Pith. sign in

REVIEW 4 major objections 6 minor 54 references

Few-Shot Test-Time Optimization Without Retraining for Semiconductor Recipe Generation and Beyond

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A five-iteration feedback loop tunes a fixed machine's inputs to new targets without retraining.

desk verdict A plausible two-loop inverse-model idea whose central '5 iterations' claim is directly contradicted by its own Table 10 (τ=200), and whose machine-loop gradient assumption is unaddressed. read the letter →

arxiv 2505.16060 v1 pith:7ZPTZCA6 submitted 2025-05-21 cs.LG

Shangding Gu , Donghao Ying , Ming Jin , Yu Joe Lu , Jun Wang , Javad Lavaei , Costas Spanos This is my paper
classification cs.LG
keywords modelfeedbacklearningtest-timeoptimizationrecipegenerationplasmaetchingsemiconductormanufacturinginversefew-shotadaptationchemicalvapordeposition
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

The paper proposes Model Feedback Learning (MFL), a way to make a deployed machine or pre-trained model hit new output targets by changing only its inputs, never its parameters. MFL trains a small reverse network that maps a desired target to the input that produces it, using feedback from a cheap emulator first and from the real machine second. In semiconductor plasma-etching recipe generation the method reaches the target ranges in five machine-loop iterations, compared with at least twenty for a published human-machine method and eighty-four for a senior engineer. It also reports five iterations for chemical vapor deposition and nine for wire bonding. If correct, MFL offers a retraining-free route to re-target costly deployed equipment when process requirements change.

What carries the argument

The central object is the reverse model $R_\theta$, a small neural network (about 7 kB) that maps a desired target output $z'$ to a candidate input $x' = R_\theta(z')$. The mechanism is gradient feedback through a fixed forward model: each iteration computes the output mismatch $z' - E(R_\theta(z'))$ or $z' - M(R_\theta(z'))$ and back-propagates it through the forward model's Jacobian $\partial E/\partial x$ (Loop A) or $\partial M/\partial x$ (Loop B) to update $\theta$. The stability-aware part is a sensitivity check $s_E(x)$ or $s_M(x)$ that switches the learning rate from $\alpha_1$ to the smaller $\alpha_2$ when the forward model is too sensitive at the current input. This division of labor makes few-shot adaptation possible: Loop A does the bulk learning cheaply, and Loop B needs only a handful of real-machine evaluations.

What would settle it

A real test would be to run Loop B on a physical plasma etcher that returns only scalar measurements and provides no input-output gradient. If the recipe cannot reach the target ranges in about five iterations under those conditions, the paper's core efficiency claim collapses.

Watch

Extended reading notes

Core claim

The central claim is that input optimization can be solved by learning the inverse map $R_\theta$ from targets to inputs and refining it in two loops. Loop A pre-trains $R_\theta$ against an emulator $E$, a supervised approximation of the machine, so that most of the learning happens cheaply. Loop B then refines $R_\theta$ against the real machine model $M$, using the gradient of $M$'s outputs with respect to its inputs; a conservative learning-rate rule that switches to a smaller step when model sensitivity is high keeps the updates stable. The paper's empirical headline is that in plasma etching, five machine-loop iterations produce a recipe with etch depth 2255.55 nm, etch rate 109.9 nm/min, mask remaining 358.95 nm, top CD 198.80 nm, $\Delta$CD 10.04, and bow CD 198.52 nm, all inside the target windows. The same few-shot pattern appears in CVD (5 iterations) and wire bonding (9 iterations).

Load-bearing premise

The five-iteration result depends on getting a gradient of the real machine's output with respect to its input settings during Loop B; the paper assumes that gradient exists and is accessible, but does not explain how a physical etcher would provide it.

Editorial extensions

If this is right

  • Deployed semiconductor equipment could be re-targeted to new specifications in roughly five machine evaluations, without retraining or hardware changes.
  • Process development would need far fewer physical trials than the current human-driven practice, reducing time and cost while keeping recipes inside domain constraints.
  • The same two-loop inverse-learning procedure transfers to other manufacturing steps, with five iterations reported for chemical vapor deposition and nine for wire bonding.
  • Stability-aware conservative updates make the optimization robust to target shifts and attack noise, which is relevant when process conditions drift in production.

