Method, program, and device for adaptively generating element characteristics on basis of artificial intelligence

Abstract

According to an embodiment of the present disclosure, disclosed are a method, program, and device for adaptively generating element characteristics on the basis of artificial intelligence, which is performed by a computing device. The method may comprise the steps of: obtaining reference function-type data related to characteristics of a semiconductor element and an adjustment ratio for an electrical parameter included in the reference function-type data; and receiving the reference function-type data and the adjustment ratio, and training a neural network model to generate target function-type data from the reference function-type data according to the adjustment ratio while maintaining a unique form of the reference function-type data.

Description

Method, program and device for adaptive generation of device characteristics based on artificial intelligence

[0001] The present disclosure relates to deep learning technology in the semiconductor field, and more specifically, to a method for adaptively adjusting device characteristic curves for semiconductor device simulation using artificial intelligence.

[0002]

[0003] With the continuous advancement of transistor technology, devices are becoming smaller and more complex, and finding optimal manufacturing parameters that achieve satisfactory performance and power consumption is one of the biggest challenges in the semiconductor industry. This process requires numerous unnecessary fabrication and measurement steps, followed by iterative feedback loops that can take weeks. Consequently, this process significantly increases development costs, making advanced technologies even more expensive. Traditionally, analytical compact models used in circuit simulations are developed to achieve high accuracy and efficient computational complexity. These models typically incorporate numerous empirical equations, including physics-based equations and fitting parameters. However, as devices become smaller, new and complex physical phenomena frequently emerge, requiring increasingly sophisticated model equations that exceed human intuition. Consequently, analytical compact modeling procedures typically require complex domain knowledge, which is difficult to understand and time-consuming to evaluate.

[0004] In this regard, learning-based neural network compact models are emerging as a promising alternative because they provide rapid evaluation and are primarily based on observational data. Furthermore, the general nature of machine learning models allows knowledge to be transferred from existing technologies to new ones, reducing the need for frequent model development. Furthermore, because these technologies are based on black-box models, they ensure the confidentiality of transistor and manufacturing details and protect foundry proprietary information. However, neural network compact models' predictions are often inaccurate due to their strict reliance on training data. This is especially true when there is a mismatch between the device characteristics in the training dataset and the target characteristics measured or predicted in real-world scenarios. Simple methods to mitigate this mismatch in neural network compact models often require meticulous tuning of individual examples or backpropagation procedures, resulting in unsatisfactory predictions and requiring multiple iterations to achieve the desired fit.

[0005]

[0006] The present disclosure is conceived in response to the aforementioned background technology, and aims to provide a framework for learning and inference of an artificial intelligence model capable of adjusting reference characteristics to target characteristics measured or expected through simulation while maintaining the inherent form of reference characteristics of semiconductor devices.

[0007] However, the problems to be solved in this disclosure are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood based on the description below.

[0008]

[0009] According to one embodiment of the present disclosure for achieving the aforementioned task, a method for adaptive generation of device characteristics based on artificial intelligence, performed by a computing device, is disclosed. The method may include the steps of: obtaining reference functional data regarding characteristics of a semiconductor device and an adjustment ratio for electrical parameters constituting the reference functional data; and a step of training a neural network model to receive the reference functional data and the adjustment ratio and generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data.

[0010] Alternatively, at least one of the reference functional data or the target functional data may be selected through random sampling from a database generated by performing a simulation based on a plurality of process variables for the semiconductor device.

[0011] Alternatively, the step of training the neural network model may include the steps of: inputting the reference functional data into a first neural network block to generate latent features; inputting the latent features and the adjustment ratio into a second neural network block to modify the latent features; and inputting the modified latent features into a third neural network block to generate modified functional data.

[0012] Alternatively, the latent feature may include a first feature part that is common to semiconductor devices, and a second feature part that can be varied depending on the type or characteristics of the semiconductor device.

[0013] Alternatively, the second neural network block can transform the latent feature by replacing the second feature part with an encoded feature for the adjustment ratio.

