JP7529293B2 - Learning device, circuit design support device, learning method, circuit design support method, learning program, and circuit design support program - Google Patents

Description

Article 30, paragraph 2 of the Patent Act applies. Published on August 26, 2021 at the Proceedings of the 34th Circuit and System Workshop. Published on August 26, 2021 at https://www.ieice.org/ess/kws/kws34/program/program_abst.html. Published on August 27, 2021 at the 34th Circuit and System Workshop.

The present invention relates to a learning device, a circuit design support device, a learning method, a circuit design support method, a learning program, and a circuit design support program.

Patent document 1 discloses a technology that easily determines the element values of circuit elements that improve circuit performance.

JP 2020-187395 A

Since designing analog circuits takes a lot of time, the rapid development of electronic devices requires eliminating the bottleneck in circuit design. To eliminate this bottleneck, technology is known for automatically determining element values for analog circuits, such as that disclosed in Patent Document 1 above.

In automating analog circuit design, circuit synthesis and element value search are often performed simultaneously. Specifically, after a circuit is synthesized, the element values of the circuit are determined, and a simulation is performed to evaluate the circuit using a combination of the circuit and element values. Then, based on the simulation results, it is determined whether the combination of the circuit and element values meets the required specifications. If the combination of the circuit and element values does not meet the required specifications, circuit synthesis and element value search are repeated.

However, there is a problem in that the above-mentioned repetition is inefficient. The technology disclosed in the above-mentioned Patent Document 1 is a technology that uses reinforcement learning to determine the element values of an analog circuit, but is not a technology that makes circuit synthesis efficient.

The present invention was made in consideration of the above circumstances, and aims to efficiently determine the placement of circuit elements that meet the required specifications.

To solve the above-mentioned problems, the learning device of the present invention is a learning device that includes a learning data acquisition unit that acquires multiple pieces of learning data in which element placement data representing the placement of elements in a circuit is associated with circuit characteristic values representing the characteristics of the circuit, and a learning unit that trains a machine learning model based on the multiple pieces of learning data acquired by the learning data acquisition unit, thereby training a trained model for outputting the element placement data from the circuit characteristic values.

The circuit design support device according to the present invention is a circuit design support device including an acquisition unit that acquires the trained model generated in advance by the above-mentioned learning device, and a generation unit that generates element placement data that represents the placement of elements in a circuit that corresponds to the required circuit characteristic value by inputting a required circuit characteristic value that represents a required characteristic of a circuit to the trained model acquired by the acquisition unit.

A learning method according to the present invention includes acquiring a plurality of learning data in which element placement data representing an arrangement of elements in a circuit is associated with a circuit characteristic value representing a characteristic of the circuit, and training a machine learning model based on the acquired plurality of learning data, thereby training a trained model for outputting the element placement data from the circuit characteristic value;
This is a learning method in which processing is carried out by a computer.

The circuit design support method according to the present invention is a circuit design support method in which a computer executes a process to acquire the trained model generated in advance by the above-mentioned learning method, input required circuit characteristic values representing required characteristics of a circuit to the acquired trained model, and generate element placement data representing the placement of elements in the circuit corresponding to the required circuit characteristic values.

The learning program of the present invention is a learning program for causing a computer to execute a process of acquiring multiple pieces of learning data in which element placement data representing the placement of elements in a circuit is associated with circuit characteristic values representing the characteristics of the circuit, and training a machine learning model based on the acquired multiple pieces of learning data, thereby training a trained model for outputting the element placement data from the circuit characteristic values.

The circuit design support program of the present invention is a circuit design support program for causing a computer to execute a process of acquiring the trained model generated in advance by the above-mentioned learning program, inputting required circuit characteristic values representing required characteristics of a circuit to the acquired trained model, and generating element placement data representing the placement of elements in a circuit corresponding to the required circuit characteristic values.

The present invention makes it possible to efficiently determine the placement of circuit elements that meet the required specifications.

