pith. machine review for the scientific record. sign in

arxiv: 2605.05615 · v2 · submitted 2026-05-07 · 💻 cs.LG · cs.CY

Recognition: no theorem link

LLMSpace: Carbon Footprint Modeling for Large Language Model Inference on LEO Satellites

Adrian Ildefonso, Daniel Loveless, Fan Chen, Lei Jiang

Authors on Pith no claims yet

Pith reviewed 2026-05-11 00:52 UTC · model grok-4.3

classification 💻 cs.LG cs.CY
keywords carbon footprintlarge language modelsLEO satellitesAI inferenceembodied carbonradiation-hardened hardwaresustainable computingspace-based AI
0
0 comments X

The pith

LLMSpace models the full lifecycle carbon emissions of large language model inference on solar-powered LEO satellites and identifies trade-offs with latency, hardware, and lifetime.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents LLMSpace as the first framework to calculate carbon emissions for running large language models on satellites in low Earth orbit. It combines the emissions from satellite manufacturing and launch with the energy used during inference operations, while including the effects of radiation-hardened components and the distinct power demands of prompt processing versus token generation. A sympathetic reader would care because the rapid growth in AI energy use on Earth has prompted interest in space-based alternatives that could draw on solar power, yet the complete environmental cost of such systems has remained unclear. The framework uses realistic satellite and GPU setups to map how different design choices affect overall carbon output.

Core claim

LLMSpace jointly models operational and embodied carbon, peripheral subsystems, radiation-hardened accelerators and memories, and LLM-specific workload characteristics such as prefill-decode behavior and token generation. Using realistic satellite and GPU configurations, LLMSpace reveals key trade-offs among carbon footprint, inference latency, hardware design, and operational lifetime for sustainable space-based LLM inference.

What carries the argument

The LLMSpace framework, which integrates models of operational carbon from LLM prefill-decode workloads with embodied carbon from satellite manufacturing, launch, and radiation-hardened hardware.

If this is right

  • Satellite hardware can be sized and configured to reduce total carbon while keeping inference latency within acceptable bounds.
  • Extending satellite operational lifetime spreads embodied carbon costs over greater volumes of inference work.
  • Separate accounting for prefill and decode phases allows targeted power optimizations that lower overall emissions.
  • Inclusion of peripheral systems such as power management and communications shows their contribution to the carbon budget.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the trade-offs favor satellites, space-based inference could complement terrestrial data centers in a decarbonization strategy for AI.
  • Real deployment data from satellites would allow calibration of the embodied carbon estimates that currently rely on modeled values.
  • Extending the framework to orbital variations and multi-satellite networks could refine guidance for large-scale space AI systems.

Load-bearing premise

Realistic satellite and GPU configurations plus models of radiation-hardened hardware produce sufficiently accurate trade-off insights even without direct validation measurements.

What would settle it

Deploying a radiation-hardened GPU on an actual LEO satellite, running representative LLM inference workloads, and comparing measured power draw and derived carbon emissions against LLMSpace predictions would test the modeled trade-offs.

Figures

Figures reproduced from arXiv: 2605.05615 by Adrian Ildefonso, Daniel Loveless, Fan Chen, Lei Jiang.

Figure 1
Figure 1. Figure 1: AI-enabled LEO satellite platforms for large-scale LLM inference. view at source ↗
Figure 2
Figure 2. Figure 2: Lifecycle carbon footprint comparison between orbital and terrestrial data centers. view at source ↗
Figure 2
Figure 2. Figure 2: The carbon footprint comparison between orbital and terrestrial data centers. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The operational energy consumption of LLM inferences on various benchmarks. view at source ↗
Figure 3
Figure 3. Figure 3: The operational energy consumption of LLM inferences on various benchmarks. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The latency and opera￾tional energy comparison between DGX-H100 and -A100 nodes (nor￾malized to DGX-H100). Latency–Carbon Tradeoff. As shown in Figure 2b, we build a configuration denoted as A100 by replacing the radiation￾hardened DGX-H100 node in the rad-hard configuration with a radiation-hardened DGX-A100 node. Compared with rad-hard, A100 reduces the total embodied carbon footprint by ∼ 30%. This redu… view at source ↗
Figure 4
Figure 4. Figure 4: The latency and operational energy comparison between DGX￾H100 and -A100 nodes (normalized to DGX-H100). The distributions of input prompt lengths and generated to￾ken counts are shown in Figure 3a and Figure 3b. Error bars denote min–max across requests. We run inference using the codellama/CodeLlama-34b-Instruct-hf model with bfloat16 precision on a single GPU of an NVIDIA DGX H100 node (batch size = 1).… view at source ↗
read the original abstract

