AlGaN solar blind APD performance prediction and optimization method and system

Abstract

The invention provides an AlGaN solar blind APD performance prediction and optimization method and system, and relates to the technical field of photoelectric devices, and the method comprises the steps: obtaining the experimental data information of an AlGaN solar blind APD; based on a deep learning algorithm, performing model training by taking the process parameters as input and the device performance parameters as output, and constructing a performance prediction model; obtaining a performance prediction result of the target device through training, and generating an optimization scheme including material component adjustment parameters, device structure improvement parameters and process condition optimization parameters; the optimization scheme is subjected to experimental verification, the optimization effect is verified, an evaluation result is obtained, and after the evaluation is qualified, the optimization scheme is fed back to the performance prediction model for model updating, so that the technical problems that the performance optimization of the AlGaN solar blind APD is low in efficiency, not accurate enough and difficult to adapt to complex process conditions are solved; the technical effects of fast and accurate performance prediction and automatic optimization, and improvement of optimization efficiency and working condition adaptability are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of optoelectronic devices, and in particular to a method and system for predicting and optimizing the performance of an AlGaN solar-blind APD. Background Art

[0002] With the rapid development of optoelectronic technology, AlGaN materials have shown great application potential in the field of ultraviolet detection due to their unique physical and chemical properties, especially detectors in the day-blind ultraviolet band (200-280nm), which have broad application prospects in environmental monitoring, medical imaging and other fields. However, the performance of AlGaN day-blind APD (avalanche photodiode) is affected by many factors. Traditional design and optimization methods require a lot of experiments and trial and error, relying on gradual experimental adjustments, with a long cycle and difficulty in obtaining the best process parameter combination, resulting in low efficiency and waste of resources. Therefore, how to optimize the performance of AlGaN day-blind APD devices under complex process conditions has become a problem that needs to be solved urgently.

[0003] In summary, the prior art has technical problems such as low efficiency, inaccuracy in performance optimization of AlGaN day-blind APD, and difficulty in adapting to complex process conditions. Summary of the invention

[0004] The present application provides a method and system for predicting and optimizing the performance of AlGaN day-blind APD, which is used to solve the technical problems in the prior art that the performance optimization of AlGaN day-blind APD is inefficient, inaccurate, and difficult to adapt to complex process conditions.

[0005] In view of the above problems, the present application provides a method and system for predicting and optimizing the performance of an AlGaN solar-blind APD.

[0006] In a first aspect, the present application provides a method for predicting and optimizing the performance of an AlGaN solar-blind APD, the method comprising:

[0007] Acquire experimental data information of AlGaN day-blind APD, the experimental data information including process parameters and device performance parameters, wherein the process parameters include material composition, device structure and process conditions, and the device performance parameters include dark current, quantum efficiency and spectral responsivity; based on a deep learning algorithm, use the process parameters as model input and the device performance parameters as model output, perform model training and build a performance prediction model; use the performance prediction model to perform performance prediction and obtain performance prediction results of a target device; based on the performance prediction results, generate an optimization scheme for the target device, the optimization scheme including material composition adjustment parameters, device structure improvement parameters and process condition optimization parameters; experimentally verify the optimization scheme, verify the optimization effect, and obtain an evaluation result. When the evaluation result is qualified, use the optimization scheme as the optimization result and feed it back to the performance prediction model for model updating.

[0008] In a second aspect, the present application provides an AlGaN solar-blind APD performance prediction and optimization system, the system comprising:

[0009] An experimental data acquisition module is used to obtain experimental data information of AlGaN day-blind APD, the experimental data information includes process parameters and device performance parameters, wherein the process parameters include material composition, device structure and process conditions, and the device performance parameters include dark current, quantum efficiency and spectral responsivity; a model construction module is used to train the model based on a deep learning algorithm, using the process parameters as model input and the device performance parameters as model output, and to construct a performance prediction model; a performance prediction module is used to use the performance prediction model to perform performance prediction and obtain performance prediction results of a target device; an optimization scheme acquisition module is used to generate an optimization scheme for the target device based on the performance prediction results, the optimization scheme includes material composition adjustment parameters, device structure improvement parameters and process condition optimization parameters; a model update module is used to experimentally verify the optimization scheme, verify the optimization effect, and obtain an evaluation result. When the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model update.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The AlGaN day-blind APD performance prediction and optimization method provided in the present application obtains experimental data information of the AlGaN day-blind APD, wherein the experimental data information includes process parameters and device performance parameters, wherein the process parameters include material composition, device structure and process conditions, and the device performance parameters include dark current, quantum efficiency and spectral responsivity; based on a deep learning algorithm, the process parameters are used as model inputs and the device performance parameters are used as model outputs to perform model training and construct a performance prediction model; the performance prediction model is used to perform performance prediction and obtain performance prediction results of a target device; based on the performance prediction results, an optimization scheme for the target device is generated, wherein the optimization scheme includes material composition adjustment parameters, device structure improvement parameters and process condition optimization parameters; the optimization scheme is experimentally verified to verify the optimization effect and obtain an evaluation result; when the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model updating, thereby solving the technical problems in the prior art that the performance optimization efficiency of the AlGaN day-blind APD is low, inaccurate and difficult to adapt to complex process conditions, and achieves fast and accurate performance prediction and automatic optimization, and improves the technical effect of optimization efficiency and working condition adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic flow chart of the AlGaN solar-blind APD performance prediction and optimization method is provided for this application.

[0013] Figure 2 A schematic diagram of the structure of an AlGaN solar-blind APD performance prediction and optimization system is provided for this application.

[0014] Figure 3 A schematic diagram of the BOOST boost circuit structure is provided for this application.

[0015] Figure 4 A structural schematic diagram of a BOOST boost circuit when the switch tube is turned on is provided for this application.

[0016] Figure 5 A structural schematic diagram of a BOOST boost circuit when the switch tube is turned off is provided for this application.

[0017] Figure 6 A MAX1932 schematic is provided for this application.

[0018] Figure 7 A schematic diagram of the structure of a transimpedance amplifier circuit is provided for this application.

