A circuit optimization method based on forward and backward simulation strategy and multi-objective optimization

Abstract

The present invention proposes a circuit optimization method based on pre- and post-simulation strategies and multi-objective optimization, which belongs to the field of multi-objective circuit optimization design. The method includes a pre-simulation process: based on a circuit with a fixed component connection structure, a multi-objective heuristic strategy is used to adjust the parameter settings of circuit components; in the calculation of the objective function and constraint function of the multi-objective optimization, a simulator is used for evaluation; a post-simulation process: an adaptive reference vector method is used to screen the solution set obtained by parameter optimization to obtain an elite solution; layout and routing optimization is performed based on the elite solution, and the simulator is called to calculate the target value and constraint value; if the result that satisfies the constraint is not obtained, the optimization process is adaptively adjusted according to the unsatisfied situation, and the optimization is continued until an excellent solution set is obtained. The present invention can save design time and energy by iteratively optimizing the analog circuit design work using an automated optimization algorithm, effectively improving the circuit design efficiency, and saving design costs.

Description

Technical Field

[0001] The present invention relates to the technical fields of EDA, analog circuit design, multi-objective optimization technology and simulation strategy, and in particular to a circuit optimization method based on forward and backward simulation strategy and multi-objective optimization. Background Art

[0002] Electronic design automation (EDA) is the core of the development of electronic design and manufacturing technology. EDA technology uses computers as tools and hardware description language to scientifically and effectively integrate databases, computational mathematics, graph theory, graphics, topological logic, optimization theory, etc., thus forming a new technology dedicated to electronic systems. It is the latest achievement of computer technology, signal processing technology, and signal analysis technology.

[0003] The emergence of EDA technology not only better guarantees the simulation, debugging and error correction at all levels of electronic engineering design, bringing strong technical support for its development, but also occupies an increasingly important position in various fields such as electronics, communications, chemical industry, aerospace, biology, etc., and greatly reduces the workload of related practitioners.

[0004] Analog circuit design is a very important technical field, involving the processing and amplification of analog signals, as well as the use and design of various analog circuit components. The difficulty of analog circuit design lies in the need to compromise between multiple factors such as speed, power consumption, gain, accuracy, power supply voltage, noise, area, etc., while digital design only needs to balance between power consumption, speed and area. In addition, analog circuits are much more sensitive to noise, crosstalk and other interference than digital circuits.

[0005] Modern circuit design is complex, and the design of high-performance analog circuits is a time-consuming process that requires the participation of highly qualified designers and frequent and tedious interactions with EDA tools. Summary of the invention

[0006] The purpose of the present invention is to provide a circuit optimization method based on pre- and post-simulation strategy and multi-objective optimization, which uses a meta-heuristic strategy design algorithm for multi-objective optimization, automatically calls EDA simulation tools, and adopts the optimization strategy of pre- and post-simulation to optimize the circuit. By designing the coordination and feedback mechanism of pre- and post-simulation, the efficiency of circuit optimization is effectively improved, and the production efficiency of the current manual analog circuit design is greatly improved. It is used to assist or even replace manual circuit design work and improve the efficiency of circuit design. In order to solve the technical problems in the current technology that modern circuit design is complex, the design of high-performance analog circuits is time-consuming, and the participation of high-quality designers and frequent and cumbersome interactions with EDA tools are required.

[0007] In order to solve the above technical problems, the circuit optimization method based on the forward and backward simulation strategy and multi-objective optimization of the present invention comprises the following steps:

[0008] S0: Obtain the circuit schematic to be optimized, circuit index requirements, simulator setting files for pre-simulation, and algorithm parameter setting files.

[0009] S1: Inputting the circuit schematic, circuit index requirements, simulator setting files of pre-simulation and algorithm parameter setting files into the pre-simulation algorithm; the above files set and define the number, type, connection mode and device properties of circuit components;

[0010] S2: Initialize the pre-simulation algorithm according to the circuit schematic, circuit index requirements and algorithm parameter setting file; specifically including the following steps:

[0011] S21: Select two or more conflicting simulation indicators as optimization targets; select several other simulation indicators of interest as optimization constraints; select circuit device parameters to be optimized as decision variables.

