CN114972327A - Semiconductor package test system and test method thereof - Google Patents

Abstract

The application relates to the field of semiconductor packaging, and particularly discloses a semiconductor packaging test system and a test method thereof, wherein an ultrasonic scanner is used for collecting an ultrasonic image of a plastic packaging microcircuit, a convolutional neural network model is further used for respectively extracting shallow implicit characteristic information of the ultrasonic image and high-dimensional implicit characteristic distribution information which focuses on chip layering, lead bonding and chip bonding areas, and characteristic graphs of the shallow implicit characteristic information and the high-dimensional implicit characteristic distribution information are subjected to scale correction processing during characteristic fusion, so that Cauchy weight probabilistic analysis is performed by performing probabilistic information interpretation on characteristic values of all positions of a global characteristic graph, robustness on information loss is enhanced, and characteristic expression capacity of the global characteristic graph is improved. Furthermore, the packaging effect of the plastic packaged microcircuit can be more accurately evaluated.

Description

Semiconductor packaging test system and test method thereof

Technical Field

The present invention relates to the field of semiconductor packaging, and more particularly, to a semiconductor package test system and a test method thereof.

Background

The plastic-packaged microcircuit is a semiconductor product prepared by a semiconductor packaging process. The particularity of the plastic-encapsulated microcircuits itself causes the internal defects to be easily generated in the early injection molding stage and the subsequent storage and use stage. The internal interface of the circuit can cause stress action in the layering and layering processes, so that the problems of device bonding degradation, metallization step degradation, chip cracking and the like are caused.

Therefore, it is desirable to test the packaged plastic-encapsulated microcircuits to determine whether their packaging quality meets the requirements.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a semiconductor package testing system and a testing method thereof, wherein an ultrasonic scanner is used for acquiring an ultrasonic image of a plastic package microcircuit, a convolutional neural network model is further used for respectively extracting shallow implicit characteristic information of the ultrasonic image and high-dimensional implicit characteristic distribution information focused on a chip layering area, a lead bonding area and a chip bonding area, and characteristic graphs of the shallow implicit characteristic information and the chip bonding area are subjected to scale correction processing during characteristic fusion, so that Cauchy weight probability is performed by performing probabilistic information interpretation on characteristic values of all positions of a global characteristic graph, robustness on information loss is enhanced, and characteristic expression capacity of the global characteristic graph is improved. Furthermore, the packaging effect of the plastic packaged microcircuit can be more accurately evaluated.

According to an aspect of the present application, there is provided a semiconductor package testing system, including: a training module comprising: the training image acquisition unit is used for acquiring an ultrasonic image of the plastic package microcircuit acquired by the ultrasonic scanner; the first convolution coding unit is used for enabling the ultrasonic image to pass through a first convolution neural network using a first convolution kernel so as to obtain an ultrasonic characteristic map; an interesting region extracting unit, configured to determine a plurality of interesting regions in the ultrasonic feature map corresponding to the chip layering, the wire bonding region, and the chip bonding region based on positions of the chip layering, the wire bonding region, and the chip bonding region in the ultrasonic image; a region-of-interest coding unit, configured to pass the plurality of regions of interest through a second convolutional neural network using a second convolutional kernel respectively to obtain a plurality of region-of-interest feature maps, where the second convolutional kernel is smaller than the first convolutional kernel; a second convolution coding unit, for passing the ultrasonic feature map through a third convolution neural network using non-overlapping convolution kernels to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; a compensation unit, configured to perform probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map, where the probabilistic information interpretation is performed based on a weighted sum of a natural exponent function value raised to the power of the feature value of each position in the global feature map and a natural exponent function value raised to the power of the feature value of each position in the global feature map; the feature map fusion unit is used for fusing the compensated global feature map and the plurality of interested region feature maps to obtain a classification feature map; the loss calculation unit is used for enabling the classification characteristic map to pass through a classifier to obtain a classification loss function value; a training unit to train the first convolutional neural network, the second convolutional neural network, and the third convolutional neural network based on the classification loss function value; and an inference module comprising: the data acquisition unit of the object to be detected is used for acquiring an ultrasonic image of the plastic package microcircuit acquired by the ultrasonic scanner; the first neural network unit is used for enabling the ultrasonic image to pass through the first convolution neural network which is trained by the training module and uses the first convolution kernel so as to obtain an ultrasonic characteristic diagram; an interesting region determining unit, configured to determine a plurality of interesting regions in the ultrasonic feature map corresponding to the chip layering, the wire bonding regions and the chip bonding regions based on positions of the chip layering, the wire bonding regions and the chip bonding regions in the ultrasonic image; a region-of-interest deep coding unit, configured to pass the multiple regions of interest through a second convolutional neural network trained by the training module and using a second convolutional kernel, respectively, to obtain multiple region-of-interest feature maps, where the second convolutional kernel is smaller than the first convolutional kernel; a second neural network unit, configured to pass the ultrasonic feature map through a third convolutional neural network using the non-overlapping convolutional kernels trained by the training module to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; the characteristic information fusion unit is used for fusing the compensated global characteristic map and the characteristic maps of the plurality of interested areas to obtain a classification characteristic map; and the packaging test result generating unit is used for enabling the classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the packaging effect of the plastic packaging microcircuit meets the preset requirement or not.

