Pith. sign in

REVIEW 4 major objections 5 minor 16 references

Carbon Per Transistor (CPT): The Golden Formula for Green Computing Metrics

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read The paper proposes a Carbon Per Transistor formula and claims chip fabrication, not daily use, dominates a processor's CO2 footprint.

desk verdict A missing unit conversion reverses the paper's central claim, and the paper contradicts its own per-transistor numbers; the per-transistor framing is the only new idea. read the letter →

arxiv 2502.02606 v1 pith:BLGVG7TK submitted 2025-02-01 cs.OH cond-mat.mtrl-sci

classification cs.OHcond-mat.mtrl-sci
keywords CarbonPerTransistorCO2emissionssemiconductorfabricationgreencomputingprocessorbenchmarkinglifecycleassessmentVLSIenergy-efficientfab
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that every transistor in a processor can be assigned a carbon cost, and it proposes a formula that adds a manufacturing term (wafer-level CO2 divided by transistor count) to an operational term (per-transistor power times lifetime hours times grid emission factor). Applying this Carbon Per Transistor (CPT) metric to Intel, AMD, and Apple chips, the authors claim that manufacturing emissions dominate a CPU's total CO2 impact, with fabrication contributing 50–66 kg per high-end processor while operational emissions over a five-year lifespan come to less than 1.5 kg. They conclude that dense, energy-efficient chips like Apple's M-series actually carry a larger total carbon footprint than power-hungry x86 processors, because their high transistor counts amplify fabrication emissions. The paper presents CPT as a universal benchmark that could inform chip design, purchasing decisions, and regulation.

What carries the argument

The CPT equation is the central object: Ctrans = C_wafer/(yield × N_trans/wafer) + (P_total/N_trans) × H_lifetime × EF. The first term spreads the CO2 of fabricating a 300mm wafer across the transistors it contains, adjusted for yield; the second term spreads the processor's total power across its transistors and multiplies by lifetime hours and the grid's emission factor. This decomposition is what lets the paper compare chips on a per-transistor basis and attribute the total footprint to manufacturing versus operation.

What would settle it

For the Intel Core i9-13900K, compute Coper = 125–253 W × 14,600 h × 0.4 kg/kWh with proper unit conversion (divide watt-hours by 1000 first). If the result is about 730–1,477 kg of CO2 rather than the 0.73–1.48 kg listed in Table 4, the paper's manufacturing-dominates claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the lifetime CO2 of a microprocessor can be decomposed per transistor and that, for current high-end chips, the fabrication part outweighs the use phase. The authors derive Ctrans = C_wafer/(yield × N_trans/wafer) + P_trans × H_lifetime × EF, and apply it to the Intel Core i9-13900K, AMD Ryzen 9 7950X, and Apple M1/M2/M3, finding per-transistor manufacturing costs of 2–5 µg and per-transistor operational costs of 60–250 µg, yet processor-level manufacturing totals of 50–66 kg versus operational totals below 1.5 kg. On this basis the paper asserts that semiconductor fabrication is the dominant factor in the carbon footprint of computing hardware, and that Apple's high-transistor-count M-series chips exceed Intel and AMD parts in total CO2 despite their energy efficiency.

Load-bearing premise

The paper's conclusion that manufacturing dominates collapses if the operational carbon calculation in Equation (8) is off by a factor of 1000 when applied to whole processors, because then operational emissions would outweigh manufacturing.

