Semiconductor packaging device electromagnetic compatibility comprehensive test method and system

Abstract

The invention discloses a semiconductor packaging device electromagnetic compatibility comprehensive test method and system, and belongs to the technical field of electromagnetic compatibility testing. The method comprises the following steps: collecting structure parameters, packaging topology and predefined function states of a to-be-tested packaging device, and constructing a polymorphic working model; establishing a disturbance injection control model according to each state and configuring disturbance source parameters; implementing dynamic disturbance injection and acquiring response data in a real working state of the device; performing time domain and frequency domain conjoint analysis on the response data, constructing an electromagnetic response dynamic feature sequence, inputting the electromagnetic response dynamic feature sequence into a machine learning model, extracting multi-dimensional coupling features and predicting tolerance; calculating performance indexes such as an interference tolerance score and a coupling strength index based on model output indexes, comparing the performance indexes with a standard, and evaluating a compatible risk level in a full state; the method realizes quantitative evaluation of the EMC performance of the packaging device with high reduction degree and multi-state coverage, and has the advantages of comprehensive test, accurate prediction, explainable attribution and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic compatibility testing, and in particular to a comprehensive electromagnetic compatibility testing method and system for semiconductor packaging devices. Background Art

[0002] As semiconductor packaging evolves toward higher integration, higher frequencies, and heterogeneous packaging, the types of electromagnetic interference (EMI) to which packaged devices are subjected during actual operation are becoming increasingly complex. Issues such as multi-band superposition, nonlinear coupling, and cross-band interference are becoming increasingly prominent. Microstructures within the package, such as the lead structure, metal layer layout, and via arrays, are inherently susceptible to becoming radiation sources or interference channels. These interferences are often transient, additive, and non-periodic.

[0003] Existing electromagnetic compatibility testing techniques typically perform independent static tests for radiated emissions, conducted emissions, radiated immunity, and conducted immunity. This makes it difficult to reflect the coupled interference effects these factors produce during actual system operation. Furthermore, existing techniques fail to account for the timing response and cumulative electromagnetic effects of packaged components in complex application scenarios (such as multi-frequency and multi-carrier communications, sudden power changes, and dynamic scheduling), resulting in evaluation results that deviate from reality.

[0004] Some studies have attempted to predict interference through finite element simulation or equivalent circuit modeling. However, due to the lack of dynamic modeling of the spatial path distribution of electromagnetic interference under different functional states, it is still impossible to complete the "full process" evaluation of the electromagnetic compatibility of packaged devices at the system level.

[0005] Therefore, there is an urgent need for a comprehensive electromagnetic compatibility testing method that integrates the relationships among structural information, functional status and environmental disturbances. Through dynamic coupling modeling and data feedback mechanism, a multi-dimensional and full-process evaluation of the electromagnetic compatibility performance of semiconductor packaging devices in real usage scenarios can be achieved to improve test accuracy and design guidance value. Summary of the Invention

[0006] The object of the present invention is to provide a method and system for comprehensive electromagnetic compatibility testing of semiconductor packaging devices to address the deficiencies in the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a comprehensive electromagnetic compatibility testing method for semiconductor packaging devices, comprising:

[0008] Collect the structural parameters, package layer topology and predefined functional states of the semiconductor package device to be tested, and build a multi-state working model of the device;

[0009] According to the functional states of the device's multi-state working model, a disturbance injection control model is constructed, and different disturbance source parameter sets are selected according to the test standards;

[0010] Under the actual working conditions of the packaged device, the disturbance injection control model is dynamically injected according to the parameter-centered disturbance source, and the device response data is collected;

[0011] Perform a joint analysis of device response data in the time and frequency domains to construct a dynamic feature sequence of the electromagnetic response. This feature sequence is then input into a trained machine learning model for multi-dimensional coupling feature extraction and tolerance prediction.

[0012] Based on the indicator vector output by the machine learning model, the comprehensive electromagnetic compatibility performance indicator set of the device in various functional states is calculated, including interference tolerance and coupling strength index;

[0013] Compare the comprehensive electromagnetic compatibility performance index set with industry standards to determine the compatibility risk level of packaged devices under all working conditions.

[0014] Preferably, constructing a device multi-state working model includes:

[0015] A multi-layer graph neural network is used to model the package layer topology of the device under test, extract the node connectivity and electromagnetic coupling potential of each package layer, and generate a topology vector representation L1;

[0016] High-resolution structural imaging data is compared and fused with CAD models to identify heterogeneous structural units, metal layer arrangements, and via clusters within the device under test, construct a structural parameter tensor matrix S1, and perform feature alignment with L1.

[0017] Based on the device's design behavior model library, the predefined functional state set F1 is called to build a state control diagram. The states in F1 are logically mapped to the trigger paths of S1 to generate a state transition diagram M1.

[0018] L1, S1 and M1 are jointly embedded through a multi-channel embedding encoder, and the multi-state working model of the device is finally output.

[0019] Preferably, the construction of a disturbance injection control model according to each functional state of the device multi-state working model and the selection of different disturbance source parameter sets according to the test standard include:

[0020] Perform state cluster analysis on the state transition graph in the polymorphic working model and extract representative state sequence sets using the time series feature compression algorithm;

[0021] Based on each representative state in the representative state sequence set, a disturbance response channel diagram is established to identify the spatial nodes and temporal nodes where electromagnetic coupling may occur in this state, and a disturbance entry point set is defined;

[0022] According to the current electromagnetic compatibility test standards, a disturbance source parameter library is constructed, including the spectrum range, disturbance energy level, waveform shape and duration. The disturbance source parameter library is projected onto each disturbance entry point set through the state disturbance mapping function to generate a disturbance injection control model.

[0023] Preferably, the step of dynamically injecting the disturbance injection control model according to the parameter-centered disturbance source under the actual working state of the packaged device and collecting the device response data comprises:

[0024] Load the real functional scenario script to the packaged device through the asynchronous collaborative driving platform to start its multi-function state loop operation;

[0025] The disturbance entry point set defined in the disturbance injection control model is called, and a multi-channel digital-analog collaborative jammer is used to implement multi-dimensional disturbance injection on the selected disturbance entry point set nodes;

[0026] During the injection process, a high-precision electromagnetic response acquisition unit is activated through a time-domain synchronous trigger mechanism. Multi-point probes are deployed in the device outer shell, internal power supply network and key signal channels to collect RF leakage, current spikes, voltage fluctuations and timing drift data in real time to form a response data set.

[0027] Preferably, performing a joint analysis of the device response data in the time domain and the frequency domain to construct an electromagnetic response dynamic feature sequence, and inputting the electromagnetic response dynamic feature sequence into a trained machine learning model comprises:

[0028] The response data is sliced ​​with a fixed window, and the time domain impact features and frequency domain bandwidth expansion features are extracted by wavelet packet transform, and a multi-scale response vector group is jointly constructed.

