Chip design method and system of silicon piezoresistive pressure sensor

Abstract

The invention provides a chip design method and system for a silicon piezoresistive pressure sensor. The method comprises the following steps: constructing a design parameter set of the silicon piezoresistive pressure sensor; obtaining a performance parameter set corresponding to the design parameter set according to preset simulation software; constructing a numerical design library according to the design parameter set and the performance parameter set; and establishing a bidirectional numerical design model based on the Bi-GRU neural network according to the numerical design library. According to the method or the system, the active design of the structural parameters and the piezoresistance can be reversely realized through the target performance parameters, the performance parameters of the pressure chip can be positively obtained through the design parameters, a result can be given without repeated simulation calculation, and the design efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pressure sensors, and in particular to a chip design method and system for a silicon piezoresistive pressure sensor. Background Art

[0002] Silicon piezoresistive pressure sensor is a sensor that uses the piezoresistive effect of single crystal silicon material to measure pressure. Its working principle is that when the single crystal silicon material is subjected to external pressure, its internal lattice structure will deform, resulting in a change in resistance value. By measuring the change in resistance, the pressure acting on the silicon material can be inferred. Silicon piezoresistive pressure sensor has been widely used in extreme environments such as industrial oil pressure measurement, industrial transportation pipelines, and aerospace ignition and flameout due to its simple structure, low cost, and strong batch manufacturing capabilities.

[0003] However, due to the physical properties of silicon-based materials, before the pressure sensor chip is prepared, it is necessary to select the appropriate chip size according to the working environment of the chip and the requirements for chip performance, and simulate the chip multiple times to meet the working needs. When the working environment and performance requirements of the chip change, it is necessary to simulate and redesign the chip structure. For P-type doped varistors, their temperature coefficient of resistance TCR is positive, that is, as the temperature increases, the varistor value increases. At different doping concentrations, the piezoresistance coefficient decreases with increasing temperature, but the reduction range is different. Therefore, the chip operating temperature is different, and different doping concentrations need to be selected to reduce the impact of temperature on chip performance. This process is time-consuming and labor-intensive. Summary of the invention

[0004] The embodiment of the present invention provides a chip design method and system for a silicon piezoresistive pressure sensor, which can overcome the technical problems existing in the above-mentioned prior art, can reversely realize the active design of structural parameters and piezoresistance through target performance parameters, and can positively obtain the performance parameters of the pressure chip through design parameters, and can provide results without multiple simulation calculations, thereby greatly improving design efficiency.

[0005] To achieve the above object, the present invention provides a chip design method for a silicon piezoresistive pressure sensor, which is characterized by comprising:

[0006] Construct a design parameter set for a silicon piezoresistive pressure sensor;

[0007] Obtaining a performance parameter set corresponding to the design parameter set according to a preset simulation software;

[0008] Constructing a numerical design library according to the design parameter set and the performance parameter set;

[0009] A bidirectional numerical design model based on a Bi-GRU neural network is established according to the numerical design library.

[0010] The present invention also provides a chip design system for a silicon piezoresistive pressure sensor, comprising a database module, a design parameter module, a performance parameter module and a model evaluation module;

[0011] The database module includes a design parameter set and a performance parameter set; the design parameter set includes: chip structure, chip size, diaphragm size, piezoresistive doping concentration, heavy doping concentration, chip operating temperature and pressure range; the performance parameter set includes: chip output voltage, sensitivity, linearity, natural frequency, maximum equivalent stress, thermal sensitivity drift and thermal zero drift;

[0012] The design parameter module is used to construct a forward numerical design model based on the Bi-GRU neural network according to the database module, and obtain a chip performance parameter result prediction according to the design parameters;

[0013] The performance parameter module is used to construct a reverse numerical design model based on the Bi-GRU neural network according to the database module, and obtain a chip design parameter result prediction according to the performance parameters;

[0014] The model evaluation module is used to evaluate model performance and adjust parameters or optimize model structure as needed.

