A semiconductor chip full life cycle continuous detection method and device

Abstract

The application discloses a kind of semiconductor chip full life cycle continuous detection method and device, it is related to chip detection field, including, obtain the electromagnetic radiation signal of monitoring chip under chaotic operation mode, construct electromagnetic radiation baseline atlas by topological signal processing, according to electromagnetic radiation baseline atlas, using chaotic sensor real-time acquisition electromagnetic signal and working condition parameter, and carry out nonlinear frequency spectrum difference analysis and generate dynamic electromagnetic anomaly index, according to the comparison result of dynamic electromagnetic anomaly index and preset chaotic threshold value triggers intelligent early warning mechanism, control chip enters self-organizing diagnostic mode and establishes holographic optical reflection benchmark map;The holographic optical reflection benchmark map of the present application is coupled with electromagnetic anomaly and optical scanning closely by phase space reconstruction, accurately identifies the topological interaction of collaborative defect cluster, eliminates the problem of positioning error amplification.

Description

Technical Field

[0001] This invention relates to the field of chip testing, and in particular to a method and apparatus for continuous testing of semiconductor chips throughout their entire lifecycle. Background Technology

[0002] The semiconductor industry is currently experiencing rapid development, particularly with Moore's Law driving feature sizes down to the sub-nanometer level. The complexity and functional density of integrated circuit chips are growing exponentially, leading to their widespread application in fields such as artificial intelligence, automotive electronics, and quantum computing. Traditional chip reliability assessment primarily relies on wafer-level testing during the manufacturing phase and post-packaging functional verification, such as using automated testing equipment for parameter scanning and boundary scanning. Meanwhile, online monitoring methods introduced during the usage phase, such as electromagnetic compatibility analysis and thermal imaging, are gradually being integrated into predictive maintenance frameworks. In recent years, the application of nonlinear dynamics and topological data analysis in signal processing has injected new vitality into chip fault diagnosis, while photoluminescence mapping and holographic interferometry have improved the resolution of optical nondestructive testing.

[0003] There are still areas for improvement in existing chip lifecycle inspection methods. First, traditional electromagnetic radiation monitoring is mostly limited to linear spectrum analysis, which makes it difficult to capture the nonlinear dynamic behavior of chips under complex operating conditions. This results in early degradation signals being submerged by noise, making it impossible to construct an accurate baseline spectrum. Consequently, high false positive rates occur in real-time differential analysis, affecting detection sensitivity. Second, the early warning mechanism lacks adaptive chaos thresholds and self-organizing diagnostic modes, leading to excessively long chip service interruption times. Furthermore, the establishment of holographic optical reflection benchmarks often ignores phase space reconstruction, causing decoupling between optical scanning and electromagnetic anomalies. This makes it difficult to accurately identify the topological interaction relationships of cooperative defect clusters, further amplifying the positioning error. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for continuous monitoring of semiconductor chips throughout their entire lifecycle to solve the problem of difficulty in capturing the nonlinear dynamic behavior of chips under complex operating conditions, which leads to early degradation signals being overwhelmed by noise.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The electromagnetic radiation signal of the chip under monitoring in chaotic working mode is acquired, and an electromagnetic radiation baseline map is constructed through topological signal processing. Based on the electromagnetic radiation baseline map, a chaotic sensor is used to collect electromagnetic signals and operating parameters in real time, and nonlinear spectrum difference analysis is performed to generate a dynamic electromagnetic anomaly index. The intelligent early warning mechanism is triggered based on the comparison result between the dynamic electromagnetic anomaly index and the preset chaos threshold. The control chip enters the self-organizing diagnosis mode and establishes a holographic optical reflection reference map. Based on the holographic optical reflection reference map, a multidimensional photoluminescence intensity distribution map is generated by performing photoluminescence scanning through a quantum dot laser scanning device, which intelligently identifies defect regions and forms a collaborative defect distribution topology map. A neuromorphic computational model was established and used as an optimization criterion to perform data fusion analysis on the collaborative defect distribution topology map, and output chip detection results.

[0007] As a preferred embodiment of the continuous monitoring method for the entire lifecycle of semiconductor chips described in this invention, the method involves: acquiring the electromagnetic radiation signal of the chip under monitoring in a chaotic operating mode, and constructing an electromagnetic radiation baseline map through topological signal processing. Specifically, The chip is put into a chaotic working mode, and the resulting broadband electromagnetic radiation signal is collected through the electromagnetic transmission window. The signal is then processed by removing interference, normalizing, and extracting phase information to obtain a preprocessed signal. The preprocessed signal is mapped to the topological space, and topological invariants are extracted. A signal phase space reconstruction model is constructed based on topological invariants. High-dimensional feature vectors are generated by the delayed embedding method and associated with the corresponding operating mode parameters to form a multi-mode electromagnetic radiation baseline map.

[0008] As a preferred embodiment of the continuous monitoring method for the entire lifecycle of semiconductor chips described in this invention, the method involves: using a chaotic sensor to collect electromagnetic signals and operating parameters in real time based on an electromagnetic radiation baseline spectrum, and performing nonlinear spectral difference analysis to generate a dynamic electromagnetic anomaly index. Specifically... A reference spectrum matching the current operating parameters is retrieved from the multi-mode electromagnetic radiation baseline map. Real-time electromagnetic signals, temperature and voltage operating parameters are synchronously acquired through a chaotic sensor. The instantaneous spectral characteristics are obtained by performing a short-time Fourier transform on the real-time acquired electromagnetic signals. The difference spectrum between the instantaneous spectral characteristics and the reference spectrum is calculated. The Lyapunov exponent and fractal dimension of the difference spectrum are analyzed using a nonlinear dynamics method and then fused to generate a dynamic electromagnetic anomaly index.

[0009] As a preferred embodiment of the continuous monitoring method for the entire lifecycle of semiconductor chips described in this invention, the method includes: triggering an intelligent early warning mechanism based on the comparison result between the dynamic electromagnetic anomaly index and a preset chaos threshold, controlling the chip to enter a self-organizing diagnostic mode and establishing a holographic optical reflection reference map. Specifically, Continuously monitor the changes in the dynamic electromagnetic anomaly index and immediately trigger an early warning signal when it exceeds the preset chaos threshold; The chip's bias voltage and clock frequency parameters are adjusted according to the warning signal, so that the chip enters a low-power self-organizing diagnostic mode. In the low-power self-organizing diagnostic mode, the chip surface is scanned by an integrated optical probe to collect the reflected light intensity distribution data. The reflected light intensity distribution data is holographically encoded to generate a hologram, and the three-dimensional optical reflection field is reconstructed based on the hologram to form a holographic optical reflection reference map.

[0010] As a preferred embodiment of the continuous inspection method for the entire lifecycle of semiconductor chips described in this invention, the method involves: based on a holographic optical reflection reference image, performing photoluminescence scanning using a quantum dot laser scanning device to generate a multidimensional photoluminescence intensity distribution map, intelligently identifying defect regions and forming a collaborative defect distribution topology map. Specifically... Extracting candidate defect regions from a holographic optical reflection reference image; A quantum dot laser scanning tool is used to emit a wavelength laser beam to scan the defect candidate region point by point, and the photoluminescence signal of each scanning point is collected simultaneously. The photoluminescence signal is mapped to a color space to generate a multidimensional photoluminescence intensity distribution map, and a clustering algorithm is used to identify abnormal brightness regions and calculate defect boundaries and size parameters. A collaborative defect distribution topology map is constructed based on defect candidate regions, photoluminescence signal data, multidimensional photoluminescence intensity distribution map, defect boundaries, and size parameters.