[0012] Without loss of generality, the multi-objective minimization problem can be formulated as:

[0013] min F (x)=[ f 1 x , f 2 x ,… f i x …, f M (x)]

[0014] And meet:

[0015]

[0016]

[0017] in, yes dimensional decision vector, is the target number, is the number of constraints; in the process of circuit parameter optimization, represents a set of circuit parameters, which can also be regarded as a solution; express No. target value; and Respectively No. Inequality constraints and equality constraints, usually set to be less than or equal to when the constraints are satisfied The value of

[0018] If there is a solution and , then the Pareto dominate relationship is defined as follows:

[0019] if:

[0020]

[0021] And there is at least one Make ,in is the number of optimization targets, then Dominate ;

[0022] The ideal result of multi-objective optimization of circuit parameters is a Pareto non-dominated solution set , that is, any two solutions in the solution set and None of them satisfy the Pareto dominance relationship;

[0023] Set other simulation indicators of interest except the target as inequality constraints Or equality constraints ; It is stipulated that the solution with fewer total constraint violations is better, and the solution with smaller violation degree is better when the number of violations is the same; therefore, the definition of The degree of violation of a constraint :

[0024]

[0025] in, is the constraint index value of the current solution, is the expected constraint value, is a weight factor, which is defined differently under different constraints, as follows:

[0026] In the inequality constraint, if Greater than or equal to Satisfy the requirements, define :

[0027]

[0028] On the contrary Less than or equal to Satisfy the requirements, define :

[0029]

[0030] In the equality constraint, define :

[0031]

[0032] Therefore, the constraint relationship or Calculated by the degree of violation; When the constraint relation is an inequality constraint, , On the contrary, when When a constraint is an equality constraint, , ;Will Summing gives the total constraint violation:

[0033]

[0034] S22: Initialize algorithm parameters, including: population size , number of targets , Constraint Number , the dimension of the solution , total number of iterations , number of subpopulations Number of elite solutions ; Then generate a random initial population;

[0035] S23: starting the pre-simulator and the pre-simulation algorithm, initializing the pre-simulator according to the pre-simulation simulator setting file, and checking whether the communication between the algorithm and the pre-simulator is good; in the pre-simulation process, it is not necessary to consider the influence of factors such as device process, spatial position, temperature, wiring constraints, etc. in the actual circuit, and only consider the operation of the circuit under the most ideal condition;

[0036] S24: The pre-simulation algorithm converts each solution The pre-simulator calculates the pre-simulation index values ​​of each solution and sends them to the pre-simulation algorithm. The pre-simulation algorithm then calculates the target value of each solution based on the pre-simulation index values. and constraint value , ;

[0037] S3: The pre-simulation algorithm uses a meta-heuristic strategy to iteratively optimize the circuit parameters; the specific steps of the meta-heuristic strategy are as follows:

[0038] S31: Select the Pareto Front solution as the parent according to the target value and constraint value of the current solution set to generate offspring;

[0039] S32: By using the crossover operator and mutation operator for parameter optimization, the offspring population is calculated, and its size is the same as that of the parent;

[0040] S33: Calculate the offspring population and constraint value , , the calculation method is the same as S24;

[0041] S34: Use non-dominated sorting to select the environment, select the non-dominated solution, and then use the adaptive reference vector method to screen the non-dominated solution to obtain the best front solutions as the population of the next generation; that is, to ensure that any The solution of the generation and The solution of the generation satisfy: Non-dominant .

[0042] The specific steps of the method for adaptive reference vector are as follows:

[0043] S341: Generate a set of uniformly distributed points on a cell hyperplane using a canonical simplex grid design method :

[0044]

[0045] Each point is a vector, the elements in the vector Belongs to discrete set ,in ,here is the number of uniformly distributed points, which is equal to the number of subpopulations, is the target number, is a positive integer designed for the simplex grid. Then, the corresponding unit reference vector can be obtained .

