Device for testing hardness and residual stress of semiconductor wafer material

Abstract

The invention discloses a device for testing hardness and residual stress of a semiconductor wafer material, which belongs to the field of material testing and comprises an ultrasonic transmitter, an ultrasonic receiver, a signal processing module, a precision positioning module, an environment control module, a touch screen display and a data management module. According to the invention, test data can be quickly processed and analyzed, the test time is shortened, the working efficiency is improved, the operation and maintenance of equipment are simplified, the use cost is reduced, powerful data support is provided, and the reliability and effectiveness of the equipment in different application scenes are proved; the transmitter and the receiver can be accurately positioned, errors are reduced, the test reliability is improved, and strong signals can be obtained under high frequency, so that the test sensitivity and precision are improved.

Description

Technical Field

[0001] The invention relates to the field of material testing, and in particular to testing equipment for hardness and residual stress of semiconductor wafer materials. Background Art

[0002] With the rapid development of the semiconductor industry, the mechanical properties testing of wafer materials has become a key link. During the manufacturing process, the hardness and residual stress of wafer materials directly affect the quality and reliability of the final product. However, traditional hardness and stress testing methods (such as indentation hardness testing and X-ray stress testing) have some shortcomings. The traditional indentation hardness test leaves an indentation on the surface of the material, which will cause damage to semiconductor wafer materials with high precision requirements; although the X-ray stress test is non-destructive, the sample needs to be pre-processed and positioned in a complex manner during the test. At the same time, the traditional testing method is complicated to operate, requires professional personnel to operate, is time-consuming and inefficient, and has limited test accuracy and sensitivity.

[0003] The existing testing equipment for the hardness and residual stress of semiconductor wafer materials has low efficiency in processing and analyzing test data, long testing time, complex operation and maintenance of the equipment, and increased cost of use; in addition, the existing testing equipment for the hardness and residual stress of semiconductor wafer materials cannot accurately locate the transmitter and receiver, which reduces the reliability of the test and cannot obtain strong signals at higher frequencies. For this reason, we propose testing equipment for the hardness and residual stress of semiconductor wafer materials. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose a testing device for the hardness and residual stress of semiconductor wafer materials.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] Testing equipment for hardness and residual stress of semiconductor wafer materials, including an ultrasonic transmitter, an ultrasonic receiver, a signal processing module, a precision positioning module, an environmental control module, a touch screen display, and a data management module;

[0007] The ultrasonic transmitter is used to generate ultrasonic waves by converting electrical signals into mechanical vibrations;

[0008] The ultrasonic receiver is used to receive ultrasonic waves by converting mechanical vibrations into electrical signals;

[0009] The signal processing module specifically includes an analog-to-digital converter, a digital signal processor, a hardness calculation unit, and a data storage and transmission unit;

[0010] Among them, the analog-to-digital converter is used to convert the electrical signal generated by the ultrasonic receiver into a digital signal; the digital signal processor is used to analyze the digital signal and extract various parameters of hardness and residual stress; the hardness calculation unit is used to calculate the hardness value of the wafer according to the ultrasonic propagation characteristics; the data storage and transmission unit is used to store the test results and transmit the data to the user interface or other external systems through the interface;

[0011] The precision positioning module specifically includes an electric mobile platform, a position sensor and a control unit;

[0012] The electric mobile platform is used for high-precision position control through multi-axis adjustment; the position sensor is used for real-time feedback of position information; the control unit is used for automatically adjusting the positions of the ultrasonic transmitter and the ultrasonic receiver;

[0013] The environmental control module is used to maintain the temperature stability of the equipment and the test environment, control the humidity of the test environment, and reduce the interference of external vibrations on the test process;

[0014] The touch screen display is used to provide an intuitive display and operation interface;

[0015] The data management module is used to provide test data classification, storage, query and analysis functions.

