X / gamma ray semiconductor detector multi-parameter intelligent optimization method and system

Abstract

The present disclosure provides an X / gamma ray semiconductor detector multi-parameter intelligent optimization method and system, relating to the technical field of semiconductor detectors, comprising: according to candidate semiconductor materials and design parameters, using a photon transport and energy deposition model to simulate X rays or gamma rays of different incident energies event by event, obtaining deposited photon data; calculating the collected charge corresponding to each energy deposition, reconstructing the detector output energy spectrum, and extracting the energy spectrum multi-performance indicators; inputting the indicator samples into a neural network proxy model to learn the nonlinear mapping relationship between the material, incident energy, detector thickness, electric field strength and multiple energy spectrum performance indicators; using the neural network model as an evaluator to search for the optimal material, detector thickness and electric field strength and the corresponding predicted performance under engineering constraints. The present disclosure can automatically obtain a detector design scheme that meets multiple target performance and engineering constraints in a continuous thickness and continuous electric field design space.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor detector technology, specifically to a multi-parameter intelligent optimization method and system for X / γ-ray semiconductor detectors. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] X-ray and gamma-ray semiconductor detectors are widely used in space satellite exploration, nuclear medicine imaging, industrial non-destructive testing, nuclear safety monitoring, and high-energy astrophysics payloads. In the actual design of such detectors, there is a clear coupling relationship between detection efficiency, energy resolution, peak position shift, charge collection efficiency, operating voltage, noise level, breakdown risk, and power consumption. Adjusting any one parameter alone often fails to obtain an engineering solution with optimal overall performance.

[0004] Existing technologies typically employ empirical material selection, experimental prototyping, or single-factor parameter scanning methods for X-ray and gamma-ray semiconductor detector design. In engineering applications, simply increasing the electric field strength is not an ideal optimization approach. While a higher electric field may improve carrier collection efficiency to a certain extent, as charge collection approaches saturation, further increasing the electric field has limited impact on spectral performance and may even lead to increased leakage current, electronic noise, increased risk of device breakdown, higher power consumption, and decreased long-term stability. For space satellite payloads, low power consumption, low operating voltage, and high reliability are particularly important. Excessively high operating voltage increases the difficulty of power module design, insulation and packaging requirements, and on-orbit operational risks. Therefore, during detector parameter optimization, the leakage current, noise, breakdown, and power consumption issues caused by high electric fields must be incorporated into engineering constraints, rather than solely focusing on charge collection efficiency or peak count as the primary objective.

[0005] In summary, the existing technology has at least the following shortcomings: (1) Photon transport and energy deposition simulation data are usually finite and discrete, making it difficult to directly cover the continuous design space composed of materials, thickness and electric field intensity.

[0006] (2) Traditional parameter scanning methods have large computational load and low efficiency, making it difficult to perform global optimization under conditions of multiple materials, multiple thicknesses, multiple electric fields and multiple incident energies.

[0007] (3) It is difficult to reflect the influence of electron and hole transport, charge trapping and charge collection efficiency on energy spectrum performance based solely on energy deposition results.

[0008] (4) Traditional optimization tends to favor higher electric fields, without fully considering the leakage current, noise, breakdown, power consumption and low voltage operation requirements of space load caused by high electric fields.

[0009] (5) Manual experience-based selection or single-target optimization cannot simultaneously take into account detection efficiency, energy resolution, peak position stability, charge collection efficiency and engineering reliability. Summary of the Invention

[0010] To address the aforementioned issues, this disclosure proposes a multi-parameter intelligent optimization method and system for X / γ-ray semiconductor detectors. For candidate semiconductor materials, it utilizes photon transport and energy deposition simulation, charge collection physical models, continuous parameter surrogate prediction, and engineering constraints for global optimization and closed-loop coupling. This ensures that the optimization process has a physical basis, significantly reduces the computational overhead of repetitive simulations, and outputs detector design parameters that meet engineering reliability requirements.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions: A multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors includes: Obtain candidate semiconductor materials and design parameters; Based on candidate semiconductor materials and design parameters, an event-by-event simulation of X-rays or γ-rays with different incident energies is performed using a photon transport and energy deposition model to obtain deposited photon data. Based on the deposited photon data, the collection charge corresponding to each energy deposition is calculated, and the detector output energy spectrum is reconstructed. Feature extraction is performed on the output energy spectrum to obtain multiple performance indicators of the energy spectrum. By combining multiple energy spectrum performance indicators to construct an indicator sample, the indicator sample is input into a neural network surrogate model to learn the nonlinear mapping relationship between material, incident energy, detector thickness and electric field strength and multiple energy spectrum performance indicators. Using a neural network model as an evaluator, the optimal material, detector thickness, electric field strength, and corresponding prediction performance are searched under engineering constraints.

[0012] As one embodiment, obtaining candidate semiconductor materials and design parameters includes: The candidate semiconductor materials include one or more of CsPbBr3, MAPbBr3, CZT, CdTe, 4H-SiC, and Si; The design parameters include material properties, detector structural parameters, incident photon energy points, and engineering constraints. The material properties include average ionization energy, electron mobility lifetime product, hole mobility lifetime product, and Fano factor; the engineering constraints include upper limit of peak position shift, lower limit of charge collection efficiency, lower limit of total collected charge, upper limit of electric field strength, upper limit of bias voltage, breakdown risk limit, leakage current limit, noise limit, and power consumption limit.

[0013] As one embodiment, the step of performing event-by-event simulations of X-rays or γ-rays with different incident energies using a photon transport and energy deposition model based on candidate semiconductor materials and design parameters to obtain deposited photon data includes: Based on candidate semiconductor materials and design parameters, a photon transport and energy deposition model for X-ray or gamma-ray photons in a semiconductor detector is established. Event-by-event simulations of X-rays or γ-rays with different incident energies were performed using a photon transport and energy deposition model. For each incident photon event, its propagation path, interaction location, interaction process, energy before and after interaction, and corresponding energy deposition in the semiconductor detector were recorded. For each interaction step, the local deposition energy is calculated based on the energy difference before and after the interaction, and the coordinates of the interaction position are retained. All deposition step data within the same event are classified and exported according to the event number to obtain the deposition photon data.

[0014] As one embodiment, the step of calculating the collection charge corresponding to each energy deposition and reconstructing the detector output energy spectrum based on the deposited photon data includes: Based on the deposited photon data, combined with the material's average ionization energy, Fano factor, electron mobility lifetime product, hole mobility lifetime product, detector thickness, and electric field strength, the collection charge corresponding to each energy deposition step is calculated. The total collected charge of the event is obtained by summing the collected charges of all steps within the same event. The total collected charge is then converted into output energy. The output energy of all incident events is statistically analyzed to obtain the output energy spectrum under the corresponding material, incident energy, detector thickness, and electric field strength.

