Pith. sign in

REVIEW 4 major objections 3 minor 52 references

Femtojoule-per-operation photonic computer for the subset sum problem

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A photonic computer solves subset sum with about 10^-15 joules per operation

desk verdict The 10^-15 J/operation headline rests on energy accounting the abstract doesn't define; the experimental claim deserves review, but the number is unverified. read the letter →

arxiv 2508.17274 v1 pith:32RYVWES submitted 2025-08-24 physics.optics quant-ph

Tian-Yu Zhang , Xiao-Yun Xu , Wen-Hao Zhou , Xiao-Wei Wang , Chu-Han Wang , Yi-Jun Chang , Ying-Yue Yang , Jie Ma
show 2 more authors
Ka-Di Zhu Xian-Min Jin
This is my paper
classification physics.opticsquant-ph
keywords photoniccomputingsubsetsumproblemenergy-efficienttime-of-flightstorageopticalcomputationNP-completegreen
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 reports a working photonic computer that tackles the subset sum problem, an NP-complete task that scales poorly on electronic machines. The device encodes the numbers of a subset-sum instance as photon arrival times, storing the evolving set of possible sums in the time-of-flight of light rather than in electronic memory. The authors report energy per operation no larger than 10^-15 J for instance size N=33, about 10^8 times less than the most energy-efficient supercomputer at medium scale. The claim matters because data movement, not logic, dominates energy use in von Neumann machines, and time-of-flight storage sidesteps that cost.

What carries the argument

Time-of-flight storage is the central mechanism: the set of achievable subset sums is encoded in the arrival times of optical pulses, with each input number implemented as a delay path. This carries the argument by replacing energy-hungry electronic memory access with photon propagation, and the extremely low energy of a single photon places the physical energy floor near 10^-19 J per operation.

What would settle it

Measure the total electrical power drawn by the complete photonic setup while solving a medium-scale subset-sum instance, count the operations, and divide; if the resulting energy per operation exceeds 10^-15 J, the central claim fails. A second decisive check is to independently verify the solver's output on the same N=33 instances, since an incorrect solver would not be energy-advantageous.

Watch

Extended reading notes

Core claim

The paper claims that the subset sum problem can be solved optically with femtojoule-scale energy per operation. The demonstration uses photons with energy around 10^-19 J and a time-of-flight storage technique: possible subset sums are represented by the times at which optical pulses arrive, avoiding the energy-intensive movement of data between memory and processor. The reported result is that energy consumption stays at or below 10^-15 J per operation for problem size N=33, and for a medium-scale instance it consumes 10^8 times less energy than the most energy-efficient supercomputer. The paper further argues that when the solver is embedded in iterative real-life computations that repeat

Load-bearing premise

The load-bearing premise is that the reported 10^-15 J per operation covers all electricity consumed by the whole machine—laser, modulators, detectors, and control electronics—and that an 'operation' means the same thing for the photonic solver and for the supercomputer baseline.

Editorial extensions

If this is right

  • If the energy accounting holds, photonic subset-sum solvers could operate at orders of magnitude lower power than electronic alternatives, making them practical for embedded or mobile settings where hard combinatorial problems arise.
  • For real-life workflows that require many subset-sum computations, the paper's stated advantage strengthens because the input setup cost is amortized over many iterative calls.
  • The demonstration at N=33 indicates the approach is not limited to tiny toy instances, though larger sizes will require longer optical delay paths and more precise timing.
  • Because the computation is carried by photon propagation rather than electronic data movement, the solver directly targets the main energy bottleneck of conventional von Neumann machines.

Reading between the lines

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

  • Editorial inference: whether the 10^-15 J per operation figure survives scrutiny depends on including all real electrical power consumed by the full system—laser, modulators, detectors, and control electronics—and on 'operation' meaning the same unit for both the photonic solver and the supercomputer baseline.
  • Editorial inference: the same time-of-flight architecture may extend to other combinatorial search problems that can be encoded as sums of delays, such as the partition problem or knapsack variants.
  • Editorial inference: the practical ceiling for the approach will likely be set by detector timing precision and the available time-bandwidth product of delay lines; improving these would raise the solvable problem size.
  • Editorial inference: this is a specialized accelerator, not a general-purpose replacement for electronic computers; it is best seen as a complementary, ultra-low-energy module for problems reducible to subset sum.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 3 minor

