DC-DC converter with intelligent controller

Abstract

The input of the DC-DC converter controller is sampled at predetermined intervals, generating two-dimensional state information with one axis representing the input physical quantity and the other representing time. This two-dimensional state information is processed by a convolutional neural network to determine and output one of multiple control signals. The AI ​​control unit can operate based on multiple operating conditions or dynamically determined operating conditions by applying different AI engines depending on the operating mode.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority from Korean Patent Application No. 10-2020-0101414 filed on August 12, 2020, in the Korean Intellectual Property Office, the disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0003] Disclosed is a technology related to a power supply device. Background Art

[0004] Traditionally, a direct current (DC)-DC converter controls the duty cycle of the switching pulses in order to provide a stable voltage even when the input power or output load fluctuates. Systems with built-in DC-DC converters have become increasingly complex and demanding in terms of specifications such as voltage stability, settling time, output ripple, power transfer efficiency, device size, load current coverage, number of output voltages, etc. In response to these demands, in recent DC-DC converters, the frequency of the switching pulses has been changed, or a multi-core structure has been adopted. The controller of a conventional DC-DC converter operates within a limited input range under each output condition. The situation in which the controller's algorithm is executed sequentially to meet and control multiple output conditions makes it difficult to achieve the target characteristics or causes time delays due to interference between the output conditions.

[0005] Meanwhile, the DC-DC converter is optimized for the operating state of the load or the demand conditions from the load, such as a change in output voltage, an increase in output current, a stable voltage, etc. However, when the demand conditions of a single load system change dynamically, the response thereto is greatly limited.

[0006] In addition, when the power management device is implemented as an integrated circuit, the conventional DC-DC conversion device uses the maximum power point tracking technology to promote the improvement of energy transfer efficiency. According to this technology, the feedback control mode or the core size is controlled according to whether the load is light or heavy. Since this method pursues the best efficiency in a limited state, it does not take into account the existing accumulated information, or is accompanied by processing speed limitations or the limitation of a fixed amount of time required due to always repeating the same control process. Specifically, since many output parameters such as efficiency, settling time, output fluctuation (ripple) cannot be considered and processed at the same time, but are considered sequentially, the time required for a control loop is fixedly increased in proportion to the number of output parameters considered. Summary of the Invention

[0007] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

[0008] The following description aims to achieve a control capable of delivering maximum power while optimizing efficiency, settling time, and output ripple in response to load demands in a direct current (DC)-DC converter.

[0009] Furthermore, the following description aims to provide a DC-DC converter capable of meeting the requirements of a load with respect to multiple output parameters.

[0010] Furthermore, the following description aims to provide a DC-DC converter that can meet the demand conditions of a load over a wide range of output parameters.

[0011] Furthermore, the following description aims to provide a fully integrated DC-DC converter capable of minimizing surrounding components.

[0012] Furthermore, the following description aims to provide a DC-DC converter that can meet the requirements of dynamically changing loads.

[0013] According to one aspect of the present disclosure, since the input of the controller of the DC-DC converter is sampled for a predetermined time, two-dimensional state information is generated, wherein one axis of the two-dimensional state information is the input physical quantity and the other axis is time. The two-dimensional state information is processed by a convolutional neural network to determine and output one of a plurality of control signals.

[0014] According to another aspect, the control signal output by the artificial intelligence control unit to the DC-DC converter may be an increase signal or a decrease signal for the current control signal value.

[0015] According to other aspects, the artificial intelligence control portion may operate according to a plurality of operating conditions or operating conditions dynamically determined by applying different artificial intelligence engines according to the operating mode.

[0016] According to other aspects, the operating mode may be determined based on two-dimensional state information generated based on sensor input during a sampling window.

[0017] According to other aspects, one of a plurality of artificial intelligence networks may be selectively applied to the artificial intelligence control portion based on whether two-dimensional state information generated from sensor input during a sampling window has been previously learned.

[0018] Other features and aspects will be apparent from the following detailed description, drawings, and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a block diagram showing a configuration of a direct current (DC)-DC conversion device according to an embodiment.