According to another aspect of the present application, there is provided a test method of a semiconductor package test system, which includes: a training phase comprising: acquiring an ultrasonic image of the plastic package microcircuit acquired by an ultrasonic scanner; passing the ultrasound image through a first convolution neural network using a first convolution kernel to obtain an ultrasound signature; determining a plurality of regions of interest in the ultrasonic signature corresponding to a chip layer, a wire bonding region, and a chip bonding region based on the positions of the chip layer, the wire bonding region, and the chip bonding region in the ultrasonic image; respectively passing the multiple regions of interest through a second convolutional neural network using a second convolutional kernel to obtain multiple region of interest feature maps, wherein the second convolutional kernel is smaller than the first convolutional kernel; passing the ultrasound signature through a third convolutional neural network using non-overlapping convolutional kernels to obtain a global signature having the same size as the plurality of region-of-interest signatures; performing probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map, wherein the probabilistic information interpretation is performed based on a weighted sum of natural exponent function values raised to the power of the feature values of the positions in the global feature map and natural exponent function values raised to the power of the feature values of the positions in the global feature map; fusing the compensated global feature map and the plurality of interested region feature maps to obtain a classification feature map; passing the classification feature map through a classifier to obtain a classification loss function value; training the first, second, and third convolutional neural networks based on the classification loss function values; and an inference phase comprising: acquiring an ultrasonic image of the plastic package microcircuit acquired by an ultrasonic scanner; passing the ultrasonic image through the first convolution neural network using the first convolution kernel trained by the training module to obtain an ultrasonic feature map; determining a plurality of regions of interest in the ultrasonic signature corresponding to a chip delamination, a wire bonding region, and a chip bonding region based on the positions of the chip delamination, the wire bonding region, and the chip bonding region in the ultrasonic image; respectively passing the multiple regions of interest through a second convolutional neural network which is trained by the training module and uses a second convolutional kernel to obtain multiple region of interest characteristic maps, wherein the second convolutional kernel is smaller than the first convolutional kernel; passing the ultrasound feature map through a third convolutional neural network using the non-overlapping convolutional kernels trained by the training module to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; fusing the compensated global feature map and the plurality of interested region feature maps to obtain a classification feature map; and enabling the classification characteristic diagram to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the packaging effect of the plastic packaging microcircuit meets the preset requirement or not.

Compared with the prior art, the semiconductor package testing system and the testing method thereof have the advantages that the ultrasonic image of the plastic package microcircuit is acquired through the ultrasonic scanner, the shallow implicit feature information of the ultrasonic image and the high-dimensional implicit feature distribution information which is more focused on the chip layering, the lead bonding and the chip bonding area are further extracted through the convolutional neural network model, and the feature graphs of the shallow implicit feature information and the high-dimensional implicit feature distribution information are subjected to scale correction processing during feature fusion, so that Cauchy weight probability is performed by performing probabilistic information interpretation on the feature value of each position of the global feature graph, the robustness on information loss is enhanced, and the feature expression capability of the global feature graph is improved. Furthermore, the packaging effect of the plastic packaged microcircuit can be more accurately evaluated.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.

Fig. 1 is a schematic view of a scenario of a semiconductor package testing system according to an embodiment of the present application.

Fig. 2 is a block diagram of a semiconductor package test system according to an embodiment of the present application.

Fig. 3A is a flowchart of a training phase in a testing method of a semiconductor package testing system according to an embodiment of the present application.

Fig. 3B is a flow chart of an inference phase in a testing method of a semiconductor package testing system according to an embodiment of the application.

Fig. 4 is a schematic diagram illustrating a training phase in a testing method of a semiconductor package testing system according to an embodiment of the present application.

Fig. 5 is a schematic diagram of an inference stage in a testing method of a semiconductor package testing system according to an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.

Overview of a scene

As mentioned above, the plastic-encapsulated microcircuit is a semiconductor product prepared by a semiconductor packaging process. The particularity of the plastic-encapsulated microcircuits itself causes the internal defects to be easily generated in the early injection molding stage and the subsequent storage and use stage. The internal interface of the circuit can cause stress action in the layering and layering processes, so that the problems of device bonding degradation, metallization step degradation, chip cracking and the like are caused.

Therefore, it is desirable to test the packaged plastic microcircuits to determine whether their packaging quality meets the requirements.

Correspondingly, the analysis and evaluation of the packaging quality of the plastic packaged microcircuit can be carried out based on the ultrasonic image analysis of the plastic packaged microcircuit, and the essence is a classification problem, namely, the evaluation of the packaging quality of the plastic packaged microcircuit is carried out based on the extraction of local implicit characteristics in the ultrasonic image, and then the classification is carried out to obtain a classification result for representing whether the packaging quality meets the requirement or not.

In particular, in the technical solution of the present application, it is desirable to determine the packaging effect of the plastic packaged microcircuit by using an ultrasonic scanning detection technology. The basic working principle of the ultrasonic scanning detection technology is that a pulse generator is used for generating a pulse signal, a piezoelectric sensor is excited to generate ultrasonic waves with a certain frequency, and the ultrasonic waves are transmitted to a sample through a decoupling medium. During propagation, the ultrasonic waves pass through a continuous and uniform medium, and once the ultrasonic waves encounter different media, reflection phenomena are caused, and defects such as delamination and cracks are discovered in such a way. The reflected ultrasonic wave is received by the sensor and is processed correspondingly to automatically generate an ultrasonic image.

Thus, in a specific embodiment, first, an ultrasonic image of the plastic encapsulated microcircuit is acquired by an ultrasonic scanner. It should be understood that, in the standard convolutional neural network, as the coding process goes deep, shallow features of the ultrasonic image of the plastic-packaged microcircuit, such as shapes, textures, lines, bumps, grooves, boundaries and the like, are submerged in the background information or become blurred as the depth goes deep, and the shallow feature information has an important significance in judging the packaging effect of the plastic-packaged microcircuit. Therefore, further, the ultrasonic image is passed through a first convolution neural network of a first convolution kernel to obtain an ultrasonic feature map. Here, the first convolutional neural network performs convolutional processing, pooling processing, and nonlinear activation processing on the ultrasound image through a plurality of convolutional layers to obtain the ultrasound feature map. The number of the coiling layers in the multilayer coiling layer is 4-6.

Moreover, when the characteristics of the ultrasonic image of the plastic packaged microcircuit are extracted, high attention should be paid to focusing on chip layering, wire bonding and chip bonding areas. Accordingly, based on the locations of the die delamination, wire bonding, and die attach regions in the ultrasound image, a corresponding plurality of regions of interest in the ultrasound signature are determined. Then, the multiple regions of interest are respectively processed in a second convolution neural network by using a second convolution kernel, so as to excavate the deep-level local implicit feature information which is more focused on the chip layering, wire bonding and chip bonding areas in the high-dimensional features of the ultrasonic image of the plastic-packaged microcircuit, and thus obtain multiple region of interest feature maps, wherein the second convolution kernel is smaller than the first convolution kernel.

