A wafer-level gas sensor chip detection method based on infrared thermal analysis

Abstract

A wafer-level gas sensor chip detection method based on infrared thermal analysis, which relates to the field of sensor chip detection; the detected gas is used to oxidize or reduce the metal oxide semiconductor surface, causing the sensor resistance value to change, and the output circuit converts the sensor resistance into a voltage output to achieve gas concentration detection and type identification; the reaction between the detected gas and the gas-sensitive material usually needs to be carried out at 200-400°C, and the semiconductor gas sensor needs a heating electrode to provide heat for the sensor. The micro-hot plate gas sensor is based on the MEMS process and is usually manufactured based on wafer-level chips. There are tens of thousands of micro-hot plate chips on the surface of each wafer, and the micro-hot plate chip has a heating electrode. By energizing the micro-hot plate chip, the temperature distribution on the surface of the micro-hot plate can be obtained. The surface of the chip with a fault will appear too cold and hot spots; combined with image acquisition, recognition, and processing technology, the faulty chip is determined, and then the detection of the wafer-level sensor chip is achieved.

Description

Technical Field

[0001] The present invention relates to the field of sensor chip detection, and in particular to a wafer-level gas sensor chip detection method based on infrared thermal analysis. Background Art

[0002] Gas sensors play a vital role in modern life, and the demand for gas sensors is increasing day by day. Traditional gas detection relies on chromatography and spectroscopy. The instruments used in these methods are large in size, expensive, and complex to operate, which limits the application of on-site detection. Micro-nano gas sensors have the advantages of high sensitivity, low cost, and good portability. Micro-nano gas sensors are widely used in hazardous chemical detection, air quality monitoring, human health monitoring, chemical production processes, and agricultural intelligent systems. Among micro-nano gas sensors, gas sensor chips based on MEMS technology are widely used in the field of gas detection due to their small size and low power consumption. For gas sensor detection chips, the existing method uses wire bonding test and microscopic observation to screen chip quality and remove unqualified chips. The existing method is inefficient, the process is cumbersome, and it is easy to cause chip contamination when transferred to a microscope for observation.

[0003] The surface of the micro-hotplate gas sensor is coated with sensitive materials. The detected gas is used to oxidize or reduce the metal oxide semiconductor surface, causing the sensor resistance value to change. The output circuit converts the sensor resistance into a voltage output to achieve gas concentration detection and type identification. The reaction between the detected gas and the gas-sensitive material usually needs to be carried out at 200-400°C. Therefore, the semiconductor gas sensor requires a heating electrode to provide heat for the sensor. The micro-hotplate gas sensor is based on MEMS technology and is usually manufactured based on wafer-level chips. There are tens of thousands of micro-hotplate chips on the surface of each wafer. The micro-hotplate chip has a heating electrode. By powering on the micro-hotplate chip, the temperature distribution on the surface of the micro-hotplate can be obtained. The surface of a faulty chip will be overcooled or overheated. Combined with image acquisition, recognition, and processing technology, the faulty chip can be identified, thereby realizing the detection of wafer-level sensor chips. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a wafer-level gas sensor chip detection method based on infrared thermal analysis;

[0005] The technical solution adopted by the present invention is:

[0006] A wafer-level gas sensor chip detection method based on infrared thermal analysis comprises the following steps:

[0007] S1: The sensor chip wafer to be tested is fixed on the chuck, and a motor is connected under the chuck to control the rotation of the chuck; K sensor chips are randomly selected and powered by the ejector technology;

[0008] S2: Use an infrared camera to take photos of K sensor chips, transfer them to the host computer via USB, obtain K clear images of the surface temperature distribution of the sensor chips, and set the over-temperature value D. high ℃ and low temperature value D low ℃, in the temperature distribution image of the sensor chip surface, select the temperature at D high ℃ to D low The temperature distribution image of the sensor chip surface in the range of ℃ takes the average temperature D℃;

[0009] S3: Set the temperature floating range D according to the average temperature of the sensor surface obtained by S2 change ℃, the temperature range of the sensor's normal operation is (D-D change ℃,D+D change ℃), label the sensor chip surface temperature distribution image obtained by S2, divide it into normal sensor chips and abnormal sensor chips, and build a sensor chip thermal image dataset;

[0010] S4: Establish a sensor image recognition convolutional neural network, which includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The sensor chip thermal image dataset constructed in S3 is divided into a training dataset and a test dataset.