Reading between the lines

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

  • Editorial inference: the method only needs a differentiable forward model, so any simulator or digital twin with accessible gradients could be re-targeted this way; the paper demonstrates this on a simulator, not on live fab tools.
  • Editorial inference: the five-iteration count covers Loop B only; the emulator still requires supervised training data, so the full deployment cost depends on how that emulator is built and validated.
  • Editorial inference: a natural stress test is to withhold the machine Jacobian and see whether finite-difference or learned-gradient variants preserve the five-iteration behavior; the paper does not address that regime.
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 / 6 minor

Summary. This paper introduces Model Feedback Learning (MFL), a test-time optimization framework that uses a lightweight reverse model R_theta to iteratively search for input values for a deployed machine or pre-trained model, without retraining the forward model. The method has two loops: Loop A pre-trains R_theta against an emulator E, and Loop B refines it against the machine M using gradients that include the machine's Jacobian. Conservative learning rates based on model sensitivity are proposed for stability. Experiments are reported on semiconductor plasma etching recipe generation, chemical vapor deposition, and wire bonding, with the headline claim that MFL generates etching recipes in only five machine-loop iterations, outperforming Bayesian optimization and human engineers (claimed 20 and 84 iterations, respectively). The paper also compares against supervised learning and a random-search baseline, and reports robustness and ablation studies.

Significance. If the empirical claims were fully supported, MFL could be a practically valuable tool for input-recipe optimization in settings where deployed equipment cannot be modified. The paper has real strengths: the convergence result in Appendix B is a standard and correct gradient-descent argument under an L-smoothness assumption, and the algorithm is coherent when run against a differentiable emulator. The appendices provide detailed experimental settings for the etching, CVD, and wire-bonding cases, and the comparison against the LSRS-LR baseline (Algorithm 2 and Table 7) is a concrete, reproducible-looking experiment. However, the paper's central claimed advantage, the 'five iterations' efficiency result, is not supported by the experimental configuration: Algorithm 1 runs a fixed tau iterations in Loop B with no early-stopping condition, and Appendix E sets tau=200. In addition, the deployment mechanism relies on the Jacobian of the machine model without specifying how it is obtained. These are load-bearing issues, not presentation concerns.

major comments (4)
  1. [§5.1, §5.2 vs. Appendix E (Table 10) and Algorithm 1 (Lines 15–26)] The central claim that 'MFL requires only five iterations to generate recipes' is contradicted by the paper's own experimental configuration. Algorithm 1's Loop B runs a fixed τ iterations with no early-stopping condition, and Table 10 sets τ=200 (with τ0=150 and T=1200). The text nowhere explains how '5 iterations' is measured, whether early stopping is applied (it is not in Algorithm 1), or why the reported results would correspond to 5 instead of 200 machine-loop iterations. If Loop B actually ran 200 iterations, the comparison with the 84-iteration senior-engineer baseline in Section 5.2 is inverted (200 > 84), and the headline few-shot efficiency advantage disappears. This needs a concrete, quantitative explanation: a modified algorithm with an explicit stopping rule, a reporting of the actual number of machine evaluations used for Tables 1–4, or a clear distinction between 'iterations to convergence' and 'fixed loop budget.' As written, the paper's primary empirical claim is not derivable from the presented algorithm and settings.
