pith. sign in

arxiv: 2606.11247 · v1 · pith:DOKTBQY3new · submitted 2026-06-08 · 💻 cs.LG · cs.AI· cs.AR

Physics-informed generative AI for semiconductor manufacturing: Enforcing hard physical constraints in generative models by construction

Yaser Mike Banad , Sarah Sharif This is my paper

Pith reviewed 2026-06-27 17:34 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.AR
keywords generative modelssemiconductor manufacturingphysics-informed AIhard physical constraintslithographydiffusion modelsTCADprocess simulation
0
0 comments X

The pith

Generative models for semiconductor manufacturing must enforce physical constraints like lithography and device physics by construction rather than filtering invalid outputs afterward.

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

The paper argues that generative AI applied to physical systems must embed hard constraints directly into model architectures because invalid samples are unusable in practice, not merely low quality. Semiconductor manufacturing serves as the test case where generated masks, layouts, and process recipes must satisfy lithography, transport, reaction, and device-physics rules. It surveys architectural approaches including physics-informed diffusion, PDE-constrained variational models, neural-operator priors, and conservation-law-respecting networks, and connects them to differentiable lithography, TCAD, and process simulation tools. Four integration patterns are identified, along with a research agenda on physics-fidelity benchmarks, differentiable simulators, and multimodal foundation models.

Core claim

Where physical validity is the binding criterion of success, architectures that enforce it by construction should be expected to outperform those that filter for it after the fact, and the fab is the setting where this distinction is sharpest.

What carries the argument

The emerging architectural toolkit of physics-informed diffusion, PDE-constrained variational models, neural-operator priors, and conservation-law-respecting generative networks that embed lithography, transport, reaction, and device-physics constraints directly into the generative process.

If this is right

  • Four integration patterns can be defined between generative models and physics-based simulators such as TCAD.
  • Physics-fidelity benchmarks are required to measure how well models respect hard constraints.
  • Differentiable simulator infrastructure is needed to train models that respect those constraints.
  • Multimodal foundation models for physical design and manufacturing become a natural next step.

Where Pith is reading between the lines

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

  • The same by-construction principle could apply to other manufacturing domains that reject invalid physical proposals outright.
  • By-construction enforcement may reduce wasted computation compared with generating then discarding large numbers of invalid samples.
  • Closed-loop autonomous experimentation systems could be built that never propose physically impossible actions.

Load-bearing premise

It is feasible to construct generative architectures that enforce the relevant hard physical constraints without loss of useful generative capacity.

What would settle it

A controlled comparison in which a post-hoc filtering generative model matches or exceeds the physical validity rate and diversity of a by-construction model on a lithography mask or process-recipe generation task would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.11247 by Sarah Sharif, Yaser Mike Banad.

Figure 1
Figure 1. Figure 1: Semiconductor manufacturing as a model system for constrained scientific generation. The general computational￾science challenge (left): a generative model proposes an output that scores well on learned plausibility yet may violate hard physical constraints, so post-hoc validation rejects it. The semiconductor testbed (center): mask and layout generation, process-recipe generation, synthetic defect data, a… view at source ↗
Figure 2
Figure 2. Figure 2: Failure mode taxonomy for generative outputs in constrained physical domains, organized along axes of learned plausibility and physical validity. Learned plausibility is whatever the generative objective rewards, whether visual realism, likelihood under the training distribution, or a preference score; in the fab it is typically the visual realism of a layout or defect image. The upper-left quadrant (plaus… view at source ↗
Figure 3
Figure 3. Figure 3: Four integration patterns between the data-driven and physics-based cultures of the fab. Each pattern combines a generative component (language model, generative prior, diffusion sampler, foundation model) with a physics-based component (process-rule and process-window checker, failure-mode simulator, differentiable lithography or TCAD simulator, simulation corpus). The integration is structural rather tha… view at source ↗
Figure 4
Figure 4. Figure 4: Illustrative three-horizon research roadmap for physics-informed generative AI in semiconductor manufacturing. The dates are indicative horizons, not predictions. Near-term (2026 to 2027): physics-fidelity benchmarks and reference solvers. Medium-term (2027 to 2030): differentiable lithography and TCAD simulators and simulator-in-the-loop training infrastructure. Long-term (2030 and beyond): multimodal fou… view at source ↗
read the original abstract