[0015] As one embodiment, the step of constructing an index sample by combining multiple energy spectrum performance indicators, inputting the index sample into a neural network surrogate model, and learning the nonlinear mapping relationship between material, incident energy, detector thickness, electric field strength, and multiple energy spectrum performance indicators includes: The obtained candidate materials, design parameters, and multiple energy spectrum performance indicators are organized into a supervised learning dataset. Each sample corresponds to a combination of a material, an incident energy, a detector thickness, and an electric field intensity. The input variables include material encoding, incident energy, detector thickness, and electric field intensity. The output variables include peak count, total collected charge, full width at half maximum (FWHM), peak position shift, and charge collection efficiency. Candidate materials are represented by unique thermal encoding. The neural network surrogate model learns the nonlinear mapping relationship between materials, incident energy, detector thickness and electric field intensity and multiple energy spectrum performance indicators, expanding discrete data into a fast prediction model in a continuous parameter space.

[0016] As one embodiment, the step of using a neural network model as an evaluator to search for the optimal material, detector thickness, and electric field strength, as well as the corresponding prediction performance, considering engineering constraints, includes: Material, detector thickness, and electric field strength are used as individual encodings in a genetic algorithm; Among them, the material is a discrete variable, while the detector thickness and electric field strength are continuous variables; The algorithm initializes a set number of candidate individuals to form a population; for each candidate individual, a neural network proxy model is called to predict its peak count, total collected charge, full width at half maximum (FWHM), peak position shift, and charge collection efficiency at multiple incident energy points; Construct a multi-objective fitness function. For each incident energy point, calculate the fitness score. For multiple incident energy points, calculate the fitness score for each energy point separately, and then take the average or weighted average as the comprehensive fitness of the candidate design. During the iteration process of the genetic algorithm, engineering constraints are introduced, and the optimal design scheme that satisfies the multi-objective evaluation and engineering constraints is output. The output results include the optimal material, the optimal detector thickness, the optimal electric field strength, the corresponding bias voltage, and the predicted performance index at multiple incident energy points.

[0017] According to some embodiments, the present disclosure adopts the following technical solutions: A multi-parameter intelligent optimization system for X / γ-ray semiconductor detectors includes: The parameter input module is used to obtain candidate semiconductor materials and design parameters; The photon transport and energy deposition module is used to perform event-by-event simulations of X-rays or γ-rays with different incident energies based on candidate semiconductor materials and design parameters, and to obtain deposited photon data. The charge collection correction module is used to calculate the collection charge corresponding to each energy deposition based on the deposited photon data and reconstruct the detector output energy spectrum; The dataset construction module is used to extract features from the output energy spectrum to obtain multiple performance indicators of the energy spectrum; The neural network proxy module is used to construct index samples by combining multiple energy spectrum performance indicators. The index samples are then input into the neural network proxy model to learn the nonlinear mapping relationship between materials, incident energy, detector thickness, electric field strength, and multiple energy spectrum performance indicators. The genetic algorithm optimization module is used to search for the optimal material, detector thickness, electric field strength, and corresponding prediction performance under engineering constraints, using a neural network model as an evaluator. The results output module is used to output the optimal design parameters and the corresponding prediction performance.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product includes a computer program that, when executed by a processor, implements the aforementioned intelligent multi-parameter optimization method for an X / γ-ray semiconductor detector.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned intelligent multi-parameter optimization method for an X / γ-ray semiconductor detector.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned intelligent multi-parameter optimization method for X / γ-ray semiconductor detectors.

[0021] Compared with the prior art, the beneficial effects of this disclosure are as follows: This disclosure presents a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors, which combines photon transport and energy deposition data with charge collection models. This allows the training data to simultaneously include photon interaction processes and semiconductor carrier transport processes, thereby improving the model's ability to characterize the actual detector response.

[0022] This invention discloses a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors. By using a neural network surrogate model, it transforms finite discrete data into a fast prediction model in a continuous parameter space, avoiding repeated complex physical simulations for each candidate thickness and electric field intensity, and significantly reducing computational costs.

[0023] This disclosure presents a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors, capable of simultaneously optimizing material, thickness, and electric field intensity. This achieves collaborative optimization across multiple materials and parameters, rather than relying solely on empirical scanning of a single material or parameter. Employing a multi-energy-point comprehensive evaluation approach, it can simultaneously consider peak values, full width at half maximum (FWHM), peak position shift, total collected charge, and charge collection efficiency at multiple X-ray or γ-ray energy points, making it suitable for multi-energy spectral applications.

[0024] This disclosure presents a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors. By introducing saturation penalties and engineering constraints, it avoids the increased leakage current noise, increased breakdown risk, and increased power consumption caused by simply pursuing high electric fields, making the optimization results more in line with the actual detector design requirements.

[0025] This disclosure presents a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors. For space satellite payloads, this method tends to select schemes with lower operating electric fields or lower operating voltages while ensuring charge collection efficiency and energy resolution. This is beneficial for reducing power consumption, alleviating thermal design pressure, and improving on-orbit reliability. The output results include material, thickness, electric field strength, and multiple performance indicators, facilitating detector structure design, parameter selection, and subsequent experimental verification by researchers or engineers. Attached Figure Description

[0026] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0027] Figure 1 This is a diagram illustrating the overall architecture of the method according to an embodiment of this disclosure; Figure 2 This is a flowchart of a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors according to an embodiment of the present disclosure; Figure 3 This is a schematic diagram of the detector geometry according to an embodiment of the present disclosure; Figure 4 This is a schematic diagram illustrating charge collection corrections according to an embodiment of the present disclosure; Figure 5 This is a structural diagram of the neural network proxy model according to an embodiment of the present disclosure; Figure 6 This is a flowchart of the genetic algorithm optimization process according to an embodiment of the present disclosure. Detailed Implementation

[0028] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0029] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0031] Example 1 One embodiment of this disclosure provides a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors, the method steps of which include: Step 1: Obtain candidate semiconductor materials and design parameters; Step 2: Based on the candidate semiconductor materials and design parameters, use the photon transport and energy deposition model to perform event-by-event simulations of X-rays or γ-rays with different incident energies to obtain deposited photon data; Step 3: Based on the deposited photon data, calculate the collection charge corresponding to each energy deposition, reconstruct the detector output energy spectrum, extract features from the output energy spectrum, and obtain multiple performance indicators of the energy spectrum; Step 4: Construct an index sample by combining multiple energy spectrum performance indicators, input the index sample into the neural network surrogate model, and learn the nonlinear mapping relationship between material, incident energy, detector thickness and electric field strength and multiple energy spectrum performance indicators; Step 5: Using a neural network model as an evaluator, search for the optimal material, detector thickness, electric field strength, and corresponding prediction performance under engineering constraints.