Summary. The paper reports an experimental photonic computer for the subset sum problem (SSP), claiming that by exploiting the low energy of individual photons (~10^-19 J) and a time-of-flight storage method, the system achieves no more than 10^-15 J per operation at problem size N=33, and consumes 10^8 times less energy than the most energy-efficient supercomputer for a medium-scale problem. The readable portion of the submission is limited to the abstract; the full text is corrupted and unreadable in the provided manuscript, so no methods, experimental setup, measurements, or baseline specifications are available for evaluation.

Significance. If the central claim is substantiated, this would be a significant experimental result: a working optical solver for a relevant NP-complete problem with an orders-of-magnitude energy advantage over electronic supercomputers. The headline energy figure is specific and falsifiable, which is a strength. However, the submission as provided gives no measurement details, no error analysis, no system-boundary definition, and no description of the comparison baseline. The significance of the result therefore remains conditional on evidence that is not currently present in the manuscript.

major comments (4)
  1. [Abstract, 3rd sentence] The claim 'energy consumption ... no larger than 10^-15 J per operation at N=33' is unverifiable from the submitted text. No wall-plug energy measurement is reported, and the system boundary is undefined. The preceding statement that photons carry ~10^-19 J is not an energy-per-operation estimate for a complete machine; the laser, modulators, detectors, transimpedance amplifiers, control electronics, ADC, and idle power during time-of-flight storage must all be included. Please provide a complete energy budget and measurement protocol.
  2. [Abstract, 4th sentence] The comparison '10^8 times less energy than the most energy-efficient supercomputer' lacks a defined baseline. Which supercomputer, which algorithm for SSP, what problem instance size, and what counts as an operation for the electronic machine? Without a common definition of 'operation' and a specific baseline configuration, the claimed factor of 10^8 is not interpretable and could reflect a mismatch in problem encoding or algorithmic complexity rather than a hardware advantage.
  3. [Full text (unreadable); Methods/Results absent] The submitted PDF contains only a corrupted full-text section. There is no experimental section, no schematic, no data table, no error bars, no scaling measurement, and no description of how N=33 SSP instances are encoded and solved. Because the central claim is experimental, these omissions are load-bearing. The authors must supply a complete, readable manuscript with measurement details, a definition of 'operation', a system-boundary diagram, and ideally raw data or a reproducibility statement.
  4. [Abstract, metric definition] The metric 'energy per operation' is self-defined and may be circular if the operation count is chosen to favor the photonic system. For example, if one SSP query is counted as one photonic operation but as many bit operations in the electronic baseline, the comparison is not apples-to-apples. Please specify the computational primitive for both systems and justify that the same problem instance and algorithmic task are being compared.
minor comments (3)
  1. [Abstract, 4th sentence] The term 'medium-scale problem' is vague; if this refers to N=33, state so explicitly. If it refers to a different instance, give its parameters.
  2. [Abstract, 5th sentence] The phrase 'photonic advantage in energy consumption' is repeated from the preceding claim and adds little; consider instead giving a concrete example of an iterative SSP-based application.
  3. [General] The manuscript has no references or related-work discussion in the readable portion; even a brief comparison with prior optical Ising machines or electronic SSP solvers would help position the contribution.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identifiable: full text is unreadable and the abstract alone does not reduce any claim to its inputs.

full rationale

The supplied full text is corrupted mojibake and contains no recoverable equations, derivations, or method descriptions. The only legible source is the abstract, which reports an experimental demonstration and an energy-efficiency claim (≤10^-15 J per operation at N=33). No derivation chain is visible, so no step can be shown to be equivalent to its inputs by construction. The abstract does not define 'operation' or specify the energy-accounting boundary, which is a verification/comparability concern rather than a circularity concern: the text does not, for example, define the energy metric in terms of the very photon energy used to compute it, nor does it fit a parameter to a subset of data and call it a prediction. There are no self-citations in the abstract that could be load-bearing. Under the hard rule that circularity must be exhibited by quotation of a specific reduction, no circular step can be identified. The absence of readable methods prevents a full audit, but that is not itself evidence of circularity. Score 0 is therefore the honest finding.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