[0019] Explanation of the reference numerals: experimental data acquisition module 11, model building module 12, performance prediction module 13, optimization solution acquisition module 14, model updating module 15. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0021] Embodiment 1, as Figure 1 As shown, the present application provides a method for predicting and optimizing the performance of AlGaN solar-blind APD, the method comprising:

[0022] Acquire experimental data information of AlGaN day-blind APD, wherein the experimental data information includes process parameters and device performance parameters, wherein the process parameters include material composition, device structure and process conditions, and the device performance parameters include dark current, quantum efficiency and spectral responsivity.

[0023] Specifically, firstly, the experimental data information of AlGaN day-blind APD is obtained. This step is to ensure that the performance of the device can be accurately predicted and optimized. The experimental data information mainly includes process parameters and device performance parameters, wherein the process parameters refer to all key variables involved in the process of manufacturing AlGaN day-blind APD, specifically including material composition, device structure and process conditions. The material composition refers to the proportion of aluminum (Al), gallium (Ga) and nitrogen (N) constituting AlGaN material, as well as other possible doping elements added to achieve specific electrical or optical properties; the device structure involves the design of each layer structure of APD, such as the specific composition and thickness of the active layer, charge layer and absorption layer, which determine the final performance of the device; the process conditions include key parameters such as temperature, pressure, gas flow rate and deposition time in the manufacturing process, which directly affect the quality of the material and the consistency of the device. The experimental data information also includes device performance parameters, which are key indicators for evaluating the performance of AlGaN day-blind APD. The device performance parameters mainly include dark current, quantum efficiency and spectral responsivity. Among them, the dark current refers to the current generated by the device under no light conditions, which is an important indicator for evaluating the noise level of the device and is directly related to the sensitivity and signal-to-noise ratio of the APD; the quantum efficiency describes the efficiency of the APD in converting incident photons into electronic signals, and is the core parameter for measuring the photoelectric conversion capability of the device; and the spectral responsivity indicates the device's ability to respond to light of a specific wavelength, which determines the application performance of the APD in different spectral regions.

[0024] By collecting these experimental data information, the necessary input parameters can be provided for subsequent performance prediction models for subsequent model training and performance prediction, thereby improving device performance and meeting predetermined application standards.

[0025] Based on a deep learning algorithm, the process parameters are used as model inputs, and the device performance parameters are used as model outputs, and model training is performed to construct a performance prediction model.

[0026] The performance prediction model is used to perform performance prediction and obtain performance prediction results of the target device.

[0027] In order to optimize the performance of AlGaN day-blind APD, a performance prediction model is constructed based on a deep learning algorithm. First, the collected process parameters are used as the input of the model. These process parameters include but are not limited to material composition, device structure and process conditions in the manufacturing process, such as temperature, pressure, gas flow rate and deposition time. These parameters directly affect the physical and chemical properties of the device, and they have a decisive influence on the final device performance. Subsequently, the model outputs device performance parameters, which include dark current, quantum efficiency and spectral responsivity. During the model training process, by using a large number of experimental data sets, the model can learn the complex nonlinear relationship between process parameters and performance output. This learning is achieved by constructing a multi-layer neural network, which includes an input layer, multiple hidden layers and an output layer. Each layer contains several neurons, which process data through activation functions to simulate the nonlinear data mapping process. Specific model training also involves feature extraction, which converts important features in the process parameters into a format that the model can effectively process. Then, these feature data are used for iterative training of the deep learning model. Commonly used training algorithms include back propagation and gradient descent, which help optimize model weights and reduce prediction errors.

[0028] After the model training is completed, the performance prediction model can be used to predict the performance of the target device. In the performance prediction stage, new or unseen process parameters are input into the model, and the model will output predicted performance parameters, such as dark current, quantum efficiency, and spectral responsivity. This prediction result can be directly used to evaluate the performance of the device in future production, and can also be used as a basis for further optimizing process parameters. In addition, this deep learning-based prediction model has the characteristics of self-learning and self-adaptation, and can continuously optimize its prediction accuracy as the amount of data increases, thereby showing higher efficiency and reliability in long-term applications. Through this method, not only the accuracy of the performance prediction of AlGaN day-blind APD is significantly improved, but also scientific data support is provided for the manufacturing and optimization of devices.

[0029] In order to achieve accurate testing of APD (avalanche photodiode) performance parameters, the present application may also have a set of dedicated hardware circuit systems. The system mainly includes a BOOST boost circuit, a transimpedance amplifier circuit (TIA) and an ADC data acquisition module, which together provide a stable working environment and accurate data acquisition capabilities for the APD. APD usually requires a higher reverse bias voltage to achieve the avalanche multiplication effect, so a stable boost circuit is an indispensable part of the test system. The present application adopts a BOOST boost circuit solution, which controls the on and off of switching elements (such as MOSFET) through PWM (pulse width modulation) to convert a lower input voltage into a higher output voltage required by the APD. And in order to more accurately control the output voltage, the present application uses a MAX1932 digitally controlled high-voltage DC-DC boost converter. The chip has a built-in DAC (digital to analog converter) and can be programmed through the SPI interface to achieve accurate adjustment of the output voltage. By adjusting the resistor divider network, the output voltage range can be set to meet the needs of different models of APD. Ensure that the selection of inductors, diodes and capacitors meets the recommended values ​​in the chip data sheet to optimize conversion efficiency and stability. Feedback control using a thermistor or temperature sensor is used to compensate for the effects of temperature changes on the output voltage.

[0030] Among them, the BOOST boost circuit, also known as a boost converter, is a DC-DC converter widely used in power supply technology. Its main function is to convert the input DC voltage into a higher DC voltage. Figure 3 As shown in the figure, the core components of the BOOST circuit include an inductor, a diode, a switch element (such as a transistor or MOSFET) and a capacitor. If the switch tube is not controlled for a long time, it is assumed that all components are in an ideal state, so Uo = Ui. Figure 4 As shown in the figure, when the switch is turned on, the current in the inductor increases linearly. The self-inductance of the inductor hinders the current from rising, and the inductor converts the electrical energy into magnetic energy for storage. The role of the diode is to prevent the capacitor from discharging to the ground when the switch is turned on. When the switch is turned off, Figure 5 As shown, the current of the inductor will begin to decrease slowly. Since the self-inductance hinders the reduction of current, the left side of the inductor is negative and the right side is positive, so the voltage at the output end becomes Uo=Ui+UL. The output voltage is greater than the input voltage. For the inductor:

[0031] U on =U i (1)

[0032] U off =U o -Ui(2)

[0033] According to the volt-second balance principle of inductance: the positive volt-second value across the inductance in a steady-state switching power supply is equal to the negative volt-second value.