[0032] As one embodiment, this disclosure discloses a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors. First, an X / γ-ray photon transport and energy deposition model is established for candidate semiconductor materials to obtain the interaction positions of photons in the detector and energy deposition data under different incident energies. Then, based on a charge collection model, the energy deposition data is converted into an output energy spectrum considering electron and hole drift, trapping, and charge contribution. Next, performance indicators such as peak count, total collected charge, full width at half maximum (FWHM), peak position shift, and charge collection efficiency are extracted from the output energy spectrum to form a material-energy-thickness-electric field-performance indicator sample dataset. A neural network surrogate model is further trained to establish a nonlinear mapping relationship between input design parameters and output performance indicators. Finally, a genetic algorithm is used to call this surrogate model as a fast evaluator to perform multi-objective global optimization in the design space composed of material, thickness, and electric field strength, and engineering constraints related to high electric field noise, breakdown, power consumption, and low voltage requirements for space load are introduced into the fitness function. Figure 2 As shown, the specific implementation process includes: S1: Input candidate materials and design parameters.

[0033] This disclosure first obtains candidate semiconductor materials and design parameters. The candidate semiconductor materials may include one or more of CsPbBr3, MAPbBr3, CZT, CdTe, 4H-SiC, and Si. The design parameters include material properties, detector structure parameters, incident photon energy points, and engineering constraints. Furthermore, the material properties include at least the average ionization energy, electron mobility lifetime product, hole mobility lifetime product, and Fano factor. The average ionization energy is used to convert energy deposition into electron-hole pairs; the electron mobility lifetime product and hole mobility lifetime product are used to calculate the effective drift length of electrons and holes under an applied electric field and the charge collection efficiency; the Fano factor is used to characterize the effect of ionization statistical fluctuations on energy spectrum broadening.

[0034] Furthermore, the engineering constraints include upper limits for peak offset, lower limits for charge collection efficiency, lower limits for total collected charge, upper limits for electric field strength, upper limits for bias voltage, limits for breakdown risk, limits for leakage current, limits for noise, and limits for power consumption.

[0035] As one example, candidate materials are not limited to CsPbBr3, MAPbBr3, CZT, CdTe, GaAs, and Si, but may also include Ge, HgI2, TlBr, PbI2, BiI3, CsPbCl3, CsPbI3, FAPbBr3, mixed halide perovskites, two-dimensional perovskites, gallium oxide, diamond, and other semiconductor materials suitable for hard X-ray or gamma-ray detection.

[0036] Furthermore, for space satellite payloads, it is preferable to set lower operating voltage or lower electric field strength constraints to reduce power consumption, simplify high-voltage power supply design, alleviate thermal management pressure, and improve long-term on-orbit reliability.

[0037] S2: Establish a photon transport and energy deposition model.

[0038] Based on the candidate materials and design parameters input in S1, a photon transport and energy deposition model for X-ray or gamma-ray photons in a semiconductor detector is established. This model is used to describe the propagation, interaction, energy transfer, and energy deposition processes of incident photons in the detector.

[0039] like Figure 3 As shown, the detector geometry includes an external space region (world volume), an envelope, an incident window or pre-contact material layer, metal electrodes, and a semiconductor sensor (semiconductor material). The incident window or pre-contact material layer can be made of materials such as Al (aluminum), the metal electrodes are made of Pt (platinum) conductive material, and the semiconductor sensor can be made of CsPbBr3, MAPbBr3, CZT, CdTe, 4H-SiC, Si, or other candidate materials. The semiconductor sensor is configured as an energy deposition statistical region to record the interaction position, interaction type, and energy deposition of incident photons in the detector on an event-by-event basis.

[0040] Furthermore, the physical processes include the photoelectric effect, Compton scattering, secondary electron generation, atomic deexcitation, and energy deposition. To improve the sampling accuracy of the energy deposition location, a small limit of 0.001 mm for the maximum particle tracking step size is set within the semiconductor, which is achieved using Geant4.

[0041] As an example, the photon transport and energy deposition simulation tool is not limited to Geant4, but can also use MCNP, FLUKA, EGSnrc, PENELOPE or other programs that can output energy deposition distribution and interaction information.

[0042] S3: Perform photon transport and energy deposition simulations.

[0043] Event-by-event simulations were performed for X-rays or gamma rays with different incident energies. During the simulation, a semiconductor sensor was set as the energy deposition statistical region, and its material composition, density, and geometric thickness were set according to the candidate materials. A monoenergetic X-ray or gamma-ray photon source was set up, allowing incident photons to be incident on the detector surface along a preset direction. For each incident photon event, during particle tracking, it was determined whether the interaction step occurred within the semiconductor sensor. If it did, the event number, energy before interaction, energy after interaction, interaction process type, position coordinates before and after interaction, and interaction time were recorded. The interaction processes included photoelectric absorption, Compton scattering, Rayleigh scattering, secondary electron generation, and energy deposition processes caused by other secondary particles.

[0044] For each event, the total energy deposition within the detector is calculated and recorded, while local energy deposition information for each interaction step is retained. Step-by-step data is used to subsequently calculate the different collection contributions of electrons and holes based on the interaction locations, while the total per-event data is used to form the output energy spectrum and statistical performance metrics.

[0045] S4: Export the deposited photon data.

[0046] The photon transport and energy deposition simulation results obtained from S3 were organized into a data format that can be directly read for subsequent charge collection corrections. For each interaction step, the local deposition energy was calculated based on the energy difference before and after the interaction, and the interaction location coordinates were preserved.

[0047] in, E dep,i For localized energy deposition, E pre,i Energy before interaction E post,i This represents the energy generated after the interaction.

[0048] For each incident event, all deposition steps within the same event are categorized according to the event number to form an event-by-event deposition sequence, and the total deposition energy of the event is obtained by summing the local deposition energies of each step:

[0049] The location coordinates are used to subsequently determine the distance from the deposition point to the electrodes on both sides, and to further calculate the electron and hole drift distances, charge collection factor, and total event-collected charge.

[0050] Furthermore, the derived data includes: event number, deposition energy, interaction location, interaction process, primary photon energy, and total number of incident photons. If needed, photoelectric effect events, Compton scattering events, and other event types can also be recorded to analyze the contribution of different interaction processes to the energy spectrum.

[0051] This step also allows for data validity filtering, such as removing invalid values, abnormal events, or deposition steps outside the semiconductor sensor region. The processed data serves as input for the S5 charge collection correction, ensuring that subsequent calculations consider not only the deposition energy but also the impact of deposition location on carrier drift and collection efficiency.