[0011] As a preferred embodiment of the continuous inspection method for the entire lifecycle of semiconductor chips described in this invention, the method includes: establishing a neuromorphic computational model and using this model as an optimization criterion to perform data fusion analysis on the cooperative defect distribution topology map, and outputting chip inspection results. Specifically, A spiking neural network model was constructed, and the topology map of the collaborative defect distribution was converted into a spiking sequence and input into the neural network. The defect evolution dynamics are simulated through neuromorphic computation, and a defect growth probability distribution map is output. The remaining lifespan of the chip is calculated based on the defect growth probability distribution map, and the weights of the spiking neural network model are optimized by integrating real-time operating parameters to generate a complete chip inspection report.

[0012] As a preferred embodiment of the continuous monitoring method for the entire lifecycle of semiconductor chips described in this invention, wherein: mapping the preprocessed signal to the topological space specifically involves, A simplex complex model of the preprocessed signal is established, the homology group generator is calculated, and the key topological structures are selected and quantized into topological feature vectors. Calculate the topological entropy based on the topological feature vectors and convert it into topological space coordinates to complete the mapping.

[0013] As a preferred embodiment of the continuous monitoring method for the entire lifecycle of semiconductor chips described in this invention, the Lyapunov exponent and fractal dimension of the difference spectrum using nonlinear dynamics refers to reconstructing the phase space of the difference spectrum between the instantaneous spectral characteristics and the reference spectrum, calculating the maximum Lyapunov exponent using the Wolf algorithm, and simultaneously calculating the fractal dimension of the difference spectrum.

[0014] As a preferred embodiment of the continuous monitoring method for the entire lifecycle of semiconductor chips described in this invention, the step of performing holographic encoding processing on the reflected light intensity distribution data to generate a hologram specifically involves: The reflected light intensity data is converted into a complex field form, and a frequency domain hologram is generated through Fourier transform. After optimizing the contrast in the frequency domain hologram, the spatial hologram is reconstructed using inverse Fourier transform.

[0015] Secondly, the present invention provides a continuous monitoring device for the entire lifecycle of a semiconductor chip, comprising, The acquisition module acquires the electromagnetic radiation signal of the chip under monitoring in chaotic working mode and constructs an electromagnetic radiation baseline map through topological signal processing. The analysis module uses a chaotic sensor to collect electromagnetic signals and operating parameters in real time based on the electromagnetic radiation baseline map, and performs nonlinear spectrum difference analysis to generate a dynamic electromagnetic anomaly index. The early warning module triggers an intelligent early warning mechanism based on the comparison result between the dynamic electromagnetic anomaly index and the preset chaos threshold, and controls the chip to enter the self-organizing diagnosis mode and establish a holographic optical reflection reference map. The identification module, based on a holographic optical reflection reference map, performs photoluminescence scanning using a quantum dot laser scanning device to generate a multidimensional photoluminescence intensity distribution map, intelligently identifying defect regions and forming a collaborative defect distribution topology map; The detection module establishes a neuromorphic computing model and uses it as an optimization criterion to perform data fusion analysis on the collaborative defect distribution topology map, outputting chip detection results.

[0016] The beneficial effects of this invention are as follows: By constructing an electromagnetic radiation baseline map through chaotic working mode excitation and topological signal processing, nonlinear dynamic behavior is captured and noise interference is decoupled, an accurate baseline is established, and high sensitivity and low false positive rate of real-time differential analysis are achieved; at the same time, an intelligent early warning mechanism with dynamic electromagnetic anomaly index and adaptive chaotic threshold is introduced to drive the switching of self-organizing diagnostic mode, significantly shortening the service interruption time; and by tightly coupling electromagnetic anomalies and optical scanning through a holographic optical reflection reference map reconstructed in phase space, the topological interaction relationship of cooperative defect clusters is accurately identified, eliminating the problem of amplified positioning error, and ultimately achieving a breakthrough in zero interruption throughout the entire life cycle, high sensitivity to weak signals, and accurate prediction of cooperative degradation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for a continuous monitoring method throughout the entire lifecycle of a semiconductor chip.

[0019] Figure 2 A flowchart for generating dynamic electromagnetic anomaly indices.

[0020] Figure 3 A flowchart for establishing a holographic optical reflection reference map.

[0021] Figure 4 This is a flowchart for defect detection and result output. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for continuous monitoring of a semiconductor chip throughout its entire lifecycle, including the following steps: S1: Acquire the electromagnetic radiation signal of the chip under monitoring in chaotic working mode, and construct the electromagnetic radiation baseline map through topological signal processing; S1.1: The monitored chip is put into a chaotic operating mode. In this mode, a specific nonlinear perturbation excitation signal is injected from an external circuit board into the chip's power supply pins and clock input via a chip test interface (e.g., JTAG or a dedicated perturbation injection pin). This signal uses a double-vortex chaotic waveform generated by a Chua circuit, characterized by non-periodicity and an amplitude range of ±10% of the chip's nominal supply voltage (e.g., ±0.1V perturbation at a nominal 1.0V). The spectrum covers a continuous wideband from 100kHz to 10MHz. The signal is modulated with a pseudo-random sequence to ensure non-periodicity and repeatability. The nonlinear perturbation excitation signal is applied to the chip externally. The chaotic signal source can be generated by an external precision signal generator or by the Chua circuit oscillator module built into the chaotic sensor, and then coupled... A network (e.g., coupling capacitors) is injected into the chip's power rails and / or clock input. The disturbance intensity is adjusted in real time by external control circuitry or embedded control logic. To maintain the chip in a chaotic state, the disturbance amplitude is dynamically adjusted through a feedback loop, keeping the frequency jitter of the internal ring oscillator within the 5%–15% range. Simultaneously, the chip temperature is monitored to ensure it does not exceed 85°C, guaranteeing that the nonlinear dynamic response induced by transistor-level electromigration and thermal stress is fully activated, thereby generating a broadband electromagnetic radiation signal that characterizes the chip's inherent electromagnetic leakage features. A high-sensitivity near-field probe array is used to collect the broadband electromagnetic radiation signal through an electromagnetic transmission window pre-set on the chip package surface. The electromagnetic transmission window is located in a predetermined area of ​​the chip package shell, formed using a packaging material with high electromagnetic wave transmittance or a locally thinned structure to enhance the efficiency of external coupling of the internal electromagnetic radiation signal. The near-field electromagnetic probe array is mounted on the outer surface of the chip package via a fixed bracket. The bracket is a rigid or elastic structure, with one end fixed to the high-sensitivity near-field probe array and the other end fixed to the chip package shell, maintaining a relatively fixed spatial position between the high-sensitivity near-field probe array and the electromagnetic transmission window. The fixed bracket is equipped with a limiting structure to limit the distance between the near-field probe array and the electromagnetic transmission window. The distance is set within the range of 0.1mm to 1mm to ensure stable near-field coupling and avoid mechanical contact. A low-dielectric-constant insulating dielectric layer can be placed between the high-sensitivity near-field probe array and the chip package to reduce the impact of environmental vibration and temperature changes on measurement stability, thereby ensuring the consistency and repeatability of electromagnetic radiation signal acquisition. The broadband electromagnetic radiation signal covers the time-domain waveform from DC to GHz, and the ring oscillator frequency at the acquisition time is recorded simultaneously to calibrate the time reference. Wavelet denoising algorithm is applied to the acquired broadband electromagnetic radiation signal to remove environmental electromagnetic interference and baseline drift, and normalization processing is performed to unify the amplitude scale. Instantaneous phase information is extracted through Hilbert transform to obtain the preprocessed broadband electromagnetic radiation signal, which is then combined into a preprocessed broadband electromagnetic radiation signal sequence.