[0020] Figure 2 is a block diagram showing a configuration of a DC-DC conversion device according to another embodiment.

[0021] Figure 3 is a block diagram showing a configuration of a DC-DC conversion device according to still another embodiment.

[0022] Figure 4 An example of two-dimensional state information is shown, which is the input to a convolutional neural network.

[0023] Figure 5 A conceptual diagram used to describe the process of preparing multiple sets of two-dimensional filters and artificial neural network weighting coefficients according to design goals.

[0024] Figure 6 is a block diagram showing a configuration of one embodiment of a DC-DC conversion device to which an artificial intelligence control portion according to the present disclosure is applied.

[0025] Figure 7 is a block diagram showing a configuration of another embodiment of a DC-DC conversion device to which the artificial intelligence control portion according to the present disclosure is applied.

[0026] Throughout the drawings and detailed description, unless otherwise described, the same drawing reference numerals will be understood to refer to the same elements, features, and structures. The relative size and depiction of these elements may be exaggerated for clarity, illustration, and convenience. DETAILED DESCRIPTION

[0027] The above-mentioned aspects and other aspects are specified by referring to the embodiments described with reference to the accompanying drawings. It should be understood that the components of each embodiment can be combined in various ways within that embodiment or with the components of other embodiments without further explanation or without contradiction. The terms used in the specification and claims should be understood as means and concepts consistent with the specification or the technical spirit proposed, and the principle is that the inventor can appropriately define the concepts of the terms to best describe his invention. Below, the preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0028] Figure 1 1 is a block diagram showing a configuration of a direct current (DC)-DC converter according to an embodiment. As shown in the figure, the DC-DC converter according to an embodiment includes a power conversion unit 100, a plurality of sensors 900, and an artificial intelligence control unit 300.

[0029] The power conversion unit 100 can be one of many well-known DC-DC converters, such as a buck converter, a boost converter, a buck-boost converter, a complex structure including these basic structures or a multi-core structure, a structure including multiple parallel structures of these structures, and the like. These can be operated in one of a pulse width modulation (PWM) method and a pulse frequency modulation (PFM) method. In one embodiment, the power conversion unit 100 can be implemented as a semiconductor integrated circuit. When the DC-DC converter is implemented as an integrated circuit, there is a problem of reducing the size of the inductor. When the switching frequency is increased to reduce the size of the inductor, the stability in the feedback loop becomes a problem, the problem of the size of the resistor or capacitor for compensation occurs, and the output ripple and the settling time increase, thereby reducing efficiency. The present disclosure can be used to effectively solve these complex controller problems.

[0030] The plurality of sensors 900 may be a circuit composed of one or more elements for sensing a physical state or a circuit state, wherein the one or more elements are connected to each input, output, or one or more nodes of the circuit of the power conversion unit 100 .

[0031] The artificial intelligence control unit 300 samples the values ​​output from the multiple sensors 900 multiple times within a predetermined time to generate two-dimensional state information and processes the two-dimensional state information through a convolutional neural network (CNN), thereby outputting multiple control signals for controlling the operation of the power conversion unit according to the classification result values.

[0032] The input data for a general artificial neural network consisting of fully connected layers is a one-dimensional vector. Convolutional neural networks (CNNs) typically propose an architecture that can utilize this fully connected artificial intelligence network while preserving spatial information, including two-dimensional image features. A CNN consists of convolutional layers, which include filters to extract image features, followed by nonlinear activation functions, and pooling layers to extract features from the output. A one-dimensional vector is generated from the output of the final pooling layer and provided as input to a fully connected artificial intelligence network, described later.

[0033] As mentioned above, in general, a convolutional neural network (CNN) can reflect the features of a two-dimensional image to classify the features through a general artificial intelligence network. The physical quantities input to the controller of a DC-DC converter (e.g., input terminal voltage, output terminal voltage, temperature, etc.) can vary in quantity, type, or range depending on the configuration and target operating specifications or characteristics of the DC-DC converter. They do not have a two-dimensional arrangement, nor do they have characteristics such as spatial information of image information.