[0052] S5: Perform charge collection correction and energy spectrum reconstruction.

[0053] Based on the step-by-step energy deposition data obtained from S4, and combined with the material's average ionization energy, Fano factor, electron mobility lifetime product, hole mobility lifetime product, detector thickness, and electric field strength, the collected charge corresponding to each energy deposition step is calculated, and the detector output energy spectrum is further reconstructed. For example... Figure 4 As shown, the charge collection correction module calculates the charge collection efficiency based on the interaction sites and material carrier transport parameters. The specific calculation process is as follows: For each deposition step, first, based on the average ionization energy of the material... Deposited energy Converted to electron-hole pairs To account for ionization statistical fluctuations in semiconductor detectors, Fano fluctuations are introduced. Let the mean number of carriers generated under the fluctuation-free condition be... The actual number of carriers generated According to the mean ,variance The statistical distribution is corrected, where The material's Fano factor.

[0054] Furthermore, the contributions of electrons and holes to the charge at the interaction sites are calculated based on the charge collection model. Let the detector thickness be... The electric field strength is The electron mobility lifetime product is The hole mobility lifetime product is Then the electron drift length is:

[0055] The hole drift length is:

[0056] When the left cathode is taken as the starting point of the coordinate system, This represents the distance from the interaction site to the left cathode. Electrons drift towards the right anode, and the drift distance is... The hole drifts to the left cathode, and the drift distance is... The charge collection factor can be expressed as:

[0057] In this formula, the first term represents the contribution of holes to the charge, and the second term represents the contribution of electrons to the charge. Given the material, incident energy, detector thickness, and electric field strength, the charge collection efficiency can be obtained by summing all effective energy deposition steps:

[0058] For each energy deposition step, the collected charge can be expressed as:

[0059] in, Represents the Fano fluctuation correction coefficient. Collected charge for all steps within the same event. Q step Summing gives the total collected charge for the event. Q event Then Q event This is converted into output energy. By statistically analyzing the output energy of all incident events, the output energy spectrum under the corresponding material, incident energy, detector thickness, and electric field strength can be obtained.

[0060] As one embodiment, this method does not directly equate energy deposition with detector output, but couples the photon energy deposition process with semiconductor carrier transport and charge collection processes, thereby making the subsequent training data closer to the response of a real semiconductor detector.

[0061] As an example, the charge collection correction model is not limited to the standard Hecht equation, but may also employ an improved Hecht model, a model that considers the space charge effect, a model that considers the polarization effect, a model that considers the trap distribution, a drift-diffusion model, or a finite element-based electric field / carrier transport model.

[0062] S6: Extract energy dispersive spectroscopy performance indicators.

[0063] This step extracts features from the output energy spectrum reconstructed by S5. Specifically, the output energy corresponding to each incident event is denoted as... The output energy spectrum histogram is then established based on the preset energy bin width. First, a coarser energy bin is used within the high-energy candidate region to determine the approximate peak center of the photoelectric peak. Then with A preset energy window is selected as the center, and the energy spectrum is re-statistically analyzed within the window using finer energy bins to obtain a subdivided bin counting sequence.

[0064] Furthermore, the subdivided bin counting sequence is smoothed using a moving average to obtain a smoothed counting sequence; the energy corresponding to the maximum smoothed count is taken as the peak center. The maximum count is taken as the peak count. Using half the peak count as the half-height count, the positions where the count drops to the half-height count are searched on both sides of the peak center. The left half-height position is obtained by linear interpolation using the counts of adjacent bins. and the right half of the high position Therefore, calculate the full width at half height. Peak position shift according to Calculation, where The incident photon energy is used, and its absolute value can be taken as an evaluation index for peak position shift. The total collected charge is obtained by summing the total collected charge of each event, and the charge collection efficiency is calculated based on the ratio of collected charge to generated charge. The extracted indices are shown in Table 1 below.

[0065] Table 1 Extracted performance indicators

[0066] The aforementioned performance indicators include information on detection efficiency and signal strength, as well as information on energy resolution, peak position stability, charge collection, and high electric field engineering risks, providing a unified evaluation basis for subsequent multi-objective optimization.

[0067] S7: Construct a neural network training dataset.

[0068] The material parameters, design parameters, and energy spectrum performance indices obtained from S1 to S6 are organized into a supervised learning dataset. Each sample corresponds to a combination of a material, an incident energy, a detector thickness, and an electric field intensity. Input variables include material encoding, incident energy, detector thickness, and electric field intensity; output variables include peak count, total collected charge, full width at half maximum (FWHM), peak offset, and charge collection efficiency.

[0069] Material names are categorical variables and cannot be directly input into the neural network using material serial numbers, otherwise non-existent size relationships would be introduced. Therefore, this disclosure uses one-heat encoding to represent candidate materials. For example, when candidate materials include CsPbBr3, MAPbBr3, CZT, CdTe, GaAs, and S... i When there are six materials, the material code can be represented as follows: to ,in:

[0070] When a material is selected, its corresponding encoding bit is 1, and the remaining encoding bits are 0. This encoding method enables the neural network to independently learn the response differences of different material systems, avoiding spurious numerical relationships caused by material numbering. The input and output of the training dataset are shown in Table 2 below.

[0071] Table 2 Inputs and outputs of the training dataset

[0072] S8: Construct a neural network proxy model.

[0073] A neural network surrogate model was trained using the training dataset constructed in S7. This surrogate model was used to learn the nonlinear mapping relationship between material, incident energy, detector thickness, electric field strength, and multiple energy spectrum performance indicators. Since the physical calculation data obtained from S2 to S6 only cover a finite number of discrete parameter points, this neural network surrogate model was used to expand the discrete data into a fast prediction model in a continuous parameter space.

[0074] like Figure 5 As shown, the neural network input includes a 6-dimensional material unique thermal encoding and three continuous variables: incident energy, detector thickness, and electric field strength, for a total of 9 input quantities; the output includes smoothed peak count, total collected charge, smoothed full width at half maximum (FWHM), peak position shift, and charge collection efficiency, for a total of 5 output quantities.

[0075] Furthermore, to enhance the ability to express the coupling relationship between material category variables and continuous physical parameters, this disclosure preferably employs a multi-output neural network surrogate model with a differentiated structure. This model is designed according to functional divisions including input feature fusion, nonlinear response mapping, performance feature compression, multi-index regression representation, and multi-index output. These functional divisions are used to illustrate the roles of different network layers or modules in the surrogate model and do not constitute a limitation on the number of network layers, the number of neurons per layer, or the type of activation function.