  2. [§4.1, Eq. (4), and Algorithm 1 (Lines 16–25)] Loop B updates the reverse model using gradients that contain the Jacobian ∂M(x)/∂x of the deployed machine model. The paper does not describe how this Jacobian is obtained for real hardware: there is no finite-difference scheme, no surrogate-gradient procedure, no sensor interface, and no discussion of noisy or black-box machine access. If a real plasma etcher cannot supply ∂M(x)/∂x, the machine loop as written cannot run, and the 'five iterations within the machine loop' deployment claim has no mechanism. The authors should either specify a practical way to compute or approximate these Jacobians (e.g., by calibrated finite differences on the physical system, with a cost/iteration budget) or replace the machine Jacobian in Loop B with a computable surrogate and justify that substitution.
  3. [§5.2 (Bayesian optimization comparison)] The claimed superiority over Bayesian optimization is not supported by any reported experiment. The text states that 'Bayesian optimization without human efforts... is unable to generate correct recipes,' but gives no data, no BO configuration, no number of trials, and no evaluation protocol. The Lam Research comparison is taken from [18] rather than measured under the same setup, so the '5 vs. 20 vs. 84 iterations' comparison is not a controlled study. To support the efficiency claim, the paper needs a BO baseline run on the same emulator/machine with the same target ranges and a reported distribution of iteration counts, not just a sentence saying BO fails.
  4. [§5.1, Table 1, and §5.4] The accuracy results are presented as a single set of output values with no standard deviation, no number of seeds, and no repetition. Given that Figure 5 presents MFL as a curve over epochs and the robustness section claims low error under noise, the paper should report mean ± std over multiple seeds for both the final recipe error and the number of iterations required. Without this, 'achieving accurate results' is an anecdotal observation from one run, and it is impossible to assess whether the method reliably meets the semiconductor targets.
minor comments (6)
  1. [Table 4] The 'Process time [s]' row contains two conflicting values, '5.05' and '144.5516', which appears to be a formatting error; please correct it.
  2. [§5.5 and abstract] The statement 'requiring only 5 iterations to converge for CVD and 9 for wire bonding' is not accompanied by a formal convergence criterion; define explicitly what counts as an iteration and what stopping threshold is used.
  3. [§3 and Eq. (2)] The notation is confusing: z' is used both for the target random variable and for the vector of targets in Eq. (2); please use consistent notation (e.g., Z' for the random variable and z' for its realization).
  4. [Figure 9] The y-axis label 'Error' has no units; indicate whether this is mean-squared error, L2 norm, or normalized error.
  5. [§5.3] The 'supervised learning approach' baseline is attributed to [24, 34], but [34] is a plasma etching review rather than a supervised-learning recipe-generation method; please cite a more appropriate baseline or clarify the relationship.
  6. [Algorithm 2] Line 13 says 'Update x with learning rate η' but does not specify the optimizer; the text mentions Adam, so either state that in the algorithm or use a concrete update rule.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: MFL's reported target matching is its optimization objective, and the '5 iterations' contradiction with \tau=200 is a consistency issue, not a circularity.

