Skip to content

ViennaFit Documentation

Welcome to ViennaFit - a Python package for optimizing and calibrating ViennaPS process simulation models by fitting parameters to experimental data.

What is ViennaFit?

ViennaFit automates the task of finding optimal process parameters that make your ViennaPS simulations match experimental results. Instead of manually tweaking parameters through trial and error, ViennaFit uses advanced optimization algorithms to systematically search the parameter space and find the best fit.

Whether you're calibrating a deposition or etch model to match measured profiles, exploring how different process conditions affect outcomes, or analyzing which parameters matter most, ViennaFit provides the tools you need.

Key Features

  • Multiple Optimization Algorithms

    Choose from dlib (global optimization), CMA-ES, Nevergrad, or Ax/BoTorch (Bayesian optimization) depending on your needs.

  • 8 Distance Metrics

    Compare simulations to targets using CA (area), CCH (Chamfer), CSF (sparse field), CCD (critical dimensions), and more.

  • Sensitivity Analysis

    Identify which parameters matter most using local (one-at-a-time) or global (Sobol indices) sensitivity analysis.

  • Multi-Domain Support

    Optimize process parameters across multiple geometries simultaneously to find universal parameter sets.

  • Custom Parameter Evaluation

    Explore parameter space with grid search, test specific combinations, or run repeatability tests to assess variance.

  • Incomplete Run Recovery

    Load and analyze optimization runs that were stopped early or didn't complete successfully.

  • Comprehensive Reporting

    Automatic generation of convergence plots, parameter evolution tracking, and detailed CSV/JSON outputs.

  • Production Ready

    Battle-tested in research and production environments with ViennaPS 4.0.0+ integration.

Quick Example

The core of the deposition example in examples/: fit a neutral-plus-ion deposition model to the profile of a filled trench.

import viennafit as fit
import viennaps as vps

# Load a project that holds the initial and the target domain
project = fit.Project()
project.load("./projects/exampleProject")

# Define the process sequence
def processSequence(domain: vps.Domain, params: dict[str, float]):
    # Run a ViennaPS process on the domain using params ...
    return domain.getLevelSets()[-1]

# Set up and run the optimization
opt = fit.Optimization(project)
opt.setProcessSequence(processSequence)
opt.setParameterNames(
    ["neutralStickP", "ionPowerCosine", "neutralRate", "ionRate", "ionEnergy"]
)
opt.setVariableParameters({
    "neutralStickP": (0.001, 0.9),
    "ionPowerCosine": (1.0, 900.0),
    "neutralRate": (1.0, 150.0),
    "ionRate": (0.1, 10.0),
})
opt.setFixedParameters({"ionEnergy": 100.0})
opt.setDistanceMetrics(primaryMetric="CCH")  # Chamfer distance
opt.setOptimizer("cma")
opt.setName("run1")
opt.apply(numEvaluations=200)

# Results are saved in the project folder

The complete, runnable scripts are in the examples/ folder of the repository. The Quick Start shows how to run them.

Getting Started

  • Quick Start

    Run the example optimization from the examples/ folder and look at its results.

    Quick Start

  • Step-by-Step Tutorials

    Walk through the example scripts step by step: optimization, custom evaluation and sensitivity analysis.

    Tutorials

  • Installation Guide

    Detailed installation instructions including prerequisites and troubleshooting.

    Installation

  • Core Concepts

    Understand the fundamental concepts of projects, domains, metrics, and optimization.

    Concepts

Typical Workflows

ViennaFit supports various workflows depending on your goals:

Initial Calibration

Start from scratch to calibrate your process model to experimental data:

  1. Annotate the profile before and after the process
  2. Build the initial and target domain from the annotations
  3. Create process sequence with parameters to optimize
  4. Run optimization with appropriate distance metric
  5. Validate results and refine if needed

Best for: First-time calibration, new process models

Parameter Exploration

After optimization, explore the parameter landscape:

  1. Load optimization results
  2. Set up parameter grid around optimal values
  3. Evaluate systematic combinations
  4. Visualize parameter relationships
  5. Identify robust parameter regions

Best for: Understanding parameter sensitivity, finding robust solutions

Sensitivity Analysis

Identify which parameters matter most:

  1. Define parameter ranges
  2. Run Sobol or local sensitivity analysis
  3. Interpret sensitivity indices
  4. Focus optimization on important parameters

Best for: Model reduction, experimental design, uncertainty quantification

Multi-Domain Optimization

Find universal parameters that work across different geometries:

  1. Add multiple initial and target domains
  2. Write multi-domain process sequence
  3. Run optimization (metrics automatically aggregated)
  4. Validate on each domain

Best for: Process window optimization, robust parameter sets

What's New in v2.0

ViennaFit 2.0 brings major improvements:

  • ViennaPS 4.0.0 Integration: Updated for latest ViennaPS API
  • Incomplete Run Support: Load and analyze runs that didn't complete
  • Repeatability Testing: Convenient API for variance analysis
  • Fixed Multi-Domain Detection: Single-domain is now the safe default
  • CSV Parameter Loading: Direct loading from progressBest.csv
  • Better Documentation: This site! Complete tutorials and guides

See the CHANGELOG for full details.

Example Use Cases

ViennaFit is used for:

  • Deposition Processes: Fit sticking coefficients and growth rates to experimental film profiles (the case used in the examples and tutorials)
  • Plasma Etching: Calibrate ion and neutral flux parameters to match trench profiles
  • Multi-Step Processes: Optimize complex sequences of etching, deposition, and planarization
  • Process Window Analysis: Find robust parameter sets that work across different geometries
  • Model Validation: Quantify how well process models reproduce experimental observations

Community and Support

Next Steps

Ready to get started? Here's what we recommend:

  1. Install ViennaFit - Set up your environment
  2. Quick Start - Run the example optimization
  3. Tutorial 1 - The optimization workflow, step by step
  4. Core Concepts - Understand the fundamentals

Happy optimizing! 🚀