[0076] In one specific embodiment, the input feature fusion layer includes a first linear transformation layer, a ReLU activation layer, and a Dropout layer. The first linear transformation layer maps the input features to 128-dimensional fused features. When there are six candidate materials, the input features include 6-dimensional material unique thermal encoding, incident energy, detector thickness, and operating electric field strength, for a total of 9 input quantities. The nonlinear response mapping layer includes a second linear transformation layer, a layer normalization layer, and a SiLU activation layer. The second linear transformation layer has both 128 input and output dimensions. The performance feature compression layer includes a third linear transformation layer and a GELU activation layer. The third linear transformation layer compresses the 128-dimensional features into 64-dimensional features. The multi-index regression characterization layer includes a fourth linear transformation layer and a ReLU activation layer. The fourth linear transformation layer compresses the 64-dimensional features into 32-dimensional features. The multi-index output layer includes a fifth linear transformation layer, which maps 32-dimensional features into five continuous output indices, including peak count, total collected charge, full width at half maximum (FWHM), peak offset, and charge collection efficiency. The multi-index output layer does not have an activation function to accommodate the regression prediction of continuous physical performance indices.

[0077] For peak counts and total collected charge with large numerical ranges, it is preferable to perform a logarithmic transformation before standardized training, followed by an inverse transformation after prediction. This approach avoids large-scale metrics dominating the loss function and improves the stability of multi-output regression training. Both input and output variables can be standardized to mitigate the impact of parameters with different dimensions on the training process.

[0078] The training and test sets can be hierarchically divided according to the combination of materials and incident energies to ensure that different materials and different energy points are representative in both the training and test sets. During training, the mean squared error loss function and adaptive optimization algorithm can be used, and an early stopping mechanism can be set to use the model parameters with the lowest loss on the test set as the final model.

[0079] After training, the neural network surrogate model is no longer just used as a fitting tool, but as a fast performance evaluator for subsequent genetic algorithms. For a large number of candidate material-thickness-electric field combinations generated by the genetic algorithm, the surrogate model can quickly predict their performance indicators at multiple energy points, thereby avoiding the need to repeatedly perform photon transport and energy deposition simulations and charge collection correction calculations for each candidate design.

[0080] As one embodiment, the neural network proxy model is not limited to the differentiated fully connected network described in this embodiment, but may also employ convolutional neural networks, residual networks, Transformers, graph neural networks, radial basis function networks, Gaussian process regression, random forests, gradient boosting trees, support vector regression, physical information neural networks, or ensemble learning models.

[0081] Training data is not limited to pure simulation data; it can also incorporate experimentally measured energy spectra, calibration source data, manufacturer material parameters, literature data, or semi-empirical model data. Alternatively, an active learning approach can be adopted, where the uncertainty of the surrogate model guides the selection of parameters for the next round of physical simulation.

[0082] S9: Genetic algorithm for multi-parameter intelligent optimization.

[0083] like Figure 6 As shown, material, detector thickness, and electric field strength are used as the encoding parameters for individuals in the genetic algorithm. Material is a discrete variable, while detector thickness and electric field strength are continuous variables. The algorithm initializes a population of candidate individuals; for each candidate individual, it calls the neural network surrogate model trained with S8 to predict its peak count, total collected charge, full width at half maximum (FWHM), peak position shift, and charge collection efficiency at multiple incident energy points.

[0084] To comprehensively evaluate detector design, this disclosure constructs a multi-objective fitness function. For each incident energy point, the fitness score can be expressed as:

[0085] in, The peak count is normalized to a score. To normalize the total collected charge score, The score is the half-height and full-width normalized score. The score is the peak position shift normalized score. The normalized score for charge collection efficiency, As a saturation penalty factor, to These are weighting coefficients. Higher peak count, total collected charge, and charge collection efficiency result in higher scores; smaller full width at half maximum (FWHM) and peak offset result in higher scores; a larger saturation penalty factor results in lower overall fitness.

[0086] For multiple incident energy points, the values ​​at each energy point can be calculated separately. The average or weighted average value is then taken as the overall fitness of the candidate design. Therefore, the genetic algorithm does not optimize only for a single energy point, but rather performs collaborative optimization of the overall response at multiple X-ray or gamma-ray energy points.

[0087] To prevent the optimization algorithm from continuing to select excessively high electric fields after the charge collection efficiency has already reached saturation, this embodiment introduces an electric field saturation penalty term. This penalty term does not deduct points for all electric field conditions, but rather gradually increases when the charge collection efficiency has already reached a high level, the benefit of further increasing the electric field is small, and the electric field itself is already high. Given the material, incident energy, detector thickness, and electric field strength E, the charge collection efficiency under the current electric field is first predicted by a neural network surrogate model. Then, keeping the material, incident energy, and thickness constant, the electric field is increased by a predetermined increment ΔE, resulting in:

[0088] The neural network surrogate model predicts the charge collection efficiency after improving the electric field:

[0089] Further calculation of the marginal gain of charge collection efficiency:

[0090] In a specific embodiment, a charge collection efficiency saturation threshold is set. The electric field increment is Charge collection efficiency marginal gain tolerance The electric field search range is... , The operating point is considered to have entered the high electric field saturation region only when both of the following conditions are met simultaneously, and the electric field saturation penalty term is calculated: (1) The current charge collection efficiency is relatively high:

[0091] (2) The gain in charge collection efficiency is relatively small after further increasing the electric field:

[0092] At this point, the electric field penalty term is defined as:

[0093] And The value of is limited to the range of 0 to 1. The specific penalty details and results are shown in Table 3.

[0094] Table 3. Penalties and Results

[0095] From the above definition, it can be seen that, Three factors are considered simultaneously: first, whether the current charge collection efficiency is already high; second, whether the gain in charge collection efficiency from further increasing the electric field is already minimal; and third, whether the current electric field is already too high within the search range. Therefore, this penalty term can suppress the unengineered result of "blindly choosing a higher electric field even when the charge collection efficiency is close to saturation".

[0096] Engineering constraints are introduced during the iterative process of the genetic algorithm. These constraints include: (1) Semiconductor material thickness constraint: The detector thickness is within a preset upper and lower limit range, for example, 1 mm to 20 mm; (2) Peak position offset constraint: The predicted peak position offset shall not exceed a preset proportion of the incident energy, such as 5%; (3) Total collected charge constraint: The total collected charge shall not be lower than the preset proportion of the reference charge of the same material and energy, for example, 10%; (4) Charge collection efficiency constraint: The charge collection efficiency at each incident energy point shall not be lower than a preset threshold, for example, 20%; (5) High electric field risk constraint: When the candidate solution is in the high electric field region and the charge collection efficiency is not significantly improved, its fitness is reduced by saturation penalty, and the design scheme with lower operating voltage and lower power consumption is preferred.