[0015] The chip design method and system of the silicon piezoresistive pressure sensor provided by the present invention uses simulation software and numerical calculation to obtain corresponding performance parameters according to design parameters, builds a numerical design library, and then uses a Bi-GRU neural network to mine the forward and reverse information of the design parameters and performance parameters; wherein the first layer of GRU obtains the forward information of the input sequence, and the second layer of GRU further extracts the reverse features of the input sequence based on the first layer, which can improve the expression ability of the model. This method realizes the bidirectional design of the design parameters and performance parameters of the pressure chip, that is, the active design of the structural parameters and piezoresistance is reversely realized through the target performance parameters, and the performance parameter indicators of the pressure chip can also be obtained through the design parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1 A schematic flow chart of a chip design method for a silicon piezoresistive pressure sensor provided in an embodiment of the present invention;

[0018] Figure 2 A schematic diagram of the numerical design structure of the Bi-GRU neural network pressure chip provided in an embodiment of the present invention;

[0019] Figure 3 A forward model diagram of pressure chip design parameters of a Bi-GRU neural network with multi-variable input and multi-variable output provided by an embodiment of the present invention;

[0020] Figure 4 A reverse model diagram of pressure chip performance parameters with multi-variable input and multi-variable output of a Bi-GRU neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In addition, the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. In the description of the following embodiments, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0023] The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. To make the technical solution of the present invention clearer, the embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0024] Figure 1 A schematic flow chart of a chip design method for a silicon piezoresistive pressure sensor provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes:

[0025] S1: Construct a set of design parameters for a silicon piezoresistive pressure sensor.

[0026] When designing a silicon piezoresistive pressure sensor according to an embodiment of the present invention, the following key design parameters need to be considered: chip structure, chip size, diaphragm size, piezoresistive doping concentration, heavy doping concentration, chip operating temperature, and pressure range.

[0027] Each design parameter has an important impact on the performance of silicon piezoresistive pressure sensors, among which: the chip structure morphology involves the geometry and internal structure design of the sensor chip, such as the layout of the bridge resistors. The structure morphology will directly affect the pressure distribution and the conversion efficiency of the electrical signal; the chip size determines the mechanical and thermal properties of the sensor. A larger chip may provide a larger contact area and better thermal stability, but may increase manufacturing costs and complexity; the diaphragm is the part of the sensor that senses pressure changes. The size of the diaphragm will affect its sensitivity to pressure and response speed; the piezoresistive effect is achieved by doping impurities in the silicon wafer. The piezoresistive doping concentration determines the rate of change of resistance and the sensitivity of the sensor; the heavily doped area is usually used to form a good ohmic contact or adjust the resistance value. The heavily doped concentration will affect the long-term stability and temperature coefficient of the sensor; the chip operating temperature range will affect the material properties and the output stability of the sensor. It is necessary to ensure that the performance of the sensor remains consistent at the expected operating temperature; the pressure range is the pressure range that the sensor can measure, from the minimum detectable pressure to the maximum pressure. The range defines the application scenario and market demand of the sensor.

[0028] S2: Obtain a performance parameter set corresponding to the design parameter set according to a preset simulation software.

[0029] Using simulation analysis and numerical calculations, such as COMSOL The performance parameter set corresponding to the above design parameter set can be obtained, and the specific steps include:

[0030] Construct a basic model of the pressure sensor based on the chip size and diaphragm size;

[0031] Through preset numerical calculations, the performance parameters of the model under different design parameters are predicted. For example, the model performance parameters can be obtained by changing the design parameters such as the piezoresistive doping concentration, the chip operating temperature (-50 to 400°C), and the pressure range.

[0032] Furthermore, in an embodiment of the present invention, the performance parameter set includes: chip output voltage, sensitivity, linearity, natural frequency, maximum equivalent stress, thermal sensitivity drift and thermal zero drift.

[0033] S3: Construct a numerical design library based on the design parameter set and the performance parameter set.

[0034] Building a numerical design library is to systematically organize the design parameter sets and performance parameter sets to facilitate query, analysis and design optimization.

[0035] Furthermore, in order to ensure the consistency and comparability of data, before building the numerical design library, it also includes: parameter standardization of the design parameter set and the performance parameter set.

[0036] Normalization usually refers to scaling the data so that it falls into a small specific range. The purpose of this is to remove the unit restrictions of the data and convert it into a dimensionless pure value, so that indicators of different units or magnitudes can be compared and weighted.