1 is a functional block diagram showing a circuit design support device according to an embodiment of the present invention; 4 is a diagram showing an example of a plurality of learning data stored in a learning data storage unit of the present embodiment; FIG. FIG. 4 is a diagram for explaining element placement data. FIG. 4 is a diagram for explaining element placement data. FIG. 4 is a diagram for explaining element placement data. FIG. 4 is a diagram for explaining element placement data. FIG. 1 is a diagram for explaining a trained model. FIG. 2 is a diagram illustrating an example of the configuration of a computer that constitutes the circuit design assistance device of the present embodiment. FIG. 4 is a diagram illustrating an example of a learning process routine executed by the circuit design assistance device. FIG. 13 is a diagram showing an example of an element placement estimation processing routine executed by the circuit design assistance device; FIG. 1 shows the results of an example.

The following describes the embodiment in detail with reference to the drawings.

<System configuration of circuit design support device>

Figure 1 is a functional block diagram showing a circuit design support device 10 of this embodiment. As shown in Figure 1, the circuit design support device 10 of this embodiment functionally includes a learning data storage unit 12, a trained model storage unit 14, a result storage unit 16, a learning data acquisition unit 20, a learning unit 22, an acquisition unit 24, and a generation unit 26.

The circuit design support device 10 of this embodiment automatically determines the placement of elements within an analog circuit that meets the required specifications.

The learning data storage unit 12 stores multiple pieces of learning data in which circuit characteristic values representing the characteristics of a circuit are associated with element placement data representing the placement of elements within the circuit. FIG. 2 is a diagram showing an example of multiple pieces of learning data stored in the learning data storage unit 12 of this embodiment. As shown in FIG. 2, the learning data storage unit 12 stores circuit characteristic values and element placement data for each analog circuit having desired characteristics in association with each other. For example, the learning data with data ID "00001" is associated with a circuit characteristic value P1 and element placement data C1.

Circuit characteristic values are information that represents the characteristics of an analog circuit, such as values of the analog circuit's power consumption, phase margin, and common-mode rejection ratio.

The element placement data is information that represents the placement of elements in an analog circuit. Figures 3 to 5 are diagrams for explaining the element placement data. In this embodiment, as shown in Figure 3, a circuit diagram C is represented by mesh data M. As shown in Figure 4, N-type MOS transistors placed at each location of the circuit diagram C are represented as N in the mesh data M, P-type MOS transistors are represented as P, and resistors are represented as R. The element placement data is multi-label data that represents multiple different types of elements. Specifically, as shown in Figure 5, label data RR representing the placement of resistors, label data PR representing the placement of PMOS, and label data NR representing the placement of NMOS are prepared in advance, and these data are stored in the learning data storage unit 12 as element placement data. Note that, as shown in Figure 5, label 1 is assigned to the location where the element is located.

Figure 6 shows an example of element layout data that represents required performance and a circuit structure that satisfies that required performance. As shown in Figure 6, element layout data R1 represents a circuit structure that corresponds to a basic topology. Furthermore, element layout data R2 represents a circuit structure that corresponds to a circuit topology in which the differential amplifier stage is a cascode. Furthermore, element layout data R3 represents a circuit structure that corresponds to a circuit topology in which the output stage is a super source follower. Furthermore, element layout data R4 represents a circuit structure that corresponds to a circuit topology in which the differential amplifier stage is rail-to-rail and the output stage is push-pull.

For example, a genetic algorithm (GA) can be used to collect learning data. For example, by setting multiple different evaluation functions, it is possible to efficiently collect data on various circuit layouts.

The trained model storage unit 14 stores trained models that output element placement data from circuit characteristic values.

Figure 7 shows a diagram for explaining a trained model. The trained model is, for example, a machine learning model such as a neural network model. Note that the trained model is not limited to a neural network model and may be other models.

As shown in FIG. 7, the trained model is configured to output element placement data when circuit characteristic values (or required circuit characteristic values representing the required characteristics of the circuit) are input.

For example, when values such as the power consumption, phase margin, and common-mode rejection ratio of an analog circuit are input to the trained model as desired required circuit characteristic values, the trained model outputs the probability of an element placement that satisfies the required circuit characteristic value. For example, the trained model outputs data such as the probability that a resistor will be placed in one position is 0.8, and the probability that a PMOS will be placed in another position is 0.6. A predetermined threshold process is performed on this data, and elements are assumed to be placed in locations with a probability of, for example, 0.8 or higher.

