BCD device process control method and device, storage medium and equipment

Abstract

The invention discloses a BCD device process control method and device, a storage medium and equipment. The control method comprises the following steps: collecting real-time fault detection and classification data in a BCD device technological process; inputting the collected fault detection and classification data into a preset deep reinforcement learning model to obtain adjusted process parameters; and performing real-time control on the technological process of the BCD device by using the adjusted technological parameters. By adopting the scheme of the invention, the fluctuation of the BCD process can be reduced, so that the stability of the performance of the BCD device is improved.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor technology, and in particular to a BCD device process control method, device, storage medium and equipment. Background Art

[0002] BCD devices are semiconductor devices manufactured using the BCD process (Bipolar-CMOS-DMOS). This process integrates three types of devices—bipolar transistors (bipolar), complementary metal-oxide semiconductor (CMOS), and double-diffused metal-oxide semiconductor (DMOS)—on the same chip. BCD devices are widely used in power management, motor drives, and RF power amplifiers.

[0003] With the further development of integrated circuit technology, the BCD process has evolved from the initial 4-micron technology to 0.11-micron technology. Line widths have continued to decrease, while more advanced multi-layer metal wiring systems have been adopted, increasing integration and narrowing the gap with pure CMOS processes. Furthermore, the BCD process is also moving towards modularization and standardization. By standardizing basic processes and combining hybrid processes, designers can flexibly adjust process steps to meet different application requirements.

[0004] However, using existing BCD process control methods to control the BCD process will cause process fluctuations, ultimately leading to unstable performance of the BCD device. Summary of the Invention

[0005] The problem to be solved by the present invention is to reduce the fluctuation of the BCD process to improve the stability of the BCD device performance.

[0006] To solve the above problems, an embodiment of the present invention provides a BCD device process control method, the method comprising:

[0007] Collect real-time fault detection and classification data during BCD device processing;

[0008] Input the collected fault detection and classification data into a pre-set deep reinforcement learning model to obtain adjusted process parameters;

[0009] The adjusted process parameters are used to control the BCD device process in real time.

[0010] In a possible embodiment, the deep reinforcement learning model includes: a policy network; the policy network includes: an input layer, two or more layers of deep neural network, and an output layer.

[0011] In one possible embodiment, each layer of the deep neural network is composed of multiple long short-term memory units.

[0012] In a possible embodiment, the deep reinforcement learning model further includes: a value network; the value network is obtained using a Q-learning algorithm.

[0013] In a possible embodiment, the deep reinforcement learning model is trained using the following method:

[0014] Establish an initial deep reinforcement learning model based on the preset deep reinforcement learning algorithm;

[0015] Acquire training samples, the training samples including: historical fault detection and classification data of a BCD device process and corresponding historical BCD device process result data, as well as simulated fault detection and classification data of a BCD device process and corresponding simulated BCD device process result data;

[0016] The initial deep reinforcement learning model is trained using the training samples to obtain the deep reinforcement learning model.

[0017] In a possible embodiment, the real-time fault detection and classification data includes:

[0018] Temperature data, current data, voltage data and humidity data.

[0019] In one possible embodiment, before inputting the collected fault detection and classification data into a preset deep reinforcement learning model, the process further includes:

[0020] The collected fault detection and classification data are preprocessed to obtain preprocessed data, so as to input the preprocessed data into a preset deep reinforcement learning model.

[0021] In a possible embodiment, the method further includes:

[0022] Obtain BCD device process result data;

[0023] The deep reinforcement learning model is adjusted using the BCD device process result data.

[0024] An embodiment of the present invention further provides a BCD device process control device, the device comprising:

[0025] Acquisition unit, suitable for collecting real-time fault detection and classification data during the BCD device process;

[0026] A parameter decision unit adapted to input the collected fault detection and classification data into a preset deep reinforcement learning model to obtain adjusted process parameters;

[0027] The control unit is adapted to use the adjusted process parameters to perform real-time control on the BCD device process.

[0028] In a possible embodiment, the deep reinforcement learning model includes: a policy network; the policy network includes: an input layer, two or more layers of deep neural network, and an output layer.

[0029] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of any of the above methods.

[0030] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor executes the steps of any of the above methods when running the computer program.