[0037] The embodiment of the present invention uniformly converts the data in the design parameter set and the performance parameter set into values ​​between [0 and 1]. In this way, the influence of the design parameters on the performance parameters can be more effectively identified, and the best design solution can be found.

[0038] S4: Establish a bidirectional numerical design model based on the Bi-GRU neural network according to the numerical design library.

[0039] Figure 2 A schematic diagram of the numerical design structure of the Bi-GRU neural network pressure chip provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, where {x1,..,x i ,...,x7} represents the forward GRU{t1,...t i ,...,t7}, {x8,...,x j ,...,x 14} represents the reverse GRU{t8,...t j ,...,t 14}, h0 represents the initial hidden vector, h7 and h 14 They are the outputs of the forward and reverse GRU at the last moment.

[0040] Furthermore, the specific steps of constructing the above model include:

[0041] S41: Build the Anaconda integrated development environment;

[0042] The embodiment of the present invention adopts the Anaconda integrated development environment, which is a popular Python data science and machine learning distribution that includes many pre-installed libraries and tools.

[0043] S42: Randomly sort and divide the data of the numerical design library;

[0044] In order to improve the generalization ability of the model and ensure the fairness and accuracy of the model evaluation, it is necessary to randomly sort and divide the data in the numerical design library. The specific steps include:

[0045] Loading data of the numerical design library;

[0046] The data of the numerical design library are randomly sorted to ensure that the order of the data does not affect the results of model training.

[0047] The randomly sorted data is divided into a training set and a test set according to a preset ratio. The embodiment of the present invention uses 80% of the data as a training set and 20% of the data as a test set.

[0048] S43: Import the pytorch learning framework.

[0049] PyTorch is a very powerful tool that can quickly implement and iterate models, which improves the efficiency of chip design to a certain extent.

[0050] S44: Build a forward numerical design model based on Bi-GRU neural network.

[0051] Figure 3 The pressure chip design parameter forward model diagram of the Bi-GRU neural network with multi-variable input and multi-variable output provided by the embodiment of the present invention is as follows Figure 3 As shown in the figure, the model consists of four parts, namely the design parameter input sequence, the Bi-GRU neural network model diagram, 7 fully connected layers and the performance parameter output.

[0052] Furthermore, the specific steps of building the above forward model include:

[0053] S441: Setting model parameters, where the model parameters include: number of hidden neurons, learning rate, batch size, epoch value and multivariate input sequence length sequence.

[0054] S442: Convert the design parameter data in the training set into a three-dimensional tensor as the input of the first layer of GRU forward neurons of the forward numerical design model.

[0055] S443: Flip the design parameter sequence as the input of the second layer GRU reverse neuron; the two-layer GRU can provide richer contextual semantic information, obtain the positive and reverse information of the design parameters, and improve the expressiveness of the model.

[0056] S444: Connect 7 fully connected layers to the output layer of the neural network, each layer corresponds to a performance parameter prediction task; decompose the multi-variable output into multi-task learning, and each task is a multi-variable single output.

[0057] S445: Obtain prediction results of seven performance parameters based on the output of the fully connected layer.

[0058] S446: Calculate the loss function. For each of the above output results, calculate its loss, and then add and average the 7 loss values ​​as the final loss result.

[0059] S447: Complete the training of the forward numerical design model based on the Bi-GRU neural network.

[0060] S45: Build a reverse numerical design model based on Bi-GRU neural network.

[0061] Figure 4 The reverse model diagram of the pressure chip performance parameters of the Bi-GRU neural network with multi-variable input and multi-variable output provided by the embodiment of the present invention is as follows Figure 4 As shown in the figure, the model consists of four parts, namely, performance parameter input sequence, Bi-GRU neural network model diagram, 7 fully connected layers and design parameter output.

[0062] Furthermore, the specific steps of building the above reverse model include:

[0063] S451: Set model parameters, where the model parameters include: number of hidden neurons, learning rate, batch size, epoch value and input sequence length.

[0064] S452: Convert the performance parameter data in the training set into a three-dimensional tensor as the input of the first layer of GRU forward neurons of the reverse numerical design model.

[0065] S453: Flip the performance parameter sequence as the input of the GRU reverse neuron in the second layer. Through this operation, the forward and reverse information of the performance parameters can be obtained, which can improve the expression ability of the model.