Large language models (LLMs) impose rapidly growing energy demands, creating an emerging energy and carbon crisis driven by large-scale inference. Solar-powered, AI-enabled low Earth orbit (LEO) satellites have been proposed to mitigate terrestrial electricity consumption, but their lifecycle carbon footprint remains poorly understood due to launch emissions, satellite manufacturing, and radiation-hardened hardware requirements. This paper presents \textit{LLMSpace}, the first carbon modeling framework for LLM inference on AI-enabled LEO satellites. LLMSpace jointly models operational and embodied carbon, peripheral subsystems, radiation-hardened accelerators and memories, and LLM-specific workload characteristics such as prefill-decode behavior and token generation. Using realistic satellite and GPU configurations, LLMSpace reveals key trade-offs among carbon footprint, inference latency, hardware design, and operational lifetime for sustainable space-based LLM inference. Source code: https://github.com/UnchartedRLab/LLMSpace.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper introduces LLMSpace, the first carbon modeling framework for LLM inference on AI-enabled LEO satellites. It jointly models operational and embodied carbon (including launch emissions, satellite manufacturing, and radiation-hardened accelerators/memories), peripheral subsystems, and LLM-specific workload traits such as prefill-decode behavior and token generation. Using realistic satellite and GPU configurations, the framework is used to reveal key trade-offs among carbon footprint, inference latency, hardware design, and operational lifetime for sustainable space-based LLM inference. Source code is provided.

Significance. If the underlying models prove accurate, LLMSpace could offer actionable quantitative guidance for hardware and operational choices in orbital AI systems, helping address the energy demands of terrestrial LLM inference. The open-source release supports reproducibility and extension, which strengthens the contribution for the green computing and space systems communities.

major comments (3)
  1. Abstract and framework overview: no equations, data sources, validation results, or error analysis are supplied, so it is impossible to determine whether the modeled carbon-latency-hardware-lifetime trade-offs are supported by the calculations rather than by untested parameter choices.
  2. Embodied-carbon and launch-emission components: the models for radiation-hardened accelerators/memories and launch emissions receive no comparison to measured space-grade hardware data or existing literature benchmarks, leaving the quantitative optima vulnerable to factor-of-two errors common in early-stage estimates.
  3. Results and sensitivity: no Monte-Carlo uncertainty propagation, ablation on the radiation-hardening overhead multiplier, or sensitivity analysis on SEU rates and power models is shown; if any of these inputs shift, the reported design recommendations and trade-off rankings could invert.
minor comments (1)
  1. Notation for prefill versus decode phases and peripheral power terms should be defined explicitly in the first use to aid readers unfamiliar with LLM serving workloads.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below with clarifications on the existing content and additional analyses incorporated in the revision. The responses focus on strengthening transparency without altering the core contributions of the LLMSpace framework.

read point-by-point responses

  1. Referee: [—] Abstract and framework overview: no equations, data sources, validation results, or error analysis are supplied, so it is impossible to determine whether the modeled carbon-latency-hardware-lifetime trade-offs are supported by the calculations rather than by untested parameter choices.

    Authors: The abstract follows standard length and style constraints for high-level overview. The framework overview and detailed models in Sections 2 and 3 provide the governing equations for operational carbon (including prefill-decode power), embodied carbon (manufacturing, launch, and radiation-hardening overheads), data sources drawn from cited satellite power models, GPU datasheets, and ESA/NASA reports, plus validation against terrestrial LLM inference measurements. We have revised the manuscript to expand the framework overview with explicit equation references and added a dedicated validation and error analysis subsection reporting bounds on key parameters. revision: yes

  2. Referee: [—] Embodied-carbon and launch-emission components: the models for radiation-hardened accelerators/memories and launch emissions receive no comparison to measured space-grade hardware data or existing literature benchmarks, leaving the quantitative optima vulnerable to factor-of-two errors common in early-stage estimates.