[0016] As a further solution of the present invention, the ultrasonic transmitter and the ultrasonic receiver are both air-coupled ultrasonic sensors; the ultrasonic transmitter is specifically composed of a piezoelectric crystal, an acoustic lens, a transmitting circuit and a housing;

[0017] The piezoelectric crystal is used to generate ultrasonic waves, which are converted into mechanical vibrations through electrical signals; the acoustic lens is used to focus the ultrasonic beam and reduce the attenuation of sound energy in the air; the transmitting circuit is used to provide a high-power pulse signal to drive the piezoelectric crystal to generate ultrasonic waves; the housing is an anti-vibration and dust-proof structure to protect the internal components;

[0018] The ultrasonic receiver is specifically composed of a piezoelectric crystal, an acoustic lens, a receiving circuit and a housing;

[0019] Among them, the piezoelectric crystal is used to receive ultrasonic waves and convert them into electrical signals through mechanical vibration; the acoustic lens is used to receive ultrasonic waves and convert them into electrical signals through mechanical vibration; the receiving circuit is used to amplify and process the electrical signals transmitted from the piezoelectric crystal; the outer shell is a vibration-proof and dust-proof structure for protecting internal components.

[0020] As a further solution of the present invention, the specific steps of extracting various parameters of hardness and residual stress by the digital signal processor are as follows:

[0021] S1.1: After the ultrasonic receiver converts the received ultrasonic signal into an electrical signal, the digital-to-analog converter samples each group of electrical signals using the Nyquist sampling theorem and converts each group of electrical signals into corresponding time domain signals. The digital signal processor receives each group of time domain signals obtained by sampling and performs filtering on each group of collected signals.

[0022] S1.2: Analyze the rise time and peak time characteristics of the signal, obtain the propagation time of the signal based on the analysis results, and calculate the propagation speed of the ultrasonic wave in the material based on the thickness of the semiconductor wafer and the propagation time;

[0023] S1.3: After the time domain analysis is completed, the digital signal processor converts the time domain signal into a frequency domain signal through FFT, analyzes the signal spectrum, correlates the attenuation of the ultrasonic wave with the attenuation of the frequency spectrum, and uses the hardness calculation unit to calculate the material hardness through the amplitude spectrum in the frequency domain. Then, the internal residual stress of the semiconductor wafer material is calculated based on the sound wave velocity and the sound velocity in the stress-free state.

[0024] As a further solution of the present invention, the specific steps of the hardness calculation unit calculating the hardness value of the wafer according to the ultrasonic propagation characteristics are as follows:

[0025] S2.1: The hardness calculation unit extracts the propagation time Δt of the ultrasonic signal based on the thickness d of the semiconductor wafer and the digital signal processor through time domain analysis or frequency domain analysis, and uses the formula Calculate the propagation speed v of the ultrasonic wave in the wafer, and then extract the corresponding signal attenuation rate α based on the correlation between the attenuation of the ultrasonic wave and the attenuation of the frequency spectrum;

[0026] S2.2: The collected propagation time, propagation speed, and signal attenuation rate are used as input data, and each group of input data is standardized or normalized, and each group of processed data is input into the local model obtained by federated learning, and then each group of input data of the local model is forward propagated;

[0027] S2.3: Each hidden layer in the local model performs weighted summation on the input data to obtain the weighted sum of each neuron in the corresponding hidden layer, and performs nonlinear transformation on the final weighted sum through the ReLU activation function;

[0028] S2.4: After processing all hidden layers, the output layer performs weighted sum and activation on the output values ​​of the last hidden layer, and obtains the corresponding prediction result, that is, the hardness value of the wafer. The output result is then verified by the mean absolute error function, and the hyperparameters of the local model are adjusted based on the evaluation results. At the same time, the wafer hardness value is output to the user through the touch screen display.

[0029] As a further solution of the present invention, the specific steps of the federated learning described in S2.2 are as follows:

[0030] S3.1: The data management module of each test device stores a local training data set, and an external central server initializes a global model. The initialized global model is downloaded by the hardness calculation unit of each test device and is trained locally on each test device;

[0031] S3.2: The hardness calculation unit aims to minimize its local loss function, uses the local training data set to train the model, calculates the gradient of its local loss function for each model parameter through the back propagation algorithm, and updates the model using the gradient descent method;

[0032] S3.3: After each round of training, the central server collects the model parameters of all devices and aggregates the model parameters of each device through weighted averaging to obtain the corresponding global model parameters. The central server then updates the original global model with the obtained global model parameters, and then distributes the updated global model to all participating devices and uses the model for the next round of local training until the accuracy of the hardness prediction of the global model reaches the preset threshold.