The result storage unit 16 stores element placement data obtained by the generation unit 26, which will be described later.

The learning data acquisition unit 20 acquires multiple pieces of learning data from the learning data storage unit 12.

The learning unit 22 trains a trained model for outputting element placement data from circuit characteristic values by subjecting the machine learning model to supervised learning based on the multiple pieces of training data acquired by the training data acquisition unit 20. The learning unit 22 then stores the trained model in the trained model storage unit 14.

The acquisition unit 24 acquires the trained model from the trained model storage unit 14.

The generation unit 26 inputs required circuit characteristic values that represent the required characteristics of the circuit into the trained model acquired by the acquisition unit 24, thereby generating element placement data that represents the placement of elements in the circuit that corresponds to the required circuit characteristic values. The generation unit 26 then stores the element placement data in the result storage unit 16. This allows element placement data that satisfies the required circuit characteristic values to be automatically acquired.

The generation unit 26 may generate circuit structure data that satisfies the required performance by generating wiring between elements using known technology based on the element placement data. Also, for example, the generation unit 26 may further calculate the element values of each element included in the circuit structure data using the technology disclosed in the above-mentioned JP2020-187395A.

The circuit design assistance device 10 can be realized, for example, by a computer 50 shown in FIG. 8. The computer 50 includes a CPU 51, a memory 52 as a temporary storage area, and a non-volatile storage unit 53. The computer 50 also includes an input/output interface (I/F) 54 to which an input/output device or the like (not shown) is connected, and a read/write (R/W) unit 55 that controls the reading and writing of data from and to a recording medium. The computer 50 also includes a network I/F 56 that is connected to a network such as the Internet. The CPU 51, memory 52, storage unit 53, input/output I/F 54, R/W unit 55, and network I/F 56 are connected to one another via a bus 57.

The storage unit 53 can be realized by a hard disk drive (HDD), a solid state drive (SSD), a flash memory, etc. The storage unit 53 as a storage medium stores a program for causing the computer 50 to function. The CPU 51 reads the program from the storage unit 53, expands it into the memory 52, and sequentially executes the processes contained in the program.

<Function of circuit design support device 10>

Next, the operation of the circuit design support device 10 of this embodiment will be described. When a user operates the circuit design support device 10 and the circuit design support device 10 receives a learning process instruction signal, the circuit design support device 10 executes the learning process routine shown in FIG. 9.

In step S100, the learning data acquisition unit 20 acquires multiple pieces of learning data from the learning data storage unit 12.

In step S102, the learning unit 22 performs supervised learning of a machine learning model based on the multiple pieces of learning data acquired in step S100, thereby learning a model for outputting element placement data from circuit characteristic values, and generating a learned model.

In step S104, the learning unit 22 stores the trained model generated in step S102 in the trained model storage unit 14.

When the learning process routine shown in FIG. 9 is executed, the trained model is stored in the trained model storage unit 14.

Next, when the user operates the circuit design support device 10 and the circuit design support device 10 receives an instruction signal for element placement estimation processing, the circuit design support device 10 executes the element placement estimation processing routine shown in FIG. 10.

In step S200, the acquisition unit 24 reads the trained model from the trained model storage unit 14.

In step S202, the generation unit 26 inputs required circuit characteristic values representing the required characteristics of the circuit to the trained model acquired in step S200, thereby generating element placement data representing the placement of elements in the circuit corresponding to the required circuit characteristic values.

In step S204, the generation unit 26 stores the element placement data generated in step S202 in the result storage unit 16.

As described above, the circuit design assistance device 10 according to this embodiment acquires multiple pieces of learning data in which element placement data representing the placement of elements in a circuit is associated with circuit characteristic values representing the characteristics of the circuit. The circuit design assistance device 10 trains a machine learning model based on the multiple pieces of learning data, thereby training a trained model for outputting element placement data from circuit characteristic values.

In addition, the circuit design assistance device 10 according to this embodiment generates element placement data representing the placement of elements in a circuit corresponding to the required circuit characteristic values by inputting required circuit characteristic values representing the required characteristics of the circuit to the trained model.