[0011] S5: Set the accuracy threshold as M, and use the test data set sensor image recognition convolutional neural network accuracy to test. When the test accuracy is greater than or equal to the accuracy threshold, encapsulate the sensor image recognition convolutional neural network for subsequent calls; when the test accuracy is lower than the accuracy threshold, reduce the learning rate of the sensor image recognition convolutional neural network, add a Dropout layer before the fully connected layer, and add a Bath Normalization layer after the convolution layer to improve the accuracy of the sensor image recognition convolutional neural network until the accuracy threshold is met;

[0012] S6: Divide the wafer surface into equal parts and power the sensor chip through ejector pin technology;

[0013] The wafer surface is equally divided into n regions, where n≥2, n is a positive integer, and the central angle of each region is 360° / n. An arbitrary dividing line is selected as the starting reference line for wafer rotation; the sensor chip on the starting dividing line is powered by ejector technology;

[0014] S7: Use an infrared camera to obtain the temperature distribution image of the sensor chip surface, call the packaged sensor image recognition convolutional neural network for recognition, and determine whether there is a faulty chip;

[0015] Use an infrared camera to take N photos of the adjacent areas after segmentation, and transfer them to the host computer via USB to obtain a clear image of the sensor chip surface temperature distribution on the segmentation line; call the sensor image recognition convolutional neural network encapsulated in S5, input the obtained sensor chip surface temperature distribution image, and identify whether there is a faulty chip;

[0016] S8: The motor drives the chuck to rotate 360° / n in the same direction, and S6-S8 are repeated. After rotating n-1 times, the fault detection of the wafer sensor chip is completed.

[0017] Beneficial technical effects

[0018] 1. The wafer-level gas sensor chip detection method based on infrared thermal analysis proposed in the present invention uses an infrared camera to obtain the temperature distribution on the wafer surface. The surface of the chip with a fault will have overcooling and overheating points. Combined with image acquisition, recognition, and processing technology, before the gas sensor is packaged, the available gas sensor chips can be directly screened on the wafer-level surface to eliminate abnormal gas sensor chips;

[0019] 2. The wafer-level gas sensor chip detection method based on infrared thermal analysis proposed in the present invention realizes the rapid acquisition of thermal images of wafer-level sensor chips by dividing the wafer into equal parts and driving the wafer to rotate with a motor. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A diagram of an infrared thermal image acquisition device for a wafer-level sensor chip provided in an embodiment of the present invention;

[0021] Figure 2 A flow chart of a wafer-level gas sensor chip detection method based on infrared thermal analysis provided by an embodiment of the present invention;

[0022] Figure 3 A top view of a sensor chip wafer provided by an embodiment of the present invention;

[0023] Figure 4 A flow chart of sensor chip wafer rotation identification provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments;

[0025] This embodiment adopts a wafer-level gas sensor chip detection system based on infrared thermal analysis, such as Figure 1 As shown, it includes a host computer, an infrared camera, a chuck, a motor, and a power module.

[0026] The infrared camera is used to photograph the sensor chip to obtain a surface temperature distribution image.

[0027] The host computer is used to display and process the temperature distribution image of the sensor chip surface taken by the infrared camera.

[0028] The chuck is used to fix the sensor chip wafer to be tested.

[0029] The motor is used to drive the chuck to rotate.

[0030] The power supply module is used for powering an infrared camera, a sensor chip wafer to be tested, and a motor.

[0031] A wafer-level gas sensor chip detection method based on infrared thermal analysis, using the wafer-level gas sensor chip detection system based on infrared thermal analysis, such as Figure 2 As shown, the following steps are included:

[0032] S1: Fix the wafer, rotate it and randomly power the sensor chip. Fix the sensor chip wafer to be tested on the chuck. Connect a motor under the chuck to control the chuck to rotate clockwise or counterclockwise. Randomly select K sensor chips and power them through the ejector technology.