[0033] As a further solution of the present invention, the specific steps of storing the test results by the data storage and transmission unit are as follows:

[0034] S3.1: After the hardness value and residual stress of the current semiconductor wafer are calculated, the data storage and transmission unit organizes the test data into a block, and the creator of each block signs the block and encrypts it using his private key;

[0035] S3.2: After the block is constructed, the data storage and transmission unit uploads the newly generated block to the blockchain. When the blockchain detects that a new block has been added, the remaining blocks of the blockchain conduct consensus verification on the newly added block through the PoW mechanism;

[0036] S3.3: After verification, the new block is broadcast to the nodes in the entire blockchain network. Each node verifies the legitimacy of the block based on the hash value, signature and data content of the block, and stores it in its own database.

[0037] As a further solution of the present invention, the specific steps of high-precision position control of the electric mobile platform are as follows:

[0038] S4.1: The position sensor collects the positions of semiconductor wafer materials, ultrasonic transmitters and ultrasonic receivers in real time. The electric mobile platform establishes a state space. Each state corresponds to a position of the electric mobile platform. The movement of each axis of the platform corresponds to a state conversion, and the movement direction and step length of each axis are obtained.

[0039] S4.2: Take the current position of the electric mobile platform as the root node, generate corresponding child nodes by simulating the movable path of the platform at the current position, so as to construct a set of search trees, and then start from the current node, calculate the UCB value of each child node of the current node, select the child node with the highest UCB value, and generate new child nodes by randomly selecting one or more groups of axial actions, and add them to the search tree;

[0040] S4.3: For the selected path, the movement of the electric mobile platform is simulated by a predefined dynamic model or control rule. During the simulation, the Euclidean distance between the current position of the platform and the target position is calculated. The smaller the distance, the more successful the path selection. After the simulation is completed, the simulation result is transmitted back and the average reward value of each node of the path in the search tree is updated;

[0041] S4.4: Repeat the selection, expansion, simulation and backtracking steps many times until the range of the average reward value of each node converges to the preset threshold, and select the node with the smallest Euclidean distance as the optimal moving path of the electric mobile platform. Then calculate the step size and movement command of each axis. After receiving the movement command, the electric drive of each axis moves so that the platform gradually approaches the target position.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. The present invention initializes a global model through an external central server. The initialized global model is downloaded by the hardness calculation unit of each test device and is locally trained on each test device. The hardness calculation unit uses a local training data set to train and optimize the model with the goal of minimizing its local loss function. After each round of training, the central server collects the model parameters of all devices and aggregates the model parameters of each device by weighted average to obtain the corresponding global model parameters. Then the central server updates the original global model by the obtained global model parameters, and then distributes the updated global model to all participating devices. After repeated training is completed, the hardness calculation unit transmits the received groups of data as input data to the local model, obtains the corresponding prediction results, i.e., the hardness value of the wafer, through forward propagation, and then verifies the output results through the mean absolute error function, and adjusts the hyperparameters of the local model according to the evaluation results. At the same time, the wafer hardness value is output to the user through the touch screen display, which can quickly process and analyze test data, shorten test time, improve work efficiency, simplify equipment operation and maintenance, reduce use costs, and provide strong data support, proving the reliability and effectiveness of the equipment in different application scenarios.

