and physical constraint correction are performed; otherwise, the current compensation output is maintained. S4.2 Incremental Federated Compensation: After the drift criterion response, the weights of the CNN convolutional layers are frozen, the fully connected layers are trained in a lightweight manner to generate local compensation coefficients to update parameters, and the parameters are uploaded to the federated server through differential privacy encryption. S4.3 Physical Constraint Correction: Receives global model parameters aggregated from the federated server, combines them with the chip FEM thermal-mechanical-electrical coupling transfer function matrix to construct a regularization term, and outputs the result while meeting accuracy requirements under physical consistency constraints.

4. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 3, characterized in that: In step S4.1, cliff Defined as: ； Where T is the temperature difference vector, μ is the mean, and σ is the standard deviation. This represents the expectation operation; Spectral entropy Defined as: ； in is the energy proportion of the i-th frequency band in the normalized vibrational power spectral density, and n is the number of frequency bands; Electromagnetic shielding effectiveness Defined as: ； in Electromagnetic field strength under reference conditions The average field strength; The relative rate of change of vibrational spectrum entropy is defined as: ； in The current frame spectral entropy, Baseline spectral entropy; Electromagnetic shielding effectiveness attenuation is defined as follows: ； in This represents the output field strength at a frequency of 5MHz. Output field strength values ​​for historical static baselines; The rate of change of relative humidity is defined as: ； in The relative humidity value being measured. This serves as the humidity baseline.

5. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 4, characterized in that: In step S4.1, the temperature gradient kurtosis >4.0 satisfies the drift criterion; >0.05 relative change rate of vibration spectrum entropy satisfies the drift criterion; >20dB attenuation of electromagnetic shielding effectiveness satisfies the drift criterion; >0.1 relative humidity change rate and duration exceeding 30s satisfy the drift criterion.

6. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 3, characterized in that: In step S4.2, the compensation coefficient is defined as: ； in It is a multidimensional physical field drift characteristic vector, composed of temperature gradient kurtosis, vibration spectrum entropy, and electromagnetic shielding attenuation factor; This is the nonlinear compensation function obtained by mapping from the CNN output layer; Update parameters for weighted average aggregation; Defined as: ； in This represents the number of incremental samples for the i-th device; This represents the change in the fully connected layer weights calculated by the i-th device. Defined as: ； in This represents the weights of the fully connected layer after training on the i-th device. This represents the weights of the original fully connected layer.

7. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 3, characterized in that: In step S4.3, the thermo-mechanical-electric coupling transfer function matrix includes the transfer function in the frequency domain. The parasitic capacitance change caused by thermal deformation and its transmission through the transfer function are characterized. The change in carrier mobility caused by vibrational stress; The model's predicted output is aligned with the theoretical output calculated based on the physical field input. Real-time correction is achieved by constructing a physical loss function, which is defined as follows: ;in =0.7、 =0.3 is the weighting coefficient for the loss term. This represents the current temperature gradient tensor. Represents the current vibrational spectrum tensor. This is the model's predicted output; The accuracy requirement is defined as: ； in This represents the FEM simulation response under thermo-mechanical-electric coupling. For multiphysics input vectors, This is the model's predicted output.

8. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 1, characterized in that: In step S5, the degradation mapping function of accumulated environmental stress is defined as: ； in The signal-to-noise ratio at the current time t. This is the zero-point drift amount. The cumulative environmental stress vector is defined as: ； in , , These represent the cumulative thermal stress, vibration stress, and humidity stress that the chip has endured.

9. The high-reliability and high-precision testing method for sensor chips in extreme environments according to claim 1, characterized in that: In step S6, the confidence interval is defined as: ； in This represents the mean of the performance parameters. Let n be the standard deviation and n be the sample size. For degrees of freedom The critical value of the t-distribution with a confidence level of 95%.

10. A high-reliability, high-precision testing system for sensor chips designed for extreme environments, characterized in that: The testing system is used in the testing method according to any one of claims 1-9, and the testing system includes a programmable environment simulation cavity, an isolated low-noise signal generation module, and a distributed high-speed data acquisition network; The programmable environment simulation cavity is used to provide temperature, mechanical vibration, electromagnetic, and humidity fields for applying to the sensor chip under test. The isolated low-noise signal generation module is used to output excitation signals and power disturbance signals to the sensor chip under test. The distributed high-speed data acquisition network is used to acquire and preprocess the captured signal data.

Images