[0034] Right now:

[0035] U on T on =U off T off (3)

[0036] have to:

[0037] U i *T D =(U o -U i )*(T s -T D ) (4)

[0038] Simplifying, we can get:

[0039]

[0040] From formula (5), we can see that since the PWM duty cycle D<1, the output voltage U O Greater than the input voltage Ui, and the output voltage can be controlled by changing the PWM duty cycle.

[0041] Among them, the MAX1932 chip is a digitally controlled high-voltage DC-DC boost converter launched by ADI, which is designed to provide the necessary working bias voltage for the avalanche photodiode in the optical receiver. The constant frequency width modulation (PWM) boost topology and unique design architecture adopted by the MAX1932 chip enable it to generate low-noise and high-voltage outputs, and achieve extremely low output ripple. In addition, the architecture incorporates an optional RC or LC post-filter in the feedback loop, thereby ensuring high accuracy and stability of the output performance.

[0042] Internally, the MAX1932 is equipped with a digital-to-analog converter (DAC) for precise control of the output voltage of the DC-DC converter. The output voltage of the DAC is adjusted via the SPI serial interface and configured using an 8-bit control byte. When the device is powered on, the DAC defaults to FF hexadecimal (equivalent to 1.25V), which represents the minimum output voltage of the boost converter. The output voltage is adjusted by coupling the voltage generated by DACOUT with FB through resistor R6, while DACOUT itself only supports a maximum current of 50μA. The schematic diagram of the MAX1932 is shown below. Figure 6 shown.

[0043] To achieve the desired voltage adjustment range, the minimum (VOUTFF) and maximum (VOUT01) output voltages need to be set. Resistor R5 sets the output voltage adjustment range using the following formula:

[0044] R5=(V OUT01 -V OUTFF )(R6 / 1.25V) (6)

[0045] At the same time, the R8 resistor is used to determine the lowest output of the adjustment range, and the calculation formula is:

[0046] R8=(1.25V×R5) / (V OUTFF ) (7)

[0047] In addition, the MAX1932 monitors the output current through resistor R1 connected between CS+ and CS-, where the typical current-limit threshold (I LIMIT ) is set to 1mA, and the value of R1 is calculated by the formula:

[0048]

[0049] The purpose of the CL pin is to indicate that the APD has reached the avalanche state through a low-level signal, thereby helping to find the best bias point for the APD. To reduce noise interference, the CL signal only changes state at the edge of the internal oscillator.

[0050] Since the gain of the APD may change with temperature, it is necessary to adjust the bias voltage appropriately to compensate for this gain change. In practical applications, the adjustment of the MAX1932 output voltage can be achieved through an on-chip DAC controlled by system software, or it can be automatically adjusted in hardware through an external thermistor or temperature sensor to ensure stable operation and optimal performance of the device under different temperature conditions.

[0051] After receiving the optical signal, the APD will generate a weak current signal. In order to accurately measure these signals, a high-gain, low-noise transimpedance amplifier (TIA) is required. The transimpedance amplifier converts the current signal into a voltage signal, which is convenient for subsequent analog-to-digital conversion and processing. This application selects the LTC6268-10 as the core operational amplifier of the TIA. The chip has extremely low input bias current and input offset voltage, which is very suitable for amplifying weak signals. Through reasonable circuit design, it can be ensured that the TIA maintains a stable gain and a low noise figure over a wide frequency range. Low-noise, high-precision resistors are used as feedback elements to improve the accuracy of the amplifier circuit, and shielding and grounding techniques are used to reduce external interference and ensure signal integrity.

[0052] Among them, Figure 7As shown in the figure, the transimpedance amplifier circuit can convert weak current signals into voltage signals, and its main advantages include providing high gain and low noise. This amplifier circuit is particularly suitable for processing the current signal output by the APD because it can effectively convert weak signals into a voltage form that is easier to measure, thereby ensuring the detectability of the signal without being masked by system noise. In the application, the input of the transimpedance amplifier circuit can be directly connected to the output of the APD. The current output of the APD is converted into a corresponding voltage output after passing through the transimpedance amplifier circuit, and the voltage signal can then be collected by the ADC. At the same time, when designing a signal processing system for an APD, the selection of a transimpedance amplifier circuit as a preamplifier circuit is the key to improving signal quality and measurement accuracy. A suitable transimpedance amplifier circuit can not only effectively improve the signal-to-noise ratio, but also ensure the responsiveness and reliability of the system to weak signals.

[0053] The voltage signal after TIA amplification needs to be converted into a digital signal through an ADC (analog-to-digital converter) for subsequent data processing and analysis. This application selects a high-performance 24-bit Δ-Σ ADC chip, such as ADS127L11, to achieve high-precision data acquisition. The chip provides high dynamic range and low noise performance, suitable for accurate measurement of weak signals. The common-mode noise can be further suppressed and the signal-to-noise ratio can be improved through the differential input mode. The programmable gain amplifier (PGA) and filter are integrated inside the chip, and the gain and bandwidth can be adjusted as needed. A high-precision reference voltage source is used to provide a stable reference voltage for the ADC, and an appropriate filtering circuit is added to the ADC input to reduce high-frequency noise interference, and the ADC parameters are configured through the SPI interface to achieve flexible data acquisition control.