[0033] S2: Take a temperature distribution image of the sensor chip surface, and calculate the temperature average value of the temperature distribution image of the sensor chip surface; as shown in the figure, use an infrared camera to take pictures of 20 sensor chips for the temperature distribution of the sensor chip surface, transfer them to the computer via USB, obtain K clear temperature distribution images of the sensor chip surface, set the high temperature value to 340°C and the low temperature value to 290°C, and select the temperature distribution images of the sensor chip surface with a temperature range of 340°C to 290°C to take the temperature average value of 320°C;

[0034] Table 1 Sensor chip surface temperature

[0035] Sensor chip number 1 2 3 4 5 Temperature value 286 296 302 301 316 Sensor chip number 6 7 8 9 10 Temperature value 304 312 300 302 301 Sensor chip number 11 12 13 14 15 Temperature value 280 275 291 297 299 Sensor chip number 16 17 18 19 20 Temperature value 301 307 299 311 313

[0036] S3: Construct a sensor chip thermal image dataset. According to the average temperature of the sensor surface obtained in S2, set the temperature floating range to 20°C. The normal working temperature range of the sensor is (300°C, 340°C). Label the sensor chip surface temperature distribution image obtained in S2, as shown in Table 2, and divide it into normal sensor chips and abnormal sensor chips, and construct a sensor chip thermal image dataset.

[0037] Table 2 Sensor chip temperature label processing

[0038] Sensor chip temperature 286 296 302 301 316 Label Abnormal chip Abnormal chip Normal chip Normal chip Normal chip Sensor chip temperature 304 312 300 302 301 Label Normal chip Normal chip Normal chip Normal chip Normal chip Sensor chip temperature 280 275 291 297 299 Label Abnormal chip Abnormal chip Abnormal chip Abnormal chip Abnormal chip Sensor chip temperature 301 307 299 311 313 Label Normal chip Normal chip Abnormal chip Normal chip Normal chip

[0039] S4: Establish a sensor image recognition convolutional neural network. In this embodiment, the VGGNet16 convolutional neural network is used; the sensor image recognition convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; in this embodiment, the VGGNet16 convolutional neural network includes 5 convolutional layers and 3 fully connected layers; the input layer is responsible for loading images, and the images come from the sensor chip thermal image dataset established in S3, and the generated output vector is used as the input of the convolutional layer; the convolutional layer extracts features from the thermal image of the sensor chip through the convolution kernel, and the convolution kernel size is 3×3 and the step size is 1; VGGNet16 convolutional neural network The Relu function is used as the activation function in the five convolutional layers included in the network; a pooling layer is set in the middle of the five consecutive convolutional layers to reduce the dimension of the thermal image of the sensor chip; after the five convolutional layers, three fully connected layers are set, and the fully connected layers are responsible for assembling the local features of the thermal image of the sensor chip extracted by the convolutional layer into a complete image through the weight matrix; after the fully connected layer, an output layer is set, and Softmax is used to classify the thermal image of the sensor chip to determine whether the sensor chip is an abnormal sensor chip; the sensor chip thermal image dataset constructed by S3 is divided into a training dataset and a test dataset;

[0040] S5: Set the accuracy threshold of the sensor image recognition convolutional neural network to 95%, use the test data set to test the accuracy of the VGGNet16 convolutional neural network, and when the test accuracy is greater than or equal to 95%, encapsulate the sensor image recognition network for subsequent calls; when the test accuracy is lower than the accuracy threshold, improve the accuracy of the sensor image recognition convolutional neural network by reducing the learning rate of the VGGNet16 convolutional neural network, adding a Dropout layer before the fully connected layer, and adding a BathNormalization layer after the convolution until the accuracy threshold is met;

[0041] S6: Divide the wafer surface into equal parts and power the sensor chip through ejector pin technology;

[0042] The wafer surface is divided into 8 equal areas, such as Figure 3 As shown, the central angle of each area is 45°; Figure 4 As shown, any dividing line is selected as the starting reference line for wafer rotation; 20 sensor chips on the starting dividing line are selected for power supply through ejector pin technology;

[0043] S7: Use an infrared camera to obtain the temperature distribution image of the sensor chip surface, call the packaged sensor image recognition convolutional neural network for recognition, and determine whether there is a faulty chip;

[0044] Use an infrared camera to take pictures of the sensor chips between the segmented adjacent areas, transfer them to a computer via USB, and obtain surface temperature distribution images of 20 sensor chips in the adjacent areas; call the sensor image recognition convolutional neural network encapsulated in S5, input the obtained surface temperature distribution image of the sensor chip, and identify whether there is a faulty chip;

[0045] S8: The motor drives the chuck to rotate 45° in the same direction, and S6-S8 are repeated. After 7 rotations, the fault detection of the wafer sensor chip is completed.