full rationale

The derivation is not circular. MFL's reverse model R is trained by gradient descent on L(theta) = (1/n') sum ||z'_j - M(R_theta(z'_j))||^2 (Eqs. (2)-(4), Algorithm 1), so Table 1's in-range outputs are the value of the objective being optimized rather than an independent prediction; that is an evaluation limitation, not a reduction of a derived claim to an input. The convergence theorem (Theorem B.1) is a standard smooth-optimization bound and does not presuppose the empirical result. Baselines, target ranges, and engineer iteration counts are taken from the externally published Nature paper [18]; although one current co-author appears there, those numbers are not produced by MFL and are externally falsifiable, so the citation is not a load-bearing self-reference. The paper's most serious problem is internal consistency: Sections 5.1-5.2 claim 'only 5 iterations within the machine loop,' while Algorithm 1 has no early stopping and Appendix E sets tau=200, which would make the published configuration run 200 Loop B iterations. This contradiction undermines the few-shot efficiency claim but is an inconsistency in the evidence, not a circular step. No step of the derivation is equivalent to its inputs by construction.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The central claim rests on unverified premises about emulator fidelity, gradient availability for real hardware, and transfer of simulation results, plus several hand-chosen hyperparameters that are not justified by sensitivity analysis.

free parameters (5)
  • sensitivity threshold δ = 0.9
    Hand-chosen threshold that switches between standard and conservative learning rates; no sensitivity analysis is reported.
  • conservative learning rate α2 = 0.0099 (0.99 × α1)
    Chosen by hand for late-stage stability; no derivation or grid search.
  • training periods T, T0, τ, τ0 = T=1200, T0=1150, τ=200, τ0=150
    Set by hand; the paper's 'five iterations' claim is not reflected in these algorithm settings.
  • domain randomization noise scale = not reported
    Gaussian noise added to emulator inputs to improve robustness, but its standard deviation is not specified.
  • neural network hyperparameters = lr=0.01, hidden=64, MLP, 700 epochs
    Standard choices, but no sensitivity or ablation on these is shown.
assumptions (6)
  • standard math The loss L(θ) is L-Lipschitz smooth and the learning rate satisfies α < 1/L.
    Theorem B.1 uses this to prove convergence to a stationary point; the smoothness is asserted but not verified for the specific networks.
  • domain assumption The emulator E trained on Gaussian-sampled data approximates the machine M well enough in the target region.
    Loop A transfers the reverse model from E to M; if E is inaccurate, the pre-trained initialization is poor.
  • domain assumption The deployed machine M is differentiable or its gradient ∂M/∂x is available for Loop B.
    Algorithm 1 and Eq. (4) require this Jacobian; no finite-difference or derivative-free mechanism is described.
  • domain assumption Target samples z'_j are available before optimization and represent the new environment.
    The loss in Eq. (2) uses target samples; MFL adapts per target group, not to unseen targets.
  • domain assumption Simulation results on a Gaussian-sampled dataset transfer to real semiconductor equipment.
    All experiments use a learned emulator or synthetic data, not a real etcher, CVD tool, or bonder.
  • ad hoc to paper The conservative learning rule (δ threshold, α1/α2) improves robustness without harming convergence.
    This mechanism is heuristic and only illustrated in two figures with no error bars.
invented entities (1)
  • Reverse model Rθ
    purpose: Maps desired outputs Z' to candidate inputs X' and is updated in two loops while the emulator and machine stay fixed.
    An internal neural network component; it makes no falsifiable prediction outside the paper's simulator.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Few-Shot Test-Time Optimization Without Retraining for Semiconductor Recipe Generation and Beyond." pith.science (2026). https://pith.science/paper/7ZPTZCA6

@misc{pith2026250516060,
  author       = {Pith},
  title        = {Pith review of: Few-Shot Test-Time Optimization Without Retraining for Semiconductor Recipe Generation and Beyond},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7ZPTZCA6}},
  note         = {Machine review of arXiv:2505.16060}
}
read the original abstract

We introduce Model Feedback Learning (MFL), a novel test-time optimization framework for optimizing inputs to pre-trained AI models or deployed hardware systems without requiring any retraining of the models or modifications to the hardware. In contrast to existing methods that rely on adjusting model parameters, MFL leverages a lightweight reverse model to iteratively search for optimal inputs, enabling efficient adaptation to new objectives under deployment constraints. This framework is particularly advantageous in real-world settings, such as semiconductor manufacturing recipe generation, where modifying deployed systems is often infeasible or cost-prohibitive. We validate MFL on semiconductor plasma etching tasks, where it achieves target recipe generation in just five iterations, significantly outperforming both Bayesian optimization and human experts. Beyond semiconductor applications, MFL also demonstrates strong performance in chemical processes (e.g., chemical vapor deposition) and electronic systems (e.g., wire bonding), highlighting its broad applicability. Additionally, MFL incorporates stability-aware optimization, enhancing robustness to process variations and surpassing conventional supervised learning and random search methods in high-dimensional control settings. By enabling few-shot adaptation, MFL provides a scalable and efficient paradigm for deploying intelligent control in real-world environments.

Figures

Figures reproduced from arXiv: 2505.16060 by the authors.

Figure 1
Figure 1. Schematic of the semiconductor manufacturing process, showing the incoming photoresist [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of standard supervised learning. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Two-loop training process for the reverse emulator model [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Illustration for the approximation errors of the emulator [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Comparison of MFL and supervised learning in high-dimensional space training, conducted across three random seeds. The y￾axis refers to the output error per epoch in a six-dimensional target space. Exp 1 Exp 2 Exp 3 Exp 4 0 20 40 60 80 100 Values 0.0 0.1 0.0 1 0.01 10 …
Figure 7
Figure 7. Figure 7: Robustness evaluation under target shifts. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Domain randomization ablation: Evaluating the impact of domain randomiza￾tion on model output error across different attack noise levels. 0 5 10 15 20 25 30 35 Epoch 0 5 10 15 20 25 30 Error MFL-without-loop-2 (Ours) MFL (Ours) [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 10
Figure 10. Figure 10: MFL without pre-training is compared to MFL with pre-training (a) and the supervised [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 42 canonical work pages

  1. [18]

    Human–machine collaboration for improving semiconductor process development.Nature, 616(7958):707–711, 2023