    Authors: Publicly available measured carbon data for radiation-hardened hardware is limited due to proprietary manufacturer information. Our models are grounded in scaling factors and emission coefficients from peer-reviewed literature on rad-hard components and launch vehicles. The revision adds explicit comparisons to available space-systems benchmarks (e.g., from prior satellite lifecycle studies) and quantifies potential error ranges, while noting the inherent uncertainties in early-stage estimates. revision: yes

  3. Referee: [—] Results and sensitivity: no Monte-Carlo uncertainty propagation, ablation on the radiation-hardening overhead multiplier, or sensitivity analysis on SEU rates and power models is shown; if any of these inputs shift, the reported design recommendations and trade-off rankings could invert.

    Authors: We agree that formal sensitivity analysis improves robustness. The revised results section now incorporates Monte-Carlo uncertainty propagation over 1000 iterations for power and SEU parameters, an ablation study varying the radiation-hardening overhead multiplier, and sensitivity plots for SEU rates and peripheral power models. These additions confirm that the primary trade-off rankings and qualitative recommendations remain stable within the explored ranges, although absolute numerical values carry quantified uncertainty bands. revision: yes

Circularity Check

0 steps flagged

Forward modeling framework with no self-referential derivations or fitted predictions

full rationale

The paper presents LLMSpace as a joint modeling framework for operational and embodied carbon, incorporating satellite configurations, radiation-hardened hardware parameters, and LLM workload characteristics (prefill/decode). No equations or steps are described that define a quantity in terms of itself, rename a fitted parameter as a prediction, or rely on self-citations for load-bearing uniqueness claims. The outputs are simulation results from input parameterizations rather than derivations that reduce to the inputs by construction. This is a standard engineering modeling exercise whose validity depends on external parameter accuracy, not internal circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The framework implicitly relies on external data for launch emissions, radiation-hardening costs, and satellite power budgets, but these are not enumerated.

pith-pipeline@v0.9.0 · 5464 in / 1206 out tokens · 76510 ms · 2026-05-11T00:52:39.629308+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 7 canonical work pages · 1 internal anchor

  1. [1]

    How hungry is ai? benchmarking energy, water, and carbon footprint of llm inference,

    Nidhal Jegham, Marwan Abdelatti, Chan Young Koh, Lassad Elmoubarki, and Abdeltawab Hendawi. How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference.arXiv preprint arXiv:2505.09598, 2025

    work page arXiv 2025

  2. [2]

    Trends in ai supercomput- ers.arXiv preprint arXiv:2504.16026, 2025

    Konstantin F Pilz, James Sanders, Robi Rahman, and Lennart Heim. Trends in AI Supercom- puters.arXiv preprint arXiv:2504.16026, 2025

    work page arXiv 2025

  3. [3]

    New azure space products enable digital resiliency and empower the industry

    Microsoft. New azure space products enable digital resiliency and empower the industry. https://azure.microsoft.com/en-us/blog/new-azure-space-products-enabl e-digital-resiliency-and-empower-the-industry/, 2025

    2025

  4. [4]

    How Starcloud Is Bringing Data Centers to Outer Space

    Angie Lee. How Starcloud Is Bringing Data Centers to Outer Space. https://blogs.nvid ia.com/blog/starcloud/, 2025. NVIDIA Blog

    2025

  5. [5]

    Exploring a Space-Based, Scalable AI Infrastructure System Design: Project Suncatcher

    Travis Beals. Exploring a Space-Based, Scalable AI Infrastructure System Design: Project Suncatcher. https://research.google/blog/exploring-a-space-based-scalabl e-ai-infrastructure-system-design/, 2025. Google Research Blog. 9

    2025

  6. [6]

    Amazon Leo: Low Earth Orbit (LEO) Satellite Network

    Amazon. Amazon Leo: Low Earth Orbit (LEO) Satellite Network. https://www.aboutama zon.com/what-we-do/devices-services/project-kuiper, 2025

    2025

  7. [7]

    A case for application-aware space radiation tolerance in orbital computing.arXiv preprint arXiv:2407.11853, 2024

    Meiqi Wang, Han Qiu, Longnv Xu, Di Wang, Yuanjie Li, Tianwei Zhang, Jun Liu, and Hewu Li. A Case for Application-Aware Space Radiation Tolerance in Orbital Computing.arXiv preprint arXiv:2407.11853, 2024

    work page arXiv 2024

  8. [8]