Only the abstract was readable. The central claim assumes that subset sums can be physically encoded in photon arrival times and that energy per operation can be fairly compared across architectures. No free parameters or invented entities are identifiable from the abstract.

assumptions (2)
  • domain assumption Subset sums can be represented and manipulated as photon time-of-flight delays, so that the photonic system's operation count corresponds to subset-sum computation.
    This is the physical encoding that makes the photonic solver solve SSP rather than a different problem; stated in the abstract as 'time-of-flight storage technique'.
  • domain assumption The energy-per-operation metric is comparable across photonic and electronic systems, with the same system boundary and operation definition for both.
    The 10^8 comparison in the abstract presupposes a fair accounting; no details are given in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Femtojoule-per-operation photonic computer for the subset sum problem." pith.science (2026). https://pith.science/paper/32RYVWES

@misc{pith2026250817274,
  author       = {Pith},
  title        = {Pith review of: Femtojoule-per-operation photonic computer for the subset sum problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/32RYVWES}},
  note         = {Machine review of arXiv:2508.17274}
}
read the original abstract

Energy-efficient computing is becoming increasingly important in the information era. However, electronic computers with von Neumann architecture can hardly meet the challenge due to the inevitable energy-intensive data movement, especially when tackling computationally hard problems or complicated tasks. Here, we experimentally demonstrate an energy-efficient photonic computer that solves intractable subset sum problem (SSP) by making use of the extremely low energy level of photons (~10^(-19) J) and a time-of-flight storage technique. We show that the energy consumption of the photonic computer maintains no larger than 10^(-15) J per operation at a reasonably large problem size N=33, and it consumes 10^(8) times less energy than the most energy-efficient supercomputer for a medium-scale problem. In addition, when the photonic computer is applied to deal with real-life problems that involves iterative computation of the SSP, the photonic advantage in energy consumption is further enhanced and massive energy can be saved. Our results indicate the superior competitiveness of the photonic computer in the energy costs of complex computation, opening a possible path to green computing.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 51 canonical work pages

  1. [1]

    Zwart, S. P. The ecological impact of high-performance computing in astrophysics. Nat. Astron. 4, 819-822 (2020)

    work page 2020

  2. [2]

    Khanna, T. M. et al. A multi-country meta-analysis on the role of behavioural change in reducing energy consumption and CO_ 2 emissions in residential buildings. Nat. Energy 6, 925-932 (2021)

    work page 2021

  3. [3]

    Recalibrating global data center energy-use estimates

    Masanet, E., Shehabi, A., Lei, N., Smith, S., & Koomey, J. Recalibrating global data center energy-use estimates. Science 367, 984-986 (2020)

    work page 2020

  4. [4]

    & Cameron, K

    Feng, W.-C. & Cameron, K. The green500 list: Encouraging sustainable supercomputing. Computer 40, 50-55 (2007)

    work page 2007

  5. [5]

    Valentini, G. L. et al. An overview of energy efficiency techniques in cluster computing systems. Cluster Comput. 16, 3-15 (2013)

    work page 2013

  6. [6]

    Barroso, L. A. & Hölzle, U. The case for energy-proportional computing. Computer 40, 33-37 (2007)

    work page 2007

  7. [7]

    Rong, H., Zhang, H., Xiao, S., Li, C. & Hu, C. Optimizing energy consumption for data centers. Renew. Sust. Energ. Rev. 58, 674-691 (2016)

    work page 2016

  8. [8]

    B., Xiao, X

    Lee, Y.-J., On, M. B., Xiao, X. & Yoo, S. B. Energy-efficient photonic spiking neural network on a monolithic silicon CMOS photonic platform. In Optical Fiber Communication Conference Tu5H-5 (Optical Society of America, 2021)

    work page 2021

Show all 52 references
  1. [9]

    Garey, M. R. & Johnson, D. S. Computers and Intractability (WH Freeman, 1979)

    1979

  2. [10]

    Xu, X.-Y. et al. A scalable photonic computer solving the subset sum problem. Sci. Adv. 6, eaay5853 (2020)

    2020

  3. [11]