[0014] Alternatively, the second neural network block can perform a cross-attention operation based on the encoded features for the latent features and the adjustment ratio.

[0015] Alternatively, the learning of the neural network model may be performed using a first loss function for calculating a reconstruction error between the transformed functional data and the reference functional data, and a second loss function for calculating a mapping error between the transformed functional data and the target functional data.

[0016] Alternatively, the overall loss function for learning the neural network model can be expressed as a weighted sum of the first loss function and the second loss function.

[0017] Meanwhile, according to another embodiment of the present disclosure for realizing the aforementioned task, a method for adaptively generating device characteristics based on artificial intelligence, performed by a computing device, is disclosed. The method may include the steps of: obtaining first functional data regarding characteristics of a semiconductor device and a first adjustment ratio for electrical parameters constituting the first functional data; and using a neural network model, generating second functional data desired by a user based on the first functional data and the first adjustment ratio. In this case, the neural network model may be pre-trained to receive reference functional data and a second adjustment ratio for electrical parameters constituting the reference functional data, and generate target functional data from the reference functional data in accordance with the second adjustment ratio while maintaining the unique form of the reference functional data.

[0018] According to one embodiment of the present disclosure for achieving the above-described task, a computer program stored in a computer-readable storage medium is disclosed. When the computer program is executed on one or more processors, it performs operations for adaptive generation of device characteristics based on artificial intelligence. At this time, the operations may include an operation of obtaining reference functional data regarding characteristics of a semiconductor device and an adjustment ratio for electrical parameters constituting the reference functional data; and an operation of training a neural network model to receive the reference functional data and the adjustment ratio and generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data.

[0019] According to one embodiment of the present disclosure for realizing the above-described task, a computing device for adaptive generation of device characteristics based on artificial intelligence is disclosed. The device may include a processor including at least one core; a memory including program codes executable by the processor; and a network unit for obtaining reference functional data regarding characteristics of a semiconductor device and an adjustment ratio for electrical parameters constituting the reference functional data. In this case, the processor may receive the reference functional data and the adjustment ratio, and train a neural network model to generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data.

[0020]

[0021] The present disclosure can solve the problem of mismatch between the reference characteristics of semiconductor devices and the target characteristics of simulation, which is encountered in existing methods, by adjusting the reference characteristics to the target characteristics by matching a series of electrical parameters while maintaining the unique form of the reference characteristics.

[0022] In particular, the present disclosure does not require fine-tuning for individual cases or a separate backpropagation process to alleviate mismatch problems, unlike existing neural network-based methods, and thus enables faster and more accurate prediction of device characteristics.

[0023]

[0024] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0025] FIG. 2 is a schematic diagram of a framework for adaptive generation of device characteristics according to one embodiment of the present disclosure.

[0026] FIG. 3 is a graph showing test results of a framework for adaptive generation of device characteristics according to one embodiment of the present disclosure.

[0027] FIG. 4 is a flowchart illustrating an artificial intelligence learning method for adaptive generation of device characteristics according to one embodiment of the present disclosure.

[0028] FIG. 5 is a flowchart illustrating an artificial intelligence inference method for adaptive generation of device characteristics according to one embodiment of the present disclosure.

[0029]

[0030] Below, embodiments of the present disclosure are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present disclosure. The embodiments presented in this disclosure are provided to enable those skilled in the art to utilize or implement the contents of the present disclosure. Accordingly, various modifications to the embodiments of the present disclosure will be apparent to those skilled in the art. That is, the present disclosure may be implemented in various different forms and is not limited to the embodiments described below.

[0031] Throughout the specification of this disclosure, identical or similar drawing numbers refer to identical or similar components. Furthermore, for clarity in the description of this disclosure, drawing numbers for parts unrelated to the description of this disclosure may be omitted in the drawings.

[0032] The term "or" as used herein is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified herein or clear from context, "X employs A or B" should be understood to mean either of the natural inclusive permutations. For example, unless otherwise specified herein or clear from context, "X employs A or B" can be interpreted to mean either X employs A, X employs B, or X employs both A and B.