Editorial extensions

If this is right

  • If the CPT metric is adopted, processors could be ranked by a single carbon number, allowing manufacturers and buyers to compare chips on sustainability alongside speed.
  • If manufacturing truly dominates, then the largest reductions in computing's carbon footprint would come from cleaning up wafer fabrication and choosing lower-emission process nodes, not just improving energy efficiency in use.
  • Under the paper's ranking, Apple's M-series chips would carry a carbon premium over x86 processors despite their low power draw, changing what 'green' computing means for laptops and data centers.
  • A standardized CO2-per-transistor disclosure—or a derived CO2-per-TFLOP label—could become a regulatory tool for semiconductor sustainability.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Applying the correct unit conversion to the paper's own numbers would raise processor-level operational emissions to roughly 730–1,477 kg over five years, which would reverse the headline conclusion and make usage dominate the lifetime footprint.
  • The per-transistor manufacturing estimate derives from wafer-level CO2 divided by transistor count, so the Apple-versus-Intel comparison would shift if yield or transistors-per-wafer differ between the 3nm M-series and the 7nm/10nm x86 parts.
  • The paper treats the grid emission factor as a global constant; plugging in regional grid mixes (e.g., coal-heavy vs. hydro-heavy) would change the operational term enough to alter the manufacturing/operational balance for different geographies.
  • A testable prediction from the corrected arithmetic is that low-power, high-transistor-count chips like the M3 would unexpectedly become the greener choice on lifetime emissions—opposite to the paper's stated finding.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes a Carbon Per Transistor (CPT) metric, defined as the sum of a manufacturing-emission term and an operational power-dissipation term, and applies it to the Intel Core i9-13900K, AMD Ryzen 9 7950X, and Apple M1/M2/M3 processors using reported transistor counts, TDP values, a 5-year / 8-hour-per-day usage model, and a grid emission factor of 0.4 kg/kWh. The paper's central claim is that manufacturing emissions dominate, contributing 60–125 kg CO2 per CPU, and that Apple's high-transistor-count M-series chips have a larger carbon footprint than traditional processors. I find that the headline result is invalidated by a factor-of-1000 unit error in the operational calculation, and that the manuscript's equations, tables, abstract, and conclusion are mutually inconsistent on the very quantities that would determine which term dominates.

Significance. A well-validated transistor-level carbon metric would be genuinely useful for green-computing benchmarking, and the paper identifies a real gap: most lifecycle assessments stop at wafer or system level rather than per-transistor granularity. The paper's strengths are its transparency about the proposed additive decomposition and its attempt to compare real commercial processors in a single table. However, the claimed empirical discovery—that manufacturing dominates—is not supported by the corrected arithmetic: the operational term as computed from the paper's own inputs is one to two orders of magnitude larger than the manufacturing term. Because the central result reverses under a simple unit correction and because the per-transistor values in the text contradict the conclusion, the paper's current contribution is not a scientifically rigorous benchmark.

major comments (4)
  1. [§3.3, Eq. (8); Table 4] The operational-emission calculation is missing the conversion from watt-hours to kilowatt-hours. With P_total = 125–253 W, H_lifetime = 14,600 h, and EF = 0.4 kg/kWh, Eq. (8) gives 730–1,477 kg for the Intel i9-13900K, not the 0.73–1.48 kg reported in Table 4. The same factor-of-1000 error affects the AMD and Apple rows. Correcting it makes operational emissions one to two orders of magnitude larger than the manufacturing estimates, reversing the paper's central claim that manufacturing dominates.
  2. [§3.3, §4, §8; Abstract] The manuscript's own per-transistor numbers contradict its conclusion. Eq. (7) gives manufacturing emissions of 2–5 µg/transistor, while Eq. (10) gives operational emissions of 60–250 mg/transistor and §4 restates the same operational figure as 60–250 µg/transistor with totals of 62–255 µg/transistor. If the µg numbers are used, operational emissions are about 97% of the total, not the 'approximately 98%' manufacturing share claimed in §8. The abstract's assertion that Apple's M-series has a larger carbon footprint also disagrees with Table 4, where the Apple M3 total (50.12–50.14 kg) is smaller than both Intel and AMD.
  3. [§5, Tables 3–4] The experimental validation contains arithmetic and unit inconsistencies. Table 3 gives AMD manufacturing emissions as 5.0 (presumably µg/transistor), which multiplied by 13.14 billion transistors gives 65.7 kg, yet Table 4 lists 565.7 kg as the manufacturing total for AMD; the row total 66.69–67.04 kg in Table 4 is consistent with 65.7 kg, not 565.7 kg. Tables 2–4 also omit units, and Table 4's Apple row ('0.125–140 kg') is not a clearly stated range.
  4. [§3.2, §3.3, §5.1] The claimed 'prediction' that manufacturing dominates is a definitional consequence of the assumed inputs rather than a validated empirical result. The manufacturing per-transistor values (2–5 µg) are asserted from 'industry reports' without citations, the wafer yield Y in Eq. (6) is never specified, the operational lifetime and grid emission factor appear without sensitivity analysis, and §5.1 cites 'official manufacturer reports' without identifying them. No error bars or uncertainty propagation are provided for any of the numerical conclusions.
minor comments (5)
  1. [Fig. 1] The caption refers to a 'CryptoDNA model architecture,' which appears unrelated to the CPT formula and is likely a leftover from a different manuscript.
  2. [§1.1, Eq. (1)] The notation for the operational term is inconsistent: Eq. (1) writes Coper, Eq. (3) writes Copr, and the text uses multiple spellings of 'per-transistor' (including 'pre-transistor').
  3. [§3.2] The bullet list repeats '350 kg CO2 per wafer for 7nm technology' twice, and the wafer-emission values are not referenced.
  4. [§7] The abstract promises accounting 'from fabrication to end-of-life,' but §7 lists end-of-life and recycling as future work; the manuscript should either include such terms or temper the abstract.
  5. [§5.1] The use of TDP as the average operational power is not justified; TDP is a thermal design limit, and the paper provides no evidence that it is a valid proxy for the 8 h/day usage profile.