[0046]

[0047] in, Representation vector The Euclidean norm of ;

[0048] S342: Target value conversion: Convert the target value of individuals in the population Convert to target vector , by the following method:

[0049] F(x)= f 1 x , f 2 x ,… f i x …, f M (x)]

[0050]

[0051] F ' (x)=[ f 1 ' x , f 2 ' x ,… f i ' x …, f M ' (x)]

[0052] in , Indicates the target number; is a dimensional vector, called the target vector.

[0053] S343: By associating each solution in the population with its nearest reference vector, The solution is called , its target vector With Reference vector The cosine between can be calculated as:

[0054]

[0055] here , , Indicates The target vector of the solution With Reference vector The angle between Represents the target vector The Euclidean norm of .

[0056] S344: The individual with the smallest angle penalty distance (APD) is selected from the subpopulation as the elite solution; The convergence standard is measured by To measure the diversity criteria, in order to balance the relationship between the two, the angle penalty distance is defined for:

[0057]

[0058] in:

[0059]

[0060] is a hyperparameter, express The angle of the other reference vectors that forms the smallest angle, Indicates the current iteration number. represents the total number of iterations, is the target number.

[0061] S345: Adaptively adjust the reference vector to make the population division more uniform:

[0062]

[0063] in , is the current iteration number, express The first A reference vector, Indicates the initial A reference vector, and Respectively The maximum and minimum values ​​of each objective function in the generation, The operator represents the Hadamard product. Representation vector The Euclidean norm of .

[0064] S35: Repeat S31 to S34 until the optimization exit condition is met, and then output the optimized population; this process is also called the pre-simulation optimization process;

[0065] S4: Using the adaptive reference vector method, select the output population of the previous simulation optimization process The elite solutions are used as the initial population of the post-simulation optimization algorithm in the post-simulation optimization process;

[0066] S5: Input component process library files, circuit index requirements, layout and wiring setting files, algorithm setting files and pre-simulation elite solutions; then initialize the simulation algorithm based on these files; the specific steps are as follows:

[0067] S51: Select one or more conflicting simulation indicators as optimization targets; select other simulation indicators of interest as optimization constraints; the principle of selecting multiple targets is the same as that described in S21; generally speaking, the targets or constraints in the post-simulation process include minimizing routing length, power consumption optimization, stacking design, thermal management, and test maintenance;

[0068] S52: Initialized simulation algorithm parameters, including: population size , number of targets , Constraint Number , the dimension of the solution , Iterations ; Then use the previous simulation Elite solutions generate the initial population;

[0069] S53: Start the post-simulator and check whether the communication between the post-simulation algorithm and the post-simulator is good;

[0070] S54: the post-simulation algorithm sends each solution to the post-simulator, the post-simulator calculates each post-simulation index value of each solution and sends it to the post-simulation algorithm, and the post-simulation algorithm calculates the target value and constraint value of each solution according to the post-simulation result;

[0071] S6: The post-simulation algorithm uses a meta-heuristic strategy to iteratively optimize the circuit layout and routing; the specific steps of the meta-heuristic strategy are as follows:

[0072] S61: Select the Pareto Front solution as the parent according to the target value and constraint value of the current solution set to generate offspring;

[0073] S62: By placing and routing the crossover mutation operator, the offspring population is calculated, and its size is the same as that of the parent;

[0074] S63: Calculate the target value and constraint value of the offspring population, the calculation method is the same as S54;

[0075] S64: When there are multiple targets, use non-dominated sorting to select the environment, select the non-dominated solution, and then use the adaptive reference vector method to screen the non-dominated solution to obtain the best front The solution is used as the population of the next generation; when there is a single target, the first solution with the smallest sum of the target value and the constraint value is selected from the parents and offspring. The solutions are used as the population for the next generation;

[0076] S65: Repeat S61 to S64 until the optimization exit condition is met, and then output the optimized solution; this process is also called the post-simulation optimization process;

