A temperature sensor mass production rapid test system

Abstract

The present application relates to a kind of temperature sensor mass production quick test system.The system uses deep learning technology, obtains the output value of the sensor to be measured at a certain temperature, quickly feeds back a group of calibration data, and directly determines whether the sensor to be measured has a large process deviation using the output value, enters the wafer screening process.In addition, the present application uses the characteristics of hardware circuit parallel operation to test multiple groups of sensors to be measured simultaneously, achieving the effect of rapid testing.The present application reduces the test process by testing a small number of specific temperatures, and can increase the number of sensors tested in a fixed time, ultimately reducing the cost during the temperature sensor mass production test.

Description

Technical Field

[0001] This invention relates to a rapid testing system for mass production of temperature sensors. Background Technology

[0002] Most integrated CMOS temperature sensors are manufactured using the temperature effect of the PN junction in semiconductor materials. Due to variations in manufacturing processes and factors such as packaging stress, the consistency of sensors produced in batches can vary. To ensure the accuracy of the switching temperature, integrated CMOS temperature sensors have a pre-built one-time programmable memory (OTP) calibration function, which is uniformly calibrated during functional testing after circuit packaging. This necessitates testing the sensor circuitry after mass production. However, current testing methods using automated test equipment (ATE) are not suitable for testing temperature sensors with different temperature measurement effects. Furthermore, ATE can only measure one sensor at a time, has long testing cycles, and is expensive—prominent drawbacks of traditional testing methods. Summary of the Invention

[0003] The purpose of this invention is to overcome the aforementioned shortcomings and provide a rapid mass production testing system for temperature sensors. This system employs the SMEClass algorithm to extract features from sensor output data, perform data matching, and adapt corresponding calibration data, thereby reducing the testing cycle. It can also simultaneously measure multiple sensors, improving testing efficiency. Furthermore, the cost of all modules in this system is lower than that of traditional ATE testing equipment, further enhancing the feasibility of mass production sensor testing and reducing the cost of testing equipment. This reduces the overall cost of mass-produced chips.

[0004] To achieve the above objectives, the technical solution of the present invention is: a rapid mass production testing system for temperature sensors, comprising a core testing circuit, a test bias circuit, a temperature chamber, a host computer, and a control core board; during operation, the test bias circuit provides a voltage bias for each temperature sensor in the core testing circuit; the core testing circuit is placed in the temperature chamber; the control core board triggers sampling at the output port of each temperature sensor in the core testing circuit, and performs push-pull output on each port of the core testing circuit during calibration; the host computer exchanges data with the control core board via serial communication.

[0005] In one embodiment of the present invention, the host computer is a computer running the SMEClass algorithm; the control core board is a Zynq system board of model XILINX XCKU3P-1FFVA676E; and the temperature chamber is a ZCTB-40DL deep well constant temperature bath.

[0006] In one embodiment of the present invention, the output ports of the temperature sensor include the BS, F1, and ADC_CHOP output ports of the temperature sensor; the ports of the test core circuit include the CLK_REF, LOAD_CYCLE, RESTN_IN, ENSET, D_IN, and RESTN_SET ports of the test core circuit.

[0007] In one embodiment of the present invention, the system's workflow is as follows: After the entire system is powered on, the initialization of each module is performed. Then, communication is established between the core test circuit, the bias test circuit, the host computer, the temperature chamber, and the control core board, and the communication is verified to be normal. After the communication between each module is normal, the system user sets the temperature chamber test temperature to 35°C and the alarm range for abnormal output of the core test circuit on the host computer. Then, the entire system starts working, and the system user can observe the set temperature of the temperature chamber and the output data of each temperature sensor through the host computer. During this process, the system sends the output data of each temperature sensor to the SMEClass algorithm for training. The SMEClass algorithm extracts the data features of the output data of each temperature sensor through the host computer and performs data matching. The matching object comes from the standard temperature-calibration database, which is a typical database formed by further filtering of the knowledge database and the historical database. After the data is matched, the matched calibration data is burned into the chip OTP under the constraints of pattern recognition and general rules. Then, the entire testing process ends.