Generative models are increasingly used to propose designs, data, and control actions for physical systems, yet many such systems are governed by hard physical constraints rather than by perceptual plausibility. Semiconductor manufacturing provides a demanding test case: generated masks, layouts, synthetic defect data, and process recipes must obey lithography, transport, reaction, and device-physics constraints, because physically invalid samples are not merely low quality but unusable. This Perspective argues that semiconductor manufacturing exposes a broader computational-science challenge, namely that generative AI for constrained physical domains must be physics-informed by construction, not corrected only through post-hoc filtering. We survey the emerging architectural toolkit, including physics-informed diffusion, PDE-constrained variational models, neural-operator priors, and conservation-law-respecting generative networks, and show how it connects to differentiable lithography, TCAD, process simulation, and autonomous experimentation. We identify four integration patterns between generative models and physics-based simulators, and we propose a research agenda centered on physics-fidelity benchmarks, differentiable simulator infrastructure, and multimodal foundation models for physical design and manufacturing. The central claim is analytical rather than rhetorical: where physical validity is the binding criterion of success, architectures that enforce it by construction should be expected to outperform those that filter for it after the fact, and the fab is the setting where this distinction is sharpest.

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

2 major / 0 minor

Summary. The manuscript is a Perspective arguing that generative AI for semiconductor manufacturing must enforce hard physical constraints (lithography, transport, reaction, device physics) by construction rather than post-hoc filtering, because invalid samples are unusable. It surveys architectures including physics-informed diffusion, PDE-constrained variational models, neural-operator priors, and conservation-law-respecting networks; identifies four integration patterns with differentiable lithography, TCAD, and process simulators; and proposes a research agenda on physics-fidelity benchmarks, simulator infrastructure, and multimodal foundation models. The central analytical claim is that where physical validity is binding, by-construction methods should outperform post-hoc approaches, with the fab as the sharpest test case.

Significance. If the analytical claim holds, the perspective could usefully orient research toward constraint-enforcing generative architectures in physical domains, potentially improving sample efficiency and reliability in manufacturing applications. The survey of the emerging toolkit and explicit identification of integration patterns constitute a constructive contribution that could serve as a reference point for connecting generative models to physics-based simulators.

major comments (2)
  1. [Abstract] Abstract and central claim: the assertion that by-construction enforcement 'should be expected to outperform' post-hoc filtering is presented as analytical, yet the manuscript supplies neither a derivation, quantitative comparison, nor analysis of trade-offs (e.g., possible reduction in expressivity or optimization difficulty when embedding hard constraints for coupled systems such as lithography and device physics). This assumption is load-bearing for the expectation of outperformance.
  2. [Survey and integration patterns] The survey of the architectural toolkit (physics-informed diffusion, PDE-constrained variational models, etc.) and the four integration patterns are described at a high level, but no concrete discussion is given of how these methods can be realized with differentiable TCAD simulators while preserving generative diversity and utility in the highly constrained semiconductor setting; this feasibility question directly affects whether the claimed advantage materializes.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the scope and claims of this Perspective. We address each major point below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract and central claim: the assertion that by-construction enforcement 'should be expected to outperform' post-hoc filtering is presented as analytical, yet the manuscript supplies neither a derivation, quantitative comparison, nor analysis of trade-offs (e.g., possible reduction in expressivity or optimization difficulty when embedding hard constraints for coupled systems such as lithography and device physics). This assumption is load-bearing for the expectation of outperformance.