[0077] S7: If the solution output by S6 meets the constraint requirements, it will be adopted; otherwise, according to the specific situation of non-compliance, the feedback strategy is adaptively selected: if the violation is minor, the reference vector of step S4 is adjusted; if the violation is serious, the initialization solution set of step S2 is adjusted, and then the optimization continues. S7 and S4 are collectively referred to as the forward and backward simulation collaborative process;

[0078] Compared with the prior art, the present invention has the following beneficial technical effects:

[0079] 1) The present invention can improve optimization efficiency and reduce design cycle: by combining pre-simulation and post-simulation strategies, the performance of the circuit can be better evaluated and predicted, unnecessary simulation calculations can be reduced, and the overall optimization efficiency can be improved.

[0080] 2) The present invention supports multi-objective optimization and can simultaneously consider multiple performance indicators during the design process, such as power consumption, area, speed, etc., so as to obtain a more balanced and high-quality circuit design solution.

[0081] 3) The present invention has strong flexibility and adaptability: the method is applicable to various types of circuit designs and is not limited to specific types of circuits, and has broad application prospects and flexibility.

[0082] 4) The present invention can reduce design costs: through efficient simulation and multi-objective optimization in the optimization process, the number of repeated experiments and modifications can be reduced, thereby reducing the overall design cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 An overall flow chart of a circuit optimization method based on forward and backward simulation strategies and multi-objective optimization provided by an embodiment of the present invention. Specific embodiments

[0084] In order to make the purpose, technical solution and beneficial effects of the present invention more clearly understood, the specific implementation methods of the present invention will be further described in detail below in conjunction with specific embodiments.

[0085] This embodiment takes the optimization process of a field effect transistor (MOSFET) amplifier circuit as an example. In the pre-simulation process, the amplifier circuit has three indicators that need to be met. Specifically, after omitting the unit of the indicator: total current Less than 1 and the smaller the better, harmonic loss Less than 1 and the smaller the better, and bandwidth is required Greater than or equal to 3. The indicators in the post-simulation process are: trace length The smaller the better; power consumption Less than or equal to 2 and the smaller the better, the temperature of each point in the circuit Less than or equal to 100 and the smaller the better, the device and the wire meet the overlap limit (i.e. the overlap area ).

[0086] Based on the above optimization requirements, this embodiment provides a circuit optimization method based on forward and backward simulation strategy and multi-objective optimization, and its process is as follows: Figure 1 As shown, the specific steps include:

[0087] S0: Obtain the circuit schematic to be optimized, circuit index requirements, simulator setting files for pre-simulation, and algorithm parameter setting files. The above files are provided by senior engineers and technicians.

[0088] S1: Input the circuit schematic, circuit index requirements, pre-simulation simulator setting files and algorithm parameter setting files into the pre-simulation algorithm; the above files set and define the number of circuit components (30 in total), types (resistors, capacitors, inductors, current sources, MOS tubes, ground, diodes and voltage sources), connection methods (such as which wire is connected to a port of a device) and device properties (i.e. fixed parameters that do not need to be optimized and the initial values ​​of parameters that need to be optimized).

[0089] S2: Initialize the pre-simulation algorithm according to the circuit schematic, circuit index requirements and algorithm parameter setting file. Specifically, it includes the following steps:

[0090] S21: Select and As the optimization target; select As optimization constraints; select the circuit device parameters to be optimized as decision variables.