[0008] In one embodiment of the present invention, in the SMEClass algorithm, S represents the data stream, St represents a fixed-size data block (i.e., the t-th data block in S), d represents the size of the data block used to train the base classifier, EC represents the ensemble classifier, K represents the capacity of the ensemble classifier, num represents the number of current classifiers in EC, Ei represents the base classifier trained on the i-th data block, L represents the labeled dataset in the data block, U represents the unlabeled dataset in the data block, and U′ represents the labeled dataset. The SMEClass algorithm divides the dataset S into data blocks S = {S1, S2, S3, ..., St, St + 1, ...} of the same size. When the number of samples in the dataset S is d, the following operation is performed:

[0009] 1) First, separate the data in the data block into labeled dataset L and unlabeled dataset U, and then build a decision tree classifier Enum based on labeled dataset L;

[0010] 2) The ensemble classifier EC and the decision tree classifier Enum vote on the data in U. If more than 50% of the classifiers are labeled as c, then the data is labeled as c and added to the labeled dataset U′.

[0011] 3) Update the decision tree classifier Enum according to L ∪ U′;

[0012] 4) Repeat steps 2) and 3) until no more labeled data can be added. If num < K, execute step 5); otherwise, execute steps 6) - 10);

[0013] 5) Add Enum to EC with the classification accuracy on the original labeled dataset L as the weight;

[0014] 6) Build a Naïve Bayes classifier E′ using the labeled data on the K nearest data blocks in the window;

[0015] 7) Classify each instance in L ∪ U′ using EC and E′. If misclassified, add this instance to the error buffer;

[0016] 8) Detect concept drift using the error rate et of EC on the current data block;

[0017] 9) If there is concept drift and num = K, read the data in the buffer into S, create a new classifier Enew, use its accuracy on S as the weight, and update the weights of the classifiers in EC using S;

[0018] 10) If the weight of Enew is greater than the weight of the smallest classifier Ek in EC, discard the Ek classifier and add Enew to form a new ensemble classifier.

[0019] Compared with the prior art, the present invention has the following beneficial effects: According to the application background designed by the present invention, when conducting mass production tests on temperature sensors, due to traditional test schemes, factors such as a small number of chips measured per single measurement, a long test cycle, and expensive test equipment will increase the overall cost of sensor mass production tests. The rapid mass production test system for temperature sensors proposed by the present invention adopts the SMEClass algorithm. Through this algorithm, feature extraction and data matching are performed on the sensor output data, and corresponding calibration data is adapted. By the above process, the process of constantly changing the temperature of the temperature chamber for testing and then determining the calibration data in the traditional method is omitted, reducing the test cycle. This system can measure multiple sensors simultaneously and improve the test efficiency. In addition, the cost of all modules of this system is lower than that of traditional ATE test equipment, further enhancing the feasibility of mass production sensor testing and reducing the test cost. It reduces the consumption of the test process. Therefore, the present invention has a huge application space in the calibration of temperature sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a structural block diagram of a rapid mass production test system for temperature sensors.

[0021] Figure 2 This is a schematic diagram for testing the core circuit.

[0022] Figure 3 This is a schematic diagram of the bias circuit for testing.

[0023] Figure 4 This is a flowchart of the core algorithm's workflow.

[0024] Figure 5 This is a flowchart of the test system workflow.