It will be appreciated that in order to scale-align the ultrasound signature with the region of interest signature, a global signature is obtained using a third convolutional neural network with non-overlapping convolution kernels, but this results in a global signature with some loss of information.

Therefore, the global feature map needs to be further modified:

wherein

Is the feature value of each location of the global feature map,

and an exponential operation representing a feature map, which is a calculation of a natural exponential function value raised to the power of the feature value at each position in the feature map.

Therefore, Cauchy weight probabilistic transformation is performed by performing probabilistic information interpretation on the feature value of each position of the global feature map, so that in the training process of the third convolutional neural network through gradient back propagation, the parameters of the third convolutional neural network can be self-adaptive to information loss caused by non-overlapping convolutional kernels along with periodic iteration, the robustness to the information loss is enhanced, and the feature expression capability of the global feature map is improved.

In this way, a compensated global feature map is obtained, which is then fused with the plurality of feature maps of interest to obtain a classification feature map for classification. And then, the classification characteristic diagram is passed through a classifier to obtain a classification result for indicating whether the packaging effect of the plastic packaging microcircuit meets the preset requirement or not.

Based on this, the present application proposes a semiconductor package test system, which includes a training module and an inference module. Wherein, the training module includes: the training image acquisition unit is used for acquiring an ultrasonic image of the plastic package microcircuit acquired by the ultrasonic scanner; the first convolution coding unit is used for enabling the ultrasonic image to pass through a first convolution neural network using a first convolution kernel so as to obtain an ultrasonic characteristic map; an interesting region extracting unit, configured to determine a plurality of interesting regions in the ultrasonic feature map corresponding to the chip layering, the wire bonding region, and the chip bonding region based on positions of the chip layering, the wire bonding region, and the chip bonding region in the ultrasonic image; a region-of-interest coding unit, configured to pass the plurality of regions of interest through a second convolutional neural network using a second convolutional kernel to obtain a plurality of region-of-interest feature maps, where the second convolutional kernel is smaller than the first convolutional kernel; a second convolution coding unit, for passing the ultrasonic feature map through a third convolution neural network using non-overlapping convolution kernels to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; a compensation unit, configured to perform probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map, where the probabilistic information interpretation is performed based on a weighted sum of a natural exponent function value raised to the power of the feature value of each position in the global feature map and a natural exponent function value raised to the power of the feature value of each position in the global feature map; the feature map fusion unit is used for fusing the compensated global feature map and the plurality of interested region feature maps to obtain a classification feature map; the loss calculation unit is used for enabling the classification characteristic map to pass through a classifier to obtain a classification loss function value; a training unit to train the first convolutional neural network, the second convolutional neural network, and the third convolutional neural network based on the classification loss function value. Wherein, the inference module comprises: the data acquisition unit of the object to be detected is used for acquiring an ultrasonic image of the plastic package microcircuit acquired by the ultrasonic scanner; the first neural network unit is used for enabling the ultrasonic image to pass through the first convolution neural network which is trained by the training module and uses the first convolution kernel so as to obtain an ultrasonic characteristic diagram; an interesting region determining unit, configured to determine a plurality of interesting regions in the ultrasonic feature map corresponding to the chip layering, the wire bonding regions and the chip bonding regions based on positions of the chip layering, the wire bonding regions and the chip bonding regions in the ultrasonic image; a region-of-interest deep coding unit, configured to pass the multiple regions of interest through a second convolutional neural network trained by the training module and using a second convolutional kernel, respectively, to obtain multiple region-of-interest feature maps, where the second convolutional kernel is smaller than the first convolutional kernel; a second neural network unit, configured to pass the ultrasonic feature map through a third convolutional neural network using the non-overlapping convolutional kernels trained by the training module to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; the characteristic information fusion unit is used for fusing the compensated global characteristic map and the characteristic maps of the plurality of interested areas to obtain a classification characteristic map; and the packaging test result generating unit is used for enabling the classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the packaging effect of the plastic packaging microcircuit meets the preset requirement or not.

Fig. 1 illustrates a schematic view of a scenario of a semiconductor package testing system according to an embodiment of the present application. As shown in fig. 1, in the training phase of the application scenario, first, an ultrasound image of a plastic microcircuit (e.g., T as illustrated in fig. 1) is acquired by an ultrasound scanner (e.g., C as illustrated in fig. 1). In this application scenario, the ultrasound scanner is used to acquire the ultrasound image after passing through different media by generating a pulse signal with a pulse generator (e.g., P as illustrated in fig. 1) and causing an excitation piezoelectric transducer (e.g., E as illustrated in fig. 1) to generate ultrasound at a certain frequency. Then, the obtained ultrasonic image is input into a server (e.g., S as illustrated in fig. 1) deployed with a semiconductor package test algorithm, wherein the server is capable of training the first convolutional neural network, the second convolutional neural network and the third convolutional neural network of the semiconductor package test system with the ultrasonic image based on the semiconductor package test algorithm.

After training is completed, in the inference phase, first, an ultrasonic image of the plastic packaged microcircuit (e.g., T as illustrated in fig. 1) is acquired by an ultrasonic scanner (e.g., C as illustrated in fig. 1). Then, the ultrasonic image of the plastic packaged microcircuit is input into a server (e.g., S as illustrated in fig. 1) deployed with a semiconductor packaging test algorithm, wherein the server can process the ultrasonic image of the plastic packaged microcircuit with the semiconductor packaging test algorithm to generate a classification result for indicating whether the packaging effect of the plastic packaged microcircuit meets a preset requirement.

Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary System

FIG. 2 illustrates a block diagram of a semiconductor package test system according to an embodiment of the present application. As shown in fig. 2, a semiconductor package testing system 200 according to an embodiment of the present application includes: a training module 210 and an inference module 220. Wherein, the training module 210 includes: a training image acquisition unit 211 for acquiring an ultrasonic image of the plastic packaged microcircuit acquired by the ultrasonic scanner; a first convolution coding unit 212, configured to pass the ultrasound image through a first convolution neural network using a first convolution kernel to obtain an ultrasound feature map; a region-of-interest extracting unit 213 configured to determine a plurality of regions of interest in the ultrasonic feature map corresponding to a chip lamination, a wire bonding region, and a chip bonding region based on positions of the chip lamination, the wire bonding region, and the chip bonding region in the ultrasonic image; a region-of-interest encoding unit 214, configured to pass the plurality of regions of interest through a second convolutional neural network using a second convolutional kernel to obtain a plurality of region-of-interest feature maps, wherein the second convolutional kernel is smaller than the first convolutional kernel; a second convolution encoding unit 215 for passing the ultrasonic feature map through a third convolution neural network using non-overlapping convolution kernels to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; a compensation unit 216, configured to perform probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map, where the probabilistic information interpretation is performed based on a weighted sum of a natural exponent function value raised to the power of the feature value of each position in the global feature map and a natural exponent function value raised to the power of the feature value of each position in the global feature map; a feature map fusion unit 217, configured to fuse the compensated global feature map and the feature maps of the multiple regions of interest to obtain a classification feature map; a loss calculating unit 218, configured to pass the classification feature map through a classifier to obtain a classification loss function value; a training unit 219, configured to train the first convolutional neural network, the second convolutional neural network, and the third convolutional neural network based on the classification loss function value. The inference module 220 includes: an object data acquisition unit 221 for acquiring an ultrasonic image of the plastic packaged microcircuit acquired by the ultrasonic scanner; a first neural network unit 222, configured to pass the ultrasound image through the first convolution neural network using the first convolution kernel trained by the training module to obtain an ultrasound feature map; a region-of-interest determining unit 223 for determining a plurality of regions of interest in the ultrasonic feature map corresponding to the chip delamination, the wire bonding region, and the chip bonding region, based on positions of the chip delamination, the wire bonding region, and the chip bonding region in the ultrasonic image; a region-of-interest deep encoding unit 224, configured to pass the multiple regions of interest through a second convolutional neural network trained by the training module and using a second convolutional kernel to obtain multiple region-of-interest feature maps, where the second convolutional kernel is smaller than the first convolutional kernel; a second neural network unit 225, configured to pass the ultrasound feature map through a third convolutional neural network using the non-overlapping convolutional kernels trained by the training module to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; a feature information fusion unit 226, configured to fuse the compensated global feature map and the feature maps of the multiple regions of interest to obtain a classification feature map; and a packaging test result generating unit 227, configured to pass the classification feature map through a classifier to obtain a classification result, where the classification result is used to indicate whether the packaging effect of the plastic package microcircuit meets a preset requirement.

Specifically, in the embodiment of the present application, in the training module 210, the training image obtaining unit 211 and the first convolution encoding unit 212 are configured to obtain an ultrasonic image of the plastic microcircuit acquired by the ultrasonic scanner, and pass the ultrasonic image through a first convolution neural network using a first convolution kernel to obtain an ultrasonic characteristic map. As previously mentioned, it should be understood that the evaluation of the packaging quality of the plastic packaged microcircuit can be performed based on the ultrasonic image analysis of the plastic packaged microcircuit, and this is essentially a classification problem, i.e. the evaluation of the packaging quality of the plastic packaged microcircuit is performed based on the extraction of local implicit features in the ultrasonic image, and then the classification is performed to obtain a classification result indicating whether the packaging quality meets the requirements. In particular, in the technical solution of the present application, it is desirable to determine the packaging effect of the plastic packaged microcircuit by using an ultrasonic scanning detection technology. The basic working principle of the ultrasonic scanning detection technology is that a pulse generator is used for generating a pulse signal, a piezoelectric sensor is excited to generate ultrasonic waves with a certain frequency, and the ultrasonic waves are transmitted to a sample through a decoupling medium. During propagation, the ultrasonic waves pass through a continuous homogeneous medium, and reflection phenomena are caused once different media are encountered, so that the defects such as delamination, cracks and the like are discovered. The reflected ultrasonic wave is received by the sensor and is processed correspondingly to automatically generate an ultrasonic image.

That is, specifically, in the technical solution of the present application, first, an ultrasonic image of the plastic packaged microcircuit is acquired by an ultrasonic scanner. It should be understood that, in the standard convolutional neural network, as the coding process goes deep, shallow features of the ultrasonic image of the plastic-packaged microcircuit, such as shapes, textures, lines, bumps, grooves, boundaries and the like, are submerged in the background information or become blurred as the depth goes deep, and the shallow feature information has an important significance in judging the packaging effect of the plastic-packaged microcircuit. Therefore, further, the ultrasonic image is passed through a first convolution neural network of a first convolution kernel to obtain an ultrasonic feature map. In one particular example, the first convolutional neural network includes a plurality of convolutional layers, wherein each of the plurality of convolutional layers is used in forward pass of a layer for convolutional processing, pooling processing, and nonlinear activation processing of input data to output the ultrasonic signature by a last convolutional layer of the plurality of convolutional layers. The number of the coiling layers in the multilayer coiling layer is 4-6.

Specifically, in the embodiment of the present application, in the training module 210, the region-of-interest extracting unit 213 and the region-of-interest encoding unit 214 are configured to determine a plurality of regions of interest in the ultrasonic feature map corresponding to a chip layer, a wire bonding region and a chip bonding region based on positions of the chip layer, the wire bonding region and the chip bonding region in the ultrasonic image, and pass the plurality of regions of interest through a second convolution neural network using a second convolution kernel to obtain a plurality of region-of-interest feature maps, where the second convolution kernel is smaller than the first convolution kernel. It should be appreciated that high emphasis should be placed on focusing on chip layering, wire bonding, and chip bonding areas when performing feature extraction on the ultrasonic image of the plastic encapsulated microcircuit. Therefore, in the technical solution of the present application, a plurality of corresponding regions of interest in the ultrasonic signature are determined further based on the positions of the die delamination, the wire bonding, and the die attach region in the ultrasonic image. Then, the multiple regions of interest are respectively processed in a second convolution neural network by using a second convolution kernel, so as to excavate the deep-level local implicit feature information which is more focused on the chip layering, wire bonding and chip bonding areas in the high-dimensional features of the ultrasonic image of the plastic-packaged microcircuit, and thus obtain multiple region of interest feature maps, wherein the second convolution kernel is smaller than the first convolution kernel.