[0091] The multi-objective minimization problem can be formulated as:

[0092] min F (x)=[ f IDC (x), f THD (x)]

[0093] And meet:

[0094]

[0095] Among them, the decision variables ; During the circuit parameter optimization process, represents a set of circuit parameters, which can also be regarded as a solution; Indicates circuit usage This set of parameters value, Indicates circuit usage This set of parameters value; Indicates an inequality constraint, which is set to be less than or equal to when the constraint is satisfied The value of and , then the Pareto dominate relationship is defined as follows:

[0096] if:

[0097]

[0098] And there is at least one Make , then it is called Dominate ;

[0099] The ideal result of multi-objective optimization of circuit parameters is a Pareto non-dominated solution set , that is, any two solutions in the solution set and None of them satisfy the Pareto dominance relationship;

[0100] Defining constraints The degree of violation :

[0101]

[0102] in, is the constraint index value of the current solution, is the constraint index value that is expected to be achieved;

[0103] because Greater than or equal to 3 meets the index requirements, so define :

[0104]

[0105] Therefore, the constraint relationship Calculated from the degree of violation:

[0106]

[0107] S22: Initialize algorithm parameters, including: population size , number of targets , Constraint Number , the dimension of the solution , Iterations , number of subpopulations Number of elite solutions ; Then generate a random initial population;

[0108] S23: starting the pre-simulator and the pre-simulation algorithm, initializing the pre-simulator according to the pre-simulation simulator setting file, and checking whether the communication between the algorithm and the pre-simulator is good; in the pre-simulation process, it is not necessary to consider the influence of factors such as device process, spatial position, temperature, wiring constraints, etc. in the actual circuit, and only consider the operation of the circuit under the most ideal condition;

[0109] S24: Calculate the target value and constraint value of the initial population: The pre-simulation algorithm converts each solution Sent to the pre-simulator, the pre-simulator calculates each solution , and The indicator is sent to the pre-simulation algorithm, which then calculates each target value of each solution based on the simulation indicator. and constraint value ;

[0110] S3: The pre-simulation algorithm uses a meta-heuristic strategy to iteratively optimize the circuit parameters; the specific steps of the meta-heuristic strategy are as follows:

[0111] S31: According to the target value of the current solution set and constraint value Select the Pareto Front solution as the parent to generate the offspring;

[0112] S32: Through the crossover operator and mutation operator, the offspring population is calculated, and its size is the same as that of the parents;

[0113] S33: Calculate the offspring population and : The calculation method is the same as S24;

[0114] S34: Use non-dominated sorting to select the environment, select the non-dominated solutions, and then use the adaptive reference vector method to screen the non-dominated solutions to obtain the top 100 solutions as the next generation population; that is, to ensure that any The solution of the generation and The solution of the generation satisfy: Non-dominant The specific steps of the method for adaptive reference vector are as follows:

[0115] S341: Generate a set of uniformly distributed points on the unit hyperplane using the canonical simplex-lattice design method .

[0116]

[0117]

[0118] Each point is a vector, the elements in the vector and Belongs to discrete set , because currently only and Two goals, so There are only two elements in , is the number of uniformly distributed points, is a positive integer designed for a simplex grid. Then, the corresponding unit reference vector can be obtained by transformation .

[0119]

[0120] in, Representation vector The Euclidean norm of ;

[0121] S342: Target value conversion: Convert the target value of individuals in the population Convert to target vector , by the following method:

[0122] F(x)= f IDC (x), f THD (x)]

[0123]

[0124] F ' (x)=[ f IDC ' x , f THD ' x ]

[0125] in , Indicates the target number; In the target space, there is a 2-dimensional vector, called the target vector.

[0126] S343: Divide the population into 10 sub-populations by associating each solution in the population with its nearest reference vector, and then adaptively and dynamically adjust the reference vector to make the partition selection uniform; The solution is called , its target vector With Reference vector The cosine between can be calculated as:

[0127]

[0128] here , , Indicates The target vector of the solution With Reference vector The angle between Represents the target vector The Euclidean norm of .

[0129] S344: Because , so in each subpopulation, select an individual with the smallest angle penalty distance (APD) as the elite solution; use The convergence standard is measured by To measure the diversity criteria, in order to balance the relationship between the two, the angle penalty distance is defined for:

[0130]

[0131] in:

[0132]

[0133] express The angle of the other reference vectors that forms the smallest angle, represents the number of iterations, is the target number.