[0026] The electromagnetic transmission window preset on the chip package surface is preset based on the electromagnetic wave transmittance of the chip package material (such as epoxy resin).

[0027] To further explain, the process of extracting instantaneous phase information using Hilbert transform is as follows: The Hilbert transform result of the preprocessed broadband electromagnetic radiation signal is calculated using the Fast Fourier Transform algorithm. In the frequency domain, positive frequency components are retained, negative frequency components are set to zero, and a 90-degree phase shift is applied before inverse transforming back to the time domain to obtain the accompanying signal, i.e., the Hilbert transform result. An analytic signal is then constructed using the accompanying signal, which is composed of the preprocessed broadband electromagnetic radiation signal sequence as the real part and the accompanying signal as the imaginary part. The instantaneous phase is extracted from the analytic signal by calculating the arctangent ratio of the accompanying signal and the original signal in four quadrants, and the phase value range is output. To ensure phase continuity, phase unwrapping processing is further applied to detect transitions between adjacent samples and cumulative corrections are performed to obtain a smooth sequence, thus completing the extraction of instantaneous phase information.

[0028] It should also be noted that the chaotic sensor is an integrated hardware module used to synchronously acquire electromagnetic radiation signals during chip operation and introduce a chaotic reference signal to enhance the perception of nonlinear dynamic characteristics. The chaotic sensor includes: a near-field electromagnetic probe array, a front-end signal conditioning module, a chaotic reference signal generation module, a nonlinear coupling and synchronization detection module, and an analog-to-digital conversion and digital processing module. The near-field electromagnetic probe array is used to acquire electromagnetic radiation signals leaked during chip operation. The output of the probe array is connected to the front-end signal conditioning module via a coaxial line or microstrip line. The front-end signal conditioning module includes a low-noise amplifier and a bandpass filter circuit, used to amplify the amplitude and select the frequency band of the acquired electromagnetic signal, and to achieve impedance matching to reduce reflection and noise interference. The chaotic reference signal generation module is implemented using a Chua circuit, which consists of nonlinear resistive elements, inductors, capacitors, and operational amplifiers. Its output generates a non-periodic chaotic oscillation signal as a reference chaotic signal output. The nonlinear coupling and synchronization detection module receives the signal-conditioned electromagnetic radiation signal and the chaotic reference signal, respectively, and achieves signal coupling through an analog multiplier or a nonlinear operational amplifier circuit to generate a detection signal reflecting the synchronization error between the two. When the nonlinear dynamic characteristics of the chip change, the amplitude and spectral characteristics of the synchronization error signal change accordingly. After being digitized by the analog-to-digital converter module, the synchronization error signal is sent to the embedded microcontroller, which performs subsequent spectral analysis, nonlinear feature extraction, and calculation of the dynamic electromagnetic anomaly index.

[0029] The above modules are integrated via printed circuit boards. The power supply and signal sections are each set with independent ground planes. The chaotic reference signal generation module and the signal conditioning module are electromagnetically shielded by metal shielding to reduce the impact of external interference on the measurement results.

[0030] A simplex model of the preprocessed broadband electromagnetic radiation signal sequence is established using the Vietoris-Rips complex construction method. In simpler terms, the preprocessed broadband electromagnetic radiation signal sequence is treated as point cloud data, and a set of simplexes is generated at selected scale parameters to capture the signal's geometric structure. The homology group generator of the simplex model is calculated by tracking the birth and death events of topological features at different filtering scales using a persistent homology method, generating persistent maps and barcodes to obtain the homology group generator. Based on the homology group generator, a persistence threshold is set to filter out key topological structures. This persistence threshold distinguishes between transient topological features induced by noise and stable topological invariants reflecting the chaotic dynamics of the chip. The persistence threshold is set based on the relative scale of the persistence length, which is the difference between the birth and death scales of the topological feature. In this embodiment, a persistence length exceeding 10% of the signal mean can be used as the persistence threshold to filter out noise-induced transient features and retain stable invariants related to the chaotic dynamics of the chip. The persistence threshold is selected based on the statistical difference in persistence length distribution between noise features and stable topological features in the chip's electromagnetic radiation signal: noise features typically exhibit short-lasting structures that decay rapidly with scale, while stable topological features exhibit long-lasting structures that persist over a wider scale range. In specific experimental sample analysis, after performing persistence coherence calculations on the electromagnetic radiation signals of normal and faulty chips, it was observed that the persistence length of noise features is mainly concentrated in a lower proportion range, while the persistence length of stable topological features is distributed in a higher proportion range. Therefore, selecting a persistence threshold on the order of ten percent of the signal mean can suppress noise features while preserving the main stable topological structure. The persistence threshold can be adjusted according to the chip type, signal-to-noise ratio level, or application scenario. The preserved stable invariants are quantized into topological feature vectors, which are multi-dimensional numerical arrays encoding the statistical distribution of the topological invariants.

[0031] Topological invariants are extracted from the topological feature vectors, normalized to probability distributions, and Shannon entropy is calculated to obtain initial topological entropy values. The evolution of topological invariants at different scales is analyzed, and the initial topological entropy is adjusted by calculating the rate of change of entropy with scale. The adjusted initial topological entropy value is normalized to the range of 0 to 1 to obtain the final entropy value, which is the topological entropy that quantifies the degree of chaos of the signal. The larger the value, the more complex and chaotic the chip's working state corresponding to the signal. The topological entropy value is directly used as a one-dimensional coordinate of the topological space to complete the mapping of the signal from the original domain to the topological space.

[0032] S1.2: The optimal embedding dimension and delay time are determined based on topological invariants using the delayed embedding theorem. The delay time is derived from the connectivity index of the topological feature vectors using the spurious nearest neighbor method, ensuring that the reconstructed phase space retains the topological equivalence of the original chaotic attractor. The derivation process is as follows: the connectivity index is extracted from the topological feature vectors and used as the input criterion for the spurious nearest neighbor method; the spurious nearest neighbor method is applied to successively increase the delay time, and the distance evolution of the nearest neighbor points in the phase space is calculated. The iteration stops when the proportion of spurious neighbors drops below 5%, yielding the optimal delay time. The embedding dimension ranges from 3 to 8 (determined through cross-validation using the spurious nearest neighbor method based on the connectivity index of the topological feature vectors); the preprocessed broadband electromagnetic radiation is then embedded using the delayed embedding method. The signal sequence is reconstructed into vector form, with each vector consisting of signal values ​​at a time delay interval, forming high-dimensional phase space trajectory points. Principal component analysis is applied to reduce the dimensionality of the high-dimensional phase space trajectory to highlight the manifold structure dominated by topological invariants. The geometric measures of the phase space trajectory, including curvature and torsion, are integrated with the coordinate mapping of topological feature vectors to generate high-dimensional feature vectors. A hash mapping mechanism is used to associate and encode the high-dimensional feature vectors with the corresponding operating mode parameters, including the injected perturbation amplitude, frequency range, and load conditions. By binding the high-dimensional feature vectors with the operating mode parameters, a multi-mode electromagnetic radiation baseline map is formed. The multi-mode electromagnetic radiation baseline map is formed by storing vector clusters from multiple modes in a database structure, serving as a benchmark reference for subsequent real-time comparisons.