[0029] Adaptive dynamic time warping algorithm is used to time align and morphologically normalize the multi-scale response vector group to form a unified structure of electromagnetic response dynamic feature sequence;

[0030] The dynamic feature sequence of the electromagnetic response is input into a machine learning model built based on the graph convolution-attention fusion architecture. Utilizing its spatial coupling modeling capability and cross-channel attention extraction mechanism, high-dimensional extraction of nonlinear interaction features between multiple perturbation channels in the package structure is performed.

[0031] Based on the response feature tensor output by the machine learning model, combined with the disturbance parameters and structural feature embedding, an electromagnetic tolerance prediction vector is generated to quantify the coupling strength and anti-interference capability boundary of the device under the set state.

[0032] Preferably, the index vector outputted by the machine learning model is used to calculate the comprehensive electromagnetic compatibility performance index set of the device in each functional state, including:

[0033] The principal component of each dimension output result in the electromagnetic tolerance prediction vector is reconstructed, and the disturbance response feature space H is constructed using nonlinear kernel function mapping, preserving the interference tolerance, power coupling rate and pulse recovery speed;

[0034] Combined with the state transition probability matrix of the device multi-state working model, a state-dependent weighting function φ is constructed, so that each electromagnetic performance indicator has a different weight contribution in the corresponding state;

[0035] Based on H and φ, a comprehensive set of electromagnetic compatibility performance indicators is calculated, including interference tolerance score, coupling strength index and cross-state stability.

[0036] Preferably, the comprehensive electromagnetic compatibility performance index set is compared with the industry standard to determine the compatibility risk level of the packaged device in a fully functional state:

[0037] Build an industry standard parameter library, which includes electromagnetic tolerance limits, coupling strength upper limits, and state fluctuation tolerances corresponding to various types of packaging structures, and perform standard matching based on packaging form and frequency band application scenarios;

[0038] The parameters of each indicator in the performance indicator set are scaled and aligned with the standard domain parameters through hierarchical normalization. The interval uncertainty mapping method is used to introduce test errors and boundary fuzziness to obtain the confidence score matrix.

[0039] A multi-state risk level assessment model is constructed based on the confidence score matrix. By integrating fuzzy reasoning and threshold clustering methods, a risk level label set for the device in all functional states is generated. The labels include three levels: "safe", "marginal", and "exceeding the standard" and their corresponding probability weights.

[0040] Map the labels to the encapsulation layer structure, identify the spatial structure areas and status trigger conditions with high risk levels, and perform visual output and generate avoidance suggestions.

[0041] The present invention also provides a semiconductor packaging device electromagnetic compatibility comprehensive test system, comprising:

[0042] The polymorphic working modeling module collects the structural parameters, package layer topology and predefined functional states of the semiconductor package device to be tested, and builds a polymorphic working model of the device;

[0043] The parameter configuration module builds a disturbance injection control model based on the functional states of the device's multi-state working model and selects different disturbance source parameter sets according to the test standards;

[0044] The response acquisition module dynamically injects disturbance injection control models according to the parameter-centered disturbance source under the actual working conditions of the packaged device and collects device response data.

[0045] The model prediction module performs a joint analysis of the device response data in the time and frequency domains to construct a dynamic feature sequence of the electromagnetic response. This feature sequence is then input into the trained machine learning model to extract multidimensional coupling features and predict tolerances.

[0046] The stability modeling module calculates the device's comprehensive electromagnetic compatibility performance indicators in various functional states, including interference tolerance and coupling strength index, based on the indicator vector output by the machine learning model;

[0047] The compatibility risk assessment module compares the comprehensive electromagnetic compatibility performance index set with industry standards to determine the compatibility risk level of the packaged device in all working conditions.

[0048] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0049] 1. This invention provides a comprehensive electromagnetic compatibility testing method for semiconductor packaged devices. It constructs a multi-state operational modeling system based on the three-dimensional fusion of structural topology, functional state, and electromagnetic behavior. It also introduces disturbance response path identification and an adaptive disturbance injection mechanism to enable dynamic disturbance response testing of devices under real-world operating conditions. Through a collaborative architecture combining multi-channel injection with multi-point probe synchronous acquisition, it significantly improves test scenario fidelity and data acquisition integrity, effectively overcoming the limitations of existing methods, which suffer from a single test mode and insufficient state coverage.

[0050] 2. This invention combines a joint feature extraction method from the time and frequency domains with a graph convolution-attention fusion machine learning model to establish a multidimensional prediction mechanism for coupling path behavior and electromagnetic tolerance. Based on the model outputs, a set of engineering-interpretable performance indicators is constructed, including an interference tolerance score, a coupling strength index, and cross-state stability. This indicator system not only supports quantitative performance evaluation but also integrates packaging structure to attribute sensitive areas and determine risk levels. This provides a highly accurate and systematic evaluation framework for electromagnetic compatibility design, verification, and optimization, demonstrating excellent engineering practicality and universal application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0052] Figure 1 Flow chart of the method of the present invention.

[0053] Figure 2 It is a flow chart of the system modules of the present invention. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0055] Example 1, please refer to Figure 1 As shown, the embodiment of the present invention provides a comprehensive electromagnetic compatibility testing method for a semiconductor package device, including:

[0056] Collect the structural parameters, package layer topology and predefined functional states of the semiconductor package device to be tested, and build a multi-state working model of the device;

[0057] Construct a disturbance injection control model for each functional state of the device's multi-state working model, and select different disturbance source parameter sets based on the test standards;

[0058] Under the actual working conditions of the packaged device, the disturbance injection control model is dynamically injected according to the parameter-centered disturbance source, and the device response data is collected;

[0059] Perform a joint analysis of device response data in the time and frequency domains to construct a dynamic feature sequence of the electromagnetic response. This feature sequence is then input into a trained machine learning model for multi-dimensional coupling feature extraction and tolerance prediction.

[0060] Based on the indicator vector output by the machine learning model, the comprehensive electromagnetic compatibility performance indicator set of the device in various functional states is calculated, including interference tolerance and coupling strength index;

[0061] Compare the comprehensive electromagnetic compatibility performance index set with industry standards to determine the compatibility risk level of packaged devices under all working conditions.

[0062] This embodiment takes a multi-layer active chip stacking package structure as the object. First, its structure, topology, and functional state information are comprehensively modeled to construct a polymorphic working model. The specific steps are as follows:

[0063] Among them, structural information includes: metal wiring layer layout, via location and density, power / ground plane configuration, packaging material and layer thickness parameters, etc.

[0064] Topological information includes: node connection relationships (such as interconnect lines, packaging layer connections), node physical coordinates, and electromagnetic coupling potential annotations;

[0065] Functional status information includes: operating voltage, current mode, main frequency, data transmission behavior, switching rate, load switching behavior, etc. in each state.

[0066] Using the acquired package design description file (e.g., in OASIS or GDSII format), we extract the device's package hierarchy, including metal routing layers, package substrate, upper and lower interconnect vias, and vertical interconnects between dies. Each package layer is abstracted as a subgraph unit within a graph structure, with each graph node corresponding to a metal block, power island, or signal trace, and each edge representing an electrical connection or a possible electromagnetic coupling channel.