Circularity Check

2 steps flagged · score 6.0 of 10

The central 'manufacturing dominates' claim is not derived: it is inserted as an unproved note in the derivation and appears in the results only after a factor-1000 unit inconsistency, so the prediction reduces to the paper's own assumptions.

  1. other [Section 3.2, after Eq. (7)]
    "Cman ≈ 2 − 5µgCO2/transistor (7) Note: for sub-7nm nodes, fabrication is the dominant source of emissions."

    The note states the paper's headline conclusion, that manufacturing/fabrication dominates, as an unproved aside inside the derivation section. The abstract and conclusion later present the same claim as an empirical discovery, but no comparison with operational data appears before the note. The 'prediction' therefore restates an input assumption; it is not obtained from the CPT formula.

  2. other [Section 3.3 Eq. (10), Section 4 summary, and Section 5.1 Table 4]
    "Coper ≈ 60 − 250mgCO2/transistor (10) ... Note: indicating that operational emissions are significant but secondary to manufacturing emissions. ... Table 4: 1 60 kg .73-1.48 kg 60.73-61.48 kg"

    The paper's own per-transistor numbers make operational emissions larger than manufacturing: Section 4 gives operational 60–250 µg and manufacturing 2–5 µg, i.e., operational is 12–125 times larger; multiplying both by the same N cannot reverse that ordering. Processor-level Table 4 shows Intel operational as only 0.73–1.48 kg, but applying the paper's own Eq. (5), Ctotal = Ntrans × Ctrans, to Table 3 (per-transistor operational 62–126 µg, N = 12e9) gives 744–1512 kg. The claimed manufacturing dominance appears only because the operational term is a factor of 1000 too small in Table 4. The headline 'finding' is thus an artifact of inconsistent unit handling, not a consequence of the stated model.

full rationale

The CPT formula itself is a definition and is not circular. The circularity lies in the empirical claim that manufacturing emissions dominate. That claim is first asserted as a note in the derivation (Section 3.2: 'for sub-7nm nodes, fabrication is the dominant source of emissions') and then repeated as a result. At the per-transistor level the paper states operational is 60–250 µg and manufacturing is 2–5 µg, so operational is 12–125 times larger; multiplying both by the same N cannot reverse that ordering. The processor-level table reverses the ordering only by listing operational totals that are 1000 times smaller than Eq. (5) implies when applied to Table 3. Correcting the arithmetic makes operational emissions an order of magnitude larger than manufacturing, reversing the conclusion. Therefore the central 'prediction' reduces to the paper's own assumption plus a unit-conversion error rather than to an independent derivation. This is partial circularity because the result is pre-loaded in the notes, though it is not a self-citation chain or a fitted-parameter renaming, so the score is 6 rather than higher.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The 'formula' combines an empirically assumed constant (2-5 ug/transistor) with a usage model; the processor-level results are direct arithmetic consequences of these inputs, not independent measurements.