[0054] Among them, the key performance parameters of ADS127L11 are as follows:

[0055] ADS127L11 performance parameters

[0056]

[0057] ADS127L11 supports single-ended input and differential input, which can effectively suppress common-mode noise and improve the dynamic range of the signal. In order to achieve flexible switching of input modes, resistor shorting is used in the design, which allows the input configuration to be freely converted between differential and single-ended modes according to actual needs. In differential input mode, the two input terminals receive the IN+ and IN- of the test signal respectively; while in single-ended mode, one input terminal receives the test signal and the other is grounded. In order to cope with the uncertainty of the power supply, the design also includes a positive and negative power supply switching mechanism. The high-precision voltage reference source of the REF34 series is used to provide a stable reference voltage of 2.5V and 4.096V. By using an inverting amplifier, the positive 2.5V power supply is converted to a negative power supply of -2.5V, thereby meeting the ADC's requirements for positive and negative power supplies and ensuring the stability of the ADC. In the process of ADS127L11 circuit design, special attention was paid to the protection of the decoupling signal path of the power supply. Decoupling capacitors are arranged near the power pins of the ADC to stabilize the power supply and reduce the ripple noise of the input power supply. Through the above design, this system can not only efficiently process the IV conversion output signal, but also adapt to the changing measurement conditions and power supply configuration, ensuring high flexibility and high reliability of data acquisition. These design schemes make full use of the performance characteristics of ADS127L11 and ensure the stability and reliability of the system in the test environment.

[0058] Through the design and implementation of the above hardware circuits, this application provides a stable and accurate working environment for the APD test system. The BOOST boost circuit ensures that the APD can work stably at the required high voltage; the transimpedance amplifier circuit effectively amplifies the weak current signal generated by the APD; the ADC data acquisition module converts these analog signals into digital signals, providing a basis for subsequent data processing and analysis. The design and optimization of these hardware circuits provide strong support for the accurate testing of APD performance.

[0059] Based on the performance prediction results, an optimization scheme for the target device is generated, wherein the optimization scheme includes material component adjustment parameters, device structure improvement parameters, and process condition optimization parameters.

[0060] Optionally, based on the performance prediction results obtained by the performance prediction model, the system automatically generates an optimization scheme for the target device. This optimization scheme comprehensively considers material composition adjustment parameters, device structure improvement parameters and process condition optimization parameters, aiming to maximize the overall performance of the AlGaN day-blind APD. Among them, the material composition adjustment parameters involve fine adjustment of the ratio of aluminum (Al), gallium (Ga) and nitrogen (N) in the AlGaN day-blind APD. This step is based on the results predicted by the model, such as the relationship between quantum efficiency and spectral responsivity and specific material components, to determine the optimal element ratio. For example, if the prediction results show that increasing the proportion of gallium can improve quantum efficiency, the material ratio is adjusted accordingly in the optimization scheme. The device structure improvement parameters include optimizing the thickness, type and order of each layer of the APD. These structural parameters directly affect the photoelectric conversion efficiency and response speed of the device. For example, by increasing the thickness of the absorption layer, the absorption of the device to light of a specific wavelength can be enhanced, thereby improving the spectral responsivity. The process condition optimization parameters focus on the temperature, gas flow rate, pressure and deposition time in the manufacturing process. The optimization of these parameters is based on the model's in-depth learning of the relationship between process conditions and device performance. For example, if the model predicts that lower deposition temperature helps reduce dark current, the corresponding optimization scheme will recommend adjusting the deposition temperature during production.

[0061] The optimization scheme is experimentally verified to check the optimization effect and obtain an evaluation result. When the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model updating.

[0062] Specifically, experimental verification of the generated optimization scheme is a key step to ensure its effectiveness and practicality. This process involves specific experimental operations on the adjustments of material components, device structures and process conditions proposed in the optimization scheme, and evaluating the actual impact of these adjustments on the performance of AlGaN day-blind APD devices. First, clear test standards and evaluation indicators are set. These usually include key performance parameters such as dark current, quantum efficiency and spectral responsivity. By comparing with the performance parameters predicted by the model, the optimization effect in the experiment can be objectively evaluated. In specific operations, the material component ratio of the device is adjusted according to the optimization scheme, the layer thickness and material type of the device structure are changed, and the process conditions such as temperature and pressure in the manufacturing process are adjusted. After the adjustment, a new device sample will be manufactured and a performance test will be carried out under standard test conditions. The test results will be further analyzed to evaluate whether they have met the preset performance standards. For example, if the optimized device shows lower dark current and higher quantum efficiency under the same lighting conditions, the optimization is considered successful.

[0063] When the evaluation results show that the optimization adjustments have reached the qualified evaluation standards, these optimization parameters will be officially adopted as the optimization results of the device. At this time, these experimentally verified data and conclusions will be fed back to the performance prediction model. In the model update phase, these newly acquired data will be used to retrain and adjust the model to ensure that the model's predictive ability can reflect the latest process and device structure changes. This feedback mechanism not only enhances the accuracy and adaptability of the performance prediction model, but also improves the efficiency of the entire manufacturing process and the performance of the device through continuous iterative optimization and verification. Through this systematic approach, it is ensured that the performance optimization of AlGaN day-blind APD devices is not only based on scientific prediction models, but also verified by practical applications, ultimately achieving continuous technological progress and steady improvement in product quality.

[0064] Furthermore, after obtaining the experimental data information of AlGaN solar-blind APD, it includes:

[0065] The process parameters of the AlGaN day-blind APD are collected, the process parameters are screened and analyzed, and a cleaned process parameter data set is obtained.

[0066] The experimental device performance parameter records corresponding to the data in the process parameter data set are extracted from the experimental data records to obtain a performance parameter data set.

[0067] The process parameter data set and the performance parameter data set are verified for consistency.

[0068] Furthermore, the process parameters of AlGaN day-blind APD are collected, including but not limited to the material component ratio, device structure design details, and temperature, pressure, gas flow rate and deposition time during the manufacturing process. These data are usually derived from the manufacturing process records and experimental logs. Subsequently, these process parameters are screened and analyzed to remove redundant data, identify and process outliers, and finally obtain a cleaned process parameter data set that can be used for further analysis. For example, through statistical analysis methods, parameters that are not significantly associated with device performance fluctuations are identified and removed from the data set to simplify the model and improve its prediction accuracy. Next, experimental device performance parameter records corresponding to the data in the screened process parameter data set, such as dark current, quantum efficiency and spectral responsivity, are extracted from a wide range of experimental data records. This step ensures that each set of process parameters has corresponding performance parameters to form a complete performance parameter data set, which will be directly used for subsequent performance prediction model training. Finally, the consistency verification of the process parameter data set and the performance parameter data set is performed. This step is to confirm whether the data correspondence between the two data sets is accurate and ensure that all data records are matched. Consistency verification can be accomplished through data cross-validation methods, for example, by comparing the performance parameter records of repeated experiments under the same process conditions and checking their consistency. If the data are found to be inconsistent, it is necessary to recheck the data entry process or conduct additional experiments to verify the accuracy of the data.