[0067] A multi-layer graph neural network is used to train the topological graph, with node degree, edge bandwidth, current density and spatial overlap as feature inputs. The electromagnetic coupling potential score vector of each packaging layer is output and combined into a topological vector representation L1 according to the order of packaging layers for subsequent modeling input.

[0068] This paper uses a multi-layer graph neural network (GNN), which is mainly used to model the node connection relationship and coupling potential between encapsulation layers. Its structure is as follows:

[0069] The input layer receives the graph structure units encapsulated by each layer;

[0070] The middle layer includes three graph convolutional layers, which perform feature aggregation on node degree, edge weight (bandwidth, resistance), and spatial coordinates respectively;

[0071] The output layer is a set of node vectors, which represent the coupling potential score of each encapsulated node.

[0072] In addition, the topology map construction process includes the following steps:

[0073] Map metal blocks, vias, power islands, etc. as nodes, and electrical connections as edges;

[0074] Node attributes include position, area, capacitance, current density, etc.;

[0075] Edge attributes include resistance, inductance, and electromagnetic coupling factor;

[0076] Construct a heterogeneous graph to support the fusion of heterogeneous node features.

[0077] Perform high-resolution structural scanning (such as laser confocal imaging or X-ray tomography) on actual device samples to obtain a physical structural map of the device, and establish positioning relationships through CAD model mapping. By calculating geometric fitting and feature similarity, the system automatically identifies heterogeneous units within the device (such as multi-chip embedded components, high-frequency bypass capacitors, and ground bounce structures). It extracts information such as metal layer area distribution, via cluster density, and package size ratio, generating a high-dimensional structural parameter tensor S1.

[0078] Specifically: using the point cloud data extracted from the three-dimensional structure atlas to construct a voxel grid;

[0079] Compare the CAD model for coordinate registration and use shape matching algorithms (such as ICP, Chamfer Distance) to calculate feature similarity;

[0080] Match heterogeneous cells based on preset feature templates (such as capacitor shape and die stacking mode) and output classification labels.

[0081] By utilizing the consistency between the node index in the graph structure and the coordinate system in the imaging data, the features of L1 and S1 are aligned to ensure the spatial consistency and expression complementarity between the topological modeling results and the physical structure parameters.

[0082] Access the design behavioral model library, which contains device operating modes under different supply voltages, load switching, communication protocol switching, and other conditions. Invoke the target device's predefined functional state set F1 (such as high-speed signal transmission and reception, low-power sleep state, frequency hopping state, etc.) and combine it with the structural coupling characteristics in S1 to establish a state control diagram.

[0083] In this invention, the structural parameter tensor S1 not only includes static structural parameters such as package layer thickness, material, metal traces and vias, but should also be expanded to include a subset of structural coupling features, which is mainly used to characterize the structural coupling strength potential between different package layers under electromagnetic interference fields. It specifically includes the following feature dimensions:

[0084] The coupling area ratio of adjacent metal layers;

[0085] Calculated parasitic capacitance / inductance between ground and power planes;

[0086] The spatial distance matrix between signal traces;

[0087] Current closed-loop impedance in the power supply decoupling path, etc.

[0088] When constructing a state control diagram, it is necessary to extract the node set related to the state change from the structural coupling characteristics. For example:

[0089] Whether to activate the high-frequency interconnection line when a certain function state is turned on;

[0090] Whether to trigger cross-zone via current switching in the power conversion state;

[0091] Will changes in package structure (such as switching of hierarchical power supply networks) form common-mode current paths?

[0092] After marking these structural coupling nodes as state-sensitive structural nodes, they are logically mapped with the functional state F1 to establish the edge connection conditions in the state control graph.

[0093] For each state in F1, a logical mapping of its triggering path to the physical structure is constructed to obtain the mapping function between the state and the interference-sensitive structure, and finally a state transition map M1 is constructed, which represents the transition sequence of functional states and their potential excitation paths.

[0094] The construction of the state transition graph MI is based on the following three parts:

[0095] Functional state set F1: derived from the chip design behavioral model (such as clock gating, power domain switching, bus switching, etc.), representing the functional nodes of the device under different operating conditions;

[0096] Edge connection logic of the state control diagram: Based on state trigger conditions, such as "reaching the temperature threshold", "data packet reception completed", "entering idle state for more than 100ms", etc., define the transition edges between states;

[0097] State-associated structure node set (extracted from S1): Which physical structures are activated or sensitive to each state during operation (e.g., turning on the high-speed interface → activating the top metal trace → affecting the electromagnetic interference path of the power ring network).

[0098] Finally, a state transition graph M1 with edge weights (indicating migration probability or activation frequency) is formed, which is a directed graph structure used for subsequent disturbance response modeling and dynamic injection scheduling.

[0099] The topology vector L1, structural parameter tensor S1, and state transition map M1 are input into a customized multi-channel embedding encoder. The encoder consists of a structural embedding module, a state logic module, and a spatial attention fusion module. Through feature compression and semantic fusion mechanisms, it generates a device state-structure coupling vector in a unified format.

[0100] The multi-channel embedding encoder consists of three modules:

[0101] Structure embedding module: performs convolution compression on the structure parameter tensor S1;

[0102] State logic module: Use GRU or Transformer to extract state transition sequence features from the state graph M1;

[0103] Fusion mechanism: The attention mechanism is used to semantically fuse the three vector sets L1, S1, and M1, and output a unified working model embedding vector W1.

[0104] The final output device's multi-state working model W1 can be used to support subsequent disturbance injection path planning, electromagnetic response simulation, and state sensitivity assessment.

[0105] This embodiment proposes a method for constructing a disturbance injection control model based on a multi-state operating model, aiming to identify electromagnetic disturbance paths and optimize disturbance source configuration under different operating states, thereby improving the electromagnetic compatibility assessment capability of semiconductor packaging devices. Specifically, the method includes the following steps:

[0106] Using the previously constructed multi-state operating model as input, the state transition graph is extracted. For all state nodes in the graph, a temporal behavior clustering algorithm is used to construct a state feature tensor, taking into account multi-dimensional attributes such as state duration, switching frequency, and current transition amplitude.

[0107] The state transition graph comes from the predefined functional state set F1 and its triggering logic; the state machine model is used to model state transition; the state edge has probability weights, timing labels and behavior labels.

[0108] For example, the time series behavior clustering algorithm uses Dynamic Time Warping (DTW) as a similarity measure between time series; constructs a time-function node sequence for each state sequence (such as S1→S2→S3→S5); uses K-means or DBSCAN to perform similarity clustering on the state time series to obtain a representative state sequence set (only the most representative central state sequence is retained for each class);

[0109] Using a time series feature compression algorithm (such as the SP-TS clustering method based on principal component dynamic time series clustering), the state transition map is divided into state clusters, and a representative state sequence set is extracted to cover the main electromagnetic behavior change trends.