    Space Microdatacenters

    Nathaniel Bleier, Muhammad Husnain Mubarik, Gary R Swenson, and Rakesh Kumar. Space Microdatacenters. InIEEE/ACM International Symposium on Microarchitecture, pages 900– 915, 2023

    2023

  9. [9]

    The Development of Carbon- Neutral Data Centres in Space.Nature Electronics, pages 1–11, 2025

    Ablimit Aili, Jihwan Choi, Yew Soon Ong, and Yonggang Wen. The Development of Carbon- Neutral Data Centres in Space.Nature Electronics, pages 1–11, 2025

    2025

  10. [10]

    Total Ionizing Dose Radi- ation Testing of NVIDIA Jetson Nano GPUs

    Windy S Slater, Nayana P Tiwari, Tyler M Lovelly, and Jesse K Mee. Total Ionizing Dose Radi- ation Testing of NVIDIA Jetson Nano GPUs. InIEEE High Performance Extreme Computing Conference, pages 1–3, 2020

    2020

  11. [11]

    Single-Event Effects on Commercial-Off-The-Shelf Edge-Processing Artificial Intelligence ASICs.IEEE Transactions on Nuclear Science, 70(8):1716–1723, 2023

    Megan C Casey, Justin S Goodwill, Edward J Wyrwas, Rebekah A Austin, Christopher M Wilson, Scott D Stansberry, Nicolas Gorius, and Shahid Aslam. Single-Event Effects on Commercial-Off-The-Shelf Edge-Processing Artificial Intelligence ASICs.IEEE Transactions on Nuclear Science, 70(8):1716–1723, 2023

    2023

  12. [12]

    Space Science with CubeSats and Nanosatellites.Nature Astronomy, 4(11):1026–1030, 2020

    J Douglas Liddle, Antony P Holt, Susan J Jason, Kathryn A O’Donnell, and Edward J Stevens. Space Science with CubeSats and Nanosatellites.Nature Astronomy, 4(11):1026–1030, 2020

    2020

  13. [13]

    Laser Intersatellite Links in a Starlink Con- stellation: A Classification and Analysis.IEEE vehicular technology magazine, 16(2):48–56, 2021

    Aizaz U Chaudhry and Halim Yanikomeroglu. Laser Intersatellite Links in a Starlink Con- stellation: A Classification and Analysis.IEEE vehicular technology magazine, 16(2):48–56, 2021

    2021

  14. [14]

    Dirty Bits in Low-Earth Orbit (LEO): The Carbon Footprint of Launching Computers.ACM SIGENERGY Energy Informatics Review, 5(2):26–33, 2025

    Robin Ohs, Gregory F Stock, Andreas Schmidt, Juan A Fraire, and Holger Hermanns. Dirty Bits in Low-Earth Orbit (LEO): The Carbon Footprint of Launching Computers.ACM SIGENERGY Energy Informatics Review, 5(2):26–33, 2025

    2025

  15. [15]

    LLMCarbon: Modeling the End-to-End Carbon Footprint of Large Language Models

    Ahmad Faiz, Sotaro Kaneda, Ruhan Wang, Rita Chukwunyere Osi, Prateek Sharma, Fan Chen, and Lei Jiang. LLMCarbon: Modeling the End-to-End Carbon Footprint of Large Language Models. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  16. [16]

    LLMCO2: Advancing accurate carbon footprint prediction for LLM inferences.ACM SIGENERGY Energy Informatics Review, 5(2):63–68, 2025

    Zhenxiao Fu, Fan Chen, Shan Zhou, Haitong Li, and Lei Jiang. LLMCO2: Advancing accurate carbon footprint prediction for LLM inferences.ACM SIGENERGY Energy Informatics Review, 5(2):63–68, 2025

    2025

  17. [17]

    Srour and J.W

    J.R. Srour and J.W. Palko. Displacement Damage Effects in Irradiated Semiconductor Devices. IEEE Transactions on Nuclear Science, 60(3):1740–1766, 2013

    2013

  18. [18]

    Modeling and Simulation of Single-Event Effect in CMOS Circuit.Journal of Semiconductors, 36(11):111002, 2015

    Suge Yue, Xiaolin Zhang, Yuanfu Zhao, Lin Liu, and Hanning Wang. Modeling and Simulation of Single-Event Effect in CMOS Circuit.Journal of Semiconductors, 36(11):111002, 2015