    & Marques-Silva, J

    Lynce, I. & Marques-Silva, J. P. Efficient haplotype inference with Boolean satisfiability. In National Conference on Artificial Intelligence 104-109 (AAAI Press, 2006)

    2006

  4. [12]

    & Naghibzadeh, M

    Eghdami, M. & Naghibzadeh, M. SSA: subset sum approach to protein -sheet structure prediction. Comput. Biol. Chem. 94, 107552 (2021)

    2021

  5. [13]

    Marques-Silva, J. P. & Sakallah, K. A. Boolean satisfiability in electronic design automation. In Proceedings of the 37th Annual Design Automation Conference 675-680 (2000)

    2000

  6. [14]

    & Lizaso, E

    Or \'u s, R., Mugel, S. & Lizaso, E. Quantum computing for finance: Overview and prospects. Rev. Phys. 4, 100028 (2019)

    2019

  7. [15]

    & Bauckhage, C

    Biesner, D., Sifa, R. & Bauckhage, C. Solving subset sum problems using binary optimization with applications in auditing and financial data analysis. Preprint at https://doi.org/10.36227/techrxiv.18994160.v1 (2022)

    2022 doi

  8. [16]

    Arle, J. E. & Carlson, K. W. Medical diagnosis and treatment is NP-complete. J. Artif. Intell. Res. 33, 297-312 (2021)

    2021

  9. [17]

    & Sakai, R

    Murakami, Y. & Sakai, R. Security of knapsack cryptosystem using subset-sum decision problem against alternative-solution attack. In 2018 International Symposium on Information Theory and Its Applications 614-617 (IEEE, 2018)

    2018

  10. [18]

    & Kasahara, M

    Murakami, Y., Hamasho, S. & Kasahara, M. A public-key cryptosystem based on decision version of subset sum problem. In 2012 International Symposium on Information Theory and Its Applications 735-739 (IEEE, 2012)

    2012

  11. [19]

    Zhang, A. et al. Quantum verification of NP problems with single photons and linear optics. Light Sci. Appl. 10, 1-11 (2021)

    2021

  12. [20]

    V \'a zquez, M. R. et al. Optical NP problem solver on laser-written waveguide platform. Opt. Express 26, 702–710 (2018)

    2018

  13. [21]

    S., Chelyapov, N., Johnson, C., Rothemund, P

    Braich, R. S., Chelyapov, N., Johnson, C., Rothemund, P. W. K. & Adleman, L. Solution of a 20-variable 3-SAT problem on a DNA computer. Science 296, 499-502 (2002)

    2002

  14. [22]

    & Kerenidis, I

    Centrone, F., Kumar, N., Diamanti, E. & Kerenidis, I. Experimental demonstration of quantum advantage for NP verification with limited information. Nat. Commun. 12, 1-11 (2021)

    2021

  15. [23]

    & Parkin, S

    Demasius, K.-U., Kirschen, A. & Parkin, S. Energy-efficient memcapacitor devices for neuromorphic computing. Nat. Electron. 4, 748-756 (2021)

    2021

  16. [24]

    Chen, Y.-H., Krishna, T., Emer, J. S. & Sze, V. Eyeriss: An energy-efficient reconfigurable accelerator for deep convolutional neural networks. IEEE J. Solid-State Circuits 52, 127-138 (2016)

    2016

  17. [25]

    Pei, J. et al. Towards artificial general intelligence with hybrid Tianjic chip architecture. Nature 572, 106-111 (2019)

    2019

  18. [26]

    & M \"o tt \"o nen, M

    Ikonen, J., Salmilehto, J. & M \"o tt \"o nen, M. Energy-efficient quantum computing. NPJ Quantum Inf. 3, 1-7 (2017)

    2017

  19. [27]

    Elsayed, N., Maida, A. S. & Bayoumi, M. A review of quantum computer energy efficiency. In 2019 IEEE Green Technologies Conference 1-3 (IEEE, 2019)

    2019

  20. [28]

    A., Mohiyaddin, F

    Britt, K. A., Mohiyaddin, F. A. & Humble, T. S. Quantum accelerators for high-performance computing systems. In 2017 IEEE International Conference on Rebooting Computing 1-7 (IEEE, 2017)

    2017

  21. [29]