free parameters (4)
  • Per-transistor manufacturing CO2 (Cman,per-transistor) = 2-5 ug/transistor
    Taken from 'industry reports' in Section 3.2 without specific sources or error bars; the processor-level manufacturing totals in Table 4 are simply this value times transistor count.
  • Wafer yield factor Y = not specified (implicitly 1)
    Appears in Eq. (6) but no value is used in the empirical estimates; if Y < 1, per-transistor manufacturing emissions would increase.
  • Lifetime hours H = 14,600 h (5 years x 8 h/day)
    Assumed usage model in Section 5; no sensitivity analysis.
  • Grid emission factor EF = 0.4 kg CO2/kWh
    Global average stated in Section 5; no regional variation or uncertainty.
assumptions (3)
  • domain assumption Transistor count is the correct normalizer for comparing chip carbon footprints.
    The paper assumes that per-transistor emissions are meaningful and comparable across architectures, but this ignores fixed non-scalable emissions.
  • domain assumption Total manufacturing emissions scale linearly with transistor count (C_total = N x C_trans).
    Used in Eq. (5) and Table 4; ignores packaging, testing, and other fixed overheads that do not scale with N.
  • ad hoc to paper TDP is a valid proxy for average operational power.
    The paper computes P_trans = TDP/N and uses it for the lifetime calculation, which overestimates (or misrepresents) actual energy use under variable workloads.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Carbon Per Transistor (CPT): The Golden Formula for Green Computing Metrics." pith.science (2026). https://pith.science/paper/BLGVG7TK

@misc{pith2026250202606,
  author       = {Pith},
  title        = {Pith review of: Carbon Per Transistor (CPT): The Golden Formula for Green Computing Metrics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BLGVG7TK}},
  note         = {Machine review of arXiv:2502.02606}
}
abstract

As computing power advances, the environmental cost of semiconductor manufacturing and operation has become a critical concern. However, current sustainability metrics fail to quantify carbon emissions at the transistor level, the fundamental building block of modern processors. This paper introduces a Carbon Per Transistor (CPT) formula -- a novel approach and green implementation metric to measuring the CO$_2$ footprint of semiconductor chips from fabrication to end-of-life. By integrating emissions from silicon crystal growth, wafer production, chip manufacturing, and operational power dissipation, the CPT formula provides a scientifically rigorous benchmark for evaluating the sustainability of computing hardware. Using real-world data from Intel Core i9-13900K, AMD Ryzen 9 7950X, and Apple M1/M2/M3 processors, we reveal a startling insight-manufacturing emissions dominate, contributing 60-125 kg CO$_2$ per CPU, far exceeding operational emissions over a typical device lifespan. Notably, Apple's high-transistor-count M-series chips, despite their energy efficiency, exhibit a significantly larger carbon footprint than traditional processors due to extensive fabrication impact. This research establishes a critical reference point for green computing initiatives, enabling industry leaders and researchers to make data-driven decisions in reducing semiconductor-related emissions and get correct estimates for the green factor of the information technology process. The proposed formula paves the way for carbon-aware chip design, regulatory standards, and future innovations in sustainable computing.

Figures

Figures reproduced from arXiv: 2502.02606 by the authors.

Figure 1
Figure 1. The proposed CryptoDNA model architecture. and other greenhouse gases (GHGs) reported by [9] and the US Environmental Protection Agency (EPA 2025), shown in Fig1. To address this, we propose a Carbon Per Transistor (CPT) formula, a math￾ematical model that systematically quantifies the CO2 footprint associated with semiconductor fabrication and operation. By applying this model to Intel, AMD, and Apple processors, w… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

  1. [1]

    Springer Science & Business Media (2011)

    Boyd, S.B.: Life-cycle assessment of semiconductors. Springer Science & Business Media (2011)

  2. [2]

    IEEE Journal of solid-state circuits9(5), 256–268 (1974)

    Dennard, R.H., Gaensslen, F.H., Yu, H.N., Rideout, V.L., Bassous, E., LeBlanc, A.R.: Design of ion-implanted mosfet’s with very small physical dimensions. IEEE Journal of solid-state circuits9(5), 256–268 (1974)