[0134] S345: Adaptively adjust the reference vector to make the population division more uniform:

[0135]

[0136] in , is the current iteration number, express The first A reference vector, Indicates the initial A reference vector, and Respectively The maximum and minimum values ​​of each objective function in the generation, The operator represents the Hadamard product. Representation vector The Euclidean norm of .

[0137] S35: Repeat S31 to S34 until the optimization exit condition is met, and then output the optimized population;

[0138] S4: Using the adaptive reference vector method, 10 elite solutions are selected from the output population of the pre-simulation as the initial population of the post-simulation algorithm;

[0139] S5: Input component process library files, circuit index requirements, layout and routing setting files, algorithm setting files, and 10 elite solutions of the pre-simulation; then initialize the post-simulation algorithm based on these files; the specific steps are as follows:

[0140] S51: Select the trace length As an optimization goal ; Select power consumption As an inequality constraint ; Select temperature As an inequality constraint ; Select overlap limit As an equality constraint

[0141] At this time, the optimization problem can be expressed as:

[0142]

[0143] And meet:

[0144]

[0145]

[0146]

[0147] Among them, the decision variables ;During the layout and routing optimization process, Represents the encoding information of layout and routing, 150 is the encoding length; Indicates circuit usage This layout and wiring method value; represents an inequality constraint, represents an inequality constraint, Indicates an equality constraint. When the constraint is satisfied, it is set to be less than or equal to The value of

[0148] by For example, define the constraints The degree of violation :

[0149]

[0150] in, is the overlap value between the device and the wire of the current solution, is the constraint index value that is expected to be achieved; is a weight factor. The specific definition of constraints is as follows:

[0151] because It is an equality constraint. When it is equal to 0, it meets the index requirements, so we define :

[0152]

[0153] Therefore, the constraint relationship Calculated from the degree of violation:

[0154]

[0155] S52: Initialized simulation algorithm parameters, including: population size , number of targets , Constraint Number , the dimension of the solution , Iterations ; Then use the 10 elite solutions from the previous simulation to generate the initial population;

[0156] S53: Start the post-simulator and check whether the communication between the post-simulation algorithm and the post-simulator is good;

[0157] S54: Calculate the target value of the initial population and constraint value , , :The post-simulation algorithm sends each solution to the post-simulator, the post-simulator calculates the simulation indicators of each solution and sends them to the post-simulation algorithm, and the post-simulation algorithm calculates the target value and constraint value of each solution based on the simulation results;

[0158] S6: The post-simulation algorithm uses a meta-heuristic strategy to iteratively optimize the circuit layout and routing; the specific steps of the meta-heuristic strategy are as follows:

[0159] S61: Select some better solutions as parents according to the target value and constraint value of the current solution set to generate offspring;

[0160] S62: By placing and routing the crossover mutation operator, the offspring population is calculated, and its size is the same as that of the parent;

[0161] S63: Calculate target value for offspring population and constraint value , , ;

[0162] S64: Since there is only one objective function at this time, the first 100 solutions with the smallest sum of objective value and constraint value are selected from the parents and offspring as the population of the next generation;

[0163] S65: Repeat S61 to S64 until the optimization exit condition is met, and then output the optimized solution set;

[0164] S7: If the solution set output by S6 exists Make , , are less than or equal to 0, then the It is adopted; otherwise, according to the specific situation that it does not meet, the feedback strategy is adaptively selected: if the violation is minor, the reference vector selection of step S4 is adjusted; if the violation is serious, the initialization solution set of step S2 is adjusted, and then the optimization continues.

[0165] The present invention supports multi-objective optimization and can simultaneously consider multiple performance indicators during the design process, such as power consumption, area, speed, etc., so as to obtain a more balanced and high-quality circuit design solution. The present invention improves the optimization efficiency and reduces the design cycle: by combining the strategies of pre-simulation and post-simulation, the performance of the circuit can be better evaluated and predicted, unnecessary simulation calculations can be reduced, and the overall optimization efficiency can be improved. In addition, the present invention has strong flexibility and adaptability, is suitable for various types of circuit designs, and is not limited to specific types of circuits, and has broad application prospects and flexibility. Through efficient simulation and multi-objective optimization during the optimization process, the present invention can reduce the number of repeated experiments and modifications and reduce the overall design cost.