[0034] Furthermore, the outputs of a DC-DC converter controller (e.g., the duty cycle and frequency of the switching pulses) may vary in quantity, type, or range depending on the configuration and target operating specifications or characteristics of the DC-DC converter. These are determined by reflecting the time-accumulated characteristics of the inputs and therefore may not be determined by classifying the controller inputs at a specific point in time.

[0035] According to one aspect of the present disclosure, two-dimensional state information (one axis of the two-dimensional state information is an input physical quantity and the other axis is time) is generated by sampling the input of a controller of a DC-DC converter for a predetermined time. This two-dimensional state information is then processed by a convolutional neural network to generate multiple control signals. The input of the controller is a physical quantity related to the environment or signal of the DC-DC converter and can therefore have a continuous characteristic. According to this aspect of the present disclosure, the output of the controller can reflect the input value within a predetermined time or predetermined period.

[0036] In one embodiment, the artificial intelligence control unit 300 can be implemented using a program instruction set running on a general-purpose processor, which is configured on the same substrate or in the same package as the power conversion unit 100. These program instructions are stored in the memory 700. In the illustrated embodiment, the memory 700 can be implemented as one or more physical memory elements. As another example, the artificial intelligence control unit 300 can be implemented by a combination of dedicated hardware, a field programmable gate array (FPGA), and a general-purpose processor.

[0037] According to other aspects, the control signal output by the artificial intelligence control unit to the DC-DC converter can be an increase signal or a decrease signal for the current control signal value. According to this aspect, the artificial intelligence control unit of the present disclosure is used to determine whether to increase or decrease the current control signal according to the input value during a predetermined period (i.e., a sampling window). The control signal of the DC-DC converter can gradually affect the output or characteristics according to the increase or decrease of physical quantities (such as duty cycle, frequency, size or capacity of the gate driver, and size of the encoder). The artificial intelligence control unit can gradually change the characteristics relative to the current state and reduce risks according to the noise contained in the input or the error contained in the erroneous judgment of the artificial intelligence engine in the form of an increase signal or a decrease signal for the current value.

[0038] As shown in the figure, in one embodiment, the artificial intelligence control unit 300 may include an image information generation unit 350, a convolutional neural network circuit unit 310 and an output signal generation unit 330.

[0039] Image information generation unit 350 samples the values ​​output from sensor 900 for a predetermined time to generate two-dimensional state information. Image information generation unit 350 sequentially stores the digital values ​​read from sensor 900 in buffer memory 730. In one embodiment, buffer memory 730 is implemented as a static random access memory (RAM) that is implemented in the same package as power conversion unit 100 and artificial intelligence control unit 300. Figure 4 An example of two-dimensional state information is shown, which is generated by the image information generation unit 350 and input to the convolutional neural network. The horizontal axis is the time axis, and the vertical axis is the input variable. When the input variables are physically adjacent to each other and have similar time fluctuations, it is advantageous to process the convolutional neural network. In this example, the size of the sampling window is (t M -t0).

[0040] The convolutional neural network circuit section 310 processes the two-dimensional state information to output a classification result value. The convolutional neural network circuit section 310 outputs a probability value for each classification result, but in the present disclosure, only the classification result value with the highest probability is selected. According to one aspect of the present disclosure, the convolutional neural network circuit section 310 determines and outputs one of multiple sets of control signals based on the input values ​​during a predetermined period (i.e., a sampling window). For example, the output of the convolutional neural network circuit section 310 can have the following form.

[0041] [Table 1]

[0042] Nuclear option 2 bits (00, 01, 10) PWM / PFM selection word 1 bit {0 (PWM) or 1 (PFM)} Duty cycle control word 4 bits Switching frequency control word 2 bits

[0043] In this example, the core selection word is a control word for selecting one of the multiple cores when the power converter has a multi-core structure. For example, when the power converter has a structure including three cores, the core selection word may be a 2-bit word that can have one of {00, 01, and 10} in the order from small core to large core.

