A comprehensive tutorial on end-to-end carbon modeling methodologies for the AI era — from IC supply chain uncertainty to software-level carbon attribution in shared cloud environments.
The rapid proliferation of AI and large-scale machine learning has placed computing at a sustainability crossroads. Computing now accounts for up to 4% of global carbon emissions — rivaling the aviation industry — with AI training and inference serving as primary drivers of this growth.
Carbon emissions have shifted from being dominated solely by operational energy to a combination of operational energy and hardware manufacturing. To prevent environmental impact from becoming the scaling bottleneck for next-generation systems, sustainability must be elevated to a first-order design target.
This tutorial introduces a suite of end-to-end modeling methodologies designed to make carbon accounting fast, accessible, and rigorous for the AI era. Building on ACT [ISCA 2022], this expanded framework addresses the specific complexities of the AI hardware-software stack.
Since ACT's introduction in 2022, a large community of computer architects and systems researchers has adopted these methodologies — now incorporated into industry frameworks such as Meta’s sustainability reporting. Our mission has shifted from providing a standalone tool to establishing a rigorous, multi-dimensional accounting standard.
A cross-stack methodology spanning recent publications at ISCA, ICCAD, and DATE.
Click any tool to view the paper. Hover to highlight.
The first fast, accessible tool for estimating embodied carbon during early hardware design space exploration. Enables carbon as a first-order design target alongside power and performance.
Uncertainty estimation in the global IC supply chain for AI chips — enabling more credible and rigorous carbon claims in research.
AI-specific hardware accounting including emerging non-volatile memories for in-memory computing architectures.
Full-system lifecycle analysis for GPUs and high-performance AI accelerators, including EcoServe for large-scale inference systems.
Software-level carbon attribution for fair accounting in shared AI cloud environments, drawing on ideas from economics and game theory.
Carbon modeling for 3D integrated circuits, capturing the unique manufacturing and assembly footprint of chiplet-based designs.
Device-level carbon emission modeling for advanced semiconductor processes, enabling fab-aware sustainability analysis.
Half-day tutorial · 9:00 AM – 1:00 PM · New sessions are new additions from the previous iteration
| Time | Topic | Presenter |
|---|---|---|
| 9:00–9:45 |
Sustainable Computing: Motivation
Overview of carbon modeling and accounting strategies
|
Udit Gupta |
| 9:45–10:15 |
Hands-on: ACT
How to use and extend ACT
|
Leo Han |
| 10:15–10:45 |
Hands-on: Modeling Uncertainty in Carbon Accounting
New
CarbonClarity — IC supply chain uncertainty estimation
|
Xuesi Chen |
| 10:45–11:15 |
Hands-on: Full System Life Cycle Analyses
New
GPUs · Mobile and embedded systems
|
Xuesi Chen |
| 11:15–11:45 |
Hands-on: Software-Level Carbon Attribution
FAIR-CO₂ — fair accounting in shared cloud environments
|
Leo Han |
| 11:45–12:00 |
Emerging Carbon Accounting Extensions
|
Udit Gupta |
| 12:00–1:00 |
Invited Talks
~6 talks × ~15 minutes each
|
See below |
We are assembling an exciting lineup of researchers and practitioners spanning the full computing stack — from device physics to AI systems — across academia and industry.
Cornell Tech
PhD Student
Research interests at the intersection of sustainable computing and computing for sustainability. Recent work focuses on fair carbon attribution in cloud ecosystems using ideas from economics and game theory. Recipient of the Digital Life Initiative Fellowship at Cornell Tech (2024).
Cornell Tech
PhD Student
Research in computer architecture, energy-efficient computing, AI at the edge, and sustainable computing. Develops mathematical models to quantify uncertainty in carbon accounting across the IC supply chain and designs energy-efficient and sustainable mobile devices.
Cornell Tech
Assistant Professor, ECE
Research at the intersection of computer architecture, systems, ML, and environmental sustainability. Recognized as IEEE MICRO Top Picks (2022, 2023), SIGARCH Outstanding Ph.D. Dissertation Honorable Mention, and SIGMICRO Outstanding Ph.D. Dissertation Honorable Mention.
A pre-configured environment (Docker images / cloud-hosted JupyterHub) will be provided so attendees can begin modeling immediately without local environment setup.
Enabling sustainable computing requires solutions across the entire computing stack. An important mission for this workshop is to foster a diverse, collaborative, and innovative environment. Our speakers and organizers represent diversity in technical expertise (application developers to data center designers), organizations (universities and industry), professional experience, ethnic backgrounds, and genders.
If you use ACT in your research, please cite:
@inproceedings{gupta2022act,
title={ACT: designing sustainable computer systems
with an architectural carbon modeling tool},
author={Gupta, Udit and Elgamal, Mariam and Hills, Gage
and Wei, Gu-Yeon and Lee, Hsien-Hsin S
and Brooks, David and Wu, Carole-Jean},
booktitle={Proceedings of the 49th Annual International
Symposium on Computer Architecture},
pages={784--799},
year={2022}
}