[0044] 2. The present invention collects the positions of semiconductor wafer materials, ultrasonic transmitters and ultrasonic receivers in real time through position sensors. The electric mobile platform establishes a state space, each state corresponds to a position of the electric mobile platform, and the movement of each axis of the platform corresponds to a state conversion, and obtains the movement direction and step length of each axis. The current position of the electric mobile platform is used as the root node, and the corresponding child nodes are generated by simulating the movable path of the platform at the current position to construct a set of search trees. Then, starting from the current node, the UCB value of each child node of the current node is calculated, and the child node with the highest UCB value is selected. A new child node is generated by randomly selecting one or more groups of axial movements and added to the search tree. For the selected path, the electric mobile platform is simulated by a predefined dynamic model or control rule. During the simulation, the Euclidean distance between the current position of the platform and the target position is calculated. If the distance is smaller, the path selection is more successful. After the simulation is completed, the simulation results are transmitted back, and the average reward value of each node in the path in the search tree is updated. The selection, expansion, simulation and backtracking steps are repeated many times until the range of the average reward value of each node converges to the preset threshold, and the node with the smallest Euclidean distance is selected as the optimal moving path of the electric mobile platform. After that, the step size and movement command of each axis are calculated. After receiving the movement command, the electric drive of each axis moves to make the platform gradually approach the target position, which can accurately locate the transmitter and receiver, reduce errors, improve the reliability of the test, and ensure that strong signals can be obtained at higher frequencies, thereby improving the sensitivity and accuracy of the test. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0046] Figure 1 A system block diagram of the testing equipment for hardness and residual stress of semiconductor wafer materials proposed by the present invention;

[0047] Figure 2 This is a structural block diagram of a signal processing module of a test device for semiconductor wafer material hardness and residual stress proposed by the present invention;

[0048] Figure 3 This is a structural block diagram of the precision positioning module of the testing equipment for semiconductor wafer material hardness and residual stress proposed by the present invention. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0050] Example 1

[0051] Reference Figure 1-Figure 3 , a testing device for hardness and residual stress of semiconductor wafer materials, including an ultrasonic transmitter, an ultrasonic receiver, a signal processing module, a precision positioning module, an environmental control module, a touch screen display and a data management module.

[0052] The ultrasonic transmitter is used to generate ultrasonic waves by converting electrical signals into mechanical vibrations; the ultrasonic receiver is used to receive ultrasonic waves by converting mechanical vibrations into electrical signals.

[0053] It should be further explained that both the ultrasonic transmitter and the ultrasonic receiver are air-coupled ultrasonic sensors;

[0054] The ultrasonic transmitter is specifically composed of a piezoelectric crystal, an acoustic lens, a transmitting circuit and a housing;

[0055] Among them, piezoelectric crystals are used to generate ultrasonic waves, which are converted into mechanical vibrations through electrical signals; acoustic lenses are used to focus ultrasonic beams and reduce the attenuation of sound energy in the air; the transmitting circuit is used to provide high-power pulse signals to drive the piezoelectric crystals to generate ultrasonic waves; the outer shell is an anti-vibration and dust-proof structure to protect the internal components;

[0056] The ultrasonic receiver is specifically composed of a piezoelectric crystal, an acoustic lens, a receiving circuit, and a housing;

[0057] Among them, the piezoelectric crystal is used to receive ultrasonic waves and convert them into electrical signals through mechanical vibration; the acoustic lens is used to receive ultrasonic waves and convert them into electrical signals through mechanical vibration; the receiving circuit is used to amplify and process the electrical signals transmitted from the piezoelectric crystal; the outer casing is a vibration-proof and dust-proof structure to protect the internal components.

[0058] The signal processing module specifically includes an analog-to-digital converter, a digital signal processor, a hardness calculation unit, and a data storage and transmission unit.

[0059] The analog-to-digital converter is used to convert the electrical signal generated by the ultrasonic receiver into a digital signal; the digital signal processor is used to analyze the digital signal and extract various parameters of hardness and residual stress.

[0060] Specifically, after the ultrasonic receiver converts the received ultrasonic signal into an electrical signal, the digital-to-analog converter uses the Nyquist sampling theorem to sample each group of electrical signals and converts each group of electrical signals into a corresponding time domain signal. The digital signal processor receives each group of time domain signals obtained by sampling, and filters each group of collected signals, analyzes the rise time and peak time characteristics of the signal, obtains the propagation time of the signal based on the analysis results, and calculates the propagation speed of the ultrasonic wave in the material based on the thickness of the semiconductor wafer and the propagation time. After the time domain analysis is completed, the digital signal processor converts the time domain signal into a frequency domain signal through FFT, analyzes the frequency spectrum of the signal, correlates the attenuation of the ultrasonic wave with the attenuation of the frequency spectrum, and uses the hardness calculation unit to calculate the material hardness through the amplitude spectrum in the frequency domain, and then calculates the internal residual stress of the semiconductor wafer material based on the sound wave velocity and the sound velocity in the stress-free state.