This makes it possible to efficiently generate element placement data within a circuit from circuit characteristic values. It also makes it possible to automatically obtain element placement data that satisfies the circuit characteristic values. Furthermore, it is possible to determine the placement of elements within a circuit that satisfies the required performance without repeatedly synthesizing the circuit and searching for element values for that circuit.

Next, an example of the circuit design support device 10 according to this embodiment will be described. In this example, element placement data that satisfies certain required circuit characteristics is automatically generated.

In this example, first, a trained model was generated using 2,682 pieces of training data. The conditions for generating the trained model are as shown in Table 1 below. The evaluation function and loss function in Table 1 below represent the mean squared error.

Next, we set the required circuit characteristic values as shown in Table 2 below.

When the required circuit characteristic values shown in Table 2 above were input to the trained model, element placement data such as that shown in the upper part of FIG. 11 was output. The circuit structure data in the lower part of FIG. 11 was created based on the element placement data. The element placement data in FIG. 11 is the element placement data predicted by the trained model, the neural network, when the required circuit characteristic values were input to design a circuit with roughly "large gain," "small output resistance," "large slew rate," and "low power consumption." In fact, the predicted circuit structure shown in FIG. 11 is a circuit structure of a cascode-connected input stage that is likely to have high gain and slew rate, and a super source follower output stage that can achieve low output resistance, which is consistent with the standard techniques for circuit design. In addition, this predicted circuit structure is a new circuit structure that is not included in the circuit structures used as training data, and this method also shows the possibility of creating a new circuit structure.

The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the spirit of the invention.

For example, in the above embodiment, the circuit design assistance device 10 executes the learning process and the element placement estimation process, but this is not limited to the above. For example, the learning device may execute the learning process, and another device different from the learning device may execute the element placement estimation process.

10 Circuit design support device 12 Learning data storage unit 14 Learned model storage unit 16 Result storage unit 20 Learning data acquisition unit 22 Learning unit 24 Acquisition unit 26 Generation unit 50 Computer

Claims (6)

a learning data acquisition unit that acquires a plurality of learning data, the learning data being element placement data representing an arrangement of elements in a circuit, and in which the element placement data is label data representing a plurality of different types of elements and a circuit characteristic value representing a characteristic of the circuit are associated with each other;
a learning unit that learns a machine learning model based on the plurality of learning data acquired by the learning data acquisition unit, thereby learning a learned model for outputting the element placement data from the circuit characteristic values;
A learning device comprising: An acquisition unit that acquires the trained model generated in advance by the learning device according to claim 1 ;
a generation unit that generates element placement data representing a placement of elements in a circuit corresponding to a required circuit characteristic value by inputting a required circuit characteristic value representing a required characteristic of a circuit to the trained model acquired by the acquisition unit;
A circuit design support device comprising: acquiring a plurality of learning data sets, the learning data sets being element placement data representing placement of elements in a circuit and being label data representing a plurality of different types of elements, and circuit characteristic values representing characteristics of the circuit;
a machine learning model is trained based on the acquired plurality of pieces of learning data to train a trained model for outputting the element placement data from the circuit characteristic values;
A learning method in which a computer performs a process. Obtaining the trained model generated in advance by the training method according to claim 3 ;
inputting required circuit characteristic values representing required characteristics of a circuit into the acquired trained model, thereby generating element placement data representing a placement of elements in the circuit corresponding to the required circuit characteristic values;
A circuit design support method in which processing is executed by a computer. acquiring a plurality of learning data sets, the learning data sets being element placement data representing placement of elements in a circuit and being label data representing a plurality of different types of elements, and circuit characteristic values representing characteristics of the circuit;
a machine learning model is trained based on the acquired plurality of pieces of learning data to train a trained model for outputting the element placement data from the circuit characteristic values;
A learning program that allows a computer to carry out processing. Acquire the trained model generated in advance by the training program according to claim 5 ,
inputting required circuit characteristic values representing required characteristics of a circuit into the acquired trained model, thereby generating element placement data representing a placement of elements in the circuit corresponding to the required circuit characteristic values;
A circuit design support program that causes a computer to execute processing.