[0110] For example, the time series feature compression algorithm uses a sliding window + principal component analysis (PCA) method to extract key time series transition points in each state sequence;

[0111] Dimensionality reduction encoding of multi-dimensional temporal features such as inter-state duration, power trajectory, and number of logic switches;

[0112] Generate a compressed state time series vector for constructing a disturbance response channel diagram.

[0113] For each representative state in the state sequence set, its structural activation path in the polymorphic model is called, and combined with the topological vector L1 and the structural tensor S1, the node group where electromagnetic coupling may occur in this state is calculated, including the high-speed signal access area, the power return path set, the ground grid resonance domain, etc.

[0114] In this paper, a structural activation path refers to a set of paths within a packaged structural unit that, under a specific functional state, is driven by logical control behavior, generates electromagnetic activity (current changes or field intensity disturbances), and may cause interference transmission. This path is obtained by mapping the functional state map with the structural topology model. The specific modeling process is as follows:

[0115] Feature state activation mapping:

[0116] Each functional state Fi∈F1 corresponds to a set of working units (such as core IP, communication interface, clock module);

[0117] Each work unit is associated with one or more package structure nodes (such as metal layer segments, via arrays, and pads) through physical interconnection;

[0118] Establish a mapping function: Fi → {vj∈VTopo}, where vj is a node in the package topology graph. VTopo represents the set of nodes in the device package topology graph, i.e., the complete set of graph neural network nodes used to represent semiconductor package structural units in the polymorphic working model.

[0119] Electromagnetic activity path marking: Identify the main power supply path (PDN path), high-speed signal traces (such as SerDes), and clock synchronization network in the active state;

[0120] Perform current path traversal from upstream to downstream along the mapped nodes in the structural topology (minimum impedance or shortest path algorithm may be used);

[0121] The continuous path node set is recorded as the structural activation path Pi⊂VTopo in this state

[0122] Output results: For each state Fi, its activated structural path set Pi is output to guide the selection of perturbation entry points and coupling node prediction.

[0123] Based on the obtained structural activation path Pi, the node groups where electromagnetic coupling may occur are further derived as follows:

[0124] Establishment of local electromagnetic coupling evaluation model: For each active path node vj∈Pi, search for its spatially adjacent nodes on the topological graph (i.e., nodes with physically close or shared conductor areas) and evaluate their coupling potential using the following formula: ;in: represents the coupling potential score between node vj and its adjacent node vk; Represents the overlapping or aligned area between two nodes; represents the square of the center distance between nodes; represents the coupling type coefficient (capacitive coupling, inductive coupling, radiation coupling); ϵ is a small constant to prevent division by zero.

[0125] Set the coupling threshold to screen the node group: set the coupling score threshold τ, when ≥τ, node vk is regarded as a node that may undergo electromagnetic coupling in state Fi;

[0126] All vk that meet the conditions are classified into node group Ei; this node group will be used to draw the perturbation response channel graph and define the injection entry point set.

[0127] The spatial nodes and trigger timing are combined to establish a disturbance response channel diagram, in which nodes represent sensitive areas and edges represent cross-domain conduction or coupling paths. According to the degree of electromagnetic excitation influence, a set of high-priority disturbance entry points D is determined. n , as the target location for disturbance injection.

[0128] Based on current electromagnetic compatibility test standards (such as IEC 61967, CISPR 25, or GB / T 17626), a structured disturbance source parameter library Pstd is constructed. This parameter library includes, but is not limited to, standard spectrum segmentation, typical disturbance energy levels (such as 1V / m, 3V / m, etc.), pulse and amplitude modulation waveform templates, and interference duration windows.

[0129] Define the state perturbation mapping function for different representative states and their corresponding entry point sets ; Wherein: statei represents the i-th functional state; dj represents the j-th disturbance entry point; Fem(statei) represents the electromagnetic behavior characteristics in the current state (such as: frequency principal component, mutation rate, synchronous switching ratio); Gtopo(dj) represents the electrical properties of the entry point in the topology diagram (such as: coupling degree, node capacitance, power path impedance); Sref represents the disturbance parameter set in the industry disturbance standard (CISPR / IEC); M represents the parameter matching mapper based on rules or machine learning models;

[0130] The disturbance parameter library includes: spectrum range: for example, 10kHz–6GHz; amplitude level: such as 1V / m, 3V / m, 10V / m, etc.; waveform type: sine, pulse, amplitude modulated wave, broadband noise;

[0131] Duration: such as 10μs, 100ms, continuous wave, etc.; Source standards: IEC 61000-4-3, IEC 61967, CISPR25.

[0132] The state perturbation mapping function is used to project the perturbation parameters in Pstd to D in the spatial domain and time domain respectively. n The position of the disturbance source is determined to achieve state-sensitive configuration. Finally, a disturbance injection control model is formed to support the scheduling and execution of the dynamic disturbance excitation process.

[0133] Disturbance injection control model: Maps disturbance parameters to entry points (such as power, ground, and signal pins). The injection method is a digital-analog collaborative jammer, using dual-channel synchronous excitation. Control mechanism: Disturbance injection is synchronized with state switching events and initiated by the test control platform.

[0134] To accurately reproduce the electromagnetic behavior response of semiconductor packaged devices under actual use conditions, this embodiment provides a method for implementing dynamic perturbation injection and collecting electromagnetic response data in a real functional state. The specific steps are as follows:

[0135] An integrated hardware and software test control environment is built using an asynchronous collaborative drive platform, which includes a script scheduling module, an interface timing controller, and a state feedback channel. Functional state scripts based on actual use cases (such as dynamic frequency modulation of multi-core processors, power switching of RF front-ends, and memory access burst sequences) are loaded onto the packaged device under test.

[0136] The asynchronous collaborative drive platform can be built based on cloud computing and edge computing architecture. The script scheduling module adopts an event-driven mechanism. When the multi-function state loop running script is ready, the module distributes the script to the interface timing controller according to the preset priority and timestamp. The interface timing controller uses a state machine model to accurately control the order and time intervals of calling different perturbation entry point sets to ensure that the multi-dimensional perturbation injection and the multi-function state loop operation of the packaged device are precisely synchronized. The state feedback channel uses message queue technology. The packaged device sends its own operating status information to the message queue in real time. Each module of the platform obtains status feedback by subscribing to the message queue to achieve dynamic adjustment. For example, in the test of a complex semiconductor package device, the state feedback channel detected that the device had a brief freeze in the high-speed data transmission state. The platform adjusted the intensity and frequency of subsequent perturbation injection in a timely manner to avoid affecting the test accuracy.

[0137] Start the device functional state cycling mechanism to automatically switch between multiple high-power, high-frequency, or interference-sensitive states during testing, keep the state behavior consistent with actual usage conditions, and ensure that the disturbance injection process covers critical operating modes.