More specifically, in an aspect of the present application, the region-of-interest encoding unit is further configured to perform, in forward pass of layers, using each layer of the second convolutional neural network, respectively on input data: performing convolution processing on the input data based on the second convolution kernel to obtain a convolution characteristic diagram; pooling the convolution characteristic map to obtain a pooled characteristic map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the second convolutional neural network is the region-of-interest feature map.

Specifically, in the embodiment of the present application, in the training module 210, the second convolution encoding unit 215 and the compensation unit 216 are configured to pass the ultrasonic feature map through a third convolution neural network using non-overlapping convolution kernels to obtain a global feature map having the same size as the plurality of region-of-interest feature maps, and perform probabilistic information interpretation on feature values of respective positions in the global feature map to obtain a compensated global feature map, where the probabilistic information interpretation is performed based on a weighted sum of natural exponent function values raised to the power of the feature values of the respective positions in the global feature map and natural exponent function values raised to the power of the feature values of the respective positions in the global feature map. It should be understood that, in order to align the ultrasound feature map with the region of interest feature map in a scale, in the technical solution of the present application, the ultrasound feature map is processed using a third convolutional neural network having non-overlapping convolutional kernels to obtain a global feature map. Accordingly, in one particular example, the input data is separately subjected to in-layer forward passes using layers of the third convolutional neural network: performing convolution processing on the input data based on a third convolution kernel to obtain a convolution characteristic map, wherein the step size of the movement of the third convolution kernel on the input data is larger than the width size of the third convolution kernel; pooling the convolution characteristic map to obtain a pooled characteristic map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the third convolutional neural network is the global feature map. However, considering that this may cause the global feature map to have a certain information loss, the global feature map is further modified by probabilistic information interpretation to obtain a compensated global feature map.

More specifically, in the embodiment of the present application, the compensation unit includes: and a local information representation subunit, configured to calculate, as the local information representation feature value of each position in the global feature map, a natural exponent function value raised by a feature value of each position in the global feature map. And a global information representation subunit for calculating a weighted sum of natural exponent function values raised by the feature values of the respective positions in the global feature map as a global information representation feature value of the global feature map. And the probabilistic interpretation subunit is used for calculating a logarithmic function value obtained by dividing the local information representation characteristic value of each position in the global characteristic diagram by the sum of the quotient and the one of the global information representation characteristic values of the global characteristic diagram as the probabilistic information interpretation value of each position in the global characteristic diagram so as to obtain the compensated global characteristic diagram. That is, the formula for performing probabilistic information interpretation on the feature values of the positions in the global feature map to obtain the compensated global feature map is as follows:

wherein

Is the feature value of each location of the global feature map,

and an exponential operation representing a feature map, which is a calculation of a natural exponential function value raised to the power of the feature value at each position in the feature map. It should be understood that, by performing cauchy weight probabilistic quantization by performing probabilistic information interpretation on the feature value of each position of the global feature map, it may enable the parameters of the third convolutional neural network to adapt to the information loss caused by the non-overlapping convolutional kernel according to the periodic iteration in the training process of the third convolutional neural network through the back propagation of the gradient, so as to enhance the robustness to the information loss and improve the feature expression capability of the global feature map.

Specifically, in this embodiment, in the training module 210, the feature map fusing unit 217, the loss calculating unit 218, and the training unit 219 are configured to fuse the compensated global feature map and the region-of-interest feature maps to obtain a classification feature map, pass the classification feature map through a classifier to obtain a classification loss function value, and train the first convolutional neural network, the second convolutional neural network, and the third convolutional neural network based on the classification loss function value. That is, in the technical solution of the present application, after the compensated global feature map is obtained, the compensated global feature map is further fused with the plurality of feature maps of interest to obtain a classification feature map for classification. In one specific example, a weighted sum by location of the compensated global feature map and the plurality of region of interest feature maps may be computed to fuse feature information of both. Further, the classification feature map is passed through a classifier to obtain a classification loss function value to train the first convolutional neural network, the second convolutional neural network and the third convolutional neural network.

More specifically, in an embodiment of the present application, the loss calculating unit includes: the classifier processes the classification feature map to generate a classification result according to the following formula, wherein the formula is as follows:

wherein

Representing the projection of the classification feature map as a vector,

to

Is a weight matrix of the fully connected layers of each layer,

to

A bias matrix representing all layers of the fully connected layer; and calculating a cross entropy value between the classification result and a real value as the classification loss function value.

After the training is completed, an inference module is carried out, namely the trained first convolutional neural network, the trained second convolutional neural network and the trained third convolutional neural network are used for the actual analysis and evaluation of the packaging effect of the plastic packaging microcircuit.

Specifically, in the embodiment of the present application, first, an ultrasonic image of the plastic packaged microcircuit acquired by the ultrasonic scanner is acquired. Then, the ultrasonic image is passed through the first convolution neural network using the first convolution kernel trained by the training module to obtain an ultrasonic feature map. Then, based on the positions of the chip delamination, the wire bonding regions and the chip bonding regions in the ultrasonic image, a plurality of regions of interest in the ultrasonic feature map corresponding to the chip delamination, the wire bonding regions and the chip bonding regions are determined. Then, the plurality of regions of interest are respectively passed through a second convolution neural network which is trained by the training module and uses a second convolution kernel to obtain a plurality of region of interest characteristic maps, wherein the second convolution kernel is smaller than the first convolution kernel. Then, the ultrasonic feature map is passed through a third convolutional neural network using the non-overlapping convolutional kernels trained by the training module to obtain a global feature map having the same size as the plurality of region-of-interest feature maps. And then, fusing the compensated global feature map and the plurality of region-of-interest feature maps to obtain a classification feature map. And finally, passing the classification characteristic diagram through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the packaging effect of the plastic package microcircuit meets the preset requirement or not.