[0097] While a high electric field may improve carrier collection efficiency to a certain extent, it also leads to increased leakage current and electronic noise, increased risk of breakdown, increased power consumption, and higher requirements for bias power supply, insulation design, and long-term stability.

[0098] For space satellite payloads, lower operating voltages help reduce power supply system complexity, alleviate thermal management stress, and improve long-term on-orbit reliability. Therefore, the method disclosed herein... By coupling the benefits of charge collection with the costs of electric field engineering, the optimization results not only have better energy spectrum performance, but also better meet the engineering requirements of low voltage, low power consumption and high reliability.

[0099] As an example, the intelligent optimization algorithm is not limited to the genetic algorithm, but may also employ particle swarm optimization, Bayesian optimization, simulated annealing, differential evolution, ant colony optimization, multi-objective NSGA-II, reinforcement learning optimization, grid search and local search combined methods, or combinations of the above algorithms.

[0100] The fitness function is not limited to the weighted sum of peak value, total collected charge, full width at half maximum (FWHM), peak position shift, charge collection efficiency, and saturation penalty. It can also incorporate indicators such as detection efficiency, peak-to-compass ratio, energy linearity, noise, dark current, leakage current, breakdown electric field, bias voltage, material cost, fabrication difficulty, stability, response time, temporal resolution, and spatial resolution, depending on the specific application.

[0101] S10: Outputs the optimal detector design results.

[0102] Based on the optimization results of the S9 genetic algorithm, the optimal or preferred detector design scheme that satisfies multi-objective evaluation and engineering constraints is output. The output results include optimal material, optimal detector thickness, optimal electric field strength, corresponding bias voltage, and predicted performance indicators at multiple incident energy points.

[0103] Predicted performance metrics include peak count, total collected charge, full width at half maximum (FWHM), peak offset, charge collection efficiency, saturation penalty factor, and overall fitness value. For space satellite payload applications, it can also simultaneously output engineering evaluation information such as predicted operating voltage, relative power consumption level, high electric field risk indicator, and whether low-voltage operating constraints are met.

[0104] As one embodiment, the present invention is not limited to hard X-ray detectors, but can also be extended to gamma-ray detectors, nuclear radiation spectrometers, photon counting detectors, pixel array detectors, flat panel detectors, space high-energy particle detectors, and other semiconductor radiation detectors that require synergistic optimization of material-thickness-electric field parameters.

[0105] Example 2 One embodiment of this disclosure provides a multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors. A set of specific simulation embodiments illustrates the implementation of this disclosure, as follows: Step 1: Candidate materials, energy points, and design space setup.

[0106] In this embodiment, candidate semiconductor detector materials include CsPbBr3, MAPbBr3, CZT, CdTe, 4H-SiC, and S. i The aforementioned materials include perovskite semiconductors, traditional compound semiconductors, and wide-bandgap semiconductors, and can be used to compare the overall performance of different material systems under hard X-ray / gamma-ray detection conditions.

[0107] The main parameters of the candidate materials include average ionization energy, material density, electron mobility lifetime product, hole mobility lifetime product, and Fano factor. The average ionization energy is used to convert energy deposition into electron-hole pairs; density affects the absorption capacity of incident photons in the material; the electron and hole mobility lifetime products are used to calculate carrier drift length and charge collection efficiency; and the Fano factor characterizes the effect of ionization statistical fluctuations on spectral broadening.

[0108] In this embodiment, the candidate material parameters are shown in Table 4 below.

[0109] Table 4 Candidate Material Parameters

[0110] In this embodiment, the incident energy points are set to 59.6 keV, 80.9 keV, 122.1 keV, and 136.5 keV; the detector thickness search range is 1 mm to 20 mm; and the electric field strength search range is 50 V / cm to 3500 V / cm. The above thickness and electric field range are merely examples and can be adjusted according to the detector's structural dimensions, material breakdown field strength, readout circuit withstand voltage capability, high-voltage power supply capability, leakage current level, and the low-power requirements of the space satellite payload.

[0111] Step 2: Photon transport and energy deposition data processing.

[0112] For each candidate material and each incident energy point, a photon transport and energy deposition model for X-rays / γ-rays in a semiconductor detector is established. The detector model includes an external space region, an incident window, metal electrodes, and a semiconductor. The semiconductor serves as the statistical region for energy deposition, used to record event-by-event interaction locations, interaction processes, and deposited energy.

[0113] In this embodiment, the simulation output data includes event-by-event photon interaction step data and primary particle data. For each interaction step, the local deposition energy is calculated based on the energy difference before and after the interaction, and the interaction location is recorded. Subsequently, all deposition steps within the same event are categorized according to event number, providing basic data for subsequent charge collection correction and energy spectrum reconstruction.

[0114] The energy spectrum feature processing flow used in this embodiment includes: reading energy deposition data under different materials, energies, thicknesses, and electric field conditions; calculating the generation and statistical fluctuations of electron-hole pairs based on the average ionization energy of the material and the Fano factor; calculating the contribution of electrons and holes to induced charges based on the charge collection model; converting the total collected charge of the event into output energy; and finally, statistically analyzing the output energy of all events to obtain the output energy spectrum under the corresponding conditions.

[0115] Step 3: Genetic Algorithm Setup.

[0116] Each individual in the genetic algorithm consists of three genes: material encoding, detector thickness, and electric field strength. Material encoding is a discrete variable, while thickness and electric field strength are continuous variables. This embodiment sets the population size to 160, the number of generations to 50, the crossover probability to 0.7, the mutation probability to 0.25, and employs tournament selection. The fitness function integrates peak count, total collected charge, full width at half maximum (FWHM), peak position shift, charge collection efficiency, and a saturation penalty factor, while simultaneously imposing constraints on thickness range, upper limit of peak position shift, lower limit of charge collection efficiency, lower limit of total collected charge, and electric field risk.

[0117] Step 4: Global hybrid material optimization results.

[0118] In the global optimization process of mixed materials, genetic algorithms can be applied to CsPbBr3, MAPbBr3, CZT, CdTe, 4H-SiC, and S. i The algorithm automatically selects from six materials while simultaneously optimizing thickness and electric field strength. Based on the optimization results, the top ten globally optimal individuals all converge to the CsPbBr3 material, with a thickness of approximately 2.00 mm, an electric field strength of approximately 215 V / cm, and a fitness of approximately 3.54742.

[0119] In this embodiment, a set of preferred solutions obtained by global hybrid material optimization is shown in Table 5.

[0120] Table 5 Optimal Solution for Global Hybrid Materials

[0121] The predicted performance of the optimal solution for the global hybrid material at various energy points is shown in Table 6 below.