[0033] Preferably, this invention introduces nonlinear perturbation excitation and topological signal processing under chaotic operating mode when constructing the electromagnetic radiation baseline map. This significantly improves the accuracy of capturing the inherent electromagnetic leakage characteristics of the chip and avoids the sensitivity of traditional linear spectrum analysis to noise and the omission of early degradation signals. By using persistent cohomology method and topological entropy mapping to achieve accurate conversion of signals from the original domain to the topological space, the baseline map's adaptability to operating condition variations and its inclusiveness of individual chip differences are enhanced. Combined with the phase space reconstruction of delayed embedding theorem and spurious nearest neighbor method, the high-dimensional feature vectors are ensured to retain the topological equivalence of chaotic attractors, significantly improving the stability of the baseline map and the reliability of real-time comparison, laying a solid foundation for subsequent anomaly detection and lifetime prediction.

[0034] S2: Based on the electromagnetic radiation baseline map, use a chaotic sensor to collect electromagnetic signals and operating parameters in real time, and perform nonlinear spectrum difference analysis to generate a dynamic electromagnetic anomaly index. A reference spectrum matching the current operating parameters is retrieved from a multi-mode electromagnetic radiation baseline map. The reference spectrum includes voltage, temperature, and load status. A chaotic sensor, an integrated hardware module, is used to synchronously acquire the chip's electromagnetic radiation signals and introduce a chaotic reference signal during chip operation, thereby enhancing the perception of the chip's nonlinear dynamic characteristics. The chaotic sensor is used to synchronously acquire real-time electromagnetic signals and operating parameters. The real-time electromagnetic signals are directly acquired through a wideband probe array, while the operating parameters are synchronously acquired through an integrated temperature sensor and voltage sampling circuit. Short-time Fourier transform is applied to the real-time electromagnetic signals and operating parameters to generate instantaneous spectral features. The instantaneous spectral features are then compared with the reference spectrum to obtain a differential spectrum. Specifically, the real-time electromagnetic signal is divided into multiple time windows, and the signal within each time window is frequency-domain transformed using short-time Fourier transform to generate instantaneous spectral features containing frequency, amplitude, and phase information. The amplitude difference and phase difference are calculated point-by-point between the frequency components of the instantaneous spectral features and the corresponding frequency components of the reference spectrum. The difference results are integrated to generate a differential spectrum, which reflects the deviation between the real-time electromagnetic signal and the reference spectrum in the frequency domain.

[0035] The embedding dimension and delay time are selected from the frequency domain feature distribution of the difference spectrum using the delay embedding theorem. The embedding dimension ranges from 3 to 8, and the delay time ranges from 2 to 10 times the signal sampling interval. The embedding dimension and delay time are determined for the frequency domain sequence of the difference spectrum using the spurious nearest neighbor method. The frequency domain sequence of the difference spectrum is converted into a high-dimensional phase space trajectory composed of delay coordinate vectors. Initial point pairs are selected from the high-dimensional phase space trajectory, and the divergence rate of the initial point pairs is tracked. The maximum Lyapunov exponent is generated by iteratively taking the logarithmic average. The frequency domain sequence of the difference spectrum is divided into multi-scale grids, and the number of grids covering the difference spectrum data at different scales is counted. The slope of the logarithmic relationship between the number of grids and the scale is fitted to generate the fractal dimension.

[0036] The maximum Lyapunov exponent and fractal dimension are combined using a linear weighting method. The weight parameters in the dynamic electromagnetic anomaly exponent are adaptively determined based on statistical characteristics. Specifically, correlation or regression analysis is performed on the nonlinear characteristic parameters of the electromagnetic signals acquired in real time or initially to calculate the degree of correlation between each characteristic parameter and the anomaly state discrimination result, and corresponding weights are assigned according to the degree of correlation. In practice, the weights can be initialized or calibrated by combining historical operating data or initial calibration data.

[0037] Preferably, this invention achieves accurate differential analysis of real-time electromagnetic signals by matching operating parameters to retrieve the reference spectrum and generating instantaneous spectral features using short-time Fourier transform, avoiding the omission of dynamic deviations in traditional static spectrum comparison. It employs phase space reconstruction using the delayed embedding theorem and the spurious nearest neighbor method, combined with the Wolf algorithm to extract the maximum Lyapunov exponent, and generates the fractal dimension using the correlation dimension method, comprehensively capturing the nonlinear chaotic characteristics of the differential spectrum and overcoming the blind spot of linear analysis for early degradation signals. Linear weighted fusion of two nonlinear indices generates a dynamic electromagnetic anomaly index, improving the sensitivity and robustness of anomaly detection and providing a high-precision quantitative benchmark for chip lifecycle monitoring.

[0038] S3: Based on the comparison result between the dynamic electromagnetic anomaly index and the preset chaos threshold, the intelligent early warning mechanism is triggered, the control chip enters the self-organizing diagnosis mode and establishes a holographic optical reflection reference map. S3.1: Continuously monitor the numerical changes of the dynamic electromagnetic anomaly index and compare it with the chaos threshold, which is determined adaptively. Specifically: Under normal chip operation, collect a dynamic electromagnetic anomaly index sequence over a period of time (preferably 3-5 minutes), analyze the statistical characteristics of the dynamic electromagnetic anomaly index sequence, and determine the chaos threshold range based on the mean and fluctuation range of the dynamic electromagnetic anomaly index sequence to distinguish between normal chaotic dynamics and abnormal degenerative states; during actual operation, the chaos threshold can be dynamically adjusted according to real-time operating parameters.

[0039] When the dynamic electromagnetic anomaly index exceeds the preset chaos threshold, an early warning signal is immediately generated. The early warning signal is a dedicated interrupt signal inside the chip, which is broadcast to the power management unit (PMU) and the embedded microcontroller (Embedded-MCU) through the interrupt controller. The change in the value of the dynamic electromagnetic anomaly index is monitored in real time by the embedded microcontroller. When the value exceeds the chaos threshold, the embedded microcontroller immediately sets the early warning flag bit in the dedicated control register (e.g., bit[7]=1). The early warning flag bit is connected to the PMU and the diagnostic circuit through the internal bus (e.g., AMBA-APB), thereby triggering the chip to enter the low-power self-organizing diagnostic mode.