[0033] The term "and / or" as used herein should be understood to refer to and include all possible combinations of one or more of the related concepts listed.

[0034] The terms "comprises" and / or "comprising" as used herein should be understood to mean the presence of certain features and / or components. However, it should be understood that the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other features, other components, and / or combinations thereof.

[0035] Unless otherwise specified in this disclosure or unless the context makes it clear that the singular form is intended to be referred to, the singular should generally be construed to include “one or more.”

[0036] The term "Nth (N is a natural number)" used in the present disclosure can be understood as an expression used to mutually distinguish components of the present disclosure based on a predetermined standard such as a functional perspective, a structural perspective, or convenience of explanation. For example, components performing different functional roles in the present disclosure can be distinguished as a first component or a second component. However, components that are substantially the same within the technical spirit of the present disclosure but must be distinguished for convenience of explanation may also be distinguished as a first component or a second component.

[0037] The term "acquisition" as used in this disclosure may be understood to mean not only receiving data through a wired or wireless communication network with an external device or system, but also generating data in an on-device form.

[0038] Meanwhile, the term "module" or "unit" used in the present disclosure can be understood as a term referring to an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a part thereof, hardware or a part thereof, or a combination of software and hardware. At this time, the "module" or "unit" may be a unit composed of a single element, or a unit expressed as a combination or set of multiple elements. For example, as a narrow concept, a "module" or "unit" may refer to a hardware element of a computing device or a set thereof, an application program that performs a specific function of software, a processing process implemented through software execution, or a set of instructions for program execution, etc. In addition, as a broad concept, a "module" or "unit" may refer to the computing device itself that constitutes the system, or an application running on the computing device, etc. However, since the above-described concept is only an example, the concept of “module” or “part” may be defined in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0039] The term "model" as used herein may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a set of software units to solve a specific problem, or an abstract model of a processing process to solve a specific problem. For example, a neural network "model" may refer to the entire system implemented as a neural network that has problem-solving capabilities through learning. In this case, the neural network can have problem-solving capabilities by optimizing the parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network or a set of neural networks that are a combination of multiple neural networks.

[0040] The term "data" used in this disclosure may include "images," signals, and the like. The term "image" used in this disclosure may refer to multidimensional data composed of discrete image elements. In other words, "image" may be understood as a term referring to a digital representation of an object visible to the human eye. For example, "image" may refer to multidimensional data composed of elements corresponding to pixels in a two-dimensional image. "Image" may refer to multidimensional data composed of elements corresponding to voxels in a three-dimensional image.

[0041] The term "block" used in this disclosure can be understood as a set of configurations categorized based on various criteria, such as type and function. Therefore, the configurations classified as a single "block" can vary depending on the criteria. For example, a neural network "block" can be understood as a set of neural networks including at least one neural network. In this case, the neural networks included in the neural network "block" can be assumed to perform specific operations identically.

[0042] The term "functional data" used in this disclosure can be understood as multivariate data having infinite dimensional values. For example, "functional data" may be data that includes a continuously changing curve, such as a curve relating to device characteristics that represents changes in voltage with respect to current.

[0043] The term "electrical parameter" as used in the present disclosure can be understood as a measurable characteristic that defines the operation and performance of an electrical component, device, or system under various conditions. For example, an "electrical parameter" with respect to the characteristics of a metal-oxide-semiconductor field-effect transistor (MOSFET) is V, which represents the voltage difference between the drain and the source. DS , V represents the voltage difference between the gate and the source. GS , the current flowing through the drain, I D It may include:

[0044] The explanation of the above terms is intended to aid understanding of the present disclosure. Therefore, unless explicitly stated as limiting the contents of the present disclosure, it should be noted that the above terms are not intended to limit the technical ideas of the contents of the present disclosure.