[0025] In the diagram, 1. Test core circuit; 101. Temperature sensor array under test; 101-1. Chip under test 1; 101-2. Chip under test 2; 101-3. Chip under test 3; ...; 101-10. Chip under test 10; 102. Input / output pin array; 103. Digital / analog isolation resistor array; 104. Four-corner copper pillars; 2. Test bias circuit; 201. Input / output pin array; 202. Bias chip functional module; 203. Power supply module; 204. Four-corner copper pillars; 205. Digital / analog isolation resistor array; 3. Temperature chamber; 4. Control core board; 5. Host computer. Detailed Implementation

[0026] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention proposes a rapid mass production testing system for temperature sensors. Employing the SMEClass algorithm, it extracts features from sensor output data, performs data matching, and adapts corresponding calibration data, reducing the testing cycle. It can also simultaneously measure multiple sensors, improving testing efficiency. Furthermore, the cost of all modules in this system is lower than that of traditional ATE testing equipment, further enhancing the feasibility of mass production sensor testing and reducing the cost of testing equipment. This reduces the overall cost of mass-produced chips.

[0028] The system of this invention mainly consists of five main parts, including a test core circuit, a test bias circuit, a temperature chamber, a host computer, and a control core board.

[0029] The host computer is a computer running the SMEClass algorithm;

[0030] The control core board is a Zynq system board of model XILINX XCKU3P-1FFVA676E;

[0031] The temperature chamber is a ZCTB-40DL deep-well constant temperature bath;

[0032] The core test circuit is integrated on a rectangular PCB board measuring 22.7cm x 19.4cm. Its circuit diagram is attached. Figure 2 As shown;

[0033] The test bias circuit is integrated on a rectangular PCB board with dimensions of 12.1cm x 6.4cm. Its circuit diagram is attached. Figure 3 As shown;

[0034] As attached Figure 1 As shown, during operation, the test bias circuit 2 provides voltage bias for each temperature sensor in the test core circuit 1; the test core circuit is placed in the temperature chamber 3; the control core board 4 triggers sampling at the BS, F1, and ADC_CHOP output ports of each temperature sensor chip in the test core circuit (in the figure, 101-1 is chip 1 under test, 101-2 is chip 2 under test, 101-3 is chip 3 under test, ..., 101-10 is chip 10 under test) and pushes-pull outputs at the CLK_REF, LOAD_CYCLE, RESTN_IN, ENSET, D_IN, and RESTN_SET ports of the test core circuit during calibration; the host computer 5 exchanges data with the control core board through serial communication.

[0035] like Figure 2 The diagram shows the circuit schematic of the core test circuit. 101 is the array of temperature sensors under test, 102 is the input / output pin array, 103 is the digital / analog isolation resistor array, and 104 is the four-corner copper pillar. The core test circuit includes the temperature sensors under test: TEMPSENSOR1, TEMPSENSOR2, ​​TEMPSENSOR3, TEMPSENSOR4, TEMPSENSOR5, TEMPSENSOR6, TEMPSENSOR7, TEMPSENSOR8, TEMPSENSOR9, TEMPSENSOR10; and 2-pin headers: CN1, CN2, CN3, CN4, CN5, CN6, CN7, CN8, CN9, CN10, CN11, CN12, CN13, CN1 4.0Ω resistors: R1, R2, R3, R4, R5, R6, R7, R8, R9, R10, R11, R12, R13, R14, R15, R16, R17, R18, R19, R20, R21, R22, R23, R24; 8-pin connectors: J2, J3, J4, J5, J6, J7, J8, J9, J10, J11; 8-pin headers: U1, U2, U3, U4, U5, U6, U7, U8, U9, U10, U11; M3 copper posts: U12, U13, U14, U15; 6-pin headers: J12; 8-pin horizontal connector: H2.

