PFAS Usage-Burden Model (Lithography of GPU manufacturing in GenAI)

1. Project Overview

This project models PFAS use and environmental burden in a lithography-focused pathway linked to server/GPU scale growth.

The workflow is organized into three computational stages:

1. Server projection: estimates quarterly server-scale growth with Monte Carlo sampling. _{This module was partially inspired by the methodology described in https://doi.org/10.1038/s43588-024-00712-6 and related materials available at https://zenodo.org/records/13790035.}
2. Emission model: converts scale into lithography PFAS usage, then maps usage to emissions across media.
3. Burden model: applies treatment/destruction burden factors to emissions and calculates burden outputs.

The model results focus on 2025 to 2030, and outputs Excel files for downstream analysis and plotting.

2. Repository Structure

At the top level:

• Core code example/
  • Main and module code for the full pipeline.
• Data input/
  • Required input Excel files used by the model.
• Excel results/
  • All model output Excel files are written here.
• Data for figures/
  • Organized data prepared for manuscript plotting.
• Premise-IAM model/
  • Materials/scripts related to IAM-driven background database updating using premise.
• Economic model/
  • Additional analysis resources for economic components.

3. Core Code Layout

Inside Core code example/:

• PFAS usage-burden prediction.py
  • Main entry point. Orchestrates all modules and controls all Excel exports.
• server_projection_module.py
  • Server projection Monte Carlo simulation and scale statistics.
• emission_module.py
  • Lithography PFAS usage and media-specific emission calculations.
• burden_module.py
  • Treatment/destruction burden calculations by medium and by process/component.

4. Input Data

Inside Data input/:

• data.xls
  • Main input workbook for server projection, lithography PFAS settings, emission factors, and toxicity inputs.
• destruction summary.xls
  • Burden factor and uncertainty inputs for treatment/destruction pathways.

5. Output Files

All outputs are written to Excel results/:

• Result.xlsx
  • Core photolithography manufacturing summary.
• PFAS_Litho_2021_2030.xlsx
  • Lithography usage and emissions by medium, plus quarterly computing power summary.
• burden-type.xlsx
  • Burden results by treatment medium (gas/solid/solvent/water).
• burden-process.xlsx
  • Burden results by lithography component/process category.
• toxicity-PFOA-PFOS.xlsx
  • Toxicity summary outputs.

6. How to Run

Run from the project root directory:

python "Core code example/PFAS usage-burden prediction.py"

If your environment uses python3, run:

python3 "Core code example/PFAS usage-burden prediction.py"

On successful execution, the script prints:

done

7. Environment Requirements

Note that the LCA results generated using premise and ecoinvent have already been integrated and provided in Excel format. Therefore, running premise or IAM models is not required here;

Typical Python dependencies:

• numpy
• pandas
• xlrd

8. Model Logic Notes

• Current scope is lithography-focused PFAS usage and burden.
• The integrated pipeline follows:
  • server projection -> emission -> burden
• Existing formulas, uncertainty handling, and computational order are preserved in modular form.
• Main script controls output writing to keep I/O management centralized.

9. Data for Figures

Data for figures/ contains curated, post-processed data used for manuscript plotting.

This folder includes the datasets for Figure 2 through Figure 5, and together they cover the core quantitative results presented in the paper.

10. Premise-IAM Model

Premise-IAM model/ documents and supports how IAM scenario information is integrated through the premise tool to update the background LCI database.

Core concept:

• Information from different IAM scenarios is harmonized via premise.
• These scenario signals are used to update the ecoinvent database.
• Activity-level results are adjusted over time according to socioeconomic development and energy system transitions.

For premise details, see:

• https://github.com/polca/premise

11. Suggested Workflow

1. Update assumptions and parameters in Data input/*.xls.
2. Run the main script.
3. Review outputs in Excel results/.
4. Use Data for figures/ for final plotting and manuscript figures.