Overview
About
The nextnanoevo Python package is interfacing minimization and evolution algorithms with nextnano simulation tools: nextnano++, nextnano³, and nextnano.NEGF. The package takes care of proper execution of simulations for optimization algorithms and definitions of problems to optimize based on standard nextnano outputs.
The optimization problem that nextnanoevo solves is defined in the following way: given a set of input parameters for the device design \([x_1, x_2,...,x_N]\) (i.e. input file variables such as thicknesses of layers, temperature, alloy contents, simulation hyperparameters etc) and their respective limits, find such input variables vector that minimizes/maximizes the metric function \(f(x_1, x_2,...,x_N)\) (i.e. a figure of merit such as quantum efficiency, bandgap, etc). The objective could be also to have a target value of the metric function \(f_{target}\). In this case, the problem is defined as a minimization problem with the objective to minimize the difference between the target value and the metric function \(|f-f_{target}|\). The nextnanoevo package also able to handle multi-objective optimization problems, where the metric function is a vector. In this case the Pareto dominant solutions are searched for.
The metric function can be a function of the simulation results (i.e. a post-processing function) or a combination of simulation results and input parameters.
The user can choose between optimization algoritms from scipy.optimize and PyGMO:
SciPy.optimize contains set of deterministic optimization algorithms to solve root and minimization problems. You can visit the documentation here.
PyGMO is a Python library developed by F. Biscani and D. Izzo [Biscani2020]. It is an interface facilitating implementation of various evolutionary algorithms such as evolution strategies or genetic algorithms. It can be used to solve many kinds of single- and multi-objective problems.
We recommend using SciPy algorithms to solve simple single-objective problems, while PyGMO is more suited for multi-objective problems as well as for problems that have convergence issues with deterministic algorithms.
Example
For example, one wants to study an LED with the goal to maximize quantum efficiency (QE) and has an initial design with a QE of 4%. The user would provide the algorithm with set of parameters that can be modified and the limits for these parameters. The metric function will be defined as a quantum efficiency and the algorithm will be run to find the optimal design.
Requirements
To run the nextnanoevo module, you need to:
Installation
We recommend installing nextnanoevo with conda package manager. Installation without conda using different python distributions is possible if requirements above are met (but not tested).
Installation with conda
-
Miniconda is a lightweight version that install only the minimum necessary liabraries).
Create and new conda environment with python 3.10
$ conda create -n nnevoEnv python=3.10
Activate newly created conda environment
$ conda activate nnevoEnv
Install pygmo (this step might take few minutes)
$ conda config --add channels conda-forge $ conda config --set channel_priority strict $ conda install pygmo
Install nextnanoevo into the activated environment with installation script
$ path/to/nextnanoevo/install.bat
License activation
The license is activated in the same way as other products, for more details visit License activation page
Config setup
In order to use nextnanoevo, at least 2 products should be set up in nextnanopy config.
nextnanoevo itself (license option only)
nextnano product you want to use it with. For example, for nextnano++ you need to set up license, exe, database and output directory
To set up nextnanoevo license, run the following script.
import nextnanopy as nn nn.config.set('nextnanoevo', 'license', 'path\to\your\License_nnevo.lic') nn.config.save()
Tutorial on how to set up other nextnano product can be found here - Tutorial 0 - Set up the configuration.