    2015

  19. [19]

    Investigation into the Degradation of DDR4 DRAM Owing to Total Ionizing Dose Effects

    Gyeongyeop Lee, Minki Suh, Minsang Ryu, Yunjong Lee, Jin-Woo Han, and Jungsik Kim. Investigation into the Degradation of DDR4 DRAM Owing to Total Ionizing Dose Effects. IEEE Access, 11:97456–97465, 2023

    2023

  20. [20]

    Total Ionizing Dose Effects on Deca-Nanometer Fully Depleted SOI Devices

    P Paillet, M Gaillardin, V Ferlet-Cavrois, A Torres, O Faynot, C Jahan, L Tosti, and S Cristoloveanu. Total Ionizing Dose Effects on Deca-Nanometer Fully Depleted SOI Devices. IEEE transactions on nuclear science, 52(6):2345–2352, 2006

    2006

  21. [21]

    Evaluation of error correcting codes for radiation-tolerant memory.The Interplanetary Network Progress Report, 42:181, 2010

    Seungjune Jeon, BVK Vijaya Kumar, Euiseok Hwang, and Michael K Cheng. Evaluation of error correcting codes for radiation-tolerant memory.The Interplanetary Network Progress Report, 42:181, 2010

    2010

  22. [22]

    Radiation Effects in Advanced and Emerging Nonvolatile Memories

    Matthew J Marinella. Radiation Effects in Advanced and Emerging Nonvolatile Memories. IEEE Transactions on Nuclear Science, 68(5):546–572, 2021. 10

    2021

  23. [23]

    A systematic characterization of LLM inference on GPUs,

    Haonan Wang, Xuxin Xiao, Mingyu Yan, Zhuoyuan Zhu, Dengke Han, Duo Wang, Wenming Li, Xiaochun Ye, Cunchen Hu, Hongyang Chen, et al. A Systematic Characterization of LLM Inference on GPUs.arXiv preprint arXiv:2512.01644, 2025

    work page arXiv 2025

  24. [24]

    Radiation Effects in VLSI Circuits–Part I: Historical Perspective.IETE Technical Review, 41(6):716–735, 2024

    Aryan Kannaujiya and Ambika Prasad Shah. Radiation Effects in VLSI Circuits–Part I: Historical Perspective.IETE Technical Review, 41(6):716–735, 2024

    2024

  25. [25]

    Srour, C.J

    J.R. Srour, C.J. Marshall, and P.W. Marshall. Review of Displacement Damage Effects in Silicon Devices.IEEE Transactions on Nuclear Science, 50(3):653–670, 2003

    2003

  26. [26]

    Radiation Failures in Intel 14nm Microprocessors

    Dobrin P Bossev, Adam R Duncan, Matthew J Gadlage, Austin H Roach, Matthew J Kay, Carl Szabo, Tammy J Berger, Darin A York, Aaron Williams, K LaBel, et al. Radiation Failures in Intel 14nm Microprocessors. InMilitary and Aerospace Programmable Logic Devices Workshop, 2016

    2016

  27. [27]

    Global 3D Rocket Launch and Re-entry Air Pollutant and CO2 Emissions at the Onset of the Megaconstellation Era.Scientific data, 11(1):1079, 2024

    Connor R Barker, Eloise A Marais, and Jonathan C McDowell. Global 3D Rocket Launch and Re-entry Air Pollutant and CO2 Emissions at the Onset of the Megaconstellation Era.Scientific data, 11(1):1079, 2024

    2024

  28. [28]

    Position, Navigation, and Timing (PNT) through Low Earth Orbit (LEO) Satellites: A Survey on Current Status, Challenges, and Opportunities.IEEE Access, 10:83971–84002, 2022

    Fabricio S Prol, R Morales Ferre, Zainab Saleem, Petri Välisuo, Christina Pinell, Elena Simona Lohan, Mahmoud Elsanhoury, Mohammed Elmusrati, Saiful Islam, and Kaan Çelikbilek. Position, Navigation, and Timing (PNT) through Low Earth Orbit (LEO) Satellites: A Survey on Current Status, Challenges, and Opportunities.IEEE Access, 10:83971–84002, 2022

    2022

  29. [29]