[0166] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A circuit optimization method based on forward and backward simulation strategy and multi-objective optimization, characterized in that: The following steps are involved: S0: Obtain the circuit schematic to be optimized, circuit index requirements, simulator setting files of the previous simulation, and algorithm parameter setting files; S1: Inputting the circuit schematic, circuit index requirements, simulator setting files of pre-simulation and algorithm parameter setting files into the pre-simulation algorithm; the above files set and define the number, type, connection mode and device properties of circuit components; S2: Initialize the pre-simulation algorithm according to the circuit schematic, circuit index requirements and algorithm parameter setting file; S3: The pre-simulation algorithm uses a meta-heuristic strategy to iteratively optimize the circuit parameters; S4: Using the adaptive reference vector method, select the output population of the previous simulation elite solutions as the initial population of the post-simulation algorithm; S5: Input component process library files, circuit index requirements, layout and wiring setting files, algorithm setting files and pre-simulation elite solutions; then initialize the simulation algorithm based on these files; S6: The post-simulation algorithm uses a meta-heuristic strategy to iteratively optimize circuit layout and routing; S7: If the solution output by S6 meets the constraint requirements, it will be adopted; otherwise, according to the specific situation of non-compliance, the feedback strategy is adaptively selected: if the violation is minor, the reference vector of step S4 is adjusted; if the violation is serious, the initialization solution set of step S2 is adjusted, and then the optimization continues; Step S5 specifically includes the following steps: S51: Select one or more conflicting simulation indicators as optimization targets; select other simulation indicators of interest as optimization constraints; S52: Initialized simulation algorithm parameters, including: population size , number of targets , number of constraints , the dimension of the solution , Iterations ; Then use the previous simulation Elite solutions generate the initial population; S53: Start the post-simulator and check whether the communication between the post-simulation algorithm and the post-simulator is good; S54: The post-simulation algorithm sends each solution to the post-simulator, the post-simulator calculates each simulation index of each solution and sends it to the post-simulation algorithm, and the post-simulation algorithm calculates the target value and constraint value of each solution according to the simulation results.

2. The circuit optimization method based on forward and backward simulation strategy and multi-objective optimization according to claim 1, characterized in that: Step S2 includes the following steps: S21: Select two or more conflicting simulation indicators as optimization targets; select other simulation indicators of interest as optimization constraints; select circuit device parameters to be optimized as decision variables; The minimization multi-objective problem is formulated as: And meet: in, yes dimensional decision vector, is the target number, is the number of constraints; in the process of circuit parameter optimization, represents a set of circuit parameters, which can also be regarded as a solution; express No. target value; and Respectively No. Inequality constraints and equality constraints, usually set to be less than or equal to when the constraints are satisfied The value of If there is a solution and , then the Pareto dominance relationship is defined as follows: if: And there is at least one Make ,in is the number of optimization targets, then Dominate ; The ideal result of multi-objective optimization of circuit parameters is a Pareto non-dominated solution set , that is, any two solutions in the solution set and None of them satisfy the Pareto dominance relationship; Set other simulation indicators of interest except the target as inequality constraints Or equality constraints ; It is stipulated that the solution with fewer total constraint violations is better, and the solution with smaller violation degree is better when the number of violations is the same; therefore, the definition of The degree of violation of a constraint : in, is the constraint index value of the current solution, is the expected constraint value, is a weight factor, which is defined differently under different constraints, as follows: In the inequality constraint, if Greater than or equal to Satisfy the requirements, define : On the contrary Less than or equal to Satisfy the requirements, define : In the equality constraint, define : Therefore, the constraint relationship or Calculated by the degree of violation; When the constraint relation is an inequality constraint, , On the contrary, when When a constraint is an equality constraint, , ;Will Summing gives the total constraint violation: S22: Initialize algorithm parameters, including: population size , number of targets , number of constraints , the dimension of the solution , total number of iterations , number of subpopulations Number of elite solutions ; Then generate a random initial population; S23: Start the pre-simulator and the pre-simulation algorithm, initialize the pre-simulator according to the pre-simulation simulator setting file, and check whether the communication between the algorithm and the pre-simulator is good; S24: The pre-simulation algorithm converts each solution The pre-simulator calculates the pre-simulation index values ​​of each solution and sends them to the pre-simulation algorithm. The pre-simulation algorithm then calculates the target value of each solution based on the pre-simulation index values. and constraint value , .