[0044] The PWM / PFM selection word is a control word for selecting the operation mode of the power converter to be one of the PWM method and the PFM method. For example, the PWM / PFM selection word may be a 1-bit word which indicates the PWM mode when 0 and the PFM mode when 1.

[0045] The duty cycle control word specifies the duty cycle value of the PWM switching pulses of the power converter. For example, when the power converter has a structure capable of changing the duty cycle in 16 steps, the duty cycle control word can be a 4-bit word.

[0046] The switching frequency control word is a control word that controls the switching frequency of the power converter. For example, the exemplary power converter may include a reference frequency generation circuit and a multiplexed reference frequency generation circuit. The multiplexed reference frequency generation circuit includes a frequency divider and a multiplexer. The frequency divider generates and outputs four switching frequencies by dividing the generated reference frequency signal. The four switching frequencies are input to the multiplexer. The multiplexer has a 2-bit selection input terminal to which the switching frequency control word is applied.

[0047] In the case of the embodiment in Table 1, the output of the convolutional neural network circuit unit 310 consists of 9 bits, which is the sum of a 2-bit core selection word, a 1-bit PWM / PFM selection word, a 4-bit duty cycle control word, and a 2-bit switching frequency control word, and the input two-dimensional state information is classified into one of 3×2×16×2=192 classes.

[0048] The output signal generation unit 330 generates and outputs a control signal for controlling the operation of the power conversion unit 100 based on the classification result value. For example, in the case of the example shown in Table 1 above, the output signal generation unit 330 separates the control word from the 9-bit output of the convolutional neural network circuit unit 310 and outputs the control word to the corresponding node of the power conversion unit 100.

[0049] According to other aspects, such as another embodiment, the convolutional neural network circuit portion 310 may be configured to determine whether to increase or decrease each current control signal based on the input value during a predetermined period (i.e., a sampling window). In this case, each control word may be a 1-bit word indicating an increase or decrease, or a signed binary word indicating an increase value or a decrease value, such as a 3-bit signed binary word.

[0050] When each control word consists of a 1-bit word indicating increase or decrease, the output of the convolutional neural network circuit unit 310 consists of 4 bits, which is the sum of a 1-bit core selection word, a 1-bit PWM / PFM selection word, a 1-bit duty cycle control word, and a 1-bit switching frequency control word, and the input two-dimensional state information is classified into one of 2×2×2×2=16 classes.

[0051] In this embodiment, the output signal generating unit 330 generates and outputs a control signal for controlling the operation of the power conversion unit 100 based on the classification result value. For example, in the case of the embodiment described above, the output signal generating unit 330 separates the control word from the output 4 bits of the convolutional neural network circuit unit 310, and increases or decreases the stored current control signal value according to the corresponding output value to output the control word to the corresponding node of the power conversion unit 100. When the current control signal value is the maximum value and the corresponding output value is "increase (1)", the output signal generating unit 330 can operate to maintain the output value. In addition, when the current control signal value is the minimum value and the corresponding output value is "decrease (0)", the output signal generating unit 330 can operate to maintain the output value. In this case, in the illustrated embodiment, the current control signal value can be stored in the memory 700.

[0052] The convolutional neural network circuit unit 310 is learned using labeled learning data. Commercially available circuit design tools provide simulation results that are close to real circuits. In the present disclosure, learning data is obtained from the simulation results, while changing the input parameters in various ways in the modeling of the power conversion unit 100 generated by the circuit design tool. The weight coefficients of the two-dimensional filter of the convolutional neural network and the fully connected artificial intelligence network determined by learning can be stored in the configuration information memory 710 of the memory 700. In one embodiment, the configuration information memory 710 is implemented on the same static RAM that constitutes the buffer memory 730.