    A Cost Roadmap for Silicon Heterojunction Solar Cells.Solar Energy Materials and Solar Cells, 147:295–314, 2016

    Atse Louwen, Wilfried Van Sark, Ruud Schropp, and André Faaij. A Cost Roadmap for Silicon Heterojunction Solar Cells.Solar Energy Materials and Solar Cells, 147:295–314, 2016

    2016

  30. [30]

    Integrating Sustainability in the Design of Space Activities: Development of Eco-Design Tools for Space Projects

    Augustin Chanoine, Y Le Guern, F Witte, J Huesing, T Soares, and L Innocenti. Integrating Sustainability in the Design of Space Activities: Development of Eco-Design Tools for Space Projects. In5th CEAS Air & Space Conference, pages 1–11. Delft University of Technology, 2015

    2015

  31. [31]

    A Review of Battery Technology in Cubesats and Small Satellite Solutions.Energies, 13(16):4097, 2020

    Vaclav Knap, Lars Kjeldgaard Vestergaard, and Daniel-Ioan Stroe. A Review of Battery Technology in Cubesats and Small Satellite Solutions.Energies, 13(16):4097, 2020

    2020

  32. [32]

    Advancing Energy Storage: The Future Trajectory of Lithium-Ion Battery Technologies.Journal of Energy Storage, 120:116511, 2025

    M.M Hasan, Rezwanul Haque, M.I Jahirul, Mohammad G Rasul, I.M.R Fattah, N.M.S Hassan, and M Mofijur. Advancing Energy Storage: The Future Trajectory of Lithium-Ion Battery Technologies.Journal of Energy Storage, 120:116511, 2025

    2025

  33. [33]

    Gilmore.Spacecraft Thermal Control Handbook, Volume I: Fundamental Technologies

    David G. Gilmore.Spacecraft Thermal Control Handbook, Volume I: Fundamental Technologies. The Aerospace Press, El Segundo, CA, 2002

    2002

  34. [34]

    Cube satellites – Small in Size with Big Flash Memory Constraints

    Tuxera. Cube satellites – Small in Size with Big Flash Memory Constraints. https://www.tu xera.com/blog/cubesats-flash-memory-storage/, 2023

    2023

  35. [35]

    Fault Tolerance in Space with Heterogeneous Hardware: Experiences from a 68-day CubeSat Deployment in LEO

    Ahmed El Yaacoub, Thiemo V oigt, Philipp Ruemmer, and Luca Mottola. Fault Tolerance in Space with Heterogeneous Hardware: Experiences from a 68-day CubeSat Deployment in LEO. InIEEE International Conference on Embedded Wireless Systems and Networks, 2025

    2025

  36. [36]

    Samuel K. Moore. U.S. Invests in Fabs That Make Radiation-Hardened Chips.IEEE Spectrum, October 2019

    2019

  37. [37]

    Scotten W. Jones. Modeling 300mm Wafer Fab Carbon Emissions. InInternational Electron Devices Meeting, pages 1–4, 2023

    2023

  38. [38]

    RAD750.https://en.wikipedia.org/wiki/RAD750, 2024

    Wikipedia. RAD750.https://en.wikipedia.org/wiki/RAD750, 2024

    2024

  39. [39]

    Hardening Techniques for MRAM-based Non-volatile Storage Cells and Logic

    Y Lakys, WS Zhao, JO Klein, and C Chappert. Hardening Techniques for MRAM-based Non-volatile Storage Cells and Logic. InIEEE European Conference on Radiation and Its Effects on Components and Systems, pages 669–674, 2011

    2011

  40. [40]

    Modeling STT-RAM Fabrication Cost and Impacts in NVSim

    Ismail Bayram, Enes Eken, Donald Kline, Nikolas Parshook, Yiran Chen, and Alex K Jones. Modeling STT-RAM Fabrication Cost and Impacts in NVSim. InIEEE Seventh International Green and Sustainable Computing Conference, pages 1–8, 2016. 11

    2016

  41. [41]

    RH3480: 480GB 3U VPX Rad-Tolerant SSDs

    Mercury Systems. RH3480: 480GB 3U VPX Rad-Tolerant SSDs. https://www.mrcy.com /products/data-management/rad-tolerant-storage-memory/model-rh3480

  42. [42]