In summary, the semiconductor package testing system 200 according to the embodiment of the present application is illustrated, which acquires an ultrasonic image of a plastic package microcircuit through an ultrasonic scanner, further uses a convolutional neural network model to respectively extract shallow implicit feature information of the ultrasonic image and high-dimensional implicit feature distribution information more focused on the chip layering, wire bonding and chip bonding areas, and performs scale matching and modification on feature maps of the shallow implicit feature information and the high-dimensional implicit feature distribution information when features are fused, so that cauchy weight probability transformation is performed by performing probabilistic information interpretation on feature values of each position of a global feature map, so as to enhance robustness to information loss and improve feature expression capability of the global feature map. Furthermore, the packaging effect of the plastic packaged microcircuit can be more accurately evaluated.

As described above, the semiconductor package test system 200 according to the embodiment of the present application may be implemented in various terminal devices, such as a server of a semiconductor package test algorithm, and the like. In one example, the semiconductor package test system 200 according to the embodiment of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the semiconductor package test system 200 may be a software module in an operating system of the terminal device, or may be an application developed for the terminal device; of course, the semiconductor package test system 200 may also be one of many hardware modules of the terminal device.

Alternatively, in another example, the semiconductor package test system 200 and the terminal device may also be separate devices, and the semiconductor package test system 200 may be connected to the terminal device through a wired and/or wireless network and transmit the mutual information in an agreed data format.

Exemplary method

Fig. 3A illustrates a flow chart of a training phase in a testing method of a semiconductor package testing system according to an embodiment of the present application. As shown in fig. 3A, the method for testing a semiconductor package testing system according to an embodiment of the present application includes: a training phase comprising the steps of: s110, acquiring an ultrasonic image of the plastic package microcircuit acquired by an ultrasonic scanner; s120, passing the ultrasonic image through a first convolution neural network using a first convolution kernel to obtain an ultrasonic characteristic map; s130, determining a plurality of interested areas corresponding to the chip layering, the lead bonding area and the chip bonding area in the ultrasonic characteristic diagram based on the positions of the chip layering, the lead bonding area and the chip bonding area in the ultrasonic image; s140, respectively passing the multiple interested areas through a second convolution neural network using a second convolution kernel to obtain multiple interested area feature maps, wherein the second convolution kernel is smaller than the first convolution kernel; s150, passing the ultrasonic characteristic diagram through a third convolution neural network using non-overlapping convolution kernels to obtain a global characteristic diagram with the same size as the plurality of region-of-interest characteristic diagrams; s160, performing probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map, wherein the probabilistic information interpretation is performed based on a weighted sum of a natural exponent function value raised to the power of the feature value of each position in the global feature map and a natural exponent function value raised to the power of the feature value of each position in the global feature map; s170, fusing the compensated global feature map and the plurality of interesting region feature maps to obtain a classification feature map; s180, enabling the classification characteristic graph to pass through a classifier to obtain a classification loss function value; s190, training the first convolutional neural network, the second convolutional neural network and the third convolutional neural network based on the classification loss function value.

FIG. 3B illustrates a flow chart of an inference phase in a method of testing a semiconductor package test system according to an embodiment of the application. Fig. 3B shows a method for testing a semiconductor package test system according to an embodiment of the present application, including: an inference phase comprising the steps of: s210, acquiring an ultrasonic image of the plastic package microcircuit acquired by an ultrasonic scanner; s220, passing the ultrasonic image through the first convolution neural network which is trained by the training module and uses the first convolution kernel to obtain an ultrasonic characteristic diagram; s230, determining a plurality of interested areas corresponding to the chip layering, the lead bonding area and the chip bonding area in the ultrasonic characteristic diagram based on the positions of the chip layering, the lead bonding area and the chip bonding area in the ultrasonic image; s240, enabling the multiple interested areas to pass through a second convolution neural network which is trained by the training module and uses a second convolution kernel respectively so as to obtain multiple interested area feature maps, wherein the second convolution kernel is smaller than the first convolution kernel; s250, passing the ultrasonic feature map through a third convolutional neural network using the non-overlapping convolutional kernel trained by the training module to obtain a global feature map with the same size as the plurality of region-of-interest feature maps; s260, fusing the compensated global feature map and the plurality of interesting area feature maps to obtain a classification feature map; and S270, enabling the classification characteristic diagram to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the packaging effect of the plastic packaging microcircuit meets the preset requirement or not.

Fig. 4 is a schematic diagram illustrating an architecture of a training phase in a testing method of a semiconductor package testing system according to an embodiment of the present application. As shown in fig. 4, in the training phase, first, the obtained ultrasound image (e.g., P1 as illustrated in fig. 4) is passed through a first convolution neural network (e.g., CNN1 as illustrated in fig. 4) using a first convolution kernel to obtain an ultrasound signature (e.g., F1 as illustrated in fig. 4); next, determining a plurality of regions of interest (e.g., P2 as illustrated in fig. 4) in the ultrasonic signature corresponding to the die attach region, the wire bond region, and the die attach region based on the positions of the die attach region, the wire bond region, and the die attach region in the ultrasonic image; then, passing the plurality of regions of interest through a second convolutional neural network (e.g., CNN2 as illustrated in fig. 4) using a second convolutional kernel, respectively, to obtain a plurality of region of interest feature maps (e.g., F2 as illustrated in fig. 4); next, passing the ultrasound signature through a third convolutional neural network (e.g., CNN3 as illustrated in fig. 4) using non-overlapping convolution kernels to obtain a global signature (e.g., F3 as illustrated in fig. 4) having the same size as the plurality of region of interest signatures; then, performing probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map (for example, as illustrated in fig. 4, F4); then, fusing the compensated global feature map and the plurality of region-of-interest feature maps to obtain a classification feature map (e.g., FC as illustrated in fig. 4); then, passing the classification signature through a classifier (e.g., a classifier as illustrated in fig. 4) to obtain a classification loss function value (e.g., a CLV as illustrated in fig. 4); finally, the first convolutional neural network, the second convolutional neural network, and the third convolutional neural network are trained based on the classification loss function values.