[0138] The disturbance injection control model constructed above is called to obtain the set of disturbance entry points selected in the current state. A multi-channel digital-analog collaborative jammer is used as the disturbance injection device. This device supports the simultaneous configuration of different disturbance waveforms, amplitudes, and frequency parameters for multiple injection channels.

[0139] Based on the configuration of the disturbance parameter set, the control system injects multidimensional disturbances, including amplitude-modulated RF signals, broadband pulse trains, and low-frequency common-mode disturbances, at various disturbance entry points dⱼ (such as power pins, ground plane vias, and high-speed I / O interfaces). The entire injection process supports real-time modulation and injection trajectory tracking, ensuring dynamic coupling between the injection process and the device's operating status.

[0140] Real-time modulation is achieved through digital signal processing (DSP) technology, dynamically adjusting the amplitude, frequency, and phase of the disturbance signal based on the device's real-time operating state. Injection trajectory tracking utilizes a Kalman filter algorithm, using device response data as observations to continuously predict and correct the location and parameters of the disturbance injection, ensuring dynamic coupling with the device's operating state. When testing an RF semiconductor package with multiple operating frequencies, the Kalman filter adjusts the disturbance injection frequency in real time based on the collected response data at different frequencies, ensuring that it closely tracks changes in the device's operating frequency and ensuring comprehensive test coverage.

[0141] During the disturbance injection synchronization, the start and pause of the high-precision electromagnetic response acquisition unit is controlled by a time domain synchronization trigger mechanism (for example, based on a global clock or edge detection interrupt), ensuring that data acquisition is strictly aligned with the injection timing.

[0142] Multiple response probe nodes are arranged on the metal shell of the device surface, inside the power supply circuit and on the key signal path. The probe types include near-field electromagnetic field antenna, current Hall probe, voltage sampling clamp, etc. The collected signal covers:

[0143] The collection locations include: power supply ports around the chip, common mode loops on the ground plane, near power decoupling capacitors, and high-speed signal entrances;

[0144] Data sampling frequency: up to 20GS / s;

[0145] Data content: RF leakage spectrum, current peak value, voltage offset, signal timing offset, etc.

[0146] It should be noted here that when selecting a probe, if the measured point is a tiny and closely spaced solder joint, it is preferred to use a pointed probe with an outer diameter of 0.11mm or less. It can accurately contact the solder joint, reduce the impact on the surrounding circuit, and ensure measurement accuracy. For measured points with concave surfaces, such as ground pins of certain special structures, a nine-claw or three-needle probe is used, which can fit well with the concave surface and ensure reliable data collection. When installing the probe, use a high-precision microscope to assist in positioning to ensure that the probe contacts the measured point vertically and accurately. At the same time, use shock absorption and shielding measures to prevent external environmental interference from affecting the accuracy of data collection.

[0147] The collected data is transmitted to the edge data compression module through a high-speed interface and cached to form a response data set for subsequent frequency domain and time domain feature extraction and analysis.

[0148] To accurately identify and quantify the electromagnetic coupling behavior and tolerance capabilities of semiconductor packaged devices under different functional states, this embodiment provides a machine learning prediction method based on multi-scale feature extraction and a graph convolutional attention fusion model. This method, combined with nonlinear feature space mapping and a state-dependent weighting mechanism, constructs a comprehensive electromagnetic compatibility performance index set to support performance evaluation and structural attribution analysis of packaged devices. The specific steps are as follows:

[0149] This embodiment uses the electromagnetic response dataset obtained above as input. First, a time window sliding interception is performed on the original signal. The fixed window length is set to Δt (for example, 50 μs) and the step size is set to δt (for example, 10 μs). Time segmentation processing is performed without losing continuity.

[0150] For the data in each window, perform wavelet packet decomposition transform, select the mother wavelet suitable for electromagnetic spike detection (such as Symlet 6 or Daubechies 4), and extract the following two-dimensional features:

[0151] Time domain impact characteristics: reflects the degree of rapid mutation of current / voltage / field strength after interference injection;

[0152] Frequency domain bandwidth expansion characteristics: reflects the energy leakage spectrum range and measures the frequency diffusion trend.

[0153] It's important to note that for semiconductor package device testing scenarios with abundant high-frequency electromagnetic spikes and dramatic signal variations, such as electromagnetic compatibility testing of high-speed digital signal processing chips, the Symlet 6 mother wavelet, with its high vanishing moment and excellent frequency localization, can more effectively capture high-frequency detail features and is therefore the preferred choice. For devices with a large proportion of low-frequency components in the signal and relatively gentle electromagnetic spikes, such as some analog power management chips, the Daubechies 4 mother wavelet, with its excellent approximation performance in the low-frequency band, can more accurately extract signal features and is therefore a more suitable choice. In practical applications, the energy entropy of the signal after different mother wavelet transformations can be calculated and the mother wavelet with the lowest energy entropy can be selected to ensure that the extracted features are most representative of the original signal.

[0154] The above features are uniformly encoded into a multi-scale response vector group, each vector representing the multi-domain linkage features in a disturbance response.

[0155] Considering the problems of state transition and disturbance timing offset between different time periods, this embodiment uses an adaptive dynamic time warping algorithm to align the multi-scale response vector group.

[0156] The algorithm uses an adaptive window matching mechanism to time synchronize and amplitude normalize all response vectors, eliminating non-responsive segments while retaining short-term drastic change information, thereby constructing a dynamic characteristic sequence of electromagnetic response with a unified structure. This is a sequence of tensors organized by state slices, and each tensor corresponds to a multidimensional disturbance response pattern under a specific state.

[0157] The parameters of the adaptive window matching mechanism are dynamically adjusted based on the average length of the response vector and the degree of feature variation. When the response vector lengths vary significantly, the initial window size is set to 1 / 3 of the average length, and the window expansion step size is set to 1 / 10 of the average length. During the matching process, if the feature difference between adjacent matching points exceeds a set threshold (for example, the Euclidean distance of the feature vector is greater than 0.5), the window is expanded by the step size to better align the features. For response vectors with relatively stable feature variations, the window expansion step size can be appropriately reduced to improve matching efficiency. For example, in the processing of electromagnetic response data for a certain communication chip, the above-mentioned dynamic parameter settings effectively solved the problem of response vector length and feature differences under different operating conditions, and improved the accuracy of feature alignment.

[0158] The dynamic feature sequence of the electromagnetic response is input into a pre-trained machine learning model. The model is constructed using a graph convolutional network and a cross-channel attention mechanism fusion architecture, and has the following capabilities:

[0159] The GCN module is used to model the perturbation path relationship in the package structure topology and learn the spatial coupling characteristics between different injection points;

[0160] The CCA module is used to mine the synchronous / asynchronous behavior differences among multiple response channels and extract the nonlinear interference cooperation features across paths.

[0161] The model output is a high-dimensional response feature tensor that represents the coupled response pattern under multi-state and multi-injection conditions. Combining the perturbation parameter set and the structural feature embedding vector (such as the topological location of the entry point and structural properties), the prediction layer is called to generate the electromagnetic tolerance prediction vector, which contains the following key quantities:

[0162] Interference tolerance amplitude, power coupling index, and immunity failure threshold; this vector is used to quantitatively predict the immunity capability boundary of the device in the current functional state.