    Authors: We agree the central claim is analytical and would benefit from explicit discussion of its basis and limitations. The argument rests on the observation that, in semiconductor manufacturing, physically invalid samples have zero utility; therefore any generative process that produces a non-zero fraction of invalids incurs an effective yield penalty relative to one that produces only valid samples. This is a first-principles efficiency argument rather than an empirical claim. However, we acknowledge the absence of a formal derivation or quantitative trade-off analysis (e.g., expressivity loss or optimization hardness in coupled lithography-device systems). We will revise the abstract and the section stating the central claim to (i) articulate the logical steps more explicitly and (ii) note potential trade-offs, while preserving the Perspective's non-empirical character. revision: partial

  2. Referee: [Survey and integration patterns] The survey of the architectural toolkit (physics-informed diffusion, PDE-constrained variational models, etc.) and the four integration patterns are described at a high level, but no concrete discussion is given of how these methods can be realized with differentiable TCAD simulators while preserving generative diversity and utility in the highly constrained semiconductor setting; this feasibility question directly affects whether the claimed advantage materializes.

    Authors: As a Perspective, the manuscript intentionally remains at the level of architectural patterns and research directions rather than providing implementation blueprints. That said, the referee correctly identifies that feasibility with differentiable TCAD is central to whether the claimed advantage can be realized. We will expand the integration-patterns section with additional paragraphs that (a) reference existing differentiable TCAD and lithography simulators in the literature, (b) discuss mechanisms (e.g., adjoint methods, surrogate operators) that can preserve diversity while enforcing constraints, and (c) flag open questions about scalability and diversity preservation that the proposed research agenda should address. These additions will remain within the Perspective format and will not introduce new empirical results. revision: yes

Circularity Check

0 steps flagged

No circularity; central claim is a domain-logical expectation without self-referential reduction

full rationale

The paper is a Perspective article whose central claim—that architectures enforcing physical constraints by construction should outperform post-hoc filtering where validity is binding—is presented as an analytical observation drawn from the semiconductor manufacturing domain, where invalid samples are unusable. No equations, derivations, fitted parameters, or predictions appear in the text. The argument surveys external methods (physics-informed diffusion, PDE-constrained models, etc.) and proposes an agenda without invoking self-citations as load-bearing justification or reducing any result to its own inputs by construction. The claim rests on the observable distinction between enforcement and filtering plus domain requirements, making the paper self-contained against external benchmarks of physical validity and generative utility.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper is a perspective relying on domain assumptions about the nature of physical constraints in manufacturing and the limitations of post-hoc approaches; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption Physical validity is the binding criterion of success in semiconductor manufacturing, making post-hoc filtering insufficient.
    Stated directly in the abstract as the reason generative models must enforce constraints by construction.

pith-pipeline@v0.9.1-grok · 5768 in / 1140 out tokens · 23675 ms · 2026-06-27T17:34:07.746232+00:00 · methodology

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

43 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    Deep learning for smart manufacturing: Methods and applications,

    J. Wang, Y. Ma, L. Zhang, R. X. Gao, and D. Wu, “Deep learning for smart manufacturing: Methods and applications,” Journal of Manufacturing Systems, vol. 48, pp. 144 –156, 2018

    2018

  2. [2]

    G. S. May and C. J. Spanos, Fundamentals of Semiconductor Manufacturing and Process Control. Hoboken, NJ, USA: Wiley-IEEE Press, 2006

    2006

  3. [3]

    Big data analytics for smart manufacturing: Case studies in semiconductor manufacturing,

    J. Moyne and J. Iskandar, “Big data analytics for smart manufacturing: Case studies in semiconductor manufacturing,” Processes, vol. 5, no. 3, p. 39, 2017

    2017

  4. [4]

    Automated quality and process control for additive manufacturing using deep convolutional neural networks,

    Y. Banadaki, N. Razaviarab, H. Fekrmandi, G. Li, P. Mensah, S. Bai, and S. Sharifi, “Automated quality and process control for additive manufacturing using deep convolutional neural networks,” Recent Progress in Materials, vol. 4, no. 1, 2021

    2021

  5. [5]