[0031] Compared with the prior art, the technical solution of the embodiment of the present invention has the following advantages:

[0032] By applying the solution of the present invention, real-time fault detection and classification data from the BCD device process is collected and input into a preset deep reinforcement learning model to obtain adjusted process parameters. The adjusted process parameters can then be used to control the BCD device process in real time, enabling real-time dynamic adjustment of the BCD device process. This allows for instant adaptation to various situations that arise during the production process, reducing fluctuations in the BCD process and improving the stability of BCD device performance. Compared to traditional process control methods, the solution of the present invention, through the intelligent decision-making capabilities of the deep reinforcement learning model, can more accurately predict and adjust process parameters, thereby significantly improving device performance consistency and production yield. Furthermore, the ability to automatically adjust process parameters not only reduces the burden of manual monitoring but also improves production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flow chart of a BCD device process control method according to an embodiment of the present invention;

[0034] Figure 2 This is a flowchart of a deep reinforcement learning model training method according to an embodiment of the present invention;

[0035] Figure 3 1 is a schematic structural diagram of a BCD device process control device according to an embodiment of the present invention;

[0036] Figure 4 Schematic diagram of a BCD device process control process in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The BCD process is a monolithic integration process capable of fabricating bipolar, CMOS, and DMOS devices on the same chip. STMicroelectronics pioneered the BCD process in 1985. The BCD process combines the high transconductance and strong load drive capability of bipolar devices, the high integration density and low power consumption of CMOS, and the high voltage and high current characteristics of DMOS, making it widely used in automotive electronics, industrial control, and consumer electronics.

[0038] With the further development of integrated circuit technology, the BCD process has evolved from the initial 4-micron technology to 0.11-micron technology. Line widths have continued to decrease, while more advanced multi-layer metal wiring systems have been adopted, increasing integration and narrowing the gap with pure CMOS processes. Furthermore, the BCD process is also moving towards modularization and standardization. By standardizing basic processes and combining hybrid processes, designers can flexibly adjust process steps to meet different application requirements.

[0039] The existing BCD process control method has the following characteristics:

[0040] 1) Multi-physics field control: The BCD process involves multiple physical effects, such as thermal effects, electromigration, breakdown voltage, etc. The BCD process control method can simultaneously consider and control the interactions of these multiple physical fields.

[0041] 2) High-precision process parameter control: To ensure device performance and consistency, the BCD process control method can accurately control process parameters such as doping concentration, oxide layer thickness, and photolithography accuracy.

[0042] 3) Reliability and stability: The BCD process control method can ensure long-term process stability and reliability to reduce device defects and improve yield.

[0043] 4) Automation and informatization: The BCD process control method is highly automated and informatized, using advanced control software and hardware systems to achieve precise control of the process.

[0044] The complexity of the BCD process leads to high manufacturing costs and difficulty in process control. Especially in mass production, issues such as N-junction leakage, DMOS leakage, and Schottky diode leakage seriously affect product reliability and yield. These issues are often related to process tolerance and need to be resolved through process parameter optimization.

[0045] With the development of technology, the BCD process is facing the challenge of how to maintain high consistency and high yield during the process while achieving higher integration.

[0046] Existing BCD process control methods primarily rely on empirically based settings and static models, which are insufficient to handle the dynamic changes inherent in the BCD process. Furthermore, due to the complexity of the BCD process, static control strategies cannot flexibly adapt to the various changes that occur during production. This can easily lead to process fluctuations, which in turn affect the stability of BCD device performance and production yield.

[0047] To address this issue, the present invention provides a BCD device process control method. This method, by acquiring real-time fault detection and classification data during the BCD device process, enables real-time control of the BCD device process. This allows for real-time dynamic adjustment of the BCD device process, enabling immediate adaptation to various production scenarios, reducing BCD process fluctuations and improving BCD device performance stability. Furthermore, by utilizing a deep reinforcement learning model to adjust process parameters, the accuracy of BCD device process control can be improved, significantly enhancing device performance consistency and production yield.

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0049] Reference Figure 1 , an embodiment of the present invention provides a BCD device process control method, the method may include:

[0050] Step 11: Collect real-time fault detection and classification data during the BCD device process.

[0051] In practice, fault detection and classification data refers to data collected and processed by the Fault Detection and Classification (FDC) system during semiconductor manufacturing. The FDC system is used to monitor and analyze equipment operating data in real time during the production process. The FDC system can obtain equipment operating status, enabling real-time monitoring and rapid identification of anomalies. The FDC system can also collect equipment sensor data to analyze the causes of failures.

[0052] In a specific implementation, the fault detection and classification data may include temperature data, current data, voltage data, and humidity data during the BCD device process, etc. The fault detection and classification data may be used to adjust the BCD device process parameters.

