Heat-generating elements on the substrate (parallel computation)

Parallelize wat_ex14_parametric with 3 Femtet instances. Other items, except for sample code and execution results, are the same as Heating element on the substrate.

Sample File

Note

Keep the sample project open in Femtet, then double-click on the sample code to execute it.

Note

For details on the FEM problem, please refer to FemtetHelp / Examples / Heat Conduction Analysis / Example 14.

Design Variables

Appearance of the Model

Variable Name	Description
substrate_w	Width of the substrate
substrate_d	Depth of the substrate

Objective Function

• Maximum temperature of the main chip (to minimize)

• Maximum temperature of the sub chip (to minimize)

• Occupied area on the substrate plane (to minimize)

Sample Code

wat_ex14_parametric_parallel.py

"""Multi-objective optimization: heating element on substrate.

Using Femtet's heat conduction analysis solver, we will design
to reduce the chip temperature and shrink the board size.

Corresponding project: wat_ex14_parametric.femprj
"""
from pyfemtet.opt import FEMOpt


def chip_temp(Femtet, chip_name):
    """Obtain the maximum temperature of the chip.

    Note:
        The objective or constraint function should take Femtet
        as its first argument and return a float as the output.

    Params:
        Femtet: An instance for manipulating Femtet with macros. For detailed information, please refer to "Femtet Macro Help".
        chip_name (str): The body attribute name defined in femprj. Valid values are 'MAINCHIP' or 'SUBCHIP'.

    Returns:
        float: The maximum temperature of the body with the specified body attribute name.
    """
    Gogh = Femtet.Gogh

    max_temperature, min_temperature, mean_temperature = Gogh.Watt.GetTemp(chip_name)

    return max_temperature  # unit: degree


def substrate_size(Femtet):
    """Calculate the occupied area on the XY plane of the substrate."""
    substrate_w = Femtet.GetVariableValue('substrate_w')
    substrate_d = Femtet.GetVariableValue('substrate_d')
    return substrate_w * substrate_d  # unit: mm2


if __name__ == '__main__':

    # Initialize the FEMOpt object.
    # (establish connection between the optimization problem and Femtet)
    femopt = FEMOpt()

    # Add design variables to the optimization problem.
    # (Specify the variables registered in the femprj file.)
    femopt.add_parameter("substrate_w", 40, lower_bound=22, upper_bound=40)
    femopt.add_parameter("substrate_d", 60, lower_bound=33, upper_bound=60)

    # Add the objective function to the optimization problem.
    # The target bending angle is 90 degrees.
    femopt.add_objective(fun=chip_temp, name='max temp. of<br>MAINCHIP (degree)', direction='minimize', args=('MAINCHIP',))
    femopt.add_objective(fun=chip_temp, name='max temp. of<br>SUBCHIP (degree)', direction='minimize', args=('SUBCHIP',))
    femopt.add_objective(fun=substrate_size, name='substrate size')

    # Run optimization.
    femopt.set_random_seed(42)
    femopt.optimize(n_trials=15, n_parallel=3)  # This line is the only difference with no parallel pattern.

Note

To parallelize, simply pass the desired number of parallelizations to the n_parallel argument of the optimize() method.

Execution Result of the Sample Code

Execution Environment

OS	windows 10
CPU	Intel Core-i7 12700 (12 cores, 20 threads)
Memory	32 GB

Execution Results

Without Parallelization	With 3 Parallelizations
117 sec	74 sec

In this demo, calculations were performed with 3 Femtet instances. For the problem in Heating element on the substrate, without using parallelization in the above execution environment, it took 117 seconds for 20 trials. However, in this demo, 21 trials were completed in 74 seconds, reducing the execution time by 37%.

Note

Generally, when parallelizing numerical calculations by N, the execution time does not simply become 1/N.

Warning

The acceleration effect of parallelization varies depending on the execution environment and analysis model.

Note

Results may vary slightly depending on the versions of Femtet, PyFemtet, and the optimization engine it depends on.