[0040] The warning signal triggers the chip's internal power management unit (PMU). Upon receiving the warning flag, the PMU immediately executes a pre-configured buck and frequency reduction script, adjusting the chip's bias voltage to 70% of its nominal value and lowering the clock frequency to 50% of its normal operating frequency to reduce power consumption and maintain basic functionality. This enters a low-power self-organizing diagnostic mode, a special operating state supported by the chip's internal hardware resources and firmware protocols: a built-in Dynamic Voltage and Frequency Scaling (DVFS) controller in the PMU; a programmable clock network (Clock-Tree) supporting multi-domain clock gating and division; a built-in ring-oscillator array for real-time frequency jitter monitoring; and redundant logic blocks such as spare registers. The system includes a power management unit (PMU) and self-test circuitry, as well as a dedicated diagnostic interface (JTAG or IEEE-1687-IJTAG network). The switching process is controlled by a firmware protocol: a warning signal triggers the power management unit (PMU) firmware to execute a pre-configured buck and frequency reduction script, while simultaneously sending gating commands to the configurable clock network via the IJTAG interface to disable unnecessary functions and enable the ring oscillator array and self-test circuitry. During operation, the distributed control algorithm is executed by the embedded microcontroller (MCU). The MCU coordinates the PMU, clock network, and redundant modules through an internal bus (such as AMBA-APB), dynamically allocating resources based on self-test feedback, disabling unnecessary functions, maintaining minimum power consumption, and enabling the internal diagnostic circuitry to perform self-test tasks. The integrated optical probe scans the chip surface in low-power self-organizing diagnostic mode to collect reflected light intensity distribution data.

[0041] To further explain, the pre-configured buck and frequency reduction script is a piece of firmware code that is embedded by the manufacturer in a non-volatile memory area (such as OTP, eFuse, or Flash) before the chip leaves the factory. It is usually no more than 512 bytes in length and is written in assembly or C language. Its purpose is to achieve a controllable and smooth reduction of voltage and frequency after receiving a warning signal, so as to avoid sudden changes that cause the chip to reset, lock up, or become unstable.

[0042] S3.2: Convert the reflected light intensity distribution data into a complex field form and apply Fourier transform to generate a frequency domain hologram. Specifically, assign a real part value to each data point of the reflected light intensity distribution data, and generate a corresponding imaginary part value through a phase estimation algorithm. The phase estimation algorithm adopts the Gerchberg-Saxton iterative algorithm, which uses the known reflected light intensity amplitude as a constraint. The initial phase is estimated using random phase or interference fringes. Through multiple Fourier transform and inverse transform iterations, the amplitude constraint is forced and the phase information is retained in each iteration. The number of iterations is 100 to 200. The phase offset of each pixel is calculated as the imaginary part value. The real and imaginary parts are combined to form a complex field form. Perform a fast Fourier transform on the data in the complex field form to convert it to the frequency domain and generate a frequency domain hologram containing amplitude and phase information.

[0043] Amplitude component data is separated from the frequency domain hologram. A logarithmic transform is applied as a nonlinear mapping function to process the amplitude component data, compressing high amplitude values ​​and stretching low amplitude values ​​to expand the dynamic range of the high-frequency and low-frequency regions of the frequency domain hologram and enhance the contrast difference between them. This data is then recombined with the phase component of the frequency domain hologram to form an optimized frequency domain hologram. An inverse fast Fourier transform is performed on the optimized frequency domain hologram to convert it back to the spatial domain, generating a spatial hologram. Phase and amplitude information are extracted from the spatial hologram, and the Fresnel diffraction integral algorithm is applied to numerically integrate the phase and amplitude information to reconstruct the three-dimensional reflection field distribution on the chip surface. A 532nm continuous-wave solid-state laser is used as the light source, and a 2048×2048 pixel CMOS image sensor with a pixel size of 3.45μm is used as the detector. The surface height and reflection intensity characteristics of the three-dimensional reflection field are extracted. A holographic optical reflection reference map is generated based on the surface height and reflection intensity characteristics of the three-dimensional reflection field. This reference map stores the three-dimensional reflection characteristics of the chip surface in a gridded data structure for subsequent comparative analysis in defect detection.

[0044] Preferably, compared with traditional detection methods, this invention achieves intelligent early warning triggering by comparing the dynamic electromagnetic anomaly index with the chaotic threshold step by step, avoiding the hysteresis response of static thresholds to early anomalies; the early warning signal drives the precise adjustment of bias voltage and clock frequency, combined with a distributed self-organizing diagnostic mode, significantly reducing power consumption and maintaining diagnostic continuity; the construction of the holographic optical reflection reference map adopts Fourier domain nonlinear contrast optimization and Fresnel reconstruction to enhance the resolution and robustness of the three-dimensional reflection field, overcome the distortion of surface microstructure by traditional optical detection, and improve defect positioning accuracy and reference stability.

[0045] S4: Based on the holographic optical reflection reference map, a multidimensional photoluminescence intensity distribution map is generated by performing photoluminescence scanning through a quantum dot laser scanning device, which intelligently identifies defect regions and forms a collaborative defect distribution topology map; S4.1: Analyze the reflection intensity characteristics of the holographic optical reflection reference map, identify areas with reflection intensity below the average value, and mark these areas as candidate defect areas; adjust the wavelength of the quantum dot laser scanning tool to match the bandgap of the chip material, illuminate the candidate defect areas point by point, and record the intensity, wavelength drift, and decay time of the photoluminescence signal at each scanning point.

[0046] The intensity, wavelength drift, and decay time of the photoluminescence signal are mapped to three-dimensional color space coordinates to generate a multidimensional photoluminescence intensity distribution map. Points with brightness values ​​lower than a brightness threshold are selected from the multidimensional photoluminescence intensity distribution map as the initial dataset, where the brightness threshold is determined by an adaptive method based on distribution sorting. Specifically, the process involves: sorting or performing quantile analysis on the photoluminescence brightness values ​​within the current scanning area; identifying data points in the low-end range of brightness distribution as candidate points for abnormal brightness, thereby enabling the identification of relatively low-luminescence regions; in practice, the quantile range can be dynamically adjusted based on scanning resolution, region size, or operating parameters; the low-end range of brightness distribution is determined through relative comparison: comparing the photoluminescence brightness of each scanning point with the average brightness or local neighborhood average brightness within the current scanning area, and identifying data points with brightness lower than the corresponding reference brightness as being in the low-end range of brightness distribution; initializing k cluster centers, for example, k=2, as these correspond to normal and abnormal brightness regions respectively; iteratively performing distance allocation from points to cluster centers and center position updates until the cluster centers no longer change, indicating convergence; marking points assigned to abnormal brightness cluster centers as abnormal brightness regions; applying a contour tracing algorithm to the point set of abnormal brightness regions, sequentially traversing along the boundary pixels of the abnormal brightness regions to generate closed boundary contours; calculating the area size parameter enclosed by the boundary contours using a pixel counting method, expressed as: ; in, This parameter represents the area enclosed by the boundary contour, in pixels, and quantifies the physical size of the defect region. Indicates the first [unit] within the abnormal brightness area 1 pixel This represents the total number of pixels in the set of points in the abnormal brightness area.

[0047] S4.2: Extract the center coordinates from the defect candidate region as location information, extract the intensity mean and wavelength drift from the photoluminescence signal data as features, extract the clustering labels of abnormal brightness regions from the multidimensional photoluminescence intensity distribution map, extract the area size parameters from the defect boundary and size parameters, and store this information as the attribute vector of the defect region.