[0053] Step 12: Input the collected fault detection and classification data into a preset deep reinforcement learning model to obtain adjusted process parameters.

[0054] In practice, a deep reinforcement learning model is built using a deep reinforcement learning algorithm. Deep reinforcement learning (DRL) is a machine learning algorithm that combines deep neural networks (DNNs) with reinforcement learning (RL). In DRL, an intelligent agent learns the optimal action strategy by interacting with its environment. DRL offers better convergence and accuracy than traditional RL, and can better capture nonlinear relationships between process parameters.

[0055] In a specific implementation, the deep reinforcement learning model includes a policy network and a value network. The policy network may include an input layer, two or more layers of deep neural networks, and an output layer. The collected fault detection and classification data is input into the input layer and then output to the multi-layer deep neural network via the input layer. The output of the previous layer of the deep neural network serves as the input to the next layer of the deep neural network, and the output of the last layer of the deep neural network is input to the output layer, which can output the adjusted process parameters. The value network is used to evaluate the process parameter prediction results of the policy network.

[0056] In one embodiment, each layer of the deep neural network is composed of multiple long short-term memory (LSTM) units. LSTM units can capture long-term dependencies in the time series of fault detection and classification data, thereby enabling more accurate determination of process parameters based on changes in historical fault detection and classification data and current data.

[0057] Specifically, multiple LSTM units process the input sequence sequentially, in time steps. Each LSTM unit accepts three inputs: the hidden state at the previous moment, the state of the LSTM unit at the previous moment, and the input at the current moment. Each LSTM unit can implement three gate calculations: the forget gate, the input gate, and the output gate. The forget gate determines how much information from the previous unit state is retained in the current unit state. The input gate determines how much information from the current input is written to the current unit state. The output gate determines how much output the current unit state has.

[0058] In a specific implementation, the collected fault detection and classification data can be preprocessed before being input into a preset deep reinforcement learning model to obtain preprocessed data, which can then be input into the preset deep reinforcement learning model. For example, advanced data processing algorithms such as wavelet denoising and principal component analysis can be used to denoise, normalize, and extract features from the collected fault detection and classification data to ensure data accuracy and consistency, providing reliable input for the deep reinforcement learning model.

[0059] In practice, preprocessed fault detection and classification data is fed into a trained policy network. Based on changes between historical and current fault detection and classification data, the policy network intelligently identifies and analyzes complex patterns and trends within the data, providing more precise process control strategies. This refined control approach helps reduce variation in the production process, ensuring that every product meets design specifications and significantly improving overall product consistency.

[0060] Step 13: Use the adjusted process parameters to control the BCD device process in real time.

[0061] In a specific implementation, the adjusted process parameters refer to process parameters that can be adjusted externally during the BCD device process, including exposure energy, doping gas concentration, ion implantation dose, etc.

[0062] In some embodiments, the adjusted process parameters may be directly output to a BCD device process tool, which then decides whether to adjust the BCD device process.

[0063] In some embodiments, after obtaining the adjusted process parameters, the adjusted process parameters may be compared with the current process parameters to output a process parameter adjustment instruction. The process parameter adjustment instruction may include information indicating whether the process parameters should be adjusted. When instructing to adjust the process parameters, the process parameter adjustment instruction may also include parameter adjustment details. The parameter adjustment details are used to indicate which process parameters to adjust and how to adjust them, including the adjustment range. The process parameter adjustment instruction may then be output to the BCD device process equipment, allowing the BCD device process equipment to directly adjust the BCD device process based on the process parameter adjustment instruction.

[0064] In one embodiment of the present invention, the method may further include: obtaining BCD device process result data, and adjusting the deep reinforcement learning model using the BCD device process result data.

[0065] By using BCD device process data, we can obtain real-time process results when controlling the process parameters determined by the deep reinforcement learning model. These process results are then fed back into the deep reinforcement learning model, continuously optimizing the deep reinforcement learning model's policy network to further improve process control accuracy. In this way, the system forms a process optimization loop through real-time data collection and feedback, continuously optimizing the deep reinforcement learning model's policy network to gradually improve process stability and product yield.

[0066] The embodiment of the present invention also provides a training method for a deep reinforcement learning model, referring to Figure 2 , the method may include the following steps:

[0067] Step 21: Establish an initial deep reinforcement learning model based on a preset deep reinforcement learning algorithm.

[0068] Deep reinforcement learning algorithms are machine learning algorithms that combine deep neural networks with reinforcement learning. The training process of deep reinforcement learning models involves the interaction between the agent and the environment, and maximizing cumulative rewards by optimizing the strategy or value function.