[0069] Through the above data processing and verification steps, it is ensured that the data set used for performance prediction model training is accurate and reliable, which not only improves the efficiency of model training, but also provides a solid data foundation for achieving high-precision performance prediction.

[0070] Furthermore, based on a deep learning algorithm, the process parameters are used as model inputs, and the device performance parameters are used as model outputs, and model training is performed to construct a performance prediction model, including:

[0071] Feature extraction is performed on the process parameter data set and the performance parameter data set to obtain a process parameter feature set and a performance parameter feature set.

[0072] Based on a deep learning algorithm, the performance prediction model is constructed, wherein the performance prediction model structure includes an input layer, a hidden layer and an output layer.

[0073] The process parameter data set is used as a model input, and the performance parameter data set is used as an output. Model training is performed based on process parameter characteristics and performance parameter characteristics to construct a performance prediction model.

[0074] Exemplarily, first, feature extraction is performed on the process parameters and performance parameters obtained from the experimental data set. Feature extraction is to identify the attributes that have a significant impact on the prediction results from the original data. For example, in the process parameters, the material component ratio, the device structure layer thickness, and the temperature and pressure in the manufacturing process can be converted into numerical features that can be recognized by the model. Performance parameters such as dark current, quantum efficiency, and spectral responsivity also need to be converted into a format suitable for model processing. This step ensures that the model can effectively learn the mapping relationship from process parameters to device performance. Then, based on the extracted features, a performance prediction model is constructed. The basic architecture of the model includes an input layer, multiple hidden layers, and an output layer, wherein the input layer receives the processed process parameter feature set, the hidden layer is responsible for processing the data and capturing the complex nonlinear relationship therein, and finally the predicted performance parameters are generated through the output layer. Among these layers, the hidden layer determines the model's ability to process data and the accuracy of the prediction. Each layer is composed of a number of neurons, which process the input data through activation functions. Finally, the processed process parameter data set is used as the model input, and the performance parameter data set is used as the model output for model training. In this process, through supervised learning, the model learns how to predict the corresponding performance parameters based on the input process parameters, and uses optimization techniques such as back propagation and gradient descent to adjust the weights and biases in the model to minimize the difference between the predicted output and the actual output. This training process includes multiple iterations, each of which further optimizes the performance of the model.

[0075] In this way, the performance prediction model can accurately predict the performance of AlGaN day-blind APD based on its process parameters, providing scientific support for the design and manufacture of the device, thereby significantly improving the reliability and efficiency of the product.

[0076] Furthermore, model training includes:

[0077] The process parameter data set and the performance parameter data set are divided into a training set, a validation set and a test set. The model is iteratively trained using the training set data, and the model performance is evaluated using the validation set.

[0078] Based on the evaluation results of the validation set, the number of hidden layers, the number of neurons and the learning rate of the model are automatically adjusted through a hyperparameter optimization algorithm.

[0079] The error between the model output and the actual value is calculated through the back-propagation algorithm, and the weights and biases of the model are updated using the gradient descent method to minimize the value of the loss function. In each iteration, the learning rate is automatically adjusted using an adaptive optimization algorithm.

[0080] The validation set is used to evaluate different hyperparameter combinations, and the optimal hyperparameter configuration is selected to obtain the performance prediction model.

[0081] Specifically, the collected process parameter data set and performance parameter data set are divided into a training set, a validation set, and a test set, so that the model can be trained and validated on different data subsets to avoid overfitting and ensure the generalization ability of the model. Among them, the training set is used for iterative training of the model, that is, the model learns how to predict device performance through this part of the data; the validation set is used to evaluate the performance of the model on unseen data to help monitor and adjust the training process. In the iterative training stage of the model, the back propagation algorithm is used to calculate the error between the model output and the actual performance parameters, and the weights and biases of the model are continuously adjusted by the gradient descent method, with the aim of minimizing the value of the loss function, where the loss function usually uses mean square error (MSE) or cross entropy loss, etc., to quantify the difference between the predicted value and the true value. At the same time, the key parameters of the model are automatically adjusted by hyperparameter optimization algorithms, such as grid search, random search, or Bayesian optimization, including the number of hidden layers, the number of neurons in each layer, and the learning rate, etc. These hyperparameters have a significant impact on the learning ability and prediction accuracy of the model. The validation set is used to evaluate the performance of the model under different hyperparameter settings to help select the optimal parameter configuration. In addition, an adaptive optimization algorithm (such as Adam or RMSprop) is used to automatically adjust the learning rate in each iteration, which helps to converge to the lowest point more effectively during training and avoid the oscillation problem caused by too large a learning rate or the slow convergence problem caused by too small a learning rate.

[0082] Through the above steps, the performance prediction model finally determined can accurately predict the key performance parameters of AlGaN day-blind APD such as dark current, quantum efficiency and spectral responsivity under various process parameter inputs, which not only significantly improves the accuracy of the prediction, but also improves the adaptability of device performance optimization.

[0083] Furthermore, the weights and biases of the model are updated using the gradient descent method, including:

[0084] During the model training process, the weights of each hidden layer and output layer are dynamically adjusted, and the weights of each connection node in the neural network are automatically adjusted by analyzing the contribution of the process parameters using a weight adaptive algorithm.

[0085] The correlation between each process parameter data and the performance parameter data is analyzed to determine the correlation coefficient, and the weight distribution of each connection node in the model is adjusted based on the correlation coefficient.

[0086] Build a callback function to monitor and record weight changes, analyze the dynamic change trend of weight distribution through data visualization tools, and adjust the model's learning rate and weight distribution according to the monitoring results to synchronously update the model parameters.