  3. [3]

    Elsevier (2011)

    Hennessy, J.L., Patterson, D.A.: Computer architecture: a quantitative approach. Elsevier (2011)

  4. [4]

    In: 2014 IEEE international solid-state circuits conference digest of technical papers (ISSCC)

    Horowitz, M.: 1.1 computing’s energy problem (and what we can do about it). In: 2014 IEEE international solid-state circuits conference digest of technical papers (ISSCC). pp. 10–14. IEEE (2014)

  5. [5]

    Huang, C.Y., Hu, A., Yin, J., Wang, H.C.: Developing a parametric carbon foot- printingtoolforthesemiconductorindustry.Internationaljournalofenvironmental science and technology13, 275–284 (2016)

  6. [6]

    Ovidius University Annals, Series Economic Sciences 21(2) (2021)

    Ivan, O.R., Harmanas, A.O., Cosma, M.: Practices of corporate social responsibil- ity reporting for semiconductor and chip manufacturing industry–a multicriterial analysis on intel and amd. Ovidius University Annals, Series Economic Sciences 21(2) (2021)

  7. [7]

    Journal of Asia Business Studies (2024)

    Kang, A.S., Arikrishnan, S.: Sustainability reporting and total quality management post-pandemic: the role of environmental, social, governance (esg), and smart tech- nology adoption. Journal of Asia Business Studies (2024)

  8. [8]

    A report by Analytical Press, completed at the request of The New York Times9(2011), 161 (2011)

    Koomey, J., et al.: Growth in data center electricity use 2005 to 2010. A report by Analytical Press, completed at the request of The New York Times9(2011), 161 (2011)

Show all 16 references
  1. [9]

    Re- sources, Conservation and Recycling182, 106289 (2022)

    Kuo, T.C., Kuo, C.Y., Chen, L.W.: Assessing environmental impacts of nanoscale semi-conductor manufacturing from the life cycle assessment perspective. Re- sources, Conservation and Recycling182, 106289 (2022)

  2. [10]

    Sustainability16(15), 6548 (2024)

    Liu, Y.Z., Lu, W.M., Tran, P.P., Pham, T.A.K.: Sustainable energy and semicon- ductors: A bibliometric investigation. Sustainability16(15), 6548 (2024)

  3. [11]

    Journal of Cleaner Pro- duction p

    Ma, S., Ding, W., Liu, Y., Zhang, Y., Ren, S., Kong, X., Leng, J.: Industry 4.0 and cleaner production: A comprehensive review of sustainable and intelligent manu- facturing for energy-intensive manufacturing industries. Journal of Cleaner Pro- duction p. 142879 (2024)

  4. [12]

    Proceedings of the IEEE 86(1), 82–85 (1998)

    Moore, G.E.: Cramming more components onto integrated circuits. Proceedings of the IEEE 86(1), 82–85 (1998)

  5. [13]

    Sustainability16(1), 218 (2023)

    Nagapurkar, P., Nandy, P., Nimbalkar, S.: Cleaner chips: Decarbonization in semi- conductor manufacturing. Sustainability16(1), 218 (2023)

  6. [14]

    In: 2024 International Conference on IoT Based Control Networks and Intelligent Systems (ICICNIS)

    Sungheetha, A., Sharma, R.R., Mahapatra, S., Rani, K.S.K., Leni, A.E.S., Tami- larasi, R.: Adaptive stream processing framework for energy-efficient smart green- houses using neuromorphic computing. In: 2024 International Conference on IoT Based Control Networks and Intelligen...

  7. [15]

    Applied energy 136, 636–648 (2014)

    Vasan, A., Sood, B., Pecht, M.: Carbon footprinting of electronic products. Applied energy 136, 636–648 (2014)

  8. [16]

    Water Cycle4, 47–54 (2023)

    Wang, Q., Huang, N., Chen, Z., Chen, X., Cai, H., Wu, Y.: Environmental data and facts in the semiconductor manufacturing industry: An unexpected high water and energy consumption situation. Water Cycle4, 47–54 (2023)

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.