    Smart additive manufacturing empowered by a closed -loop machine learning algorithm,

    N. Razaviarab, S. Sharifi, and Y. M. Banadaki, “Smart additive manufacturing empowered by a closed -loop machine learning algorithm,” in Nano-, Bio-, Info-Tech Sensors and 3D Systems III, vol. 10969, pp. 50–57, SPIE, 2019

    2019

  6. [6]

    On the use of machine learning for additive manufacturing technology in Industry 4.0,

    Y. Banadaki, “On the use of machine learning for additive manufacturing technology in Industry 4.0,” Journal of Computer Science and Information Technology, vol. 7, pp. 61–68, 2019

    2019

  7. [7]

    Toward digital twins in 3D IC packaging: A critical review of physics, data, and hybrid architectures,

    G. Datta, S. S. Sharif, and Y. M. Banad, “Toward digital twins in 3D IC packaging: A critical review of physics, data, and hybrid architectures,” arXiv:2601.23226, 2026

    work page arXiv 2026

  8. [8]

    Moyne, E

    J. Moyne, E. Del Castillo, and A. M. Hurwitz, Eds., Run -to-Run Control in Semiconductor Manufacturing. Boca Raton, FL, USA: CRC Press, 2001

    2001

  9. [9]

    Thermomechanical challenges of 2.5 -D packaging: A review of warpage and interconnect reliability,

    H. Kim, J. Y. Hwang, S. E. Kim, Y. C. Joo, and H. Jang, “Thermomechanical challenges of 2.5 -D packaging: A review of warpage and interconnect reliability,” IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 13, no. 10, pp. 1624–1641, 2023. Physics-informed generative AI for semiconductor manufacturing 13

    2023

  10. [10]

    Generative adversarial nets,

    I. J. Goodfellow, J. Pouget -Abadie, M. Mirza, B. Xu, D. Warde -Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial nets,” in Advances in Neural Information Processing Systems 27 (NeurIPS), 2014, pp. 2672–2680

    2014

  11. [11]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in Advances in Neural Information Processing Systems 33 (NeurIPS), 2020

    2020

  12. [12]

    How can large language models help humans in design and manufacturing?

    L. Makatura, M. Foshey, B. Wang, F. Hähnlein, P. Ma, B. Deng, M. Tjandrasuwita, A. Spielberg, C. E. Owens, P. Y. Chen, A. Zhao, A. Zhu, W. J. Norton, E. Gu, J. Jacob, Y. Li, A. Schulz, and W. Matusik, “How can large language models help humans in design and manufacturing?” arXiv:2307.14377, 2023

    work page arXiv 2023

  13. [13]

    Deep generative models in engineering design: A review,

    L. Regenwetter, A. Heyrani Nobari, and F. Ahmed, “Deep generative models in engineering design: A review,” Journal of Mechanical Design, vol. 144, no. 7, p. 071704, 2022

    2022

  14. [14]

    Diffusion models beat GANs on topology optimization,

    F. Mazé and F. Ahmed, “Diffusion models beat GANs on topology optimization,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 8, 2023, pp. 9108 –9116

    2023

  15. [15]

    GAN -OPC: Mask optimization with lithography -guided generative adversarial nets,

    H. Yang, S. Li, Y. Ma, B. Yu, and E. F. Y. Young, “GAN -OPC: Mask optimization with lithography -guided generative adversarial nets,” in Proceedings of the 55th Annual Design Automation Conference (DAC), 2018

    2018

  16. [16]

    Wafer map defect pattern classification and image retrieval using convolutional neural network,

    T. Nakazawa and D. V. Kulkarni, “Wafer map defect pattern classification and image retrieval using convolutional neural network,” IEEE Transactions on Semiconductor Manufacturing, vol. 31, no. 2, pp. 309–314, 2018

    2018

  17. [17]

    Deep learning for the design of photonic structures,

    W. Ma, Z. Liu, Z. A. Kudyshev, A. Boltasseva, W. Cai, and Y. Liu, “Deep learning for the design of photonic structures,” Nature Photonics, vol. 15, no. 2, pp. 77–90, 2021