[0045] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0046] A computing device (100) according to one embodiment of the present disclosure may be a hardware device or a part of a hardware device that performs comprehensive processing and calculation of data, or may be a software-based computing environment connected to a communication network. For example, the computing device (100) may be a server that performs intensive data processing functions and shares resources, or may be a client that shares resources through interaction with a server. In addition, the computing device (100) may be a cloud system in which multiple servers and clients interact to comprehensively process data. Since the above description is only one example related to the type of computing device (100), the type of computing device (100) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0047] Referring to FIG. 1, a computing device (100) according to one embodiment of the present disclosure may include a processor (110), a memory (120), and a network unit (130). However, FIG. 1 is merely an example, and the computing device (100) may include other components for implementing a computing environment. In addition, only some of the disclosed components may be included in the computing device (100).

[0048] The processor (110) according to one embodiment of the present disclosure may be understood as a configuration unit including hardware and / or software for performing computing operations. For example, the processor (110) may read a computer program to perform data processing for machine learning. The processor (110) may process computational processes such as processing input data for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. The processor (110) for performing such data processing may include a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application specific integrated circuit (ASIC), or a field programmable gate array (FPGA). The above-described type of the processor (110) is only one example, and thus, the type of the processor (110) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0049] The processor (110) can train a neural network model that transforms functional data regarding device characteristics for the simulation of semiconductor devices. The processor (110) can train the neural network model so that the neural network model generates functional data in which electrical parameters are transformed at a predetermined ratio while maintaining the original form of the input.

[0050] For example, the processor (110) may input reference functional data regarding the characteristics of a semiconductor device and an adjustment ratio for electrical parameters constituting the reference functional data into the neural network model. The reference functional data may be extracted from a database generated as a result of performing a simulation based on a plurality of process variables for the semiconductor device. Extraction of the reference functional data may be performed by random sampling so that the neural network model can learn various cases. Specifically, the reference functional data may be I D -V DS , I D -V GS , C GG -V GS It may include a curve expressing the relationship between electrical parameters regarding the device characteristics, such as the back. In addition, the adjustment ratio is a value indicating the ratio at which the electrical parameters constituting the curve included in the reference functional data should be adjusted, and may be determined through user input or random selection during the learning process.

[0051] The neural network model can generate transformed functional data by transforming reference functional data based on the adjustment ratio. The neural network model aims to perform learning so that the transformed functional data corresponds to target functional data that reflects the adjustment ratio while maintaining the original form of the reference functional data. The target functional data can be extracted from a database generated as a result of performing a simulation based on a plurality of process variables for a semiconductor device. The extraction of the target functional data can be performed by random sampling so that the neural network model can learn various cases. The processor (110) can calculate an error for the transformed functional data generated by the neural network model using a loss function. At this time, the loss function can include a first loss function for calculating a reconstruction error between the transformed functional data and the reference functional data, and a second loss function for calculating a mapping error between the transformed functional data and the target functional data. When the target functional data matches the reference functional data (i.e., the adjustment ratio is 0), the neural network model must accurately reconstruct the reference functional data, so the first loss function is a loss function for this purpose, and can be expressed as in [Mathematical Formula 1].

[0052]

[0053] Here, Mθ represents the neural network model and r represents the reference functional data.

[0054] When the target functional data is different from the reference functional data (i.e., when the adjustment ratio is not 0), the neural network model must accurately map the reference functional data to the target functional data, so the second loss function is a loss function to achieve this, and can be expressed as in [Mathematical Formula 2].

[0055]

[0056] Here, t represents the target functional data.

[0057] In summary, the overall loss function for training a neural network model can be expressed as a weighted sum of the first loss function and the second loss function, as in [Mathematical Equation 3].

[0058]

[0059] Here, N is the number of selections of s and t from the database P, and α is a coefficient that controls the balance between the first loss function and the second loss function.

[0060] When an error is calculated based on this loss function, the processor (110) can train the neural network model by adjusting the parameters of the neural network model so that the error is minimized. That is, the neural network model can be trained to receive reference functional data and an adjustment ratio as input, and generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data.

