Power chip defect detection system
Through multimodal data acquisition and deep learning analysis technology, the problem of incomplete and inaccurate power chip defect detection in the existing technology is solved, and accurate detection and judgment of chip defects is achieved, which significantly improves chip quality and the reliability of electronic equipment.
Patent Information
- Application Number
- CN202510035972.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-09
- Publication Date
- 2025-05-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has incomplete and inaccurate detection of power chip defects, resulting in unreliable chip quality and high failure rate of electronic equipment.
The multimodal data acquisition module is used to collect the I-V characteristic curve of the power chip, the chip surface temperature distribution and the electromagnetic radiation data inside the chip, and normalized the processing through the data processing module, combined with the deep learning architecture of the intelligent analysis module (convolutional neural network, recurrent neural network and fully connected layer) to analyze the defect type, and determine the probability distribution, location and severity of the output defects of the output module through the defect.
Comprehensive and accurate detection of power chip defects has been achieved, which significantly improves chip quality reliability, reduces product failure rate, and reduces economic losses and quality risks of manufacturers.
Smart Images
Figure CN119936620A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of chip detection, and in particular to a power chip defect detection system. Background Art
[0002] In today's highly developed electronics industry, chips are core components, and their quality is directly related to the performance and stability of various electronic devices. Power chips play a key role in the power conversion and control links of many electronic devices, such as in the power system of electric vehicles, power management of industrial automation equipment, and power amplifiers of communication base stations. The present invention focuses on the field of chip detection technology and is committed to developing a new power chip defect detection system to address the shortcomings of traditional detection methods in ensuring the quality of power chips and ensure the reliable operation of electronic equipment.
[0003] The present invention aims to provide a power chip defect detection system, which, through innovative detection methods and architectures, effectively solves the problem of incomplete and inaccurate power chip defect detection in the prior art, improves the quality reliability of power chips, and reduces product failure rates. Summary of the invention
[0004] The main purpose of the present invention is to provide a power chip defect detection system, which can effectively solve the problems in the background technology.
[0005] To achieve the above object, the technical solution adopted by the present invention is:
[0006] A power chip defect detection system specifically includes: a multi-modal data acquisition module: used to collect data of the power chip under different working conditions, specifically including IV characteristic curve data, chip surface temperature distribution data and electromagnetic radiation data inside the chip;
[0007] Data processing module: used to normalize the collected current-voltage characteristic curve data, chip surface temperature distribution data and electromagnetic radiation data inside the chip;
[0008] Intelligent analysis module: It uses a deep learning-based convolutional neural network layer, recurrent neural network layer, and fully connected layer architecture to analyze the defect types of power chips and output the probability distribution of defects;
[0009] Defect determination output module: Determine the type, possible location and severity of the defect based on the defect probability distribution output by the intelligent analysis module.
[0010] Preferably, the specific manner in which the multimodal data acquisition module acquires power chip data is:
[0011] Using a current-voltage test instrument, when applying different levels of current to the power chip, the voltage value at both ends of the power chip is tested, and an IV characteristic curve is formed according to the applied voltage;
[0012] Use infrared thermal imager to obtain chip surface temperature;
[0013] The electromagnetic sensor is used to collect the electromagnetic radiation signal emitted by the chip when it is working. The frequency coverage range of the electromagnetic sensor is 100KHz-10GHz.
[0014] Preferably, the specific manner in which the data processing unit performs normalization processing on the collected multimodal data is:
[0015] For IV characteristic curve data, the following formula is used:
[0016]
[0017] Among them, I min is the minimum current value collected, I max is the maximum current value collected, I is the input current value, and V is the actual voltage value measured under the condition of input current value I;
[0018] For temperature distribution data and electromagnetic radiation, linear interpolation is used to normalize them to the range of 0-1.
[0019] Preferably, the convolutional neural network layer is used to automatically extract local features in the IV characteristic curve and the temperature distribution image, and its convolution kernel size is 3×3 and 5×5, and the step size is 1 or 2.
[0020] Preferably, the recurrent neural network layer is used to model the time series data of electromagnetic signals, capture the changing patterns of electromagnetic signals over time, adopt long short-term memory network units, and shift the number of hidden layer nodes by 128-512 to process long-term dependencies in long sequence data.
