optimizeDesign

R2026b

Optimize circuit design in simulation environment

Since R2024b

Syntax

[sol,metric] = obj.optimizeDesign

[sol,metric] = optimizeDesign(obj,NumParallelSims=8,MaxNumberSims=100)

Description

[sol,metric] = obj.optimizeDesign optimizes the circuit parameters of the msbOptimizer object obj.

[sol,metric] = optimizeDesign(obj,NumParallelSims=8,MaxNumberSims=100) optimizes the circuit parameters by running 8 simulations per iteration and limiting the maximum number of simulations to 100.

Examples

Optimize Parameters for Second-Order DSM in Simulink

Define the performance specifications of a second-order DSM.

outputTable=table();
outputTable.Test=["ACMeas";"ACMeas";"ACMeas";"ACMeas";"ACMeas"];
outputTable.Name={'SNR';'SFDR';'SINAD';'ENOB';'NoiseFloor'};
outputTable.Units={'dB';'dB';'dB';'bits';'dB'};
outputTable.Spec={'> 72';'> 74';'> 72';'maximize 11.5';'< -78'}

outputTable = 5×4 table
      Test           Name          Units            Spec       
    ________    ______________    ________    _________________

    "ACMeas"    {'SNR'       }    {'dB'  }    {'> 72'         }
    "ACMeas"    {'SFDR'      }    {'dB'  }    {'> 74'         }
    "ACMeas"    {'SINAD'     }    {'dB'  }    {'> 72'         }
    "ACMeas"    {'ENOB'      }    {'bits'}    {'maximize 11.5'}
    "ACMeas"    {'NoiseFloor'}    {'dB'  }    {'< -78'        }

Define the variables to optimize.

variableTable=table();
variableTable.parameters={'a1';'a2';'b1';'b2'};
variableTable.values=["0.15:0.005:0.16";"0.55:0.005:0.7";"0.15:0.005:0.16";"0.55:0.005:0.7"]

variableTable = 4×2 table
    parameters         values      
    __________    _________________

      {'a1'}      "0.15:0.005:0.16"
      {'a2'}      "0.55:0.005:0.7" 
      {'b1'}      "0.15:0.005:0.16"
      {'b2'}      "0.55:0.005:0.7" 

Create the msbOptimizer object.

moptimizer = msbOptimizer(SimulationEnvironment='simulink',OutputsSetup=outputTable,VariableSetup=variableTable,DesignName='DSM2ndOrder')

moptimizer = 
  msbOptimizer with properties:

               DesignName: 'DSM2ndOrder'
                   Solver: "surrogateopt"
    SimulationEnvironment: 'simulink'
             BestSolution: []
              BestMetrics: []
     FinalOptimizerStatus: []
             OutputsSetup: [5×4 table]
          ParametersSetup: [4×2 table]
           ParameterNames: ["a1"    "a2"    "b1"    "b2"]
          ParameterValues: ["0.15:0.005:0.16"    "0.55:0.005:0.7"    "0.15:0.005:0.16"    "0.55:0.005:0.7"]
                    Eflag: []
                   Trials: []
              Constraints: [5×9 table]
                  Corners: []

Optimize the parameters.

[sol,metric] = moptimizer.optimizeDesign

Maximum number of simulations: 100
Number of parallel simulations: 1

Optimizer was able to meet all the specifications.

sol = 4×2 table
    Name    Value
    ____    _____

    "a1"     0.15
    "a2"    0.575
    "b1"    0.155
    "b2"    0.635

metric = 5×4 table
        Name        FinalMetrics         Specs         Units 
    ____________    ____________    _______________    ______

    "SNR"              72.519       "> 72"             "dB"  
    "SFDR"             80.869       "> 74"             "dB"  
    "SINAD"            72.519       "> 72"             "dB"  
    "ENOB"             11.754       "maximize 11.5"    "bits"
    "NoiseFloor"      -80.755       "< -78"            "dB"  

As you can see, the function meets the required specifications. You can use the check point file to continue further optimization from the current state.

Input Arguments

obj — Circuit design to optimize

msbOptimizer object

Circuit design to optimize, specified as an msbOptimizer object. The object includes the name of the simulation environment, the name of the circuit design, the performance specifications, and the variables to optimize.

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Example: [sol,metric] = obj.optimizeDesign(ProgressPlot='true') optimizes the circuit design defined in the obj object and shows the optimization progress in a plot.

NumParallelSims — Number of parallel simulations per iteration

1 (default) | nonnegative integer scalar

Number of parallel simulations per iteration, specified as a nonnegative integer scalar.

Data Types: double

MaxNumberSims — Maximum number of simulations to run

max(100,10*nvar) (default) | nonnegative integer scalar

Maximum number of simulations to run, specified as a nonnegative integer scalar. The default is max(100,10*nvar), where nvar is the number of problem variables.

Data Types: double

MinInitialRandomPoints — Minimum number of random sample points to create at the start of optimization

max(20,2*nvar) (default) | nonnegative integer scalar

Minimum number of random sample points to create at the start of the optimization phase, specified as a nonnegative integer scalar. The default is max(20,2*nvar), where nvar is the number of problem variables. For more information, see Surrogate Optimization Algorithm (Global Optimization Toolbox).