[0122] Table 6 Prediction performance at various energy points

[0123] As the results above show, the proposed scheme maintains a charge collection efficiency of approximately 0.99 at multiple energy points, while exhibiting small full width at half maximum (FWHM) and peak offset. The preferred electric field strength is approximately 215.16 V / cm, falling within the lower electric field operating range, which is beneficial for reducing leakage current and noise, lowering breakdown risk, and reducing power consumption. Therefore, this preferred scheme not only possesses good spectral performance but also better meets the design requirements of low-voltage, low-power, and high-reliability detectors, making it particularly suitable for applications sensitive to power supply and thermal management, such as space satellite payloads.

[0124] Furthermore, to compare the comprehensive performance of different materials under their respective optimal thickness and electric field strength, the material types were further fixed, and genetic algorithms were used for optimization. The optimization results for fixed materials are shown in Table 7 below.

[0125] Table 7 Results of Fixed Material Optimization

[0126] As can be seen from the comparison of fixed materials, there are significant differences in the optimal thickness and electric field strength among different materials. CsPbBr3 and MAPbBr3 can achieve high overall adaptability under lower electric fields; CZT and CdTe require higher electric fields to improve charge collection; 4H-SiC needs to reach the upper limit of the electric field under the conditions of this embodiment, and has lower peak count and total collected charge; Si can achieve high charge collection efficiency under low electric fields, but requires a larger thickness to compensate for its lower absorption capacity.

[0127] The results show that the method disclosed in this paper can automatically compare the comprehensive performance of different material systems under multiple energy points, multiple thicknesses, and multiple electric fields, and can output the global optimal solution or the local optimal solution under fixed material conditions according to different application requirements.

[0128] To further explain the electric field saturation penalty term The role of this will be illustrated below using four candidate operating points under the conditions of a certain candidate material, a certain incident energy, and a certain detector thickness. Let the electric field search range be... to Charge collection efficiency saturation threshold Electric field increment Charge collection efficiency marginal gain tolerance .

[0129] According to the electric field saturation penalty term calculation method described in S9 of Example 1, different candidate operating points were compared, and the results are shown in Table 8 below.

[0130] Table 8 Comparison results of different candidate operating points

[0131] As shown in Table 8, the charge collection efficiency at operating points A and B has not yet reached the saturation threshold of 0.95, and increasing the electric field may still improve carrier collection; therefore, no electric field saturation penalty is imposed. The charge collection efficiencies at operating points C and D are already relatively high, and further increasing the electric field yields minimal gain, indicating that both are close to the charge collection saturation region. However, the electric field at operating point C is only 200 V / cm, which is a relatively low electric field operating point; therefore... The electric field at the operating point D reaches 1500 V / cm, which is relatively high even though the benefit of charge collection is already small. It significantly increases and reduces the overall score of the candidate solution in the fitness function.

[0132] Therefore, this embodiment shows that, It's not simply a matter of "the higher the electric field, the more points are deducted." Instead, it suppresses higher electric field schemes when charge collection efficiency is already close to saturation and further increasing the electric field yields limited benefits. This mechanism enables the genetic algorithm to prioritize design schemes with "energy spectrum performance close to saturation but lower operating electric field," thereby reducing leakage current, electronic noise, breakdown risk, and power consumption, better meeting the engineering requirements of low voltage, low power consumption, and high reliability.

[0133] The results of global optimization and fixed material optimization show that this method does not simply select a single material with the highest density, the highest carrier lifetime product, or the highest charge collection efficiency. Instead, it makes a comprehensive trade-off between peak count, total collected charge, full width at half maximum (FWHM), peak position shift, charge collection efficiency, saturation penalty, and engineering constraints.

[0134] The global hybrid material optimization results selected CsPbBr3, with an optimal electric field of approximately 215 V / cm. This indicates that under this material system, a lower electric field is sufficient to achieve high charge collection efficiency and superior energy spectrum performance. Compared to CZT, CdTe, or 4H-SiC solutions that require electric fields in the kilovolt-per-cm range, the low-electric-field solution reduces leakage current and noise, lowers the risk of breakdown, and reduces power consumption and high-voltage power supply burden. For space satellite payloads, the lower operating voltage can also improve power supply reliability, reduce thermal management stress, and improve long-term on-orbit stability.

[0135] On the other hand, although Si can achieve high charge collection efficiency and small peak position shift at 50V / cm, its optimal thickness reaches 20mm, indicating that a larger thickness is required to compensate for insufficient absorption capacity under the energy range and objective function settings of this embodiment. 4H-SiC has potential advantages such as wide bandgap, high temperature resistance, and radiation resistance, but under the hard X-ray energy point and fitness function set in this embodiment, its peak count and total collected charge are low, resulting in a lower final fitness than other materials.

[0136] Therefore, this embodiment demonstrates that the present disclosure can automatically obtain detector design schemes that meet multiple performance objectives and engineering constraints within a continuous thickness and continuous electric field design space by utilizing limited photon transport and energy deposition simulation data, through charge collection correction, neural network surrogate models, and genetic algorithm optimization. This method is applicable to material selection, thickness design, electric field setting, and low-voltage optimization design of space payloads for hard X-ray / γ-ray semiconductor detectors.

[0137] Example 3 One embodiment of this disclosure provides a multi-parameter intelligent optimization system for X / γ-ray semiconductor detectors, comprising: The parameter input module is used to obtain candidate semiconductor materials and design parameters; The photon transport and energy deposition module is used to perform event-by-event simulations of X-rays or γ-rays with different incident energies based on candidate semiconductor materials and design parameters, and to obtain deposited photon data. The charge collection correction module is used to calculate the collection charge corresponding to each energy deposition based on the deposited photon data and reconstruct the detector output energy spectrum; The dataset construction module is used to extract features from the output energy spectrum to obtain multiple performance indicators of the energy spectrum; The neural network proxy module is used to construct index samples by combining multiple energy spectrum performance indicators. The index samples are then input into the neural network proxy model to learn the nonlinear mapping relationship between materials, incident energy, detector thickness, electric field strength, and multiple energy spectrum performance indicators. The genetic algorithm optimization module is used to search for the optimal material, detector thickness, electric field strength, and corresponding prediction performance under engineering constraints, using a neural network model as an evaluator. The results output module is used to output the optimal design parameters and the corresponding prediction performance.

[0138] Example 4 One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned intelligent multi-parameter optimization method for X / γ-ray semiconductor detectors.

[0139] Example 5 One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the aforementioned intelligent multi-parameter optimization method for X / γ-ray semiconductor detectors.