[0087] Optionally, the weights and biases of the model are updated by gradient descent, which is an optimization algorithm used to minimize prediction errors. During the model training process, the weights of each hidden layer and output layer are dynamically adjusted. Using a weight adaptive algorithm, the algorithm automatically adjusts the weights of each connection node in the neural network by analyzing the contribution of process parameters to the prediction of performance parameters. For example, if a certain process parameter has a greater impact on performance, the corresponding connection weight will be increased to reflect its importance. Further, by analyzing the correlation between process parameter data and performance parameter data, the correlation coefficient between these parameters is determined. The correlation coefficient is a numerical value that quantifies the strength of the relationship between input parameters and output parameters. This value will be used to adjust the weight distribution of each connection node in the model, which means that the model not only learns the mapping between input and output, but also learns the relative importance of each input feature to the output, making the model's prediction more accurate.

[0088] In addition, in order to effectively monitor and adjust these processes, a callback function system is built. These callback functions are triggered at each stage of model training. They are responsible for recording changes in weights and biases, and analyzing the dynamic change trends of these weights through data visualization tools. In this way, the weight adjustment during model training can be monitored in real time, and the learning rate and weight distribution can be adjusted accordingly to optimize the learning efficiency and accuracy of the model.

[0089] Through the above method, the performance prediction model can automatically learn and adapt to the ever-changing data characteristics, thereby improving the adaptability and generalization ability of the model while maintaining high accuracy. The application of this technology has significantly improved the production efficiency and device performance of AlGaN day-blind APD.

[0090] Furthermore, based on the performance prediction results, an optimization scheme is generated, including:

[0091] Based on the performance prediction results, the impact of current process parameters is analyzed to determine the process parameters that need to be adjusted.

[0092] A preset adjustment step is set to gradually adjust and optimize the process parameters.

[0093] Based on the process parameters adjusted each time, the performance prediction is re-performed to generate optimized performance prediction results.

[0094] It is determined whether the optimized performance prediction result meets the preset performance standard. If not, the process parameters are continuously adjusted and optimized until the performance prediction result meets the preset standard.

[0095] When the performance prediction results meet the preset standards, the final optimization scheme is determined and the optimized process parameter combination is output.

[0096] Specifically, based on the results of the performance prediction model, the influence of the current process parameters is analyzed in detail. This analysis mainly focuses on understanding the specific influence of various process parameters such as material component ratio, device structure layer thickness, temperature and pressure in the manufacturing process on performance parameters (such as dark current, quantum efficiency and spectral responsivity). For example, if the performance prediction results show that the dark current increases abnormally under a certain pressure, this may indicate that the pressure setting needs to be adjusted. In this way, it is determined which process parameters are the key variables affecting the performance, thereby guiding the subsequent adjustment direction. Next, a preset adjustment step is set to gradually adjust and optimize these key process parameters. The adjustment step is a pre-defined parameter change amount, which determines the magnitude of the change in the process parameter in each adjustment. Reasonable setting of the adjustment step can ensure that the process adjustment is neither too rough to miss the optimal point, nor too small to cause the optimization process to be too long. After each adjustment of the process parameters, the performance prediction model is used again to perform performance prediction to generate optimized performance prediction results. This is an iterative process. Through continuous adjustment and prediction, the optimal process parameter configuration is gradually approached. After that, we evaluate whether these optimized performance prediction results have reached the preset performance standards. If the results show that the performance does not meet the expected standards, we will continue to adjust the process parameters and optimize. This process will be repeated until the performance prediction results meet the preset standards. This method ensures that the performance is gradually optimized through systematic analysis and iterative adjustments. Finally, when the performance prediction results meet the preset standards, the final optimization plan will be determined and solidified. This plan will include the optimal process parameter combination determined after multiple rounds of optimization. These parameter configurations will be used in the actual production process to ensure that the produced devices have the best performance.

[0097] Through the above steps, not only the performance of the device is improved, but also the production efficiency and cost-effectiveness are significantly optimized, providing scientific guidance and support for the production of high-performance optoelectronic devices.

[0098] Furthermore, the optimization scheme is experimentally verified, the optimization effect is verified, and an evaluation result is obtained. After the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model updating, including:

[0099] The optimization scheme is experimentally verified to obtain experimental data of actual process parameters and device performance, and compared with the predicted results of the performance prediction model, and the deviation between the two is calculated to generate an evaluation result. If the deviation is less than a preset threshold, the evaluation is qualified.

[0100] When the evaluation is qualified, the actual data in the experimental verification is added to the training data set to update the performance prediction model.

[0101] During the model updating process, the weight distribution of the training data is dynamically adjusted according to the size of the deviation, the data set with larger training deviation is trained first, and the model parameters are optimized through iterative training times to perform model updating.

[0102] After completing the model update, re-prediction is performed based on the updated performance prediction model to verify the updated optimization effect and accuracy of the model.

[0103] Specifically, the optimization scheme is experimentally verified, which involves applying the optimized process parameters in an actual production environment and measuring the device performance parameters. These experimental data include, but are not limited to, material component ratios, device structure layer thickness, and parameters such as temperature and pressure during the manufacturing process, as well as corresponding device performance such as dark current, quantum efficiency, and spectral responsivity. These actual measurements will be compared with the predicted results of the performance prediction model, and the deviation between the two will be calculated. If the deviation is less than the preset threshold, the optimization scheme is considered to have reached the predetermined performance standard and is evaluated as qualified. Once the optimization scheme is evaluated as qualified, the actual process parameters and device performance data obtained in the experiment will be added to the original training data set. This step is to enable the performance prediction model to learn the latest data patterns and improve the accuracy and reliability of its predictions.

[0104] Next, during the model update process, the weight distribution of the training data will be dynamically adjusted according to the size of the deviation. Specifically, data sets with larger deviations will be given higher weights so that they receive more attention during the model training process, which helps the model learn more effectively and correct those parts that are under-predicted. In this way, the model parameters can be optimized, and the performance of the model can be further refined through iterative training to perform model updates. Finally, after the model update is completed, the performance prediction model will be used to re-perform performance prediction to verify the effect and accuracy of the model update optimization, which can not only ensure the effectiveness of the model after adjustment, but also ensure that the model maintains high performance and high accuracy in the long term through continuous verification and adjustment cycles.