[0069] Therefore, the initial deep reinforcement learning model includes an initial policy network and an initial value network. The initial policy network can include an input layer, two or more deep neural network layers, and an output layer. The initial value network can be derived using a Q-learning algorithm and has a structure consisting of an input layer, a hidden layer, and an output layer (e.g., a Q value).

[0070] Subsequently, the initial strategy network and the initial value network are continuously trained using the training samples until the strategy network parameters and the value network parameters meet the preset stability conditions, and the final strategy network and the final value network are obtained, so as to use the final strategy network to predict the process parameters of BCD devices.

[0071] Step 22: Obtain training samples.

[0072] The training samples include: historical fault detection and classification data of the BCD device process and corresponding historical BCD device process result data, as well as simulated fault detection and classification data of the BCD device process and corresponding simulated BCD device process result data.

[0073] Based on historical data and simulation data, the initial deep reinforcement learning model is continuously optimized through repeated process simulation and actual production data training to adapt to dynamic process changes.

[0074] Step 23: Use the training samples to train the initial deep reinforcement learning model to obtain the deep reinforcement learning model.

[0075] Specifically, a set of fault detection and classification data at the same moment is input into the initial policy network to obtain initial prediction parameters. The value network is then used to calculate the Q value or state value of the initial prediction parameters. The parameters of the initial value network are then updated based on the error between the actual value and the prediction parameter. The updated initial value network is then used to evaluate the value of the initial policy network's output. The policy gradient update algorithm provides feedback to the policy network and updates the policy network's parameters. This cycle continues until the policy network and value network parameters converge and meet the preset stability conditions. The final policy network and value network are then obtained, which can then be used to predict BCD device process parameters.

[0076] As can be seen from the above, the solution of the present invention collects fault detection and classification data in real time and inputs it into a deep reinforcement learning model, which then outputs the optimal process parameter adjustment strategy. According to the strategy of the deep reinforcement learning model, the process parameters are adjusted in real time to ensure the stability of device performance during the production process. In addition, the solution of the present invention also has a real-time monitoring and feedback mechanism that can instantly evaluate the process effect and feed this information back to the deep reinforcement learning model to achieve continuous adjustment and optimization of the process control strategy.

[0077] In order to enable those skilled in the art to better understand and implement the present invention, the apparatus, test system, electronic device and computer-readable storage medium corresponding to the above method are described in detail below.

[0078] Reference Figure 3 The embodiment of the present invention further provides a BCD device process control device 30, which may include: a collection unit 31, a parameter decision unit 32, and a control unit 33.

[0079] The acquisition unit 31 is adapted to acquire real-time fault detection and classification data during the BCD device process;

[0080] The parameter decision unit 32 is adapted to input the collected fault detection and classification data into a preset deep reinforcement learning model to obtain adjusted process parameters;

[0081] The control unit 33 is adapted to use the adjusted process parameters to perform real-time control on the BCD device process.

[0082] In one embodiment of the present invention, the deep reinforcement learning model includes: a policy network and a value network; the policy network includes: an input layer, two or more layers of deep neural network and an output layer, and the value network includes: an input layer, a hidden layer and an output layer.

[0083] Figure 4FIG1 is a flow chart of the process control process of a BCD device in one embodiment of the present invention. Figure 4 The fault detection and classification data collected by the acquisition unit (such as temperature, humidity, current, and voltage data from the BCD device process) undergoes data preprocessing to ensure accuracy and consistency. This preprocessed data is then fed into a pre-set deep reinforcement learning model to generate a process parameter adjustment strategy, which includes the adjusted process parameter information. The control unit then uses this process parameter adjustment strategy to control the BCD device process in real time.

[0084] Regarding the acquisition unit 31 , the parameter decision unit 32 and the control unit 33 , they can be specifically implemented with reference to the above description of steps 11 to 13 , which will not be repeated here.

[0085] The BCD device process control device 30 in this embodiment of the present invention is capable of responding to process changes in real time, enabling immediate adaptation to various situations arising during the production process. Compared to traditional process control solutions, the BCD device process control device 30 leverages the intelligent decision-making capabilities of a deep learning model to more accurately predict and adjust process parameters, significantly improving device performance consistency and production yield. This automated process parameter adjustment capability not only reduces the burden of manual monitoring but also improves production efficiency and product quality.