    2021

  18. [18]

    Improvement of TCAD augmented machine learning using autoencoder for semiconductor variation identification and inverse design,

    K. Mehta, S. S. Raju, M. Xiao, B. Wang, Y. Zhang, and H. Y. Wong, “Improvement of TCAD augmented machine learning using autoencoder for semiconductor variation identification and inverse design,” IEEE Access, vol. 8, pp. 143519–143529, 2020

    2020

  19. [19]

    Physics -informed machine learning,

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics -informed machine learning,” Nature Reviews Physics, vol. 3, no. 6, pp. 422–440, 2021

    2021

  20. [20]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” Journal of Computational Physics, vol. 378, pp. 686–707, 2019

    2019

  21. [21]

    A physics -informed diffusion model for high -fidelity flow field reconstruction,

    D. Shu, Z. Li, and A. Barati Farimani, “A physics -informed diffusion model for high -fidelity flow field reconstruction,” Journal of Computational Physics, vol. 478, p. 111972, 2023

    2023

  22. [22]

    MxDiffusion: A physics-aware Maxwell’s law-guided diffusion model strategy for inverse photonic metasurface design,

    S. Mondal, T. Park, S. Biswas, A. X. Wang, and W. Cai, “MxDiffusion: A physics-aware Maxwell’s law-guided diffusion model strategy for inverse photonic metasurface design,” Nano Letters, vol. 26, no. 14, pp. 4897–4905, 2026

    2026

  23. [23]

    Inverse lithography physics -informed deep neural level set for mask optimization,

    X. -Y. Ma and S. Hao, “Inverse lithography physics -informed deep neural level set for mask optimization,” Applied Optics, vol. 62, no. 33, pp. 8769–8779, 2023

    2023

  24. [24]

    DiffTaichi: Differentiable programming for physical simulation,

    Y. Hu, L. Anderson, T. -M. Li, Q. Sun, N. Carr, J. Ragan -Kelley, and F. Durand, “DiffTaichi: Differentiable programming for physical simulation,” in International Conference on Learning Representations (ICLR), 2020

    2020

  25. [25]

    Physical modeling of graphene nanoribbon field effect transistor using non -equilibrium Green function approach for integrated circuit design,

    Y. M. M. Banadaki, “Physical modeling of graphene nanoribbon field effect transistor using non -equilibrium Green function approach for integrated circuit design,” Ph.D. dissertation, Louisiana State University, 2016

    2016

  26. [26]

    A novel graphene nanoribbon field effect transistor for integrated circuit design,

    Y. M. Banadaki and A. Srivastava, “A novel graphene nanoribbon field effect transistor for integrated circuit design,” in 2013 IEEE 56th International Midwest Symposium on Circuits and Systems (MWSCAS), pp. 924 – 927, IEEE, 2013

    2013

  27. [27]

    Fourier neural operator for parametric partial differential equations,

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, and A. Anandkumar, “Fourier neural operator for parametric partial differential equations,” in International Conference on Learning Representations (ICLR), 2021

    2021

  28. [28]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators,

    L. Lu, P. Jin, G. Pang, Z. Zhang, and G. E. Karniadakis, “Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators,” Nature Machine Intelligence, vol. 3, no. 3, pp. 218–229, 2021

    2021

  29. [29]

    Hamiltonian neural networks,

    S. Greydanus, M. Dzamba, and J. Yosinski, “Hamiltonian neural networks,” in Advances in Neural Information Processing Systems 32 (NeurIPS), 2019

    2019

  30. [30]

    ChipNeMo: Domain-Adapted LLMs for Chip Design

    M. Liu, T. -D. Ene, R. Kirby, C. Cheng, N. Pinckney, R. Liang, et al., “ChipNeMo: Domain -adapted LLMs for chip design,” arXiv:2311.00176, 2023. Physics-informed generative AI for semiconductor manufacturing 14

    work page Pith review arXiv 2023

  31. [31]

