EPiCarbon

EPiCarbon is a Python-based tool for estimating the carbon footprint of electro-photonic systems. It calculates the Embodied Carbon Footprint (ECF), Operational Carbon Footprint (OCF), and the Carbon Footprint (CF) of a system based on its architecture and energy usage.

Installation

1. Ensure you have Python 3.8 or higher installed. You can check your Python version with:

   python --version
   

2. Clone the repository:

   git clone https://github.com/bu-icsg/EPiCarbon.git
   cd EPiCarbon

3. Install the required dependencies:

   pip install --upgrade pip		# Optional
   pip install -r requirements.txt

File Structure

• epicarbon.py: Main script for running the program.
• src/: Contains chiplet models, data files, and utility functions.
• archs/: Contains example architecture description files in JSON format.

Usage

Run the program using the command line:

python epicarbon.py --estimate <metric> [--arch <arch_file>] [--energy <energy_per_inf>] [--verbose]

Arguments

• --estimate: Select the metric to estimate. Options:
  • ECF: Embodied Carbon Footprint.
  • OCF: Operational Carbon Footprint.
  • CF: Carbon Footprint.
• --arch: Path to the architecture description file in JSON (required for ECF and CF)
• --energy: Energy per inference in joules (required for OCF and CF).
• --verbose: Enable detailed logging.

Examples

1. Calculate ECF:

   python epicarbon.py --estimate ECF --arch archs/adept.json --verbose

   Alternatively, call the get_carbon_embodied() API by importing epicarbon from another script.

2. Calculate OCF:

   python epicarbon.py --estimate OCF --energy 1e-3 --verbose

   Alternatively, call the get_carbon_operational() API by importing epicarbon from another script.

3. Calculate CF:

   python epicarbon.py --estimate CF --arch archs/adept.json --energy 1e-3 --verbose

   Alternatively, call the get_carbon_footprint() API by importing epicarbon from another script.

Architecture Files

Architecture files describe the system's chiplets and packaging. Example files are located in the archs/ directory, such as:

• archs/lt.json
• archs/adept.json

Structure of an Architecture File

An architecture file contains the following fields:

• chiplets: A list of chiplets in the system. Each chiplet includes:
  • type: The type of the chiplet (Options: cmos-logic or pic-logic).
  • tech: The technology node of the chiplet in nanometers (Options: for pic-logic use -1, for cmos-logic choose among 28, 20, 14, 10, 8, 7, 5, or 3 nm).
  • area: The area of the chiplet in square centimeters.
  • num_chiplets: The number of identical chiplets of this type.
• package: The packaging type of the system (Options: monolithic, 3D, 2.5D-active, 2.5D-passive).

Example Architecture File

Below is an example architecture file for a system named "ADEPT":

{
    "chiplets": [
        {
            "type": "cmos-logic",
            "tech": 20,
            "area": 6.51,
            "num_chiplets": 1
        },
        {
            "type": "pic-logic",
            "tech": -1,
            "area": 0.56,
            "num_chiplets": 4
        }
    ],
    "package": "3D"
}

Configurable Parameters

Several default parameters are defined in the program that can be changed in code (epicarbon.py):

• Fabrication (ci_fab): 820 gCO2/kWh (coal-based electricity).
• Operation (ci_op): 11 gCO2/kWh (wind-based electricity).
• Number of inferences per day (num_inf_per_day): 1e9.
• Lifetime (lifetime_days): 5*365 days.

Extending EPiCarbon

EPiCarbon can be easily extended to support a custom chiplet.

• Simply add your new chiplet inside src/chiplet_models folder by inheriting the Chiplet class. You can follow the example of cmos_logic_chiplet or pic_logic_chiplet.
• Implement the get_manufacturing_carbon() method of your new chiplet.
• Update the build_chiplet() method in src/utils.py to handle your create your new chiplet while parsing the architecture file.

Publications

• Fayza F, Demirkiran C, Papa Rao S, Bunandar D, Gupta U, Joshi A. EPiCarbon: A Carbon Modeling Tool for Electro-Photonic Accelerators. ICCAD 2025 (https://bu-icsg.github.io/publications/2025/Fayza_ICCAD_2025.pdf)
• Fayza F, Demirkiran C, Papa Rao S, Bunandar D, Gupta U, Joshi A. Photonics for sustainable AI. Communications Physics. 2025 Oct 14;8(1):403. (https://www.nature.com/articles/s42005-025-02300-0)

Contact

For questions or feedback, please contact [[email protected]].