[0061] As described above, the processor (110) can transform functional data regarding the characteristics of a semiconductor device according to the simulation direction desired by the user using a pre-trained neural network model. The processor (110) can input functional data regarding the characteristics of a semiconductor device into the pre-trained neural network model, and transform the input functional data so that the device characteristics targeted by the user are generated. At this time, the processor (110) can input an adjustment ratio regarding electrical parameters constituting the functional data into the pre-trained neural network model together with the functional data. The adjustment ratio can be determined by user input. Through this, the processor (110) can effectively resolve the discrepancy between the device characteristics input into the neural network model and the device characteristics targeted by the user, thereby providing an environment in which diverse and stable simulations can be performed.

[0062] The memory (120) according to one embodiment of the present disclosure may be understood as a configuration unit including hardware and / or software for storing and managing data processed in the computing device (100). That is, the memory (120) may store any type of data generated or determined by the processor (110) and any type of data received by the network unit (130). For example, the memory (120) may include at least one type of storage medium among a flash memory type, a hard disk type, a multimedia card micro type, a card type memory, a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk. In addition, the memory (120) may also include a database system that controls and manages data in a predetermined system. The type of memory (120) described above is only one example, and thus the type of memory (120) can be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0063] The memory (120) can structure and organize and manage data, combinations of data, and program codes executable by the processor (110) required for the processor (110) to perform operations. For example, the memory (120) can store functional data regarding semiconductor device characteristics received through the network unit (130) described below. The memory (120) can store program codes that operate a neural network model to perform learning by receiving functional data as input, program codes that operate a neural network model to perform inference according to the purpose of use of the computing device (100) by receiving functional data as input, and processed data generated as the program codes are executed.

[0064] The network unit (130) according to one embodiment of the present disclosure may be understood as a component that transmits and receives data through any type of known wired or wireless communication system. For example, the network unit (130) may perform data transmission and reception using a wired or wireless communication system such as a local area network (LAN), wideband code division multiple access (WCDMA), long term evolution (LTE), wireless broadband internet (WiBro), fifth generation mobile communication (5G), ultra wide-band, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity, near field communication (NFC), or Bluetooth. Since the above-described communication systems are only examples, the wired and wireless communication system for data transmission and reception of the network unit (130) may be applied in various ways other than the above-described examples.

[0065] The network unit (130) can receive data necessary for the processor (110) to perform calculations through wired or wireless communication with any system or any client, etc. In addition, the network unit (130) can transmit data generated through calculations of the processor (110) through wired or wireless communication with any system or any client, etc. For example, the network unit (130) can receive functional data regarding semiconductor device characteristics through communication with a database, a cloud server, or a client, etc. The network unit (130) can transmit output data of a neural network model, intermediate data derived from the calculation process of the processor (110), processed data, etc. through communication with the aforementioned database, server, or client, etc.

[0066] FIG. 2 is a schematic diagram of a framework for adaptive generation of device characteristics according to one embodiment of the present disclosure.

[0067] Referring to FIG. 2, a neural network model according to an embodiment of the present disclosure may include a first neural network block encoding input functional data, a second neural network block transforming a feature generated by the first neural network block, and a third neural network block decoding a feature transformed by the second neural network block. The first neural network block may generate a latent feature (10) by encoding a curve of the input functional data into a latent space. At this time, the latent feature (10) may include a first feature part (15) and a second feature part (19). The first feature part (15) may be maintained relatively constant across the device characteristics and may represent a general form of the device characteristics. For example, the first feature part (15) may be V DS When I increases DThe second characteristic part (19) may include general characteristics that are common to semiconductor devices, such as a characteristic of a device that monotonically increases and gradually saturates at a certain threshold voltage. The second characteristic part (19) may indicate a unique characteristic rather than a general characteristic of the device. That is, the second characteristic part (19) may include dependent characteristics of the device itself that may vary depending on the type of device, etc.