[0021] Preferably, the features extracted by the convolutional neural network layer and the recurrent neural network layer are fused through the fully connected layer, and the defects of the power chip are predicted using a softmax classifier to output the probability distribution of the defects.
[0022] Preferably, according to the defect probability distribution output by the intelligent analysis module, the specific method for determining the type, possible location and severity of the defect is:
[0023] A threshold is set. When the probability of a certain defect type is greater than the threshold, the power chip is judged to have the defect and a detailed defect report is output.
[0024] Preferably, in the defect report, the possible location of the defect is determined by analyzing the abnormal points of the IV characteristic curve and the abnormal area of the temperature distribution, and the severity is divided into three levels: high, medium and low according to the probability value.
[0025] Compared with the prior art, the present invention has the following beneficial effects:
[0026] 1. By collecting multimodal data, including current-voltage characteristic curves, chip surface temperature distribution, and chip internal electromagnetic radiation data, the present invention can detect tiny internal defects such as micro cracks inside the chip, uneven impurity distribution, etc., compared with the traditional method that only relies on wafer testing and finished product testing, effectively avoiding chip failures in actual work caused by potential defects, significantly reducing the risk of performance degradation or damage of the entire electronic equipment, and greatly reducing the economic losses and quality risks of manufacturers.
[0027] 2. The intelligent analysis module of the present invention combines a unique convolutional neural network-recurrent neural network architecture to conduct in-depth analysis of the data, and can accurately determine the possible location and severity of the defect through the defect determination output module. For example, by analyzing the abnormal points of the IV characteristic curve and the abnormal areas of the temperature distribution, the approximate location of the defect inside the chip can be accurately determined, and the severity can be divided into three levels: high, medium, and low according to the probability of the defect, providing a detailed and reliable basis for subsequent targeted processing, which helps to improve the efficiency and quality of chip repair or screening. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 It is a module diagram of the overall workflow of the present invention. DETAILED DESCRIPTION
[0029] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.
[0030] like Figure 1 As shown, the present invention discloses a power chip defect detection system, which specifically includes: a multi-modal data acquisition module: used to collect data of the power chip under different working conditions, specifically including IV characteristic curve data, chip surface temperature distribution data and electromagnetic radiation data inside the chip;
[0031] Specifically, a current-voltage test instrument is used to test the voltage value at both ends of the power chip when applying different levels of current to the power chip, and an IV characteristic curve is formed according to the applied voltage;
[0032] Use infrared thermal imager to obtain chip surface temperature;
[0033] The electromagnetic sensor is used to collect the electromagnetic radiation signal emitted by the chip when it is working. The frequency coverage range of the electromagnetic sensor is 100KHz-10GHz;
[0034] Data processing module: used to normalize the collected current-voltage characteristic curve data, chip surface temperature distribution data and electromagnetic radiation data inside the chip;
[0035] Specifically, the data processing unit performs normalization processing on the collected multimodal data in the following specific manner:
[0036] For IV characteristic curve data, the following formula is used:
[0037]
[0038] Among them, I min is the minimum current value collected, I max is the maximum current value collected, I is the input current value, and V is the actual voltage value measured under the condition of input current value I;
[0039] For temperature distribution data and electromagnetic radiation, linear interpolation is used to normalize them to the range of 0-1.
[0040] Intelligent analysis module: It uses a deep learning-based convolutional neural network layer, recurrent neural network layer, and fully connected layer architecture to analyze the defect types of power chips and output the probability distribution of defects;
[0041] Defect determination output module: Determine the type, possible location and severity of the defect based on the defect probability distribution output by the intelligent analysis module.
[0042] It should be noted that the convolutional neural network is a deep learning model specifically used to process data with a grid structure (such as images). In this patent, it is used to process IV characteristic curves and temperature distribution image data. Its core operation is convolution, which sets convolution kernels of different sizes (such as 3×3 and 5×5) to slide on the data, perform weighted summation on the local area, and extract local features of the data. For example, for the IV characteristic curve image, the convolution operation can identify features such as slope changes and inflection points in the curve. The step size (1 or 2) determines the interval between the convolution kernels at each move, affecting the refinement and computational efficiency of feature extraction. Pooling operations further reduce the data dimension. For example, maximum pooling selects the maximum value in the local area as the output, reducing the amount of data while retaining important features, thereby extracting key feature information and preparing for subsequent analysis.