[0066] S454: Connect 7 fully connected layers to the output layer of the neural network, each layer corresponds to a prediction task of a design parameter; decompose the multi-variable output into multi-task learning, and each task is a multi-variable single output.

[0067] S455: Obtain prediction results of 7 design parameters according to the output of the fully connected layer.

[0068] S456: Calculate the loss function. For each of the above output results, calculate its loss, and then add up the 7 loss values ​​and take the average as the final loss result.

[0069] S457: Complete the reverse numerical design model training based on Bi-GRU neural network.

[0070] Furthermore, after completing the construction of the bidirectional numerical design model based on the Bi-GRU neural network, it is also necessary to evaluate the model performance based on the test set data; and adjust parameters or optimize the model structure as needed.

[0071] Specifically, the design parameter data in the test set is used as the model input to evaluate the effect of the forward numerical design model; the performance parameter data in the test set is used as the model input to evaluate the effect of the reverse numerical design model.

[0072] In the above evaluation process, the model evaluation indicators include: mean absolute error MAE, root mean square error RMSE and mean absolute percentage error MAPE;

[0073]

[0074] Among them, m is the number of chips of different sizes, x i Input data to the model, f(x i ) is the model prediction result, y i is the true value corresponding to the predicted result.

[0075] S46: Save the trained bidirectional numerical design model based on the Bi-GRU neural network.

[0076] In order to avoid repeated training of the model and to further analyze and deploy the model, it is necessary to save the trained bidirectional numerical design model based on the Bi-GRU neural network for use in different applications and environments.

[0077] An embodiment of the present invention further provides a chip design system for a silicon piezoresistive pressure sensor, the system comprising: a database module, a design parameter module, a performance parameter module and a model evaluation module.

[0078] Among them, the database module includes a design parameter set and a performance parameter set; the design parameter set includes: chip structure, chip size, diaphragm size, piezoresistive doping concentration, heavy doping concentration, chip operating temperature and pressure range; the performance parameter set includes: chip output voltage, sensitivity, linearity, natural frequency, maximum equivalent stress, thermal sensitivity drift and thermal zero drift.

[0079] The design parameter module is used to construct a forward numerical design model based on the Bi-GRU neural network according to the database module, and obtain the chip performance parameter result prediction according to the design parameters.

[0080] The performance parameter module is used to construct a reverse numerical design model based on the Bi-GRU neural network according to the database module, and obtain a chip design parameter result prediction according to the performance parameters.

[0081] The model evaluation module is used to evaluate model performance and adjust parameters or optimize model structure as needed.

[0082] The chip design method and system of the silicon piezoresistive pressure sensor provided by the present invention uses simulation software and numerical calculation to obtain corresponding performance parameters according to design parameters, builds a numerical design library, and then uses a Bi-GRU neural network to mine the forward and reverse information of the design parameters and performance parameters; wherein the first layer of GRU obtains the forward information of the input sequence, and the second layer of GRU further extracts the reverse features of the input sequence based on the first layer, which can improve the expression ability of the model. This method realizes the bidirectional design of the design parameters and performance parameters of the pressure chip, that is, the active design of the structural parameters and piezoresistance is reversely realized through the target performance parameters, and the performance parameter indicators of the pressure chip can also be obtained through the design parameters.

[0083] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention is indicated by the following claims.

[0084] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A chip design method for a silicon piezoresistive pressure sensor, characterized in that: include: Construct a design parameter set for a silicon piezoresistive pressure sensor; Obtaining a performance parameter set corresponding to the design parameter set according to a preset simulation software; Constructing a numerical design library according to the design parameter set and the performance parameter set; A bidirectional numerical design model based on a Bi-GRU neural network is established according to the numerical design library.

2. The method according to claim 1, characterized in that The design parameter set includes: chip structure morphology, chip size, diaphragm size, piezoresistive doping concentration, heavy doping concentration, chip operating temperature and pressure range.

3. The method according to claim 2, characterized in that The step of obtaining the performance parameter set corresponding to the design parameter set according to the preset simulation software specifically includes: Constructing a basic model of a pressure sensor according to the chip size and the diaphragm size; Through preset numerical calculations, the performance parameters of the model under different design parameters are predicted.