[0163] Among them, interference tolerance refers to the maximum disturbance intensity that a packaged device can withstand from external electromagnetic disturbance without functional failure under a specific functional state. It is expressed in field strength (V / m) or injection power (dBm), reflecting the device's immunity threshold to radiated / conducted interference.

[0164] The power coupling rate indicates the proportion of disturbance energy transmitted into the device through the coupling path, quantifies the energy transmission efficiency between the external injection signal and key internal nodes, and is an important indicator for evaluating the electromagnetic shielding effect of the package and layout optimization.

[0165] Pulse recovery speed refers to the shortest time required for the device output signal to recover from an abnormal disturbance state to a normal stable state under the action of a strong transient electromagnetic pulse, usually measured in nanoseconds (ns).

[0166] It should be noted that the training data is derived from electromagnetic response data of a large number of semiconductor packaged devices of different types and batches under various operating conditions. The data is first cleaned to remove obvious errors and abnormal data points. The data is then divided into training, validation, and test sets with a ratio of 70%, 15%, and 15%. During training, the machine learning model based on the graph convolution-attention fusion architecture uses a stochastic gradient descent algorithm with an initial learning rate of 0.001, which is decayed to 0.9 after every 10 training cycles. During model training, the validation set is used to evaluate model performance. Training is stopped when the loss function on the validation set stops decreasing for five consecutive cycles to prevent overfitting. By training on this large amount of training data, the model effectively learns the complex mapping relationship between the dynamic characteristic sequence of electromagnetic response and electromagnetic tolerance.

[0167] In order to establish a performance evaluation model for the device in all states, the electromagnetic tolerance prediction vectors in multiple states are further calculated by index fusion. The specific method is as follows:

[0168] The principal components of the indicators in each dimension of the electromagnetic tolerance prediction vector are reconstructed, and the radial basis kernel function is used to map the indicators to the disturbance response feature space H. This space retains the nonlinear change trend, and the characteristic axes include but are not limited to interference tolerance, power coupling rate, and pulse recovery speed.

[0169] Introducing the state transition probability matrix (such as Markov chain form) in the device multi-state working model, constructing the state dependency weighting function φ(s), which is used to adjust the contribution of each state indicator to the overall performance score according to the state importance. .For example: ; Indicates the i-th functional state, which comes from the state set defined in the multi-state working model (such as high-speed operation, sleep, radio frequency on, etc.); Indicates the corresponding status Under the circumstance, the electromagnetic tolerance prediction vector output by the machine learning model is a vector containing multiple dimensions (such as coupling strength, tolerance amplitude, etc.); Indicates that the tolerance prediction vector The fusion index score obtained after projecting into the disturbance response feature space H; Represents the state dependency weighting function, used to describe the current state The importance or proportion in actual work reflects its impact on the overall compatibility score.

[0170] It's important to note that the state transition probability matrix is ​​derived by statistically analyzing the state transitions of packaged devices under a wide range of operating conditions. During testing, the device's states at various times (such as idle, data processing, and signal transmission) are recorded, and the number of transitions from one state to another is counted. For example, if, after 1000 test records, the device transitioned from the idle state to the data processing state 300 times, while the idle state occurred 400 times in total, the probability of transitioning from the idle state to the data processing state is 300 / 400 = 0.75. Similarly, the transition probabilities between all states are calculated to construct the state transition probability matrix.

[0171] Index set calculation and normalization: Calculate the device's comprehensive electromagnetic compatibility performance index set, including interference tolerance score, coupling strength index, and cross-state stability;

[0172] The above indicators are all standardized through normalization functions to eliminate the scale effect.

[0173] The interference tolerance score measures a device's ability to withstand electromagnetic disturbances under specific operating conditions. It is calculated based on the ratio between the maximum disturbance intensity a device can withstand under fault-free operating conditions during actual testing and the industry reference standard limit. The score reflects the device's immunity level under these conditions.

[0174] The Immunity Score is defined as:

[0175] If the electromagnetic field strength is used as the benchmark: ;in: Indicates the maximum electromagnetic disturbance intensity without functional failure in the device test under the i-th functional state (unit: V / m); Indicates the reference field strength specified in the standard for this type of device (for example, 10 V / m in IEC 61000-4-3). The interference tolerance score ranges from 0 to 1, with higher values ​​indicating higher immunity to interference.

[0176] The coupling strength index is used to quantify the coupling efficiency of external disturbances propagating through the packaging structure to internal key nodes, reflecting the coupling ability of electromagnetic interference on the spatial path. Its calculation expression can be based on the amplitude response ratio or the spectrum coupling ratio.

[0177] The coupling strength index CSI based on the amplitude ratio is calculated as: ;in: represents the injection amplitude at the disturbance entry point; represents the maximum response amplitude of the i-th node under disturbance; Indicates the background noise amplitude of the node when there is no disturbance. A larger CSI value indicates stronger coupling, while a smaller CSI value indicates that the package path has good electromagnetic shielding or isolation characteristics.

[0178] Cross-state stability is used to measure the stability of the electromagnetic compatibility performance indicators (such as interference tolerance score) of packaged devices when switching between different functional states, and is mainly described in the form of statistical distribution. Definition of the degree of dispersion based on the interference tolerance score: ; Where: Var represents the variance of the interference tolerance score in all states; BImmunity Score represents the mean of the interference tolerance score in all states; the closer the cross-state stability value is to 1, the higher the consistency of the device's anti-interference ability in different states; if it approaches 0, it indicates the existence of a state-dependent vulnerable area.

[0179] To further locate weak areas of electromagnetic performance and support structural optimization design, this embodiment introduces a structural attribution mechanism after the index calculation:

[0180] Establishing a package structure mapping function to map the correlation between each node in the structure and the high-frequency coupling feature in the response feature tensor;

[0181] Perform distribution fitting (e.g., Gaussian mixture modeling) on ​​the normalized interference tolerance score, coupling strength index, and cross-state stability;

[0182] According to the degree of overlap between the probability density peak position and the structural coordinates, electromagnetic weak areas (such as ground grid return ports and power node aggregation points at specific layers) are identified.

[0183] The final output is a structural sensitivity heat map, which is used to assist in local reconstruction of the packaging structure or interference isolation design.

[0184] Table 1: Comparison of electromagnetic weak area identification accuracy (verified by comparison with actual fault points)

[0185] Conclusion: The high-risk structural areas identified by the present invention are highly consistent with the actual electromagnetic interference fault points, and have reliable prediction capabilities.