[0043] A recurrent neural network is a neural network used to process sequence data. Its characteristic is that it takes into account previous input information when processing the current input and has a memory function. In this patent, it is used to process the time series data of the electromagnetic radiation signal inside the chip. Because the electromagnetic radiation signal changes over time, there is a correlation between the data before and after it, and the recurrent neural network can capture this dynamic change pattern in time. For example, by analyzing the changes in the intensity of electromagnetic radiation over a period of time, it is possible to determine whether there is an abnormal fluctuation pattern.
[0044] Long short-term memory network is a variant of recurrent neural network, which solves the long-term dependency problem that may occur in recurrent neural network through special gate structure (input gate, forget gate, output gate). In this patent, the number of hidden layer nodes is between 128 and 512. These nodes selectively update and retain information through gate structure, so that the network can better handle long-term dependencies in long sequence data. For example, in the long-term monitoring of electromagnetic radiation signals, it is possible to remember early signal features and combine them with current signals for analysis, so as to more accurately identify potential defect-related patterns and avoid the loss of important information due to long time intervals.
[0045] The fully connected layer plays the role of integrating and transforming features in the neural network. In this patent, it receives image features extracted from the convolutional neural network and time series features extracted from the recurrent neural network (through the long short-term memory network unit). The features from these different sources are deeply fused, and the fused features are mapped to a new space by multiplying each input feature by the corresponding weight and summing them, so that they can be subsequently classified using the softmax classifier. It can learn the complex combination relationship between different features and provide support for the final accurate prediction of the defect type of the chip. For example, the IV curve features extracted by the convolutional neural network and the electromagnetic radiation signal features extracted by the recurrent neural network are comprehensively analyzed to determine whether these feature combinations conform to the characteristic pattern of a certain defect, thereby determining whether the chip has defects and the type of defects. The severity is divided into three levels: high, medium, and low according to the probability value.
[0046] The present invention is further disclosed below in conjunction with specific embodiments:
[0047] Preparation stage: Before testing, the power chip to be tested needs to be properly installed on a specially customized test fixture. This test fixture is carefully designed to ensure a stable and reliable electrical connection between the chip and various test instruments, while ensuring that the chip surface is fully exposed to the effective detection range of the infrared thermal imager and electromagnetic sensor, laying a solid foundation for subsequent data collection;
[0048] Data acquisition: After starting the multimodal data acquisition module, apply current to the chip according to the preset precise current level sequence. While applying current, synchronously collect IV characteristic curve data, temperature distribution data and electromagnetic radiation data, and transmit them to the data preprocessing unit in real time through high-speed data transmission lines. For example, when testing a power chip with a rated current of 10A, start from 1A and gradually increase to 50A in steps of 1A. Data is collected at each current level to ensure that the collected data can fully reflect the performance of the chip under different working conditions;
[0049] Data processing and analysis: After receiving the incoming data, the data preprocessing unit immediately processes it according to the established normalization method. After the processing is completed, the intelligent analysis module starts working. The convolutional neural network layer first extracts features from the IV characteristic curve and temperature distribution image. After complex operations of multiple convolutional layers and pooling layers, the feature map is flattened and input into the fully connected layer. The recurrent neural network layer processes the time series of the electromagnetic radiation signal, deeply learns the dynamic change law of the signal through the long short-term memory network unit, and outputs the final hidden state to the fully connected layer. The fully connected layer fuses the features from the convolutional neural network and the recurrent neural network, and uses the softmax classifier to calculate the defect probability distribution;
[0050] Defect determination and reporting: The defect determination output module receives the output results of the intelligent analysis module. When the defect probability is greater than 0.5, a defect report is generated in combination with the characteristic information of the data. For example, if an abnormal inflection point is found in the current at a certain voltage in the IV characteristic curve, combined with the physical structure and circuit design of the chip, it can be analyzed that the possible defect location is near the junction area of a certain transistor; for the local high temperature area in the temperature distribution, further judge its association with the power module layout inside the chip. If the high temperature area happens to be near the power amplifier module, there may be a heat dissipation problem or internal short circuit defect in the module, thereby determining the specific situation of the defect.