[0140] Example 6 One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned X / γ-ray semiconductor detector multi-parameter intelligent optimization method.

[0141] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0142] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0143] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors, characterized in that, include: Obtain candidate semiconductor materials and design parameters; Based on candidate semiconductor materials and design parameters, an event-by-event simulation of X-rays or γ-rays with different incident energies is performed using a photon transport and energy deposition model to obtain deposited photon data. Based on the deposited photon data, the collection charge corresponding to each energy deposition is calculated, and the detector output energy spectrum is reconstructed. Feature extraction is performed on the output energy spectrum to obtain multiple performance indicators of the energy spectrum. By combining multiple energy spectrum performance indicators to construct an indicator sample, the indicator sample is input into a neural network surrogate model to learn the nonlinear mapping relationship between material, incident energy, detector thickness and electric field strength and multiple energy spectrum performance indicators. Using a neural network model as an evaluator, we search for the optimal material, detector thickness, electric field strength, and corresponding prediction performance under engineering constraints. The process of using a neural network model as an evaluator to search for the optimal material, detector thickness, electric field strength, and corresponding prediction performance under engineering constraints includes: Material, detector thickness, and electric field strength are used as individual encodings in a genetic algorithm; Among them, the material is a discrete variable, while the detector thickness and electric field strength are continuous variables; The algorithm initializes a set number of candidate individuals to form a population; for each candidate individual, a neural network proxy model is called to predict its peak count, total collected charge, full width at half maximum (FWHM), peak position shift, and charge collection efficiency at multiple incident energy points; Construct a multi-objective fitness function. For each incident energy point, calculate the fitness score. For multiple incident energy points, calculate the fitness score for each energy point separately, and then take the average or weighted average as the comprehensive fitness of the candidate individual. During the iteration process of the genetic algorithm, engineering constraints are introduced, and the optimal design scheme that satisfies the multi-objective evaluation and engineering constraints is output. The output results include the optimal material, the optimal detector thickness, the optimal electric field strength, the corresponding bias voltage, and the predicted performance index at multiple incident energy points.

2. The multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors as described in claim 1, characterized in that, The acquisition of candidate semiconductor materials and design parameters includes: The candidate semiconductor materials include one or more of CsPbBr3, MAPbBr3, CZT, CdTe, 4H-SiC, and Si; The design parameters include material properties, detector structural parameters, incident photon energy points, and engineering constraints. The material properties include average ionization energy, electron mobility lifetime product, hole mobility lifetime product, and Fano factor; the engineering constraints include upper limit of peak position shift, lower limit of charge collection efficiency, lower limit of total collected charge, upper limit of electric field strength, upper limit of bias voltage, breakdown risk limit, leakage current limit, noise limit, and power consumption limit.

3. The multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors as described in claim 1, characterized in that, Based on candidate semiconductor materials and design parameters, an event-by-event simulation of X-rays or gamma rays with different incident energies is performed using a photon transport and energy deposition model to obtain deposited photon data, including: Based on candidate semiconductor materials and design parameters, a photon transport and energy deposition model for X-ray or gamma-ray photons in a semiconductor detector is established. Event-by-event simulations of X-rays or γ-rays with different incident energies were performed using a photon transport and energy deposition model. For each incident photon event, its propagation path, interaction location, interaction process, energy before and after interaction, and corresponding energy deposition in the semiconductor detector were recorded. For each interaction step, the local deposition energy is calculated based on the energy difference before and after the interaction, and the coordinates of the interaction position are retained. All deposition step data within the same event are classified and exported according to the event number to obtain the deposition photon data.

4. The multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors as described in claim 1, characterized in that, The calculation of the collection charge corresponding to each energy deposition based on the deposited photon data, and the reconstruction of the detector output energy spectrum, includes: Based on the deposited photon data, combined with the material's average ionization energy, Fano factor, electron mobility lifetime product, hole mobility lifetime product, detector thickness, and electric field strength, the collection charge corresponding to each energy deposition step is calculated. The total collected charge of the event is obtained by summing the collected charges of all steps within the same event. The total collected charge is then converted into output energy. The output energy of all incident events is statistically analyzed to obtain the output energy spectrum under the corresponding material, incident energy, detector thickness, and electric field strength.

5. The multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors as described in claim 1, characterized in that, The method involves constructing an index sample by combining multiple energy spectrum performance indicators, inputting the index sample into a neural network surrogate model, and learning the nonlinear mapping relationship between material, incident energy, detector thickness, electric field strength, and multiple energy spectrum performance indicators, including: The obtained candidate materials, design parameters, and multiple energy spectrum performance indicators are organized into a supervised learning dataset. Each sample corresponds to a combination of a material, an incident energy, a detector thickness, and an electric field intensity. The input variables include material encoding, incident energy, detector thickness, and electric field intensity. The output variables include peak count, total collected charge, full width at half maximum (FWHM), peak position shift, and charge collection efficiency. Candidate materials are represented by unique thermal encoding. The neural network surrogate model learns the nonlinear mapping relationship between materials, incident energy, detector thickness and electric field intensity and multiple energy spectrum performance indicators, expanding discrete data into a fast prediction model in a continuous parameter space.

6. A multi-parameter intelligent optimization system for X / γ-ray semiconductor detectors, characterized in that, Specifically, the method for intelligent multi-parameter optimization of an X / γ-ray semiconductor detector as described in any one of claims 1-5 includes: The parameter input module is used to obtain candidate semiconductor materials and design parameters; The photon transport and energy deposition module is used to perform event-by-event simulations of X-rays or γ-rays with different incident energies based on candidate semiconductor materials and design parameters, and to obtain deposited photon data. The charge collection correction module is used to calculate the collection charge corresponding to each energy deposition based on the deposited photon data and reconstruct the detector output energy spectrum; The dataset construction module is used to extract features from the output energy spectrum to obtain multiple performance indicators of the energy spectrum; The neural network proxy module is used to construct index samples by combining multiple energy spectrum performance indicators. The index samples are then input into the neural network proxy model to learn the nonlinear mapping relationship between materials, incident energy, detector thickness, electric field strength, and multiple energy spectrum performance indicators. The genetic algorithm optimization module is used to search for the optimal material, detector thickness, electric field strength, and corresponding prediction performance under engineering constraints, using a neural network model as an evaluator. The results output module is used to output the optimal design parameters and the corresponding prediction performance.

7. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-parameter intelligent optimization method for X / γ-ray semiconductor detectors as described in any one of claims 1-5.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the multi-parameter intelligent optimization method for an X / γ-ray semiconductor detector as described in any one of claims 1-5.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform a multi-parameter intelligent optimization method for an X / γ-ray semiconductor detector as described in any one of claims 1-5.

Images