[0053] Figure 2 is a block diagram showing the configuration of a DC-DC converter according to another embodiment. According to other aspects, the DC-DC converter may further include a control selection unit 370. The control selection unit 370 applies a set of corresponding two-dimensional filters and artificial neural network weighting coefficients from the configuration information memory 710 to the convolutional neural network circuit unit 310 in response to a setting instruction. In one embodiment, the setting instruction can be manually input via an operation mode indication button. To this end, multiple sets of two-dimensional filters and artificial neural network weighting coefficients are prepared and stored in the configuration information memory 710. Each set can be prepared according to a predetermined design goal. Figure 5This is a conceptual diagram describing the process of preparing multiple sets of two-dimensional filter and artificial neural network weighting coefficients based on design objectives. In this figure, efficiency refers to the ratio of output power to input power, settling time refers to the time required for the output to reach and remain within a predetermined fluctuation range of a step input voltage, and output fluctuation (ripple) refers to the fluctuation range of the output voltage. The illustrated example represents these three design target parameters, but in expanded embodiments, more design target parameters may also be considered. Points 41, 43, and 45 indicated in the figure refer to the weights given to these design target parameters during design. The design target indicated by reference numeral 41 is a design scheme that uniformly considers efficiency, settling time, and output fluctuation. Reference numeral 43 is a design scheme in which efficiency and settling time are given equal weight, with the weighting for output fluctuation set to a smaller value. Reference numeral 45 is a design scheme in which efficiency and output fluctuation are given equal weight, with the weighting for settling time set to a smaller value.

[0054] The weight of this design goal can be achieved by annotating the learning data of the convolutional neural network. For example, the design target parameters are obtained by circuit simulation, and the weight of the design goal is obtained by annotating the learning data of the convolutional neural network. Figure 4 The classification result value corresponding to the input parameter that meets the desired design goal is assigned by changing the input parameter by the sensor in a manner such as duty cycle / core selection / frequency.

[0055] According to other aspects, the DC-DC converter device according to one aspect may further include an operating mode determination unit 390. The operating mode determination unit 390 determines the operating mode based on the outputs of the plurality of sensors 900 and outputs a setting instruction. For example, when a sudden load fluctuation causes a sharp fluctuation in the output voltage, a setting instruction may be output to select an operating mode optimized for reducing the fluctuation range of the output voltage. In addition, when a stable load state is maintained for a long time, a setting instruction may be output to select an operating mode optimized for efficiency.

[0056] Figure 3 1 is a block diagram showing a configuration of a DC-DC converter according to another embodiment. As shown in the figure, the DC-DC converter according to other aspects may further include a radial basis function neural network (RBFNN) circuit section 340 and an intelligent network engine selection section 320.

[0057] The radial basis function neural network circuit unit 340 is an artificial neural network that uses the radial basis function as an activation function. The output is a linear combination of the neuron parameters and the radial basis function values ​​of the input values. The radial basis function neural network circuit unit 340 can read and initialize the neuron parameters from the second configuration information storage unit 750. The intelligent network engine selection unit 320 controls so that when the generated two-dimensional state information is previously learned state information, the generated two-dimensional state information is processed by the radial basis function neural network, and when the generated two-dimensional state information is new state information, the generated two-dimensional state information is processed by the convolutional neural network.

[0058] Figure 6 This is a block diagram showing the configuration of an embodiment of a DC-DC converter device to which the artificial intelligence control unit according to the present disclosure is applied. The DC-DC converter device of this embodiment is disclosed as one of the embodiments of patent No. 2,135,873 filed by the present applicant on December 10, 2019 and registered on July 14, 2020. In this patent, a conventional rule-based controller is applied. The referenced block diagram is based on the disclosure of this patent. Figure 3 The embodiment shown in FIG applies an embodiment of the artificial intelligence controller of the present disclosure. In order to simultaneously meet the output fluctuation or the settling time, it is necessary to fix the time in the control process, but by applying the present disclosure, the control cycle can be shortened, thereby improving performance.