Note

The function ignores this argument if the Solver is set to essabopt.

Data Types: double

ProgressPlot — Show optimization progress

true (default) | false

Show optimization progress in a plot.

Data Types: logical

MetricPlot — Metric name to plot on progress plot

string | character vector

Metric name to plot on the progress plot, specified as a string or character vector. The default is the first metric on the constraint table.

CheckPointFileName — Name of file to create checkpoints and restart optimization

myCheckPoint (default) | string | character vector

Name of the file to create checkpoints and restart the optimization process, specified as a string or character vector.

UseCheckPoint — Use checkpoint file for optimization

true (default) | false

Use the checkpoint file for optimization. The function can use this argument only after the creation of a checkpoint file at the end of a simulation run.

If the Solver is set to surrogateopt, the checkpoint file saves the internal surrogate solver data.

If the Solver is set to essabopt, the checkpoint file saves the entire population, performance history, and optimizer state.

Data Types: logical

Solver — Optimization solver

surrogateopt (default) | essabopt

Optimization solver used to optimize circuit parameters. You can choose between surrogateopt and essabopt.

Note

To use the essabopt optimization technique, you need a license for Deep Learning Toolbox™.

ObjectiveFunction — Handle to custom objective function

object handle

Handle to a custom objective function that runs simulation and returns the objective and constraint violations, specified as an object handle.

You can use this argument to support non-standard simulation workflows such as running Simulink^® simulations or extracting custom metrics from the logged signals. For more information, see Custom Objective Functions for Simulink Simulation.

InitialPoints — Initial points for surrogate optimization

[] (default) | matrix | structure

Initial points for surrogate optimization, specified as a matrix or structure.

When you set Solver to essabopt, the function ignores this argument. The function rather sets the internal population by max(20,5*nvar).

Data Types: double

RngSeed — Random number generator seed

1 (default) | nonnegative integer scalar

Random number generator seed, specified as a nonnegative integer scalar.

Data Types: double

RngAlgorithm — Random number generator algorithm used by optimization solver

twister (default) | string

Random number generator algorithm used by optimization solver, specified as a string.

Data Types: char

Output Arguments

sol — Optimized variables or parameters

table

Optimized variables or parameters, returned as a table. The elements of the table match the variables or parameters defined in the VariableSetup argument in msbOptimizer object.

metric — Performance metrics using optimized variables or parameters

table

Performance metrics using optimized variables or parameters, returned as a table. The elements of the table match the variables or parameters defined in the OutputsSetup argument in msbOptimizer object.

More About

Custom Objective Functions for Simulink Simulation

To run non-standard Simulink simulations, you can use the ObjectiveFunction argument to create custom objective functions. The custom function must accept parameter indices as a matrix of variable values.

Argument Name	Argument Type	Description
x	Input	Matrix of variable indices, specified as integers. The elements are mapped to variablevalues.
variableNames	Input	String array of parameter names.
contraintTable	Input	Internal constraint table derived from OutputsSetup.
model	Input	Name of the Simulink model.
variableValues	Input	Cell array of value vectors, one per parameter.
out.Fval	Output	Scalar objective per design. A negative value means the function meets all constraints.
out.Ineq	Output	Constraint violations.

You can use this template to create your custom objective function:

% Custom Objective Function Template
function out = myCustomObjFcn(x, variableNames, constraintTable, model, variableValues)
    out.Fval = NaN(size(x,1), 1);
    out.Ineq = NaN(size(x,1), height(constraintTable));

    for ii = 1:size(x,1)
        % Map indices to physical values
        for jj = 1:numel(variableNames)
            paramValue = variableValues{jj}(x(ii,jj));
            % ... assign parameter, run simulation, extract metrics ...
        end
    end

    % Helper computes objective + constraint violations from results
    [out.Fval, out.Ineq] = msblks.optimization.surrogateOptHelper.getFvalIneq( ...
        size(x,1), resultsTable, constraintTable);
end

objFcn = @(x) myCustomObjFcn(x, varNames, constrTable, modelName, varValues);
[sol, met] = optimizer.optimizeDesign('ObjectiveFunction', objFcn);

ESSAB Optimization

The Efficient Surrogate Model-assisted Sizing Method for High-performance Analog Building Blocks (ESSAB) is a machine learning-based surrogate global optimization approach designed specifically for analog IC design. It combines differential evolution (DE) for design generation with an artificial neural network (ANN) for performance prediction. It uses a novel probability of further improvement (PFI) infill sampling criterion based on beta-distribution ranking.

ESSAB is designed for optimizing high-dimensional problems (more than 10 variables). But extremely high dimensions can degrade ANN prediction accuracy. Try to limit your design variables between 10 and 40.

Tips

• Set Solver to surrogateopt for general-purpose optimization problems with short simulation times or quick design verifications. This works best when the number of design variables is less than 10 or problems have discontinuous transitions.

• Set Solver to essabopt for analog IC sizing problems with long simulation times or when accurate final design values are critical. It works best when the number of design variables is between 10 and 40 or problems have multiple constraints with stringent specifications.

Version History

Introduced in R2024b