[0105] The above steps ensure that the performance prediction model can be continuously evolved and optimized through continuous experimental verification and data feedback, providing scientific and precise guidance for the production of AlGaN day-blind APDs, and ultimately achieving efficient and high-quality production of optoelectronic devices.

[0106] Through the technical solutions of the above-mentioned embodiments, the AlGaN day-blind APD performance prediction and optimization method provided in the present application solves the technical problems in the prior art that the performance optimization of the AlGaN day-blind APD is inefficient, inaccurate, and difficult to adapt to complex process conditions, and achieves fast and accurate performance prediction and automatic optimization, thereby improving the technical effect of optimization efficiency and adaptability to working conditions.

[0107] Embodiment 2, based on the same inventive concept as the AlGaN solar-blind APD performance prediction and optimization method in the above embodiment, Figure 2 As shown, the present application provides an AlGaN solar-blind APD performance prediction and optimization system, the system comprising:

[0108] The experimental data acquisition module 11 is used to obtain experimental data information of the AlGaN day-blind APD, wherein the experimental data information includes process parameters and device performance parameters, wherein the process parameters include material composition, device structure and process conditions, and the device performance parameters include dark current, quantum efficiency and spectral responsivity.

[0109] The model building module 12 is used to perform model training and build a performance prediction model based on a deep learning algorithm, with the process parameters as model input and the device performance parameters as model output.

[0110] The performance prediction module 13 is used to perform performance prediction using the performance prediction model to obtain a performance prediction result of the target device.

[0111] The optimization scheme obtaining module 14 is used to generate an optimization scheme for the target device based on the performance prediction result, wherein the optimization scheme includes material component adjustment parameters, device structure improvement parameters and process condition optimization parameters.

[0112] The model updating module 15 is used to experimentally verify the optimization scheme, check the optimization effect, and obtain an evaluation result. When the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model updating.

[0113] Furthermore, the experimental data acquisition module 11 is also used to perform the following steps:

[0114] The process parameters of the AlGaN day-blind APD are collected, the process parameters are screened and analyzed, and a cleaned process parameter data set is obtained.

[0115] The experimental device performance parameter records corresponding to the data in the process parameter data set are extracted from the experimental data records to obtain a performance parameter data set.

[0116] The process parameter data set and the performance parameter data set are verified for consistency.

[0117] Furthermore, the model building module 12 is also used to perform the following steps:

[0118] Feature extraction is performed on the process parameter data set and the performance parameter data set to obtain a process parameter feature set and a performance parameter feature set.

[0119] Based on a deep learning algorithm, the performance prediction model is constructed, wherein the performance prediction model structure includes an input layer, a hidden layer and an output layer.

[0120] The process parameter data set is used as a model input, and the performance parameter data set is used as an output. Model training is performed based on process parameter characteristics and performance parameter characteristics to construct a performance prediction model.

[0121] Furthermore, the model building module 12 is also used to perform the following steps:

[0122] The process parameter data set and the performance parameter data set are divided into a training set, a validation set and a test set. The model is iteratively trained using the training set data, and the model performance is evaluated using the validation set.

[0123] Based on the evaluation results of the validation set, the number of hidden layers, the number of neurons and the learning rate of the model are automatically adjusted through a hyperparameter optimization algorithm.

[0124] The error between the model output and the actual value is calculated through the back-propagation algorithm, and the weights and biases of the model are updated using the gradient descent method to minimize the value of the loss function. In each iteration, the learning rate is automatically adjusted using an adaptive optimization algorithm.

[0125] The validation set is used to evaluate different hyperparameter combinations, and the optimal hyperparameter configuration is selected to obtain the performance prediction model.

[0126] Furthermore, the model building module 12 is also used to perform the following steps:

[0127] During the model training process, the weights of each hidden layer and output layer are dynamically adjusted, and the weights of each connection node in the neural network are automatically adjusted by analyzing the contribution of the process parameters using a weight adaptive algorithm.

[0128] The correlation between each process parameter data and the performance parameter data is analyzed to determine the correlation coefficient, and the weight distribution of each connection node in the model is adjusted based on the correlation coefficient.

[0129] Build a callback function to monitor and record weight changes, analyze the dynamic change trend of weight distribution through data visualization tools, and adjust the model's learning rate and weight distribution according to the monitoring results to synchronously update the model parameters.

[0130] Furthermore, the optimization solution obtaining module 14 is further configured to perform the following steps:

[0131] Based on the performance prediction results, the impact of current process parameters is analyzed to determine the process parameters that need to be adjusted.

[0132] A preset adjustment step is set to gradually adjust and optimize the process parameters.

[0133] Based on the process parameters adjusted each time, the performance prediction is re-performed to generate optimized performance prediction results.

[0134] It is determined whether the optimized performance prediction result meets the preset performance standard. If not, the process parameters are continuously adjusted and optimized until the performance prediction result meets the preset standard.

[0135] When the performance prediction results meet the preset standards, the final optimization scheme is determined and the optimized process parameter combination is output.

[0136] Furthermore, the model updating module 15 is further configured to perform the following steps:

[0137] The optimization scheme is experimentally verified to obtain experimental data of actual process parameters and device performance, and compared with the predicted results of the performance prediction model, and the deviation between the two is calculated to generate an evaluation result. If the deviation is less than a preset threshold, the evaluation is qualified.

[0138] When the evaluation is qualified, the actual data in the experimental verification is added to the training data set to update the performance prediction model.

[0139] During the model updating process, the weight distribution of the training data is dynamically adjusted according to the size of the deviation, the data set with larger training deviation is trained first, and the model parameters are optimized through iterative training times to perform model updating.

[0140] After completing the model update, re-prediction is performed based on the updated performance prediction model to verify the updated optimization effect and accuracy of the model.