[0048] Using the attribute vector of each defect region as a node, the spatial distance between nodes is calculated using the Euclidean distance formula, and the brightness similarity is calculated using cosine similarity. The spatial distance and brightness similarity are fused to generate edge weights, which are the weighted sum of the inverse of the spatial distance and the brightness similarity. Nodes whose edge weights exceed the connection threshold form a node network. The connection threshold is set based on statistical analysis of historical defect distribution data. The historical defect distribution data was collected by running continuously for 1000 hours under high temperature, high humidity, and high voltage stress conditions, collecting defect location, brightness similarity, and spatial distribution data of more than 500 sets of degraded samples. The distribution of edge weights between nodes was calculated through statistical analysis (the mean of noisy connections is about 0.35, and the mean of true cooperative defect connections is about 0.78). The connection threshold is set to 0.6 (corresponding to the optimal cutoff point of the ROC curve), which can effectively filter out weak connections induced by noise and avoid excessively dense node networks that would increase computational complexity. If the value is higher than 0.6, it may disconnect weak interactions between early cooperative defects, leading to an increased false negative rate, while if it is lower than 0.6, it may retain too many noisy connections, leading to network redundancy and analysis errors.

[0049] The Louvain algorithm is used to optimize the node network, which is divided into multiple tightly connected defect clusters. Each defect cluster represents a group of collaboratively degenerate defect regions. The connectivity is obtained by dividing the total number of edges between nodes in each defect cluster by the number of node pairs within the cluster. The node with the highest connectivity in each defect cluster is used as the centrality index. The node network is arranged with the defect cluster as the central node. The node size represents the defect area and the edge thickness represents the edge weight. The defect clusters are distinguished by color coding to generate a collaborative defect distribution topology map, highlighting the spatial proximity and functional collaboration between defects.

[0050] To further explain, the spatial registration between the holographic optical reflection reference map and the quantum dot laser scanning device ensures precise physical correspondence through the following methods: The quantum dot laser scanning device employs a piezoelectric ceramic-driven two-dimensional robotic arm positioning system with a robotic arm accuracy of 0.1 μm. The scanning resolution matches the gridded data structure of the holographic optical reflection reference map. Coordinate system calibration is completed before the chip leaves the factory. Specifically, using known reference marker points (pre-etched cross alignment marks on the chip surface, 5 μm × 5 μm in size), a three-dimensional coordinate system is reconstructed from the holographic optical reflection reference map. Rigid body transformation calibration is performed with the robotic arm coordinate system, and the least squares method is used to fit the transformation matrix, with the calibration error controlled within 0.2 μm. During scanning, the center coordinates of the defect candidate region are directly mapped to the robotic arm coordinate system, with a point-to-point illumination position error of less than 0.3 μm. This ensures that the photoluminescence signal acquisition point and the three-dimensional reflection field of the holographic optical reflection reference map are precisely corresponding in physical position, avoiding misjudgment of defects caused by coordinate offset.

[0051] Preferably, this invention accurately extracts candidate defect regions from holographic optical reflection reference maps and combines them with multidimensional photoluminescence mapping via quantum dot laser scanning to achieve high-resolution defect identification, avoiding blind spots in weak light emission from traditional optical detection. It generates precise boundary and size parameters through k-means clustering and contour tracking, improving the geometric accuracy of defect localization. Furthermore, it integrates multi-source features to construct attribute vectors and uses graph theory and the Louvain algorithm to form a collaborative defect distribution topology map, revealing the spatial and functional interactions between defects. This overcomes the neglect of collaborative degradation by isolated detection and provides comprehensive topological insights for chip reliability assessment.

[0052] S5: Establish a neuromorphic computing model and use the neuromorphic computing model as an optimization criterion to perform data fusion analysis on the collaborative defect distribution topology map and output chip detection results.

[0053] A spiking neural network model was built using the Leaky-Integrate-and-Fire neuron model. The number of nodes in the input layer was set to be equal to the total number of nodes in the collaborative defect distribution topology. The hidden layer contained 3 layers, with 64 neuron clusters in each layer. Each neuron cluster contained 128 Leaky-Integrate-and-Fire neurons to simulate the dynamic interaction of defects. The number of nodes in the output layer was 5, corresponding to the predicted defect growth category (no growth, low-speed growth, medium-speed growth, high-speed growth, critical failure), generating defect growth predictions.

[0054] Each node in the collaborative defect distribution topology is encoded as a pulse firing rate. The pulse interval is modulated based on the node edge weights, mapping the topology of the node network into a pulse sequence. This sequence is then input into the spiking neural network model via an event-driven approach. After receiving the pulse sequence, each neuron in the spiking neural network model accumulates the pulse intensity from the input pulse sequence through synaptic weights. When the accumulated value exceeds the neuron threshold, a firing event is triggered. This firing event is used to simulate the expansion behavior of the defect region. The statistical characteristics of the input pulse sequence are determined by the neuron threshold, specifically: statistical analysis is performed on the accumulated intensity of the input pulse sequence within a preset time window, which is 10–200 milliseconds, preferably 20–100 milliseconds; the neuron threshold is determined based on the mean and fluctuation range of the accumulated value to represent the normal statistical level of the input pulse sequence; when the accumulated value of subsequent input pulse sequences exceeds the neuron threshold within the corresponding time window, a neuron firing event is triggered, thereby simulating the expansion behavior of the defect region. The output pulses generated by the firing event propagate to adjacent neurons through synaptic connections, iteratively executing the accumulation and firing process to form the spatiotemporal propagation of the pulse sequence, generating the spatiotemporal evolution trajectory of defect growth; based on the spatiotemporal evolution trajectory, the pulse firing frequency and expansion range of each defect region are statistically analyzed and converted into a probability density distribution, outputting a defect growth probability distribution map.

[0055] The maximum growth rate region is extracted from the defect growth probability distribution map, and Weibull distribution parameters are fitted in combination with historical degradation data to calculate the chip's remaining lifetime index. Specifically, the chip operates continuously for 1000 hours under high temperature, high humidity, and high voltage stress conditions, collecting photoluminescence signals and defect growth sequences from over 500 sets of degradation samples. Weibull distribution parameters are fitted through statistical analysis. Historical degradation data is stored in a database of multi-mode electromagnetic radiation baseline maps, which can be obtained through firmware updates or cloud synchronization during chip use. Real-time operating parameters, including voltage, temperature, and load status, are encoded as additional pulse inputs. Specifically, after normalizing voltage, temperature, and load status, they are mapped to pulse firing rates. A higher voltage corresponds to a higher firing rate, a higher temperature corresponds to a shorter pulse interval, and an increased load corresponds to an increased pulse density, forming an additional pulse input sequence. When the additional pulse input sequence precedes or follows the neuron firing event of the spiking neural network model, the synaptic weights are adjusted according to the time difference between the two, where the time difference is determined based on a preset time window. The preset time window is determined based on the temporal correlation of the input pulse sequence and the statistical stability of synaptic weight adjustment, preferably set within the range of 10–200 milliseconds to achieve a balance between capturing the temporal influence of operating parameters on defect evolution and suppressing noise interference. When the time difference falls within the preset time window, the corresponding synaptic weight is increased; when it exceeds the preset time window, the corresponding synaptic weight is decreased. A complete chip inspection report is generated based on the adjusted synaptic weights, the defect growth probability distribution map, and real-time operating parameters. The chip inspection report includes the chip's remaining lifetime indicators, defect distribution status, and condition-based early warning suggestions.