[0036] like Figure 3The circuit diagram shown is for testing the bias circuit. In the diagram, 201 is the input / output pin array, 202 is the bias chip functional module, 203 is the power supply module, 204 is the four-corner copper pillar, and 205 is the digital-analog isolation resistor array. The test bias circuit includes: LDO chip LT3042: U1, U2, U3, U4; toggle switch: SW3; 10kΩ resistor: R13; 0Ω resistors: R4, R5, R7, R8, R9, R10, R11, R12, R14, R15, R16, R17, R18; 2MΩ ​​resistors: R1, R2, R3, R6; 2-pin connectors: P1, P2, P3, P4; 10-pin vertical connector: J12; 6-pin header: J1, J2, J3, J4, J5, J6, J7, J8, J9, J10, J11; 8-pin horizontal connector: H1; 8-pin vertical connector: U5; power supply socket: CN16. ; 2-pin header: CN5, CN6, CN7, CN8, CN9, CN10, CN11, CN12, CN13, CN14, CN17, CN18, CN19, CN20; 100uF capacitor: C34; 104 capacitor: C23; 0.47uF capacitor: C3, C4, C9, C12; 4.7uF capacitor: C1, C2, C5, C6, C7, C8, C10, C11; M3 copper pillar: U6, U7, U8, U9; LED light: U12.

[0037] The SMEClass algorithm running on the host computer can balance concept drift and noise effects while ensuring classification accuracy even when there are few labeled samples.

[0038] The SMEClass algorithm classification process is shown in the attached figure. Figure 4 As shown in the diagram. Here, S represents the data stream, St represents a fixed-size data block (the t-th data block in S), d represents the size of the data block used to train the base classifier, EC represents the ensemble classifier, K represents the capacity of the ensemble classifier, num represents the number of classifiers in EC, Ei represents the base classifier trained on the i-th data block, L represents the labeled dataset in the data block, U represents the unlabeled dataset in the data block, and U′ represents the labeled dataset.

[0039] Divide the dataset S into data blocks of the same size S = {S1, S2, S3, ..., St, St + 1, ...}.

[0040] When the number of samples in data stream S is d, perform the following operation:

[0041] 1) First, separate the data in the data block into a labeled dataset L and an unlabeled dataset U, and then build a decision tree classifier Enum based on the labeled dataset L;

[0042] 2) Vote on the data in U using this integrated classifier EC and the just-trained classifier Enum together. If more than 50% of the classifiers label it as c, then add this data with label c to the labeled dataset U′ after marking it;

[0043] 3) Update the classifier Enum according to L∪U′;

[0044] 4) Repeat steps 2) and 3) until no more data can be labeled. If num < K, execute step 5), otherwise execute steps 6)-10);

[0045] 5) Add Enum to EC with the classification accuracy rate on the original labeled dataset L as the weight;

[0046] 6) Build a NaïveBayes classifier E′ using the labeled data on the K nearest data blocks in the window;

[0047] 7) Classify each instance in L∪U′ using EC and E′. If it is misclassified, add this instance to the error buffer;

[0048] 8) Detect concept drift using the error rate et of EC on the current data block;

[0049] 9) If there is concept drift and num = K, then read the data in the buffer into S, create a new classifier Enew, use its accuracy rate on S as the weight, and at the same time update the weights of the classifiers in EC using S;

[0050] 10) If the weight of Enew is greater than the weight of the smallest classifier Ek in EC, then discard the Ek classifier and add Enew to form a new integrated classifier.

[0051] The working process of the system of the present invention is as attached Figure 5As shown, after the entire system is powered on, the initialization of each module is required. Then, communication must be established between the core test circuit, the bias test circuit, the host computer, the temperature chamber, and the control core board, and the communication must be verified to be normal. After the communication between each module is normal, the system user starts to set the temperature of the temperature chamber to 35℃ and the alarm range for abnormal output of the core test circuit on the host computer. Then the entire system starts to work, and the user can observe the set temperature of the temperature chamber and the output data of each temperature sensor through the host computer. During this process, the system sends the output data of each temperature sensor to the SMEClass algorithm for training. The SMEClass algorithm mainly extracts the data features of the output data of each temperature sensor through the host computer and performs data matching. The matching object comes from the standard temperature-calibration database, which is a typical database formed by further filtering of the knowledge database and the historical database.

[0052] Once the data is matched, the matched calibration data can be programmed into the chip's OTP (Optical Point of Application) through pattern recognition and general rule constraints. The entire testing process then concludes.