[0141] Through the above-mentioned detailed description of the AlGaN day-blind APD performance prediction and optimization method in this specification, those skilled in the art can clearly understand the AlGaN day-blind APD performance prediction and optimization system in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0142] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. AlGaN solar-blind APD performance prediction and optimization method, characterized in that: The method comprises: Acquire experimental data information of AlGaN day-blind APD, wherein the experimental data information includes process parameters and device performance parameters, wherein the process parameters include material composition, device structure and process conditions, and the device performance parameters include dark current, quantum efficiency and spectral responsivity; Based on a deep learning algorithm, the process parameters are used as model inputs, and the device performance parameters are used as model outputs, to perform model training and build a performance prediction model; Use the performance prediction model to perform performance prediction and obtain performance prediction results of the target device; Based on the performance prediction results, generating an optimization scheme for the target device, the optimization scheme including material component adjustment parameters, device structure improvement parameters and process condition optimization parameters; The optimization scheme is experimentally verified to check the optimization effect and obtain an evaluation result. When the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model updating.

2. The AlGaN solar-blind APD performance prediction and optimization method according to claim 1, characterized in that: After obtaining the experimental data information of AlGaN solar-blind APD, including: Collecting process parameters of the AlGaN day-blind APD, screening and analyzing the process parameters, and obtaining a cleaned process parameter data set; Extracting experimental device performance parameter records corresponding to the data in the process parameter data set from the experimental data records to obtain a performance parameter data set; The process parameter data set and the performance parameter data set are verified for consistency.

3. The AlGaN solar-blind APD performance prediction and optimization method according to claim 2, characterized in that: Based on the deep learning algorithm, the process parameters are used as model inputs, and the device performance parameters are used as model outputs, and model training is performed to construct a performance prediction model, including: Performing feature extraction on the process parameter data set and the performance parameter data set to obtain a process parameter feature set and a performance parameter feature set; Based on the deep learning algorithm, construct the performance prediction model, wherein the performance prediction model structure includes an input layer, a hidden layer and an output layer; The process parameter data set is used as a model input, and the performance parameter data set is used as an output. Model training is performed based on process parameter characteristics and performance parameter characteristics to construct a performance prediction model.

4. The AlGaN solar-blind APD performance prediction and optimization method according to claim 3, characterized in that: Perform model training, including: The process parameter data set and the performance parameter data set are divided into a training set, a validation set and a test set, the model is iteratively trained using the training set data, and the model performance is evaluated using the validation set; Based on the evaluation results of the validation set, the number of hidden layers, the number of neurons and the learning rate of the model are automatically adjusted through a hyperparameter optimization algorithm; The error between the model output and the actual value is calculated through the back-propagation algorithm, and the weights and biases of the model are updated using the gradient descent method to minimize the value of the loss function. In each iteration, the learning rate is automatically adjusted using an adaptive optimization algorithm. The validation set is used to evaluate different hyperparameter combinations, and the optimal hyperparameter configuration is selected to obtain the performance prediction model.

5. The AlGaN solar-blind APD performance prediction and optimization method according to claim 4, characterized in that: The weights and biases of the model are updated using the gradient descent method, including: During the model training process, the weights of each hidden layer and output layer are dynamically adjusted, and the weights of each connection node in the neural network are automatically adjusted by analyzing the contribution of the process parameters using a weight adaptive algorithm; Analyze the correlation between each process parameter data and performance parameter data, determine the correlation coefficient, and adjust the weight distribution of each connection node in the model based on the correlation coefficient; Build a callback function to monitor and record weight changes, analyze the dynamic change trend of weight distribution through data visualization tools, and adjust the model's learning rate and weight distribution according to the monitoring results to synchronously update the model parameters.

6. The AlGaN solar-blind APD performance prediction and optimization method according to claim 1, characterized in that: Based on the performance prediction results, an optimization plan is generated, including: Based on the performance prediction results, analyzing the impact of current process parameters and determining process parameters that need to be adjusted; Setting a preset adjustment step to gradually adjust and optimize the process parameters; Combined with the process parameters adjusted each time, re-predict the performance and generate optimized performance prediction results; Determine whether the optimized performance prediction result meets the preset performance standard. If not, continue to adjust the process parameters and optimize until the performance prediction result meets the preset standard. When the performance prediction results meet the preset standards, the final optimization scheme is determined and the optimized process parameter combination is output.

7. The AlGaN solar-blind APD performance prediction and optimization method according to claim 1, characterized in that: The optimization scheme is experimentally verified to check the optimization effect and obtain an evaluation result. After the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model updating, including: The optimization scheme is experimentally verified to obtain experimental data of actual process parameters and device performance, and compared with the prediction results of the performance prediction model, and the deviation between the two is calculated to generate an evaluation result. If the deviation is less than a preset threshold, the evaluation is qualified; When the evaluation is qualified, the actual data in the experimental verification is added to the training data set to update the performance prediction model; During the model update process, the weight distribution of the training data is dynamically adjusted according to the size of the deviation, the data set with larger deviation is trained first, and the model parameters are optimized through iterative training times to perform model update; After completing the model update, re-prediction is performed based on the updated performance prediction model to verify the updated optimization effect and accuracy of the model.

8. AlGaN solar-blind APD performance prediction and optimization system, characterized in that: For implementing the AlGaN solar-blind APD performance prediction and optimization method according to any one of claims 1 to 7, the system comprises: An experimental data acquisition module is used to acquire experimental data information of AlGaN day-blind APD, wherein the experimental data information includes process parameters and device performance parameters, wherein the process parameters include material composition, device structure and process conditions, and the device performance parameters include dark current, quantum efficiency and spectral responsivity; A model building module, for performing model training based on a deep learning algorithm, taking the process parameters as model input and the device performance parameters as model output, and building a performance prediction model; A performance prediction module, used to perform performance prediction using the performance prediction model to obtain a performance prediction result of a target device; An optimization scheme obtaining module, used to generate an optimization scheme for the target device based on the performance prediction result, wherein the optimization scheme includes material component adjustment parameters, device structure improvement parameters and process condition optimization parameters; The model updating module is used to experimentally verify the optimization scheme, check the optimization effect, and obtain an evaluation result. When the evaluation result is qualified, the optimization scheme is used as the optimization result and fed back to the performance prediction model for model updating.

Images