[0061] The hardness calculation unit is used to calculate the hardness value of the wafer according to the ultrasonic propagation characteristics.

[0062] Specifically, the hardness calculation unit extracts the propagation time Δt of the ultrasonic signal according to the thickness d of the semiconductor wafer and the digital signal processor through time domain analysis or frequency domain analysis, and uses the formula The propagation speed v of ultrasound in the wafer is calculated, and then the corresponding signal attenuation rate α is extracted according to the correlation between the attenuation of ultrasound and the attenuation of the frequency spectrum. The collected propagation time, propagation speed and signal attenuation rate are used as input data, and each group of input data is standardized or normalized, and each group of processed data is input into the local model obtained by federated learning. Then, each group of input data of the local model is forward propagated, and each hidden layer in the local model performs weighted summation on the input data to obtain the weighted sum of each neuron in the corresponding hidden layer. The final weighted sum is nonlinearly transformed by the ReLU activation function. After all hidden layers are processed, the output layer performs weighted sum and activation on the output value of the last hidden layer, and obtains the corresponding prediction result, that is, the hardness value of the wafer. Then, the output result is verified by the mean absolute error function, and according to the evaluation result, the hyperparameters of the local model are adjusted, and the wafer hardness value is output to the user through the touch screen display.

[0063] It should be further explained that the data management module of each test device stores a local training data set, and an external central server initializes a global model. The initialized global model is downloaded by the hardness calculation unit of each test device and locally trained on each test device. The hardness calculation unit aims to minimize its local loss function, uses the local training data set to train the model, and calculates the gradient of its local loss function for each model parameter through the back propagation algorithm, and updates the model using the gradient descent method. After each round of training, the central server collects the model parameters of all devices and aggregates the model parameters of each device through weighted averaging to obtain the corresponding global model parameters. After that, the central server updates the original global model with the obtained global model parameters, and then distributes the updated global model to all participating devices, and uses the model for the next round of local training until the accuracy of the hardness prediction of the global model reaches the preset threshold.

[0064] The data storage and transmission unit is used to store the test results and transmit the data to the user interface or other external systems through the interface.

[0065] Specifically, after the current semiconductor wafer hardness value and residual stress calculation are completed, the data storage and transmission unit organizes the test data into a block. At the same time, the creator of each block signs the block and encrypts it with its private key. After the block is built, the data storage and transmission unit uploads the newly generated block to the blockchain. When the blockchain detects that a new block has been added, the remaining blocks of the blockchain use the PoW mechanism to perform consensus verification on the newly chained block. After the verification is passed, the newly chained block is broadcast to the nodes in the entire blockchain network. Each node verifies the legitimacy of the block based on the hash value, signature and data content of the block, and stores it in its own database.

[0066] Example 2

[0067] Reference Figure 1-Figure 3 , a testing device for hardness and residual stress of semiconductor wafer materials, including an ultrasonic transmitter, an ultrasonic receiver, a signal processing module, a precision positioning module, an environmental control module, a touch screen display and a data management module.

[0068] The precision positioning module specifically includes an electric mobile platform, a position sensor and a control unit;

[0069] The electric mobile platform is used for high-precision position control through multi-axis adjustment; the position sensor is used for real-time feedback of position information; and the control unit is used for automatically adjusting the positions of the ultrasonic transmitter and the ultrasonic receiver.