    Inverse design in nanophotonics,

    S. Molesky, Z. Lin, A. Y. Piggott, W. Jin, J. Vučković, and A. W. Rodriguez, “Inverse design in nanophotonics,” Nature Photonics, vol. 12, no. 11, pp. 659–670, 2018

    2018

  32. [32]

    Physics-guided and fabrication-aware inverse design of photonic devices using diffusion models,

    D. Seo, S. Um, S. Lee, J. C. Ye, and H. Chung, “Physics-guided and fabrication-aware inverse design of photonic devices using diffusion models,” ACS Photonics, 2026, doi: 10.1021/acsphotonics.5c00993

    work page doi:10.1021/acsphotonics.5c00993 2026

  33. [33]

    EDMNet: Unveiling the power of machine learning in regression modeling of powder mixed-EDM,

    M. A. Ilani and Y. M. Banad, “EDMNet: Unveiling the power of machine learning in regression modeling of powder mixed-EDM,” The International Journal of Advanced Manufacturing Technology, vol. 135, no. 5, pp. 2555–2570, 2024

    2024

  34. [34]

    LithoBench: Benchmarking AI computational lithography for semiconductor manufacturing,

    S. Zheng, H. Yang, B. Zhu, B. Yu, and M. D. F. Wong, “LithoBench: Benchmarking AI computational lithography for semiconductor manufacturing,” in Advances in Neural Information Processing Systems 36 (NeurIPS), Datasets and Benchmarks Track, 2023

    2023

  35. [35]

    Closed -loop optimization of fast -charging protocols for batteries with machine learning,

    P. M. Attia, A. Grover, N. Jin, K. A. Severson, et al., “Closed -loop optimization of fast -charging protocols for batteries with machine learning,” Nature, vol. 578, no. 7795, pp. 397 –402, 2020

    2020

  36. [36]

    An autonomous laboratory for the accelerated synthesis of novel materials,

    N. J. Szymanski, B. Rendy, Y. Fei, R. E. Kumar, et al., “An autonomous laboratory for the accelerated synthesis of novel materials,” Nature, vol. 624, no. 7990, pp. 86–91, 2023

    2023

  37. [37]

    On -the-fly closed-loop materials discovery via Bayesian active learning,

    A. G. Kusne, H. Yu, C. Wu, H. Zhang, et al., “On -the-fly closed-loop materials discovery via Bayesian active learning,” Nature Communications, vol. 11, p. 5966, 2020

    2020

  38. [38]

    Autonomous experimentation systems for materials development: A community perspective,

    E. Stach, B. DeCost, A. G. Kusne, J. Hattrick-Simpers, et al., “Autonomous experimentation systems for materials development: A community perspective,” Matter, vol. 4, no. 9, p. 2702, 2021

    2021

  39. [39]

    A Tutorial on Bayesian Optimization

    P. I. Frazier, “A tutorial on Bayesian optimization,” arXiv:1807.02811, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  40. [40]

    Taking the human out of the loop: A review of Bayesian optimization,

    B. Shahriari, K. Swersky, Z. Wang, R. P. Adams, and N. de Freitas, “Taking the human out of the loop: A review of Bayesian optimization,” Proceedings of the IEEE, vol. 104, no. 1, pp. 148 –175, 2016

    2016

  41. [41]

    R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, 2nd ed. Cambridge, MA, USA: MIT Press, 2018

    2018

  42. [42]

    Characterising the digital twin: A systematic literature review,

    D. Jones, C. Snider, A. Nassehi, J. Yon, and B. Hicks, “Characterising the digital twin: A systematic literature review,” CIRP Journal of Manufacturing Science and Technology, vol. 29, pp. 36 –52, 2020

    2020

  43. [43]

    Score-based generative modeling through stochastic differential equations,

    Y. Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differential equations,” in International Conference on Learning Representations (ICLR), 2021

    2021