[0186] To determine the electromagnetic compatibility compliance and risk classification of semiconductor packaged devices in full-function operation, this embodiment provides a risk level assessment method based on industry standard comparison and confidence model-driven analysis. Combined with a structural mapping mechanism, it outputs visual high-risk structural areas and design avoidance suggestions. The specific steps are as follows:

[0187] First, we establish an industry standard parameter library Sref for different package types and application scenarios. The data sources include:

[0188] IEC 61967 series (chip-level EMC radiated and conducted emission standards);

[0189] CISPR 25 (EMC standard for on-board equipment);

[0190] JEDEC, MIL-STD-461, ISO 11452 and other field-specific regulations.

[0191] The parameter library stores the following data items:

[0192] Electromagnetic tolerance limit (unit: V / m or dBμV);

[0193] Upper limit of coupling strength (unit: dB);

[0194] State stability tolerance (e.g., maximum acceptable response fluctuation rate for frequency hopping / power switching);

[0195] Each standard entry is associated with the following key index fields:

[0196] Package type (such as FCBGA, SiP, 2.5D, 3D IC, etc.);

[0197] Typical application frequency bands (e.g., <1 GHz, 5G sub-6, mmWave);

[0198] Power consumption level (high, medium, low);

[0199] Is there heterogeneous integration or multi-chip stacking?

[0200] Table 2: Improvement effect of electromagnetic compatibility performance indicators before and after design optimization (using the method of the present invention to guide design optimization)

[0201] Conclusion: After using the method described in the present invention to identify risk areas and guide structural adjustments, the electromagnetic tolerance of the device is significantly improved and the risk level is effectively reduced.

[0202] In this embodiment, for a target device (such as a 64-layer 2.5D packaged SiP chip), a matching parameter set Sref∗ is selected from a standard library as a comparison benchmark.

[0203] The comprehensive electromagnetic compatibility performance index set obtained by the above calculation is input into the comparison module. Since the index dimensions and numerical range are not consistent with the standard values, the following processing is required:

[0204] Each dimension in the indicator set (such as interference tolerance score, coupling strength index, and cross-state stability) is hierarchically normalized and mapped to the [0,1] interval;

[0205] At the same time, a unified scale is used to represent the corresponding index Sref∗ in the standard domain;

[0206] An interval uncertainty mapping mechanism is introduced to address boundary fluctuations and measurement errors in actual tests, introducing fuzzy processing into the scoring process:

[0207] A confidence interval (e.g. ±5% test bandwidth) is constructed for each indicator, and its “compliance level” is defined using a Bell-type membership function. The confidence score matrix Q for each state-indicator pair is obtained, and its elements are defined as: ; Where: eij represents the jth performance indicator under state i; sij represents the corresponding industry standard value; μij represents the uncertainty mapping membership function; Qij represents the confidence level that the indicator meets the standard under the current state.

[0208] The confidence score matrix Q is input into a multi-state risk assessment model that combines a fuzzy inference engine with a clustering algorithm:

[0209] Use the fuzzy rule base to map each indicator combination to the fuzzy risk factor, such as:

[0210] If the interference tolerance is low and the coupling index is high → the risk is high;

[0211] If cross-state stability is medium tolerance is high → risk is medium;

[0212] The model outputs a fuzzy risk score Ri∈[0,1] for each state;

[0213] Threshold clustering (such as K-means or fuzzy C-means) is used to divide all state labels into three categories:

[0214] “Safety”: Ri<0.3;

[0215] "Boundary": 0.3≤Ri<0.6;

[0216] “Exceeding the standard” or “high risk”: Ri≥0.6.

[0217] Finally, a risk level label set corresponding to a state is generated, and a confidence weight is assigned to each label to support probabilistic compliance judgment.

[0218] It should be noted that the fuzzy rule base is constructed based on expert experience and extensive experimental data. For example, when the interference tolerance score is "high" and the coupling strength index is "low," the risk level is determined to be "low"; when the interference tolerance score is "medium" and the coupling strength index is "medium," the risk level is determined to be "medium"; and when the interference tolerance score is "low" and the coupling strength index is "high," the risk level is determined to be "high." Rules are expressed in an if-then form, with confidence levels set. The final fuzzy rule base is determined through verification and adjustment of a large amount of test data with known risk levels, ensuring the accuracy and consistency of risk level assessments.

[0219] Link the risk level label set with the encapsulation layer structure mapping function:

[0220] Through the binding relationship between each entry point, coupling path node and state behavior in the structural model, find the spatial structural area corresponding to the high-risk state;

[0221] Generate 3D package heat maps and mark electromagnetic weak areas (such as overly dense coupling nodes, uneven power distribution, and ineffective decoupling);

[0222] Output avoidance suggestions, including local stacking adjustment suggestions (such as adjusting the position relationship between the ground and power planes); coupling path disconnection suggestions (such as adding distributed capacitors or shielding structures); and state switching timing optimization (such as extending the state switching interval).

[0223] Table 3: Project example (simplified)

[0224] Table 4: Comparison between the method of the present invention and the traditional EMC test method (interference tolerance extraction accuracy)

[0225] Conclusion: The present invention is significantly superior to traditional methods in terms of accuracy and coverage of multi-state tolerance feature extraction.

[0226] Example 2, please refer to Figure 2 As shown, the embodiment of the present invention provides a comprehensive electromagnetic compatibility test system for semiconductor packaging devices, including:

[0227] The polymorphic working modeling module collects the structural parameters, package layer topology and predefined functional states of the semiconductor package device to be tested, and builds a polymorphic working model of the device;

[0228] The parameter configuration module builds a disturbance injection control model based on the functional states of the device's multi-state working model and selects different disturbance source parameter sets according to the test standards;

[0229] The response acquisition module dynamically injects disturbance injection control models according to the parameter-centered disturbance source under the actual working conditions of the packaged device and collects device response data.

[0230] The model prediction module performs a joint analysis of the device response data in the time and frequency domains to construct a dynamic feature sequence of the electromagnetic response. This feature sequence is then input into the trained machine learning model to extract multidimensional coupling features and predict tolerances.

[0231] The stability modeling module calculates the device's comprehensive electromagnetic compatibility performance indicators in various functional states, including interference tolerance and coupling strength index, based on the indicator vector output by the machine learning model;

[0232] The compatibility risk assessment module compares the comprehensive electromagnetic compatibility performance index set with industry standards to determine the compatibility risk level of the packaged device in all working conditions.

[0233] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A comprehensive electromagnetic compatibility testing method for semiconductor packaging devices, characterized by: include: Collect the structural parameters, package layer topology and predefined functional states of the semiconductor package device to be tested, and build a multi-state working model of the device; According to the functional states of the device's multi-state working model, a disturbance injection control model is constructed, and different disturbance source parameter sets are selected according to the test standards; Under the actual working conditions of the packaged device, the disturbance injection control model is dynamically injected according to the parameter-centered disturbance source, and the device response data is collected; Perform a joint analysis of device response data in the time and frequency domains to construct a dynamic feature sequence of the electromagnetic response. This feature sequence is then input into a trained machine learning model for multi-dimensional coupling feature extraction and tolerance prediction. Based on the indicator vector output by the machine learning model, the comprehensive electromagnetic compatibility performance indicator set of the device in various functional states is calculated, including interference tolerance and coupling strength index; Compare the comprehensive electromagnetic compatibility performance index set with industry standards to determine the compatibility risk level of packaged devices under all working conditions.