[0086] In addition, the BCD device process control device 30 in the embodiment of the present invention also provides a real-time monitoring and feedback mechanism. Through this mechanism, the BCD device process control device 30 can continuously collect data from the production process, evaluate the current process effects, and feed this information back to the deep reinforcement learning model. This closed-loop feedback system enables the deep learning model to learn and adapt to new process conditions, continuously optimize the process control strategy, and ensure that the production process is always in an optimal state. This intelligent control strategy optimization not only improves the flexibility and adaptability of production, but also provides strong technical support for the realization of intelligent manufacturing and industrial automation.

[0087] Subsequently, the BCD device process control device 30 can be integrated into the production line, so that the process control system can achieve real-time monitoring and instant feedback, which not only improves the intelligence and automation level of the production process, but also significantly improves production efficiency and product quality.

[0088] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of any of the above methods.

[0089] In a specific implementation, the computer-readable storage medium may include: ROM, RAM, magnetic disk or optical disk, etc.

[0090] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor executes the steps of any of the above methods when running the computer program.

[0091] Regarding the various modules / units included in the various devices and products described in the above embodiments, they can be software modules / units, hardware modules / units, or partly software modules / units and partly hardware modules / units. For example, for various devices and products applied to or integrated into a chip, the various modules / units included therein can all be implemented in the form of hardware such as circuits, or at least part of the modules / units can be implemented in the form of software programs, which run on the processor integrated inside the chip, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits; for various devices and products applied to or integrated into a chip module, the various modules / units included therein can all be implemented in the form of hardware such as circuits, and different modules / units can be located in the same component (such as a chip, circuit module, etc.) or different components of the chip module, or at least part of the modules / units can be implemented in the form of hardware such as circuits. The element can be implemented in the form of a software program, which runs on the processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits; for various devices and products applied to or integrated in the terminal, the various modules / units contained therein can all be implemented in the form of hardware such as circuits, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or different components in the terminal, or, at least some modules / units can be implemented in the form of a software program, which runs on the processor integrated inside the terminal, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits.

[0092] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope defined by the claims.

Claims

1. A BCD device process control method, characterized in that: include: Collect real-time fault detection and classification data during BCD device processing; Input the collected fault detection and classification data into a pre-set deep reinforcement learning model to obtain adjusted process parameters; The adjusted process parameters are used to control the BCD device process in real time.

2. The BCD device process control method according to claim 1, wherein: The deep reinforcement learning model includes: a policy network; the policy network includes: an input layer, two or more layers of deep neural network and an output layer.

3. The BCD device process control method according to claim 2, wherein: Each layer of a deep neural network consists of multiple long short-term memory units.

4. The BCD device process control method according to claim 2, wherein: The deep reinforcement learning model also includes: a value network; the value network is obtained using the Q-learning algorithm.

5. The BCD device process control method according to claim 3, wherein: The deep reinforcement learning model is trained using the following method: Establish an initial deep reinforcement learning model based on the preset deep reinforcement learning algorithm; Acquire training samples, the training samples including: historical fault detection and classification data of a BCD device process and corresponding historical BCD device process result data, as well as simulated fault detection and classification data of a BCD device process and corresponding simulated BCD device process result data; The initial deep reinforcement learning model is trained using the training samples to obtain the deep reinforcement learning model.

6. The BCD device process control method according to claim 1, wherein: The real-time fault detection and classification data includes: Temperature data, current data, voltage data and humidity data.

7. The BCD device process control method according to claim 1, wherein: Before inputting the collected fault detection and classification data into the preset deep reinforcement learning model, the following steps are also included: The collected fault detection and classification data are preprocessed to obtain preprocessed data, so as to input the preprocessed data into a preset deep reinforcement learning model.

8. The BCD device process control method according to claim 1, wherein: Also includes: Obtain BCD device process result data; The deep reinforcement learning model is adjusted using the BCD device process result data.

9. A BCD device process control device, characterized in that: include: Acquisition unit, suitable for collecting real-time fault detection and classification data during the BCD device process; A parameter decision unit adapted to input the collected fault detection and classification data into a preset deep reinforcement learning model to obtain adjusted process parameters; The control unit is adapted to use the adjusted process parameters to perform real-time control on the BCD device process.

10. The BCD device process control device according to claim 9, wherein: The deep reinforcement learning model includes: a policy network; the policy network includes: an input layer, two or more layers of deep neural network and an output layer.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement the steps of the method according to any one of claims 1 to 8.

12. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor runs the computer program, the steps of the method according to any one of claims 1 to 8 are performed.

Images