Fig. 5 illustrates an architecture diagram of an inference stage in a testing method of a semiconductor package testing system according to an embodiment of the present application. As shown in fig. 5, in the inference phase, firstly, the obtained ultrasound image (e.g., P1 as illustrated in fig. 5) is passed through the first convolution neural network (e.g., CN1 as illustrated in fig. 5) using the first convolution kernel, which is trained by the training module, to obtain an ultrasound feature map (e.g., F1 as illustrated in fig. 5); next, determining a plurality of regions of interest (e.g., P2 as illustrated in fig. 5) in the ultrasonic signature corresponding to the chip delamination, the wire bonding regions, and the chip bonding regions based on the positions of the chip delamination, the wire bonding regions, and the chip bonding regions in the ultrasonic image; then, passing the multiple regions of interest through a second convolutional neural network (for example, CN2 as illustrated in fig. 5) using a second convolutional kernel, which is trained by the training module, respectively, to obtain multiple region of interest feature maps (for example, F2 as illustrated in fig. 5); then, passing the ultrasound signature through a third convolutional neural network (e.g., CN3 as illustrated in fig. 5) using the non-overlapping convolutional kernels completed by training by the training module to obtain a global signature (e.g., F3 as illustrated in fig. 5) having the same size as the plurality of region of interest signatures; then, fusing the compensated global feature map and the plurality of region-of-interest feature maps to obtain a classification feature map (e.g., FC as illustrated in fig. 5); and finally, passing the classification feature map through a classifier (e.g., a classifier as illustrated in fig. 5) to obtain a classification result, wherein the classification result is used for indicating whether the packaging effect of the plastic packaged microcircuit meets a preset requirement.

In summary, the testing method of the semiconductor package testing system based on the embodiment of the present application is clarified, which includes acquiring an ultrasonic image of a plastic package microcircuit by an ultrasonic scanner, further extracting shallow implicit feature information of the ultrasonic image and high-dimensional implicit feature distribution information more focused on the chip layer, the wire bonding and the chip bonding region by using a convolutional neural network model, and performing scale matching and modification on feature maps of the shallow implicit feature information and the high-dimensional implicit feature distribution information when the features are fused, so that cauchy weight probabilistic transformation is performed by performing probabilistic information interpretation on feature values of each position of a global feature map, so as to enhance robustness on information loss and improve feature expression capability of the global feature map. Furthermore, the packaging effect of the plastic packaged microcircuit can be more accurately evaluated.

The basic principles of the present application have been described above with reference to specific embodiments, but it should be noted that advantages, effects, etc. mentioned in the present application are only examples and are not limiting, and the advantages, effects, etc. must not be considered to be possessed by various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.

The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

It should also be noted that in the devices, apparatuses, and methods of the present application, each component or step can be decomposed and/or re-combined. These decompositions and/or recombinations are to be considered as equivalents of the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A semiconductor package test system, comprising: a training module comprising: the training image acquisition unit is used for acquiring an ultrasonic image of the plastic package microcircuit acquired by the ultrasonic scanner; the first convolution coding unit is used for enabling the ultrasonic image to pass through a first convolution neural network using a first convolution kernel so as to obtain an ultrasonic characteristic map; an interesting region extracting unit, configured to determine a plurality of interesting regions in the ultrasonic feature map corresponding to the chip layering, the wire bonding region, and the chip bonding region based on positions of the chip layering, the wire bonding region, and the chip bonding region in the ultrasonic image; a region-of-interest coding unit, configured to pass the plurality of regions of interest through a second convolutional neural network using a second convolutional kernel respectively to obtain a plurality of region-of-interest feature maps, where the second convolutional kernel is smaller than the first convolutional kernel; a second convolution coding unit, for passing the ultrasonic feature map through a third convolution neural network using non-overlapping convolution kernels to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; a compensation unit, configured to perform probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map, where the probabilistic information interpretation is performed based on a weighted sum of a natural exponent function value raised to the power of the feature value of each position in the global feature map and a natural exponent function value raised to the power of the feature value of each position in the global feature map; the feature map fusion unit is used for fusing the compensated global feature map and the plurality of interesting area feature maps to obtain a classification feature map; the loss calculation unit is used for enabling the classification characteristic map to pass through a classifier to obtain a classification loss function value; a training unit to train the first convolutional neural network, the second convolutional neural network, and the third convolutional neural network based on the classification loss function value; and an inference module comprising: the data acquisition unit of the object to be detected is used for acquiring an ultrasonic image of the plastic package microcircuit acquired by the ultrasonic scanner; a first neural network unit, configured to pass the ultrasound image through the first convolution neural network using the first convolution kernel trained by the training module to obtain an ultrasound feature map; an interesting region determining unit, configured to determine a plurality of interesting regions in the ultrasonic feature map corresponding to the chip layering, the wire bonding regions and the chip bonding regions based on positions of the chip layering, the wire bonding regions and the chip bonding regions in the ultrasonic image; a region-of-interest deep coding unit, configured to pass the multiple regions of interest through a second convolutional neural network trained by the training module and using a second convolutional kernel, respectively, to obtain multiple region-of-interest feature maps, where the second convolutional kernel is smaller than the first convolutional kernel; a second neural network unit, configured to pass the ultrasonic feature map through a third convolutional neural network using the non-overlapping convolutional kernels trained by the training module to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; the characteristic information fusion unit is used for fusing the compensated global characteristic map and the characteristic maps of the plurality of interested areas to obtain a classification characteristic map; and the packaging test result generating unit is used for enabling the classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the packaging effect of the plastic packaging microcircuit meets the preset requirement or not.

2. The semiconductor package test system of claim 1, wherein the first convolutional neural network comprises a plurality of convolutional layers, wherein each of the plurality of convolutional layers is used in forward pass of layer for convolutional processing, pooling processing, and nonlinear activation processing of input data to output the ultrasonic signature by a last convolutional layer of the plurality of convolutional layers.