2. The method for comprehensive electromagnetic compatibility testing of a semiconductor package device according to claim 1, wherein: The constructing of the device polymorphic working model comprises: A multi-layer graph neural network is used to model the package layer topology of the device under test, extract the node connectivity and electromagnetic coupling potential of each package layer, and generate a topology vector representation L1; High-resolution structural imaging data is compared and fused with CAD models to identify heterogeneous structural units, metal layer arrangements, and via clusters within the device under test, construct a structural parameter tensor matrix S1, and perform feature alignment with L1. Based on the device's design behavior model library, the predefined functional state set F1 is called to build a state control diagram. The states in F1 are logically mapped to the trigger paths of S1 to generate a state transition diagram M1. L1, S1 and M1 are jointly embedded through a multi-channel embedding encoder, and the multi-state working model of the device is finally output.

3. The method for comprehensive electromagnetic compatibility testing of a semiconductor package device according to claim 1, wherein: The construction of a disturbance injection control model according to each functional state of the device multi-state working model and the selection of different disturbance source parameter sets according to the test standard include: Perform state cluster analysis on the state transition graph in the polymorphic working model and extract representative state sequence sets using the time series feature compression algorithm; Based on each representative state in the representative state sequence set, a disturbance response channel diagram is established to identify the spatial nodes and temporal nodes where electromagnetic coupling may occur in this state, and a disturbance entry point set is defined; According to the current electromagnetic compatibility test standards, a disturbance source parameter library is constructed, including the spectrum range, disturbance energy level, waveform shape and duration. The disturbance source parameter library is projected onto each disturbance entry point set through the state disturbance mapping function to generate a disturbance injection control model.

4. The method for comprehensive electromagnetic compatibility testing of a semiconductor package device according to claim 1, wherein: The method of dynamically injecting the disturbance injection control model according to the parameter-centered disturbance source under the actual working state of the packaged device and collecting the device response data includes: Load the real functional scenario script to the packaged device through the asynchronous collaborative driving platform to start its multi-function state loop operation; The disturbance entry point set defined in the disturbance injection control model is called, and a multi-channel digital-analog collaborative jammer is used to implement multi-dimensional disturbance injection on the selected disturbance entry point set nodes; During the injection process, a high-precision electromagnetic response acquisition unit is activated through a time-domain synchronous trigger mechanism. Multi-point probes are deployed in the device outer shell, internal power supply network and key signal channels to collect RF leakage, current spikes, voltage fluctuations and timing drift data in real time to form a response data set.

5. The method for comprehensive electromagnetic compatibility testing of a semiconductor package device according to claim 1, wherein: The joint analysis of the device response data in the time domain and the frequency domain to construct an electromagnetic response dynamic feature sequence, and inputting the electromagnetic response dynamic feature sequence into the trained machine learning model includes: The response data is sliced ​​with a fixed window, and the time domain impact features and frequency domain bandwidth expansion features are extracted by wavelet packet transform, and a multi-scale response vector group is jointly constructed. Adaptive dynamic time warping algorithm is used to time align and morphologically normalize the multi-scale response vector group to form a unified structure of electromagnetic response dynamic feature sequence; The dynamic feature sequence of the electromagnetic response is input into a machine learning model built based on the graph convolution-attention fusion architecture. Utilizing its spatial coupling modeling capability and cross-channel attention extraction mechanism, high-dimensional extraction of nonlinear interaction features between multiple perturbation channels in the package structure is performed. Based on the response feature tensor output by the machine learning model, combined with the disturbance parameters and structural feature embedding, an electromagnetic tolerance prediction vector is generated to quantify the coupling strength and anti-interference capability boundary of the device under the set state.

6. The method for comprehensive electromagnetic compatibility testing of a semiconductor package device according to claim 5, wherein: The indicator vector output by the machine learning model is used to calculate the comprehensive electromagnetic compatibility performance indicator set of the device in various functional states, including: The principal component of each dimension output result in the electromagnetic tolerance prediction vector is reconstructed, and the disturbance response feature space H is constructed using nonlinear kernel function mapping, preserving the interference tolerance, power coupling rate and pulse recovery speed; Combined with the state transition probability matrix of the device multi-state working model, a state-dependent weighting function φ is constructed, so that each electromagnetic performance indicator has a different weight contribution in the corresponding state; Based on H and φ, a comprehensive set of electromagnetic compatibility performance indicators is calculated, including interference tolerance score, coupling strength index and cross-state stability.

7. The method for comprehensive electromagnetic compatibility testing of a semiconductor package device according to claim 1, wherein: The comprehensive electromagnetic compatibility performance index set is compared with industry standards to determine the compatibility risk level of the packaged device in a fully functional state: Build an industry standard parameter library, which includes electromagnetic tolerance limits, coupling strength upper limits, and state fluctuation tolerances corresponding to various types of packaging structures, and perform standard matching based on packaging form and frequency band application scenarios; The parameters of each indicator in the performance indicator set are scaled and aligned with the standard domain parameters through hierarchical normalization. The interval uncertainty mapping method is used to introduce test errors and boundary fuzziness to obtain the confidence score matrix. A multi-state risk level assessment model is constructed based on the confidence score matrix. By integrating fuzzy reasoning and threshold clustering methods, a risk level label set for the device in all functional states is generated. The labels include three levels: "safe", "marginal", and "exceeding the standard", and their corresponding probability weights. Map the labels to the encapsulation layer structure, identify the spatial structure areas and status trigger conditions with high risk levels, and perform visual output and generate avoidance suggestions.

8. A semiconductor package device electromagnetic compatibility comprehensive test system, used to implement the semiconductor package device electromagnetic compatibility comprehensive test method according to any one of claims 1 to 7, characterized in that: include: The polymorphic working modeling module collects the structural parameters, package layer topology and predefined functional states of the semiconductor package device to be tested, and builds a polymorphic working model of the device; The parameter configuration module builds a disturbance injection control model based on the functional states of the device's multi-state working model and selects different disturbance source parameter sets according to the test standards; The response acquisition module dynamically injects disturbance injection control models according to the parameter-centered disturbance source under the actual working conditions of the packaged device and collects device response data. The model prediction module performs a joint analysis of the device response data in the time and frequency domains to construct a dynamic feature sequence of the electromagnetic response. This feature sequence is then input into the trained machine learning model to extract multidimensional coupling features and predict tolerances. The stability modeling module calculates the device's comprehensive electromagnetic compatibility performance indicators in various functional states, including interference tolerance and coupling strength index, based on the indicator vector output by the machine learning model; The compatibility risk assessment module compares the comprehensive electromagnetic compatibility performance index set with industry standards to determine the compatibility risk level of the packaged device in all working conditions.

Images