[0070] Specifically, the position sensor collects the positions of semiconductor wafer materials, ultrasonic transmitters and ultrasonic receivers in real time. The electric mobile platform establishes a state space, each state corresponds to a position of the electric mobile platform, and the movement of each axis of the platform corresponds to a state conversion, and the movement direction and step length of each axis are obtained. The current position of the electric mobile platform is used as the root node, and the corresponding child nodes are generated by simulating the movable path of the platform at the current position to construct a set of search trees. Then, starting from the current node, the UCB value of each child node of the current node is calculated, and the child node with the highest UCB value is selected. New child nodes are generated by randomly selecting one or more groups of axial actions and added to the search tree. For the selected The path is simulated by a predefined dynamic model or control rule to simulate the movement of the electric mobile platform. During the simulation, the Euclidean distance between the current position of the platform and the target position is calculated. The smaller the distance, the more successful the path selection is. After the simulation is completed, the simulation result is transmitted back, and the average reward value of each node in the path in the search tree is updated. The selection, expansion, simulation and backtracking steps are repeated many times until the range of the average reward value of each node converges to the preset threshold, and the node with the smallest Euclidean distance is selected as the optimal movement path of the electric mobile platform. Then, the step size and movement command of each axis are calculated. After receiving the movement command, the electric drive of each axis moves to gradually approach the target position.

[0071] The environmental control module is used to maintain the temperature stability of the equipment and test environment, control the humidity of the test environment, and reduce the interference of external vibration on the test process; the touch screen display is used to provide an intuitive display and operation interface; the data management module is used to provide test data classification, storage, query and analysis functions.

Claims

1. A test device for hardness and residual stress of semiconductor wafer materials, characterized in that: It includes an ultrasonic transmitter, an ultrasonic receiver, a signal processing module, a precision positioning module, an environmental control module, a touch screen display, and a data management module; The ultrasonic transmitter is used to generate ultrasonic waves by converting electrical signals into mechanical vibrations; The ultrasonic receiver is used to receive ultrasonic waves by converting mechanical vibrations into electrical signals; The signal processing module specifically includes an analog-to-digital converter, a digital signal processor, a hardness calculation unit, and a data storage and transmission unit; Among them, the analog-to-digital converter is used to convert the electrical signal generated by the ultrasonic receiver into a digital signal; the digital signal processor is used to analyze the digital signal and extract various parameters of hardness and residual stress; the hardness calculation unit is used to calculate the hardness value of the wafer according to the ultrasonic propagation characteristics; the data storage and transmission unit is used to store the test results and transmit the data to the user interface or other external systems through the interface; The precision positioning module specifically includes an electric mobile platform, a position sensor and a control unit; The electric mobile platform is used for high-precision position control through multi-axis adjustment; the position sensor is used for real-time feedback of position information; the control unit is used for automatically adjusting the positions of the ultrasonic transmitter and the ultrasonic receiver; The environmental control module is used to maintain the temperature stability of the equipment and the test environment, control the humidity of the test environment, and reduce the interference of external vibrations on the test process; The touch screen display is used to provide an intuitive display and operation interface; The data management module is used to provide test data classification, storage, query and analysis functions.

2. The testing device for semiconductor wafer material hardness and residual stress according to claim 1, characterized in that: The specific steps of the digital signal processor to extract various parameters of hardness and residual stress are as follows: S1.1: After the ultrasonic receiver converts the received ultrasonic signal into an electrical signal, the digital-to-analog converter samples each group of electrical signals using the Nyquist sampling theorem and converts each group of electrical signals into corresponding time domain signals. The digital signal processor receives each group of time domain signals obtained by sampling and performs filtering on each group of collected signals. S1.2: Analyze the rise time and peak time characteristics of the signal, obtain the propagation time of the signal based on the analysis results, and calculate the propagation speed of the ultrasonic wave in the material based on the thickness of the semiconductor wafer and the propagation time; S1.3: After the time domain analysis is completed, the digital signal processor converts the time domain signal into a frequency domain signal through FFT, analyzes the signal spectrum, correlates the attenuation of the ultrasonic wave with the attenuation of the frequency spectrum, and uses the hardness calculation unit to calculate the material hardness through the amplitude spectrum in the frequency domain. Then, the internal residual stress of the semiconductor wafer material is calculated based on the sound wave velocity and the sound velocity in the stress-free state.