    Nicolau, D. V. et al. Parallel computation with molecular-motor-propelled agents in nanofabricated networks. Proc. Natl. Acad. Sci. U.S.A. 113, 2591-2596 (2016)

    2016

  22. [30]

    Adleman, L. M. Molecular computation of solutions to combinatorial problems. Science 266, 1021-1024 (1994)

    1994

  23. [31]

    & Hashimoto, T

    Nakajima, M., Tanaka, K. & Hashimoto, T. Scalable reservoir computing on coherent linear photonic processor. Commun. Phys. 4, 1-12 (2021)

    2021

  24. [32]

    & Jin, X.-M

    Xu X.-Y. & Jin, X.-M. Integrated Photonic Computing beyond the von Neumann Architecture. ACS Photonics 10, 1027--1036 (2023)

    2023

  25. [33]

    Wang, T. et al. An optical neural network using less than 1 photon per multiplication. Nat. Commun. 13, 1-8 (2022)

    2022

  26. [34]

    Uppu, R. et al. Scalable integrated single-photon source. Sci. Adv. 6, eabc8268 (2020)

    2020

  27. [35]

    Zhong, H.-S. et al. Quantum computational advantage using photons. Science 370, 1460-1463 (2020)

    2020

  28. [36]

    Yin, P. et al. Improving the precision of optical metrology by detecting fewer photons with biased weak measurement. Light Sci. Appl. 10, 1-8 (2021)

    2021

  29. [37]

    Superconducting nanowire single-photon detectors for quantum information

    You, L. Superconducting nanowire single-photon detectors for quantum information. Nanophotonics 9, 2673-2692 (2020)

    2020

  30. [38]

    Karp, R. M. Complexity of Computer Computations (Springer, 1972)

    1972

  31. [39]

    Zhou, W.-H. et al. Timestamp boson sampling. Appl. Phys. Rev. 9, 031408 (2022)

    2022

  32. [40]

    & Walter, R

    Schwerdfeger, S. & Walter, R. A fast and effective subset sum based improvement procedure for workload balancing on identical parallel machines. Comput. Oper. Res. 73, 84-91 (2016)

    2016

  33. [41]

    Xu, X.-Y., Wang, X.-W., Chen, D.-Y., Smith, C. M. & Jin, X.-M. Quantum transport in fractal networks. Nat. Photon. 15, 703-710 (2021)

    2021

  34. [42]

    Gao, J. et al. Quantum advantage with membosonsampling. Chip 1, 100007 (2022)

    2022

  35. [43]

    Ruiz-Vanoye, J. A. et al. Survey of polynomial transformations between NP-complete problems. J. Appl. Math. Comput. 235, 4851-4865 (2011)

    2011

  36. [44]

    Y., Zhang, T

    Xu, X. Y., Zhang, T. Y., Wang, Z. W., Wang, C. H., & Jin, X. M. Reconfigurable integrated photonic processor for NP-complete problems. Adv. Photonics 6, 056011 (2024)

    2024

  37. [45]

    Miller, D. A. B. Are optical transistors the logical next step? Nat. Photon. 4, 3-5 (2010)

    2010

  38. [46]

    Miller, D. A. B. Attojoule optoelectronics for low-energy information processing and communications. J. Light. Technol. 35, 346-396 (2017)

    2017

  39. [47]

    Shen, Y. et al. Deep learning with coherent nanophotonic circuits. Nat. Photon. 11, 441-446 (2017)

    2017

  40. [48]

    Noh, J. et al. Braiding photonic topological zero modes. Nat. Phys. 16, 989-993 (2020)

    2020

  41. [49]

    Green500 list, https://www.top500.org/lists/green500/2022/06/

    2022

  42. [50]

    Intel Core i7-920, https://gadgetversus.com/processor/intel-core-i7-920-specs/

  43. [51]

    NVIDIA GeForce GTX 1650, https://gadgetversus.com/graphics-card/nvidia-geforce-gtx-1650-specs/

  44. [52]

    Kestor, G., Gioiosa, R., Kerbyson, D. J. & Hoisie, A. Quantifying the energy cost of data movement in scientific applications. In 2013 IEEE international symposium on workload characterization 56-65 (IEEE, 2013)

    2013

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.