    Swamit Tannu and Prashant J. Nair. The Dirty Secret of SSDs: Embodied Carbon.ACM SIGENERGY Energy Informatics Review, 3:4–9, October 2023

    2023

  43. [43]

    Product Carbon Footprint (PCF) Summary for NVIDIA HGX H100

    NVIDIA Corporation. Product Carbon Footprint (PCF) Summary for NVIDIA HGX H100. https://images.nvidia.com/aem-dam/Solutions/documents/HGX-H100-PCF-Sum mary.pdf, July 2025

    2025

  44. [44]

    Expanding CubeSat Capabilities with A Low Cost Transceiver

    Scott Palo, Darren O’Connor, Elizabeth DeVito, Rick Kohnert, Scott H Schaire, Steve Bundick, Gary Crum, Serhat Altunc, and Thomas Winkert. Expanding CubeSat Capabilities with A Low Cost Transceiver. InAnnual AIAA/USU Conference on Small Satellites, 2014

    2014

  45. [45]

    Status on Laser Communication Activities at Tesat-Spacecom

    Mark Gregory, Frank Heine, Andrej Brzoska, Klaus Oestreich, Robert Mahn, Patricia Martin Pimentel, and Herwig Zech. Status on Laser Communication Activities at Tesat-Spacecom. In Free-Space Laser Communications XXXVI, volume 12877, page 1287704. SPIE, 2024

    2024

  46. [46]

    Satellite networks: Past present and future challenges-part 2: Broadband systems.IEEE Aerospace and Electronic Systems Magazine, 2025

    Riccardo De Gaudenzi. Satellite networks: Past present and future challenges-part 2: Broadband systems.IEEE Aerospace and Electronic Systems Magazine, 2025

    2025

  47. [47]

    Fifteen Years of ISO 14040: A Review.Journal of Cleaner Production, 57:115–123, 2013

    Jonathan Pryshlakivsky and Cory Searcy. Fifteen Years of ISO 14040: A Review.Journal of Cleaner Production, 57:115–123, 2013

    2013

  48. [48]

    Quantifying the energy consumption and carbon emissions of LLM inference via simulations,

    Miray Özcan, Philipp Wiesner, Philipp Weiß, and Odej Kao. Quantifying the Energy Consump- tion and Carbon Emissions of LLM Inference via Simulations.arXiv preprint arXiv:2507.11417, 2025

    work page arXiv 2025

  49. [49]

    Falcon 9.https://en.wikipedia.org/wiki/Falcon_9, 2026

    Wikipedia. Falcon 9.https://en.wikipedia.org/wiki/Falcon_9, 2026

    2026

  50. [50]

    Empirically-Calibrated H100 Node Power Models for Accurate AI Training Energy Estimation.Environmental Research: Energy, 2(4):045016, 2025

    Alex C Newkirk, Jared Fernandez, Jonathan Koomey, Imran Latif, Emma Strubell, Arman Shehabi, and Constantine Samaras. Empirically-Calibrated H100 Node Power Models for Accurate AI Training Energy Estimation.Environmental Research: Energy, 2(4):045016, 2025

    2025

  51. [51]

    CO2-Meter: A Comprehensive Carbon Footprint Estimator for LLMs on Edge Devices.Proceedings of the AAAI Conference on Artificial Intelligence, 2026

    Zhenxiao Fu, Fan Chen, and Lei Jiang. CO2-Meter: A Comprehensive Carbon Footprint Estimator for LLMs on Edge Devices.Proceedings of the AAAI Conference on Artificial Intelligence, 2026

    2026

  52. [52]

    Holistic Evaluation of Language Models

    Percy Liang, Rishi Bommasani, Tony Lee, Dimitris Tsipras, Dilara Soylu, Michihiro Yasunaga, Yian Zhang, Deepak Narayanan, Yuhuai Wu, Ananya Kumar, et al. Holistic Evaluation of Language Models.arXiv preprint arXiv:2211.09110, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  53. [53]

    Slo-aware gpu frequency scaling for energy efficient llm inference serving

    Andreas Kosmas Kakolyris, Dimosthenis Masouros, Petros Vavaroutsos, Sotirios Xydis, and Dimitrios Soudris. SLO-aware GPU Frequency Scaling for Energy Efficient LLM Inference Serving.arXiv preprint arXiv:2408.05235, 2024. 12

    work page arXiv 2024