[0053] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A temperature sensor production volume rapid test system, characterized by, It comprises a test core circuit, a test bias circuit, a temperature box, an upper computer and a control core board; during operation, the test bias circuit provides voltage bias for each temperature sensor in the test core circuit; the test core circuit is placed in the temperature box; the control core board triggers and samples the output port of each temperature sensor in the test core circuit and performs push-pull output on each port of the test core circuit during calibration; the upper computer exchanges data with the control core board through serial communication; The upper computer is a computer running the SMEClass algorithm; the system sends the output data of each temperature sensor into the SMEClass algorithm for training; the SMEClass algorithm extracts the data features of the output data of each temperature sensor through the upper computer and performs data matching, with the matching object coming from the standard temperature-calibration database; In the SMEClass algorithm, S represents a data stream, St represents a data block with a fixed size, i.e. the tth data block in S, d represents the data block size for training a basic classifier, EC represents an integrated classifier, K represents the capacity of the integrated classifier, num represents the number of the current classifier in EC, Ei represents a basic classifier trained on the ith data block, L represents a labeled data set in the data block, U represents an unlabeled data set in the data block, and U' represents a labeled data set; the SMEClass algorithm divides the data set S into data blocks with the same size S ={S1,S2,S3,……,St,St + 1, ……}; when the sample number of the data set S is d, the following operations are performed: 1) first separate the data in the data block into a labeled data set L and an unlabeled data set U, and then establish a decision tree classifier Enum according to the labeled data set L; 2) vote for the data in U according to the integrated classifier EC and the decision tree classifier Enum, and if more than 50% of the classifiers are labeled as c, the data is labeled as c and added to the labeled data set U'; 3) update the decision tree classifier Enum according to L∪U'; 4) repeat steps 2) and 3) until no data can be labeled, if num < K, perform step 5), otherwise perform steps 6)-10); 5) add Enum to EC with the classification accuracy on the original labeled data set L as the weight; 6) establish a NaïveBayes classifier E' using the labeled data on the last K data blocks in the window; 7) classify each instance in L∪U' using EC and E', if it is misclassified, add the instance to the error buffer; 8) use the error rate et of EC on the current data block to detect concept drift; 9) if there is concept drift and num = K, read the data in the buffer into S, create a new classifier Enew, use the accuracy of Enew on S as the weight, and update the weights of the classifiers in EC using S; 10) if the weight of Enew is greater than the weight of the smallest classifier Ek in EC, discard Ek and add Enew to the new integrated classifier.

2. The temperature sensor mass production quick test system according to claim 1, wherein, The control core board is a zynq system board of model XILINX XCKU3P-1FFVA676E; and the temperature box is a ZCTB-40DL deep-well constant-temperature tank.

3. The temperature sensor mass production quick test system according to claim 1, wherein, The output ports of the temperature sensors include BS, F1 and ADC_CHOP output ports of the temperature sensors; and the ports of the test core circuit include CLK_REF, LOAD_CYCLE, RESTN_IN, ENSET, D_IN and RESTN_SET ports of the test core circuit.

4. The temperature sensor mass production quick test system according to claim 1, wherein, The working process of the system is as follows: after the whole system is powered on, the initialization of each module of the system is performed, then the communication between the test core circuit, the test bias circuit, the upper computer, the temperature box and the control core board is established, and whether the communication is normal is verified; after the communication of each module is normal, the system user starts to set the test temperature of the temperature box to 35 DEG C and the alarm range of the abnormal output of the test core circuit on the upper computer, then the whole system starts to work, and the system user can observe the set temperature of the temperature box and the output data of each temperature sensor through the upper computer; the standard temperature-calibration database is a typical database formed by further screening of the knowledge database and the historical database; after the data matching, the matched calibration data is burned into the chip OTP under the mode recognition and general rule constraint; then the whole test process is ended.

Images