[0068] The second neural network block can replace the second feature part (19) of the latent feature (10) with the corresponding target feature part (25) by injecting information of a given target electrical parameter (a point marked near the curve). Specifically, the second neural network block can encode an adjustment ratio for the electrical parameters of the input functional data. Then, the second neural network block can generate a modified latent feature (20) by replacing the second feature part (19) with the encoded feature (25) for the adjustment ratio. For example, the second neural network block can generate the modified latent feature (20) by performing a cross-attention operation based on the latent feature (10) and the encoded feature (25) for the adjustment ratio of the electrical parameters. The second neural network block performs an attention operation using the latent feature (10) as a key and value and the encoded feature for the adjustment ratio as a query, thereby preserving the first feature part (15) and converting the second feature part (19) into the encoded feature (25) for the adjustment ratio.

[0069] The third neural network block decodes the deformation latent feature (20) into the observation curve space, thereby generating deformation functional data adjusted according to the adjustment ratio of the electrical parameters. Through this computational process, the neural network model of the present disclosure can manipulate the input characteristics while preserving the inherent form of the input characteristics and matching the electrical parameters, thereby corresponding the output characteristics to the target characteristics set as the learning objective. Therefore, the neural network model of the present disclosure can achieve stable performance for various cases that are free from learning bias without repeating numerous learning cycles.

[0070] Figure 3 is a graph illustrating test results of a framework for adaptive generation of device characteristics according to one embodiment of the present disclosure. Tests were conducted to evaluate the performance of a neural network model according to the framework of the present disclosure, and the experimental results are summarized below with reference to Figure 3.

[0071] (1) Data set

[0072] We trained and tested the SPICE model using a large dataset using the industry-standard BSIM4 model. This process considered various process variables, including the device's channel width (W), channel length (L), temperature (T), and other process variables. The channel width was 10 -7 10 in m -4 m, with channel lengths of 4.5 × 10 -8 10 in m -5 Data were generated for 270 WLT combinations at temperatures -10°C, 25°C, and 125°C. For each WLT combination, the oxide thickness (t ox ), threshold voltage (v t0 ), bulk effect coefficient 1 (k1), width mask adjustment (x w ), adjust length mask (x l ), gate-source overlap capacitance (c gso ), gate-drain overlap capacitance (c gdo) and other process variability (see the upper graph of Figure 1), 200 simulations were performed using Monte Carlo (MC) sampling. [Table 1] summarizes the range of each process variable, which is approximately ±5% of the typical value. The data set consists of a total of 108,000 samples of device characteristics. For each device characteristic, I dsat , I dlin , V tsat , V tlin , V text A set of predefined electrical parameters was extracted, where V text is the maximum slope of the ID-VGS curve gm max The threshold voltage is extracted from the JEDEC (Joint Electron Devices Engineering Council) global standard. The specifications of these electrical parameters are detailed in [Table 2], but the framework of the present disclosure is not limited to these specific values.

[0073] Process variableNMOS rangePMOS rangeUnit Oxide thickness1.19e-9∼1.31e-91.24e-9∼1.37e-9mThreshold voltages0.445∼0.492-0.467∼-0.516VBulk effect coefficient 10.38∼0.420.38∼ 0.42√VWidth mask adjustment-1.00e-9∼1.00e-9-1.00e-9∼1.00e-9mLength mask adjustment-1.9e-8∼-2.1e-8-1.9e-8∼-2.1e-8mGate-source overlap cap.1.05e-10∼1.16e-101.05e-10∼1.16e-10F / mGate-drain overlap cap.1.05e-10∼1.16e-101.05e-10∼1.16e-10F / m

[0074] Electrical parameters V DS (V)V GS (V)V BS (V)I DS (A)I dsat 1.01.00.0Idsat I dlin 0.11.00.0I dlin V tsat 1.0V tsat 0.0(W / L)*1e-7V tlin 0.1V tlin 0.0(W / L)*1e-7V text 0.1V text 0.0-

[0075] (2) The evaluation results are also 3 I D -V DS , I D -V GS and C GG -V GS The target curve results are shown. In Figure 3, the reference curve was adjusted to match the target curve by considering the adjustment ratio of the electrical parameters (see Table 3). The predicted target values ​​were found to be quite consistent with the actual target curve values. This demonstrates that the framework of the present disclosure effectively adjusts the reference curve to the desired curve based on a series of electrical characteristics, thereby resolving potential discrepancies between the reference and actual measurements.