[0059] As shown in the figure, the power supply device according to the embodiment includes a multiplexed reference clock generation unit 130, a switching pulse generation unit 120, and a DC-DC conversion unit 110. The multiplexed reference clock generation unit 130 outputs one of a plurality of reference clock signals having different frequencies. For example, the multiplexed reference clock generation unit 130 can generate ten reference clock signals having a frequency of 100 kHz intervals at a frequency of 1 MHz to 1.9 MHz. One of the generated reference clock signals is selected and output. As another example, the multiplexed reference clock generation unit 130 can include a single clock generation circuit that generates and outputs a reference clock signal having a specified frequency. For example, the reference clock can be divided and output by a controllable variable divider.

[0060] The switching pulse generating section 120 generates and outputs switching pulses obtained by varying the duty cycle of the reference clock signal output from the multiplexed reference clock generating section 130. The DC-DC converter section 110 switches input power using the switching pulses output by the switching pulse generating section 120, then converts the input power into DC power and outputs the DC power. In one embodiment, the DC-DC converter section 110 may be a buck converter. The switching pulse generating section 120 generates and outputs switching pulses that vary the duty cycle of the reference clock. PWM modulation, which modulates an input clock signal to have a given duty cycle, is a known technique. The duty cycle can be indicated by a voltage or digitally encoded information.

[0061] Switching pulse generator 120 may include a multiplexed duty cycle signal generator 121 and a duty cycle selector 123. Multiplexed duty cycle signal generator 121 generates multiple switching pulse signals in which a reference clock is modulated to have different duty cycles. In one embodiment, multiplexed duty cycle signal generator 121 includes nine signal generators that output switching pulse signals with duty cycles ranging from 10% to 90%, each synchronized with the reference clock signal. Techniques for modulating an input pulse signal to a PWM modulation with a given duty cycle are well known.

[0062] Duty cycle selector 123 outputs one of the multiple switching pulse signals selected by the duty cycle control signal output by artificial intelligence control unit 300. In one embodiment, duty cycle selector 123 is a multiplexer. In the illustrated embodiment, the multiplexer's selection input is a digital byte output by duty cycle tracking control unit 193. DC-DC converter 110 switches input power using switching pulses, converts the input power into a DC voltage, and outputs the DC voltage. In one embodiment, DC-DC converter 110 may be a step-down converter.

[0063] In the illustrated embodiment, the control signal output by the artificial intelligence control unit 300 includes a duty cycle control signal and a switching frequency control signal. The duty cycle control signal is input as a selection input of the multiplexer constituting the duty cycle selection unit 123. The switching frequency control signal is input to the multiplexed reference clock generation unit 130 to select the frequency of the reference clock provided to the switching pulse generation unit 120. For example, the switching frequency control signal can be input as a control word for the variable frequency divider of the multiplexed reference clock generation unit 130. Figure 7 The specific operation of the illustrated embodiment has been described in detail in this patent publication, and thus a detailed description will be omitted.

[0064] Figure 71 is a block diagram illustrating the configuration of another embodiment of a DC-DC converter device employing an artificial intelligence control unit according to the present disclosure. The DC-DC converter device shown includes a multi-core DC-DC converter unit 110. Specifically, the DC-DC converter device includes multiple core circuits, each of which includes a main switching transistor, an inductor, and an output circuit. The artificial intelligence control unit 300 implemented according to aspects of the present disclosure detects the voltage at the input terminal V via an input current sensor 911 and an input voltage sensor 913. IN In addition, the artificial intelligence control unit 300 detects the output terminal V through the output current sensor 951 and the output voltage sensor 953. OUT In this case, the current sensor may include a temperature compensation circuit. In one embodiment, the artificial intelligence control unit 300 may sense the temperature through the temperature sensor 930 and reflect the temperature in the control.

[0065] The artificial intelligence control unit 300 implemented according to aspects of the present disclosure, for example, outputs the core selection word described in the above-described embodiment to the core selection controller 173 to select one of the multiple core circuits. In this case, the artificial intelligence control unit 300 selects a gate driver suitable for driving the main switching transistor via the variable gate driver controller 171. Furthermore, the artificial intelligence control unit 300 outputs a control word to the PWM / PFM mode controller 151 to control the selection between the two operating modes, and outputs the control word to the dead time controller 153 to control the selection of the optimal dead time. Furthermore, by outputting the control word to the soft start controller 155 and controlling the flexible startup operation of the bandgap reference voltage 137, operational stability can be improved.