[0051] For example, a power chip used in an electric vehicle motor drive controller is tested. This chip needs to withstand large current and voltage stresses when running at high power. During the test, the IV characteristic curve collected by the multimodal data acquisition module showed abnormal current fluctuations in a certain higher voltage range, and the temperature distribution on the chip surface showed local high temperature points in the power tube area. After processing by the intelligent analysis module and analysis by the defect judgment output module, it was determined that the chip had a potential defect of power tube breakdown, and the severity was high. Further anatomical analysis found that the power tube inside the chip had photolithography defects during the manufacturing process, resulting in excessive local electric field strength, which was prone to breakdown under high voltage, which was consistent with the test results.
[0052] Based on the above, the present invention can realize comprehensive and accurate detection of power chip defects, effectively improve the quality inspection level of power chips, and significantly reduce the failure rate of products in practical applications. It has outstanding innovation and extremely high practical value in the field of chip detection.
[0053] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A power chip defect detection system, characterized in that: include: Multimodal data acquisition module: used to collect data of power chips under different working conditions, including IV characteristic curve data, chip surface temperature distribution data and electromagnetic radiation data inside the chip; Data processing module: used to normalize the collected current-voltage characteristic curve data, chip surface temperature distribution data and electromagnetic radiation data inside the chip; Intelligent analysis module: It uses a deep learning-based convolutional neural network layer, recurrent neural network layer, and fully connected layer architecture to analyze the defect types of power chips and output the probability distribution of defects; Defect determination output module: Determine the type, possible location and severity of the defect based on the defect probability distribution output by the intelligent analysis module.
2. The power chip defect detection system according to claim 1, characterized in that: The specific method of the multi-modal data acquisition module to collect power chip data is: Using a current-voltage test instrument, when applying different levels of current to the power chip, the voltage value at both ends of the power chip is tested, and an IV characteristic curve is formed according to the applied voltage; Use infrared thermal imager to obtain chip surface temperature; The electromagnetic sensor is used to collect the electromagnetic radiation signal emitted by the chip when it is working. The frequency coverage range of the electromagnetic sensor is 100KHz-10GHz.
3. The power chip defect detection system according to claim 1, characterized in that: The specific manner in which the data processing unit performs normalization processing on the collected multimodal data is: For IV characteristic curve data, the following formula is used: Among them, I min is the minimum current value collected, I max is the maximum current value collected, I is the input current value, and V is the actual voltage value measured under the condition of input current value I; For temperature distribution data and electromagnetic radiation, linear interpolation is used to normalize them to the range of 0-1.
4. The power chip defect detection system according to claim 1, characterized in that: The convolutional neural network layer is used to automatically extract local features in the IV characteristic curve and the temperature distribution image, and its convolution kernel size is 3×3 and 5×5, and the step size is 1 or 2.
5. The power chip defect detection system according to claim 4, characterized in that: The recurrent neural network layer is used to model the time series data of electromagnetic signals, capture the changing patterns of electromagnetic signals over time, adopt long short-term memory network units, shift the number of hidden layer nodes by 128-512, and process long-term dependencies in long sequence data.
6. The power chip defect detection system according to claim 5, characterized in that: The features extracted by the convolutional neural network layer and the recurrent neural network layer are fused through the fully connected layer, and the defects of the power chip are predicted using the softmax classifier to output the probability distribution of the defects.
7. The power chip defect detection system according to claim 1, characterized in that: According to the defect probability distribution output by the intelligent analysis module, the specific method for determining the type, possible location and severity of the defect is as follows: A threshold is set. When the probability of a certain defect type is greater than the threshold, the power chip is judged to have the defect and a detailed defect report is output.
8. The power chip defect detection system according to claim 7, characterized in that: In the defect report, the possible location of the defect is determined by analyzing the abnormal points of the IV characteristic curve and the abnormal area of the temperature distribution, and the severity is divided into three levels: high, medium and low according to the probability value.