[0056] To further explain, the training process of the spiking neural network model is as follows: Samples of the cooperative defect distribution topology and corresponding defect growth sequences are extracted from historical chip defect evolution data. The synaptic weights of the spiking neural network model are initialized with random values ​​ranging from -0.5 to 0.5. Based on the time difference between the pulse sequence received by the spiking neural network model and the defect growth sequence, the synaptic weights are adjusted using a peak-time-dependent plasticity rule. Specifically, when the presynaptic pulse is emitted before the postsynaptic pulse (time difference...), the synaptic weights are adjusted accordingly. When the value is greater than 0, the synaptic weight increases, and the increase is... When the presynaptic pulse is followed by the postsynaptic pulse ( When <0, the synaptic weight decreases, and the decrease is... ,in and It is an amplitude constant. and Using a time constant, synaptic weight adjustments are iteratively performed, and the mean squared error between the defect growth probability distribution map output by the spiking neural network model and the corresponding defect growth sequence is monitored on the validation dataset. When the relative decrease in the mean squared error between two adjacent iterations is less than 1% to 5%, preferably less than 2%, in five consecutive iterations, the model training is considered to have entered a stable state and iteration is stopped to avoid overfitting or underfitting. The prediction accuracy of the trained spiking neural network model is verified by testing the consistency between the prediction results of the spiking neural network model and the actual defect growth sequence on a simulated working condition dataset. After obtaining the verification results, the learning rate (preferably set in the range of 0.001 to 0.1) and neuron threshold of the spiking neural network model are adjusted to adapt to the dynamic changes of real-time working condition parameters, ensuring the accuracy of the spiking neural network model in simulating defect evolution, thus completing the pre-training of the spiking neural network model.

[0057] Preferably, a spiking neural network model is constructed using the Leaky-Integrate-and-Fire neuron model to achieve an accurate mapping from the collaborative defect distribution topology map to the spiking sequence, simulate the spatiotemporal dynamics of defect evolution, avoid the simplification of nonlinear interactions by traditional numerical models, overcome the lag of static analysis on dynamic degradation, and improve the predictability and intervention efficiency of full life cycle monitoring.

[0058] To further explain, the integration of the various subsystems of this invention into a single physical device is as follows: The entire system is centered around a SoC chip, which integrates sampling and control circuitry (including temperature / pressure sampling circuitry, synchronous control logic, and an interface with the embedded MCU) that works in conjunction with the chaotic sensor. The near-field probe array and support / shielding structure of the chaotic sensor are mounted on the outer surface of the chip package and registered with the electromagnetic transmission window. The embedded microcontroller (MCU) serves as the central scheduler, and includes a power management unit (PMU), a programmable clock network, redundant diagnostic circuitry, an integrated optical probe interface, and a spiking neural network accelerator (a dedicated IP core based on a Leaky-Integrate-and-Fire neuron array). External auxiliary hardware includes a precision signal generator (for injecting chaotic perturbations) and a quantum dot laser scanning device (a piezoelectric robotic arm + a 532nm laser + a 2048×2048 CMOS detector), which connects to the SoC via a JTAG / IJTAG interface or a dedicated GPIO.

[0059] Data Interaction and Bus: All subsystems are interconnected via the internal AMBA-AXI bus (high-speed data transmission) and AMBA-APB bus (register configuration); real-time electromagnetic signals and operating parameters are acquired by the chaotic sensor and transmitted to the embedded MCU via AXI-DMA; the dynamic electromagnetic anomaly index calculation results are written to shared memory by the MCU and broadcast as a warning signal via the interrupt controller; in self-organizing diagnostic mode, the reflected light intensity distribution data acquired by the optical probe is transmitted to the MCU via the I2C interface and then sent to the pulse neural network accelerator via the AXI bus; the final detection report is output by the MCU via the SPI or UART interface.

[0060] Timing Control and Synchronization Mechanism: The embedded MCU runs the main state machine, responsible for global timing scheduling: S1 Baseline Construction Phase, the MCU controls the external signal generator to inject disturbances and synchronously acquires them; S2 Real-time Monitoring Phase, the MCU drives the chaotic sensor to acquire data every 10ms via interrupt and calculates the anomaly index; S3 Early Warning Trigger Phase, after the MCU detects that the threshold has exceeded the limit, it completes the PMU register setting and clock / voltage adjustment within 1ms; S4 Optical Scanning Phase, the MCU synchronizes the robotic arm positioning and laser excitation via the IJTAG interface to ensure that the scan is aligned with the coordinates of the holographic reference map; S5 Neuromorphic Calculation Phase, the MCU transmits the topology map pulse sequence to the accelerator via the AXI bus DMA, and after the calculation is completed, the MCU interrupts to read the result; All key timing nodes are synchronized using the chip's unified system clock (e.g., 100MHz), and data synchronization uses FIFO buffering and handshake signals to avoid data loss or timing errors.

[0061] The above integration method is based on standard SoC design practices. All interfaces, buses, and synchronization mechanisms are existing commercial IP cores, ensuring the feasibility and reliability of the actual physical device.

[0062] This embodiment also provides a semiconductor chip lifecycle continuous monitoring device, including: The acquisition module acquires the electromagnetic radiation signal of the chip under monitoring in chaotic working mode and constructs an electromagnetic radiation baseline map through topological signal processing. The analysis module uses a chaotic sensor to collect electromagnetic signals and operating parameters in real time based on the electromagnetic radiation baseline map, and performs nonlinear spectrum difference analysis to generate a dynamic electromagnetic anomaly index. The early warning module triggers an intelligent early warning mechanism based on the comparison result between the dynamic electromagnetic anomaly index and the preset chaos threshold, and controls the chip to enter the self-organizing diagnosis mode and establish a holographic optical reflection reference map. The identification module, based on a holographic optical reflection reference map, performs photoluminescence scanning using a quantum dot laser scanning device to generate a multidimensional photoluminescence intensity distribution map, intelligently identifying defect regions and forming a collaborative defect distribution topology map; The detection module establishes a neuromorphic computing model and uses it as an optimization criterion to perform data fusion analysis on the collaborative defect distribution topology map, outputting chip detection results.

[0063] This embodiment also provides a computer device applicable to the continuous detection method for the entire lifecycle of semiconductor chips, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the continuous detection method for the entire lifecycle of semiconductor chips as proposed in the above embodiment.

[0064] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0065] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0066] In summary, this invention constructs an electromagnetic radiation baseline map through chaotic operating mode excitation and topological signal processing, captures nonlinear dynamic behavior and decouples noise interference, establishes an accurate baseline, and achieves high sensitivity and low false positive rate in real-time differential analysis. Simultaneously, it introduces an intelligent early warning mechanism with a dynamic electromagnetic anomaly index and an adaptive chaotic threshold, driving self-organizing diagnostic mode switching, significantly shortening service interruption time. Furthermore, it tightly couples electromagnetic anomalies and optical scanning through a holographic optical reflection reference map reconstructed in phase space, accurately identifying the topological interaction relationships of cooperative defect clusters, eliminating the problem of amplified positioning errors, and ultimately achieving breakthroughs in zero-interruption throughout the entire lifecycle, high sensitivity to weak signals, and accurate prediction of cooperative degradation.