[0076] Electrical Parameter Reference Target Ratio I dsat 0.83 mA0.74 mA-10.7%I dlin 0.24 mA0.23 mA-5.4%V tsat 0.16 V0.22 V36.5%V tlin 0.30 V0.35 V17.6%V text 0.37 V0.42 V14.0%

[0077] Additionally, we performed a quantitative comparison of the neural network model according to the present disclosure with existing models, including a simple multilayer perceptron (MLP)-based regression model and a fully attention-based transformer (see Table 4). Multilayer perceptron-based models typically attempt to solve problems by assimilating target electrical parameters, but they often require numerous iterations and yield unsatisfactory results in achieving the goal. Fully attention-based transformers are often impractical due to their extensive parameters for sequence length and quadratic complexity. The pre-trained model of the present disclosure outperforms other models in terms of feasible complexity.

[0078] Model Targeting Error Parameter Number of Inferences Time Multilayer Perceptron-based Model 4.67e-2349, 329e-379.58 Transformer 7.10e-52, 483e-3416.45e-1 Neural Network Model of this Disclosure 6.79e-5435, 8532.91e-1

[0079] FIG. 4 is a flowchart illustrating an artificial intelligence learning method for adaptive generation of device characteristics according to an embodiment of the present disclosure. Referring to FIG. 4, a computing device (100) according to an embodiment of the present disclosure can obtain reference functional data regarding characteristics of a semiconductor device and adjustment ratios for electrical parameters constituting the reference functional data (S110). For example, when the computing device (100) is a cloud server, the computing device (100) can receive data regarding the reference functional data and the adjustment ratio from a client. The computing device (100) may receive the reference functional data from a separate database according to a user command, and may also receive the adjustment ratio as user input data through the client. In this case, the reference functional data may be selected through random sampling from a database generated by performing a simulation based on a plurality of process variables for a semiconductor device according to a user command.

[0080] The computing device (100) can receive reference functional data and an adjustment ratio, and train a neural network model to generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data (S120). At this time, the target functional data can be selected through random sampling from a database generated by performing a simulation based on a plurality of process variables for a semiconductor device according to a user command. Specifically, the computing device (100) can input reference functional data into a first neural network block of the neural network model to generate latent features. The computing device (100) can input latent features and an adjustment ratio into a second neural network block of the neural network model to modify the latent features. The computing device (100) can input latent features modified by the second neural network block into a third neural network block of the neural network model to generate modified functional data. The computing device (100) can calculate an error for the modified functional data generated in this way using a loss function. And, the computing device (100) can train the neural network model by adjusting the parameters that configure the neural network model in a direction in which the error is minimized.

[0081] FIG. 5 is a flowchart illustrating an artificial intelligence inference method for adaptive generation of device characteristics according to one embodiment of the present disclosure.

[0082] Referring to FIG. 5, a computing device (100) according to an embodiment of the present disclosure can obtain first functional data regarding characteristics of a semiconductor device and a first adjustment ratio for electrical parameters constituting the first functional data (S210). For example, when the computing device (100) is a cloud server, the computing device (100) can receive data regarding the first functional data and the first adjustment ratio from a client. The computing device (100) can receive the first functional data from a separate database according to a user command, and the first adjustment ratio can also be received as user input data through the client. That is, the computing device (100) can obtain device characteristic data for simulation through interaction with a user.