3. The testing device for semiconductor wafer material hardness and residual stress according to claim 2, characterized in that: The specific steps of the hardness calculation unit calculating the hardness value of the wafer according to the ultrasonic propagation characteristics are as follows: S2.1: The hardness calculation unit extracts the propagation time Δt of the ultrasonic signal based on the thickness d of the semiconductor wafer and the digital signal processor through time domain analysis or frequency domain analysis, and uses the formula Calculate the propagation speed v of the ultrasonic wave in the wafer, and then extract the corresponding signal attenuation rate α based on the correlation between the attenuation of the ultrasonic wave and the attenuation of the frequency spectrum; S2.2: The collected propagation time, propagation speed, and signal attenuation rate are used as input data, and each group of input data is standardized or normalized, and each group of processed data is input into the local model obtained by federated learning, and then each group of input data of the local model is forward propagated; S2.3: Each hidden layer in the local model performs weighted summation on the input data to obtain the weighted sum of each neuron in the corresponding hidden layer, and performs nonlinear transformation on the final weighted sum through the ReLU activation function; S2.4: After processing all hidden layers, the output layer performs weighted sum and activation on the output values ​​of the last hidden layer, and obtains the corresponding prediction result, that is, the hardness value of the wafer. The output result is then verified by the mean absolute error function, and the hyperparameters of the local model are adjusted based on the evaluation results. At the same time, the wafer hardness value is output to the user through the touch screen display.

4. The testing device for semiconductor wafer material hardness and residual stress according to claim 3, characterized in that: The specific steps of federated learning described in S2.2 are as follows: S3.1: The data management module of each test device stores a local training data set, and an external central server initializes a global model. The initialized global model is downloaded by the hardness calculation unit of each test device and is trained locally on each test device; S3.2: The hardness calculation unit aims to minimize its local loss function, uses the local training data set to train the model, calculates the gradient of its local loss function for each model parameter through the back propagation algorithm, and updates the model using the gradient descent method; S3.3: After each round of training, the central server collects the model parameters of all devices and aggregates the model parameters of each device through weighted averaging to obtain the corresponding global model parameters. The central server then updates the original global model with the obtained global model parameters, and then distributes the updated global model to all participating devices and uses the model for the next round of local training until the accuracy of the hardness prediction of the global model reaches the preset threshold.

5. The testing device for semiconductor wafer material hardness and residual stress according to claim 3, characterized in that: The specific steps of storing the test results in the data storage and transmission unit are as follows: S3.1: After the hardness value and residual stress of the current semiconductor wafer are calculated, the data storage and transmission unit organizes the test data into a block, and the creator of each block signs the block and encrypts it using his private key; S3.2: After the block is constructed, the data storage and transmission unit uploads the newly generated block to the blockchain. When the blockchain detects that a new block has been added, the remaining blocks of the blockchain conduct consensus verification on the newly added block through the PoW mechanism; S3.3: After verification, the new block is broadcast to the nodes in the entire blockchain network. Each node verifies the legitimacy of the block based on the hash value, signature and data content of the block, and stores it in its own database.

6. The testing device for semiconductor wafer material hardness and residual stress according to claim 1, characterized in that: The specific steps of high-precision position control of the electric mobile platform are as follows: S4.1: The position sensor collects the positions of semiconductor wafer materials, ultrasonic transmitters and ultrasonic receivers in real time. The electric mobile platform establishes a state space. Each state corresponds to a position of the electric mobile platform. The movement of each axis of the platform corresponds to a state conversion, and the movement direction and step length of each axis are obtained. S4.2: Take the current position of the electric mobile platform as the root node, generate corresponding child nodes by simulating the movable path of the platform at the current position, so as to construct a set of search trees, and then start from the current node, calculate the UCB value of each child node of the current node, select the child node with the highest UCB value, and generate new child nodes by randomly selecting one or more groups of axial actions, and add them to the search tree; S4.3: For the selected path, the movement of the electric mobile platform is simulated by a predefined dynamic model or control rule. During the simulation, the Euclidean distance between the current position of the platform and the target position is calculated. The smaller the distance, the more successful the path selection. After the simulation is completed, the simulation result is transmitted back and the average reward value of each node of the path in the search tree is updated; S4.4: Repeat the selection, expansion, simulation and backtracking steps many times until the range of the average reward value of each node converges to the preset threshold, and select the node with the smallest Euclidean distance as the optimal moving path of the electric mobile platform. Then calculate the step size and movement command of each axis. After receiving the movement command, the electric drive of each axis moves so that the platform gradually approaches the target position.

Images