[0067] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for continuous monitoring of a semiconductor chip throughout its entire lifecycle, characterized in that: include, The electromagnetic radiation signal of the chip under monitoring in chaotic working mode is acquired, and an electromagnetic radiation baseline map is constructed through topological signal processing. Based on the electromagnetic radiation baseline map, a chaotic sensor is used to collect electromagnetic signals and operating parameters in real time, and nonlinear spectrum difference analysis is performed to generate a dynamic electromagnetic anomaly index. The intelligent early warning mechanism is triggered based on the comparison result between the dynamic electromagnetic anomaly index and the preset chaos threshold. The control chip enters the self-organizing diagnosis mode and establishes a holographic optical reflection reference map. Based on the holographic optical reflection reference map, a multidimensional photoluminescence intensity distribution map is generated by performing photoluminescence scanning through a quantum dot laser scanning device, which intelligently identifies defect regions and forms a collaborative defect distribution topology map. A neuromorphic computational model was established and used as an optimization criterion to perform data fusion analysis on the collaborative defect distribution topology map, and output chip detection results.

2. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 1, characterized in that: The electromagnetic radiation signal of the chip under test in chaotic operating mode is acquired, and an electromagnetic radiation baseline map is constructed through topological signal processing. Specifically, The chip is put into a chaotic working mode, and the resulting broadband electromagnetic radiation signal is collected through the electromagnetic transmission window. The signal is then processed by removing interference, normalizing, and extracting phase information to obtain a preprocessed signal. The preprocessed signal is mapped to the topological space, and topological invariants are extracted. A signal phase space reconstruction model is constructed based on topological invariants. High-dimensional feature vectors are generated by the delayed embedding method and associated with the corresponding operating mode parameters to form a multi-mode electromagnetic radiation baseline map.

3. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 1, characterized in that: Based on the electromagnetic radiation baseline map, a chaotic sensor is used to collect electromagnetic signals and operating parameters in real time. Nonlinear spectral difference analysis is then performed to generate a dynamic electromagnetic anomaly index. Specifically... A reference spectrum matching the current operating parameters is retrieved from the multi-mode electromagnetic radiation baseline map. Real-time electromagnetic signals, temperature and voltage operating parameters are synchronously acquired through a chaotic sensor. The instantaneous spectral characteristics are obtained by performing a short-time Fourier transform on the real-time acquired electromagnetic signals. The difference spectrum between the instantaneous spectral characteristics and the reference spectrum is calculated. The Lyapunov exponent and fractal dimension of the difference spectrum are analyzed using a nonlinear dynamics method and then fused to generate a dynamic electromagnetic anomaly index.

4. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 1, characterized in that: The intelligent early warning mechanism is triggered based on the comparison result between the dynamic electromagnetic anomaly index and the preset chaos threshold. The control chip enters a self-organizing diagnostic mode and establishes a holographic optical reflection reference map. Specifically, Continuously monitor the changes in the dynamic electromagnetic anomaly index and immediately trigger an early warning signal when it exceeds the preset chaos threshold; The chip's bias voltage and clock frequency parameters are adjusted according to the warning signal, so that the chip enters a low-power self-organizing diagnostic mode. In the low-power self-organizing diagnostic mode, the chip surface is scanned by an integrated optical probe to collect the reflected light intensity distribution data. The reflected light intensity distribution data is holographically encoded to generate a hologram, and the three-dimensional optical reflection field is reconstructed based on the hologram to form a holographic optical reflection reference map.

5. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 1, characterized in that: Based on a holographic optical reflection reference image, a multidimensional photoluminescence intensity distribution map is generated by performing photoluminescence scanning using a quantum dot laser scanning device. This intelligently identifies defect regions and forms a collaborative defect distribution topology map. Specifically... Extracting candidate defect regions from a holographic optical reflection reference image; A quantum dot laser scanning tool is used to emit a wavelength laser beam to scan the defect candidate region point by point, and the photoluminescence signal of each scanning point is collected simultaneously. The photoluminescence signal is mapped to a color space to generate a multidimensional photoluminescence intensity distribution map, and a clustering algorithm is used to identify abnormal brightness regions and calculate defect boundaries and size parameters. A collaborative defect distribution topology map is constructed based on defect candidate regions, photoluminescence signal data, multidimensional photoluminescence intensity distribution map, defect boundaries, and size parameters.

6. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 1, characterized in that: A neuromorphic computational model was established and used as an optimization criterion to perform data fusion analysis on the collaborative defect distribution topology map, outputting chip detection results. Specifically... A spiking neural network model was constructed, and the topology map of the collaborative defect distribution was converted into a spiking sequence and input into the neural network. The defect evolution dynamics are simulated through neuromorphic computation, and a defect growth probability distribution map is output. The remaining lifespan of the chip is calculated based on the defect growth probability distribution map, and the weights of the spiking neural network model are optimized by integrating real-time operating parameters to generate a complete chip inspection report.

7. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 2, characterized in that: The process of mapping the preprocessed signal to the topological space specifically involves... A simplex complex model of the preprocessed signal is established, the homology group generator is calculated, and the key topological structures are selected and quantized into topological feature vectors. Calculate the topological entropy based on the topological feature vectors and convert it into topological space coordinates to complete the mapping.

8. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 3, characterized in that: The method of analyzing the Lyapunov exponent and fractal dimension of the difference spectrum using nonlinear dynamics refers to reconstructing the phase space of the difference spectrum between the instantaneous spectral characteristics and the reference spectrum, calculating the maximum Lyapunov exponent using the Wolf algorithm, and simultaneously calculating the fractal dimension of the difference spectrum.

9. The method for continuous monitoring of a semiconductor chip throughout its entire lifecycle as described in claim 4, characterized in that: The step of performing holographic encoding on the reflected light intensity distribution data to generate a hologram specifically involves... The reflected light intensity data is converted into a complex field form, and a frequency domain hologram is generated through Fourier transform. After optimizing the contrast in the frequency domain hologram, the spatial hologram is reconstructed using inverse Fourier transform.

10. A semiconductor chip lifecycle continuous testing device, based on the semiconductor chip lifecycle continuous testing method according to any one of claims 1 to 9, characterized in that: include, The acquisition module acquires the electromagnetic radiation signal of the chip under monitoring in chaotic working mode and constructs an electromagnetic radiation baseline map through topological signal processing. The analysis module uses a chaotic sensor to collect electromagnetic signals and operating parameters in real time based on the electromagnetic radiation baseline map, and performs nonlinear spectrum difference analysis to generate a dynamic electromagnetic anomaly index. The early warning module triggers an intelligent early warning mechanism based on the comparison result between the dynamic electromagnetic anomaly index and the preset chaos threshold, and controls the chip to enter the self-organizing diagnosis mode and establish a holographic optical reflection reference map. The identification module, based on a holographic optical reflection reference map, performs photoluminescence scanning using a quantum dot laser scanning device to generate a multidimensional photoluminescence intensity distribution map, intelligently identifying defect regions and forming a collaborative defect distribution topology map; The detection module establishes a neuromorphic computing model and uses it as an optimization criterion to perform data fusion analysis on the collaborative defect distribution topology map, outputting chip detection results.

Images