[0066] According to the present disclosure, complex inputs to a DC-DC converter can be processed simultaneously to meet design goals or system requirements. Furthermore, by processing multiple output parameters simultaneously, fixed processing delays that increase proportionally with the number of output parameters considered can be avoided. Consequently, since high-speed control is possible, the accuracy of the optimal value can be improved. Furthermore, system requirements in which efficiency, settling time, and output ripple may conflict with each other can be met. Furthermore, dynamically changing requirements in the system can be met.

[0067] In the above description, the present disclosure is described with reference to the embodiments of the accompanying drawings, but is not limited thereto, and should be construed to include various modified examples that can be clearly derived by those skilled in the art. The claims are intended to include these modified examples.

Claims

1. A DC-DC converter, comprising: power conversion unit; a plurality of sensors configured to detect a state of each portion of the power conversion unit; as well as An artificial intelligence control unit is configured to sample the values ​​output by the multiple sensors for a predetermined time to generate two-dimensional state information, and process the two-dimensional state information through a convolutional neural network to output multiple control signals for controlling the operation of the power conversion unit based on classification result values, wherein one axis of the two-dimensional state information is an input physical quantity and the other axis of the two-dimensional state information is time.

2. The DC-DC converter according to claim 1, wherein: At least one of the plurality of control signals is an increase signal or a decrease signal for a current control signal value.

3. The DC-DC converter according to claim 1, wherein: The artificial intelligence control unit includes an image information generation unit, a convolutional neural network circuit unit, and an output signal generation unit. The image information generation unit is configured to sample the values ​​output by the multiple sensors for the predetermined time to generate the two-dimensional state information. The convolutional neural network circuit unit is configured to process the two-dimensional state information to output the classification result value. The output signal generation unit is configured to generate the multiple control signals for controlling the operation of the power conversion unit based on the classification result value.

4. The DC-DC converter according to claim 1, further comprising: a configuration information memory in which at least one two-dimensional filter and an artificial neural network weighting coefficient are stored; as well as A control selection unit is configured to apply a set of corresponding two-dimensional filters and artificial neural network weighting coefficients from the configuration information memory to the convolutional neural network circuit unit in response to a setting instruction.

5. The DC-DC converter according to claim 4, further comprising: The operation mode determination unit is configured to determine an operation mode based on outputs of the plurality of sensors to output the setting instruction.

6. The DC-DC converter according to claim 3, wherein: The artificial intelligence control unit also includes a radial basis function neural network (RBFNN) circuit unit and an intelligent network engine selection unit. The intelligent network engine selection unit controls so that in the case of pre-learned state information, the generated two-dimensional state information is processed by the radial basis function neural network, and in the case of new state information, the generated two-dimensional state information is processed by the convolutional neural network.

7. The DC-DC converter according to claim 1, wherein: The multiple sensors include an input voltage sensor configured to detect the input voltage of the power conversion unit, an input current sensor configured to detect the input current of the power conversion unit, an output voltage sensor configured to detect the output voltage of the power conversion unit, and an output current sensor configured to detect the output current of the power conversion unit.

8. The DC-DC converter according to claim 1, wherein: The control signal generated by the control signal generating section includes a duty cycle control signal that controls a duty cycle during a pulse width operation of the power converting section and a switching frequency control signal that controls a switching frequency of the power converting section.

9. The DC-DC converter according to claim 7, wherein: The control signal generated by the control signal generating unit further includes a core selection signal, a gate driver selection signal, a dead time control signal, and a mode selection signal. The core selection signal selects a core of the power conversion unit according to the classification result value, the gate driver selection signal selects a gate driver of the power conversion unit, the dead time control signal controls the dead time of the power conversion unit, and the mode selection signal selects one of a pulse width modulation (PWM) operation mode and a pulse frequency modulation (PFM) operation mode of the power conversion unit.

Images