4. The method according to claim 1, characterized in that: The performance parameter set includes: chip output voltage, sensitivity, linearity, natural frequency, maximum equivalent stress, thermal sensitivity drift and thermal zero drift.

5. The method according to claim 1, characterized in that Before constructing the numerical design library according to the design parameter set and the performance parameter set, the method further includes: Parameter standardization is performed on the design parameter set and the performance parameter set.

6. The method according to claim 1, characterized in that The step of establishing a bidirectional numerical design model based on a Bi-GRU neural network according to the numerical design library specifically includes: Build the Anaconda integrated development environment; Randomly sorting and dividing the data in the numerical design library; Import the pytorch learning framework; Build a forward numerical design model based on Bi-GRU neural network; Build a reverse numerical design model based on Bi-GRU neural network; Save the trained bidirectional numerical design model based on Bi-GRU neural network.

7. The method according to claim 6, characterized in that The random sorting and data division of the numerical design library specifically includes: Loading data of the numerical design library; Randomly sorting the data in the numerical design library; The randomly sorted data is divided into a training set and a test set according to a preset ratio.

8. The method according to claim 7, characterized in that The construction of the forward numerical design model based on the Bi-GRU neural network specifically includes: Setting model parameters, wherein the model parameters include: number of hidden neurons, learning rate, batch size, epoch value and multivariate input sequence length; Convert the design parameter data in the training set into a three-dimensional tensor as the input of the first layer of GRU forward neurons of the forward numerical design model; Flipping the designed parameter sequence as the input of the second layer of GRU reverse neurons; Seven fully connected layers are connected to the output layer of the neural network, and each layer corresponds to a prediction task of a performance parameter; The prediction results of 7 performance parameters are obtained based on the output of the fully connected layer; Calculate the loss function; Complete the training of the forward numerical design model based on the Bi-GRU neural network.

9. The method according to claim 7, characterized in that: The construction of the reverse numerical design model based on the Bi-GRU neural network specifically includes: Setting model parameters, wherein the model parameters include: number of hidden neurons, learning rate, batch size, epoch value and input sequence length; Converting the performance parameter data in the training set into a three-dimensional tensor as the input of the first layer of GRU forward neurons in the reverse numerical design model; Flipping the performance parameter sequence as input to the second layer of GRU reverse neurons; Seven fully connected layers are connected to the output layer of the neural network, and each layer corresponds to a prediction task of a design parameter; The prediction results of 7 design parameters are obtained based on the output of the fully connected layer; Calculate the loss function; Complete the reverse numerical design model training based on Bi-GRU neural network.

10. The method according to claim 7, characterized in that After completing the construction of the bidirectional numerical design model based on the Bi-GRU neural network, it also includes: Evaluate model performance based on the test set data; Adjust parameters or optimize model structure as needed.

11. The method according to claim 10, characterized in that The evaluating the model performance based on the test set data specifically includes: Using the design parameter data in the test set as model input to evaluate the effect of the forward numerical design model; The performance parameter data in the test set is used as model input to evaluate the effect of the reverse numerical design model.

12. The method according to claim 10, characterized in that The model evaluation indicators include: mean absolute error MAE, root mean square error RMSE and mean absolute percentage error MAPE; Among them, m is the number of chips of different sizes, x i Input data to the model, f(x i ) is the model prediction result, y i is the true value corresponding to the predicted result.

13. A chip design system for a silicon piezoresistive pressure sensor, characterized in that: include: Database module, design parameter module, performance parameter module and model evaluation module; The database module includes a design parameter set and a performance parameter set; the design parameter set includes: chip structure, chip size, diaphragm size, piezoresistive doping concentration, heavy doping concentration, chip operating temperature and pressure range; the performance parameter set includes: chip output voltage, sensitivity, linearity, natural frequency, maximum equivalent stress, thermal sensitivity drift and thermal zero drift; The design parameter module is used to construct a forward numerical design model based on the Bi-GRU neural network according to the database module, and obtain a chip performance parameter result prediction according to the design parameters; the performance parameter module is used to construct a reverse numerical design model based on the Bi-GRU neural network according to the database module, and obtain a chip design parameter result prediction according to the performance parameters; The model evaluation module is used to evaluate model performance and adjust parameters or optimize model structure as needed.

Images