3. The circuit optimization method based on forward and backward simulation strategy and multi-objective optimization according to claim 2 is characterized in that: The metaheuristic strategy of step S3 includes the following steps: S31: Select the Pareto frontier solution as the parent according to the target value and constraint value of the current solution set to generate offspring; S32: By using the crossover operator and mutation operator for parameter optimization, the offspring population is calculated, and its size is the same as that of the parent; S33: Calculate the offspring population and constraint value , ; S34: Use non-dominated sorting to select the environment, select the non-dominated solution, and then use the adaptive reference vector method to screen the non-dominated solution to obtain the best front solutions as the population of the next generation; that is, to ensure that any The solution of the generation and The solution of the generation satisfy: Non-dominant .

4. The circuit optimization method based on forward and backward simulation strategy and multi-objective optimization according to claim 3 is characterized in that: The method of adaptive reference vector comprises the following steps: S341: Generate a set of uniformly distributed points on a cell hyperplane using a canonical simplex grid design method : Each point is a vector, the elements in the vector Belongs to discrete set ,in ,here is the number of uniformly distributed points, which is equal to the number of subpopulations, is the target number, is a positive integer designed for a simplex grid; the unit reference vector is then calculated by normalizing ; in, Representation vector The Euclidean norm of ; S342: Target value conversion: Convert the target value of individuals in the population Convert to target vector , by the following method: in , Indicates the target number; is an M-dimensional vector in the target space, called the target vector; S343: By associating each solution in the population with its nearest reference vector, The solution is called , its target vector With Reference vector The cosine between is calculated as: here , , Indicates The target vector of the solution With Reference vector The angle between Represents the target vector The Euclidean norm of ; S344: The individual with the smallest angle penalty distance is selected as the elite solution in the subpopulation; The convergence standard is measured by To measure the diversity criteria, in order to balance the relationship between the two, the angle penalty distance is defined for: in: is a hyperparameter, express The angle of the other reference vectors that forms the smallest angle, Indicates the current iteration number. represents the total number of iterations, is the target number; S345: Adaptively adjust the reference vector to make the population division more uniform: in , is the current iteration number, express The first A reference vector, Indicates the initial A reference vector, and Respectively The maximum and minimum values ​​of each objective function in the generation; The operator represents the Hadamard product; Representation vector The Euclidean norm of ; S35: Repeat S31 to S34 until the optimization exit condition is met, and then output the optimized population.

5. The circuit optimization method based on forward and backward simulation strategy and multi-objective optimization according to claim 4 is characterized in that: The meta-heuristic strategy in step S6 specifically includes the following steps: S61: Select the Pareto frontier solution as the parent according to the target value and constraint value of the current solution set to generate offspring; S62: By placing and routing the crossover mutation operator, the offspring population is calculated, and its size is the same as that of the parent; S63: Calculate the target value and constraint value of the offspring population; S64: When there are multiple targets, use non-dominated sorting to select the environment, select the non-dominated solution, and then use the adaptive reference vector method to screen the non-dominated solution to obtain the best front The solution is used as the population of the next generation; when there is a single target, the first solution with the smallest sum of the target value and the constraint value is selected from the parents and offspring. The solutions are used as the population for the next generation; S65: Repeat S61 to S64 until the optimization exit condition is met, and then output the optimized solution.

Images