3. The semiconductor package test system of claim 2, wherein the number of convolutional layers in the multi-layer convolutional layer is 4-6.

4. The semiconductor package test system of claim 3, wherein the region of interest encoding unit is further configured to perform, in layer forward pass, using the layers of the second convolutional neural network, respectively on the input data: performing convolution processing on the input data based on the second convolution kernel to obtain a convolution characteristic diagram; pooling the convolution characteristic map to obtain a pooled characteristic map; and carrying out nonlinear activation on the pooling feature map to obtain an activation feature map; wherein the output of the last layer of the second convolutional neural network is the region-of-interest feature map.

5. The semiconductor package test system of claim 4, wherein the second convolutional encoding unit is further configured to perform, on the input data in a layer forward pass using the layers of the third convolutional neural network, respectively: performing convolution processing on the input data based on a third convolution kernel to obtain a convolution characteristic map, wherein the step size of the movement of the third convolution kernel on the input data is larger than the width size of the third convolution kernel; pooling the convolution characteristic map to obtain a pooled characteristic map; and carrying out nonlinear activation on the pooling feature map to obtain an activation feature map; wherein the output of the last layer of the third convolutional neural network is the global feature map.

6. The semiconductor package test system of claim 5, wherein the compensation unit comprises: a local information representation subunit, configured to calculate, as a local information representation feature value of each position in the global feature map, a natural exponent function value raised by a feature value of each position in the global feature map; a global information representation subunit for calculating a weighted sum of natural exponent function values raised by the feature values of the respective positions in the global feature map as a power, as a global information representation feature value of the global feature map; and the probabilistic interpretation subunit is used for calculating a logarithmic function value obtained by dividing the local information representation characteristic value of each position in the global characteristic diagram by the sum of the quotient and the one of the global information representation characteristic values of the global characteristic diagram as the probabilistic information interpretation value of each position in the global characteristic diagram so as to obtain the compensated global characteristic diagram.

7. The semiconductor package test system of claim 6, wherein the loss calculation unit comprises: the classifier processes the classification feature map to generate a classification result according to the following formula, wherein the formula is as follows:

wherein

Representing the projection of the classification feature map as a vector,

to

Is a weight matrix of the fully connected layers of each layer,

to

A bias matrix representing the layers of the fully-connected layer; and calculating the classification result and the true valueThe cross entropy value between them is taken as the classification loss function value.

8. A method of testing a semiconductor package test system, comprising: a training phase comprising: acquiring an ultrasonic image of the plastic package microcircuit acquired by an ultrasonic scanner; passing the ultrasound image through a first convolution neural network using a first convolution kernel to obtain an ultrasound signature; determining a plurality of regions of interest in the ultrasonic signature corresponding to a chip delamination, a wire bonding region, and a chip bonding region based on the positions of the chip delamination, the wire bonding region, and the chip bonding region in the ultrasonic image; respectively passing the multiple regions of interest through a second convolutional neural network using a second convolutional kernel to obtain multiple region of interest feature maps, wherein the second convolutional kernel is smaller than the first convolutional kernel; passing the ultrasound signature through a third convolutional neural network using non-overlapping convolutional kernels to obtain a global signature having the same size as the plurality of region-of-interest signatures; performing probabilistic information interpretation on the feature values of the positions in the global feature map to obtain a compensated global feature map, wherein the probabilistic information interpretation is performed based on a weighted sum of natural exponent function values raised to the power of the feature values of the positions in the global feature map and natural exponent function values raised to the power of the feature values of the positions in the global feature map; fusing the compensated global feature map and the plurality of interested region feature maps to obtain a classification feature map; passing the classification feature map through a classifier to obtain a classification loss function value; training the first, second, and third convolutional neural networks based on the classification loss function values; and an inference phase comprising: acquiring an ultrasonic image of the plastic package microcircuit acquired by an ultrasonic scanner; passing the ultrasonic image through the first convolution neural network using the first convolution kernel trained by the training module to obtain an ultrasonic feature map; determining a plurality of regions of interest in the ultrasonic signature corresponding to a chip delamination, a wire bonding region, and a chip bonding region based on the positions of the chip delamination, the wire bonding region, and the chip bonding region in the ultrasonic image; respectively passing the multiple regions of interest through a second convolutional neural network which is trained by the training module and uses a second convolutional kernel to obtain multiple region of interest characteristic maps, wherein the second convolutional kernel is smaller than the first convolutional kernel; passing the ultrasound feature map through a third convolutional neural network using the non-overlapping convolutional kernels trained by the training module to obtain a global feature map having the same size as the plurality of region-of-interest feature maps; fusing the compensated global feature map and the plurality of interested region feature maps to obtain a classification feature map; and enabling the classification characteristic diagram to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the packaging effect of the plastic packaging microcircuit meets the preset requirement or not.

9. The method for testing a semiconductor package test system of claim 8, wherein passing the ultrasonic image through a first convolution neural network using a first convolution kernel to obtain an ultrasonic signature comprises: the first convolutional neural network comprises a plurality of convolutional layers, wherein each convolutional layer in the plurality of convolutional layers is used for carrying out convolutional processing, pooling processing and nonlinear activation processing on input data in forward transmission of the layers so as to output the ultrasonic characteristic map by the last convolutional layer in the plurality of convolutional layers.

10. The method for testing the semiconductor package test system according to claim 9, wherein the performing probabilistic information interpretation on the feature values of the respective positions in the global feature map to obtain the compensated global feature map comprises: calculating natural exponent function values taking the feature values of all the positions in the global feature map as powers as local information representation feature values of all the positions in the global feature map; calculating a weighted sum of natural exponent function values raised by the feature values of the respective positions in the global feature map as powers as global information representation feature values of the global feature map; and calculating a logarithm function value of a sum of a quotient and a unity of the local information representation characteristic value of each position in the global characteristic diagram, which is divided by the local information representation characteristic value of each position in the global characteristic diagram, as a probabilistic information interpretation value of each position in the global characteristic diagram to obtain the compensated global characteristic diagram.

Images