[0083] The computing device (100) can generate second functional data desired by the user based on the first functional data and the first adjustment ratio using a pre-trained neural network model as shown in FIG. 4 (S220). The computing device (100) can input the first functional data and the first adjustment ratio into the pre-trained neural network model, and generate second functional data transformed from the first functional data according to the first adjustment ratio. At this time, the second functional data may be data that has the same original form as the first functional data but has been transformed to match the first adjustment ratio. Through this framework, the user can effectively utilize it in the design of semiconductor devices by adjusting the device characteristics according to the simulation direction desired by the user.

[0084] The various embodiments of the present disclosure described above can be combined with additional embodiments and modified within the scope understood by those skilled in the art in light of the detailed description above. It should be understood that the embodiments of the present disclosure are illustrative in all respects and not restrictive. For example, each component described as a single component may be implemented in a distributed manner, and likewise, components described as distributed may be implemented in a combined manner. Accordingly, all changes or modifications derived from the meaning, scope, and equivalent concepts of the claims of the present disclosure should be construed as being included within the scope of the present disclosure.

Claims

1. An adaptive generation method of device characteristics based on artificial intelligence, performed by a computing device including at least one processor, A step of obtaining reference functional data on the characteristics of a semiconductor device and an adjustment ratio for electrical parameters constituting the reference functional data; and A step of training a neural network model to receive the reference functional data and the adjustment ratio and generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data; including, method.

2. In paragraph 1, At least one of the above reference functional data or the above target functional data, A database generated by performing a simulation based on multiple process variables for the semiconductor device is selected through random sampling. method.

3. In paragraph 1, The step of training the above neural network model is: A step of generating latent features by inputting the above reference functional data into the first neural network block; A step of transforming the latent feature by inputting the latent feature and the adjustment ratio into a second neural network block; and A step of generating transformed functional data by inputting the transformed latent features into a third neural network block; including, method.

4. In paragraph 3, The above potential features are, A semiconductor device comprising a first characteristic part representing general characteristics of the semiconductor device, and a second characteristic part representing unique characteristics of the semiconductor device. method.

5. In paragraph 4, The above second neural network block is, By replacing the second feature part with an encoded feature for the adjustment ratio, the latent feature is transformed. method.

6. In paragraph 5, The above second neural network block is, Performing a cross-attention operation based on the encoded features for the above latent features and the above adjustment ratio, method.

7. In paragraph 3, The learning of the above neural network model is as follows: A first loss function for calculating a reconstruction error between the transformed functional data and the reference functional data, and a second loss function for calculating a mapping error between the transformed functional data and the target functional data, method.

8. In paragraph 7, The overall loss function for learning the above neural network model is, Expressed as a weighted sum of the first loss function and the second loss function, method.

9. An adaptive generation method of device characteristics based on artificial intelligence, performed by a computing device including at least one processor, A step of obtaining first functional data on characteristics of a semiconductor device and a first adjustment ratio for electrical parameters constituting the first functional data; and A step of generating second functional data desired by the user based on the first functional data and the first adjustment ratio using a pre-learned neural network model; Including, The above neural network model is, A method for generating target functional data from reference functional data by inputting reference functional data and a second adjustment ratio for electrical parameters constituting the reference functional data, while maintaining the unique form of the reference functional data and aligning the second adjustment ratio, is pre-trained. method.

10. A computer program stored in a computer-readable storage medium, wherein the computer program, when executed on one or more processors, performs operations for adaptive generation of device characteristics based on artificial intelligence. The above actions are, An operation of obtaining reference functional data on the characteristics of a semiconductor device and an adjustment ratio for electrical parameters constituting the reference functional data; and An operation of training a neural network model to receive the reference functional data and the adjustment ratio and generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data; including, Computer program.

11. A computing device for adaptive generation of device characteristics based on artificial intelligence, A processor comprising at least one core; A memory containing program codes executable by the processor; and A network unit for obtaining reference functional data on the characteristics of a semiconductor device and an adjustment ratio for electrical parameters constituting the reference functional data; Including, The above processor, By inputting the above reference functional data and the above adjustment ratio, a neural network model is trained to generate target functional data from the reference functional data in accordance with the adjustment ratio while maintaining the unique form of the reference functional data. device.

Images