Claims

1. A wafer-level gas sensor chip detection method based on infrared thermal analysis, characterized in that: The following steps are involved: S1: Fix the sensor chip wafer to be tested on the chuck, connect a motor under the chuck to control the rotation of the chuck; randomly select K sensor chips for power supply; S2: Use an infrared camera to take photos of K sensor chips, transmit them to the host computer, obtain K clear images of the sensor chip surface temperature distribution, and set the over-temperature value D. high ℃ and low temperature value D low ℃, in the temperature distribution image of the sensor chip surface, select the temperature at D high ℃ to D low The temperature distribution image of the sensor chip surface in the range of ℃ takes the average temperature D℃; S3: Set the temperature floating range D according to the average temperature of the sensor surface obtained by S2 change ℃, the temperature range of the sensor's normal operation is (D-D change ℃,D+D change ℃), label the sensor chip surface temperature distribution image obtained by S2, divide it into normal sensor chips and abnormal sensor chips, and build a sensor chip thermal image dataset; S4: Establish a sensor image recognition convolutional neural network and divide the sensor chip thermal image dataset constructed in S3 into a training dataset and a test dataset; S5: Use the test data set to test the accuracy of the sensor image recognition convolutional neural network; S6: dividing the wafer surface into equal parts and supplying power to the sensor chip; The specific process of S6 is: dividing the wafer surface into n equal areas, n≥2, n is a positive integer, the center angle of each area is 360° / n, and any dividing line is taken as the starting reference line for wafer rotation; and supplying power to the sensor chip on the starting dividing line; S7: Use an infrared camera to obtain the temperature distribution image of the sensor chip surface, call the packaged sensor image recognition convolutional neural network for recognition, and determine whether there is a faulty chip; S8: The motor drives the chuck to rotate 360° / n in the same direction, and S6-S8 are repeated. After rotating n-1 times, the fault detection of the wafer sensor chip is completed.

2. The wafer-level gas sensor chip detection method based on infrared thermal analysis according to claim 1, characterized in that: The sensor chip is powered via ejector pin technology.

3. The wafer-level gas sensor chip detection method based on infrared thermal analysis according to claim 1, characterized in that: After the infrared camera takes pictures of the K sensor chips, the pictures are transmitted to the host computer via USB.

4. The wafer-level gas sensor chip detection method based on infrared thermal analysis according to claim 1, characterized in that: The sensor image recognition convolutional neural network includes input layer, convolution layer, pooling layer, fully connected layer and output layer.

5. The wafer-level gas sensor chip detection method based on infrared thermal analysis according to claim 1, characterized in that: The specific process of S5 is: when the test accuracy is greater than or equal to the accuracy threshold, the sensor image recognition convolutional neural network is encapsulated for subsequent calls; the accuracy threshold is set to M, and when the test accuracy is lower than the accuracy threshold, the learning rate of the sensor image recognition convolutional neural network is reduced, a Dropout layer is added before the fully connected layer, and a Bath Normalization layer is added after the convolution layer to improve the accuracy of the sensor image recognition convolutional neural network until the accuracy threshold is met.

6. The wafer-level gas sensor chip detection method based on infrared thermal analysis according to claim 1, characterized in that: The specific process of S7 is: use an infrared camera to shoot N times between adjacent areas after segmentation, transfer to a computer via USB, and obtain a clear sensor chip surface temperature distribution image on the segmentation line; call the sensor image recognition convolutional neural network encapsulated in S5, input the obtained sensor chip surface temperature distribution image, and identify whether there is a faulty chip.

Images