    Keren J Kanarik, Wojciech T Osowiecki, Yu Lu, Dipongkar Talukder, Niklas Roschewsky, Sae Na Park, Mattan Kamon, David M Fried, and Richard A Gottscho. Human–machine collaboration for improving semiconductor process development.Nature, 616(7958):707–711, 2023

    work page 2023

  2. [1]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    arXiv 2023

  3. [2]

    The surprising effectiveness of test-time training for few-shot learning

    Ekin Akyürek, Mehul Damani, Adam Zweiger, Linlu Qiu, Han Guo, Jyothish Pari, Yoon Kim, and Jacob Andreas. The surprising effectiveness of test-time training for few-shot learning. arXiv preprint arXiv:2411.07279, 2024

    arXiv 2024

  4. [3]

    Analyzing inverse problems with invertible neural networks

    Lynton Ardizzone, Jakob Kruse, Carsten Rother, and Ullrich Köthe. Analyzing inverse problems with invertible neural networks. InInternational Conference on Learning Representations, 2019

    work page 2019

  5. [4]

    Mt3: Meta test- time training for self-supervised test-time adaption

    Alexander Bartler, Andre Bühler, Felix Wiewel, Mario Döbler, and Bin Yang. Mt3: Meta test- time training for self-supervised test-time adaption. InInternational Conference on Artificial Intelligence and Statistics, pages 3080–3090. PMLR, 2022

    work page 2022

  6. [5]

    The novel stress simulation method for contemporary dram capacitor arrays

    Kyu-Baik Chang, Yun Young Kim, Jiwoong Sue, Hojoon Lee, Won-Young Chung, Keun-Ho Lee, Young-Kwan Park, EunSeung Jung, and Ilsub Chung. The novel stress simulation method for contemporary dram capacitor arrays. In2013 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD), pages 424–427. IEEE, 2013. 12

    work page 2013

  7. [6]

    Springer, 2014

    Preeti S Chauhan, Anupam Choubey, ZhaoWei Zhong, Michael G Pecht, Preeti S Chauhan, Anupam Choubey, ZhaoWei Zhong, and Michael G Pecht.Copper wire bonding. Springer, 2014

    work page 2014

  8. [7]

    Etch model based on machine learning

    Rui Chen, Haoru Hu, Xiaoting Li, Ying Chen, SU Xiaojing, Lisong Dong, Lei Qu, Chen Li, Jiang Yan, and Yayi Wei. Etch model based on machine learning. In2020 China Semiconductor Technology International Conference (CSTIC), pages 1–4. IEEE, 2020

    work page 2020

Show all 54 references
  1. [8]

    Exploring machine learning for semiconductor process optimization: a systematic review.IEEE Transactions on Artificial Intelligence, 2024

    Ying-Lin Chen, Sara Sacchi, Bappaditya Dey, Victor Blanco, Sandip Halder, Philippe Leray, and Stefan De Gendt. Exploring machine learning for semiconductor process optimization: a systematic review.IEEE Transactions on Artificial Intelligence, 2024

    2024

  2. [9]

    A tutorial on bayesian optimization.arXiv preprint arXiv:1807.02811, 2018

    Peter I Frazier. A tutorial on bayesian optimization.arXiv preprint arXiv:1807.02811, 2018

    2018 arXiv

  3. [10]

    Test-time training with masked autoencoders.Advances in Neural Information Processing Systems, 35:29374–29385, 2022

    Yossi Gandelsman, Yu Sun, Xinlei Chen, and Alexei Efros. Test-time training with masked autoencoders.Advances in Neural Information Processing Systems, 35:29374–29385, 2022

    2022

  4. [11]

    Internal model control

    Carlos E Garcia and Manfred Morari. Internal model control. a unifying review and some new results.Industrial & Engineering Chemistry Process Design and Development, 21(2):308–323, 1982

    1982

  5. [12]

    A survey of deep learning techniques for autonomous driving.Journal of field robotics, 37(3):362–386, 2020

    Sorin Grigorescu, Bogdan Trasnea, Tiberiu Cocias, and Gigel Macesanu. A survey of deep learning techniques for autonomous driving.Journal of field robotics, 37(3):362–386, 2020

    2020

  6. [13]

    Safe multi-agent reinforcement learning for multi-robot control.Artificial Intelligence, 319:103905, 2023

    Shangding Gu, Jakub Grudzien Kuba, Yuanpei Chen, Yali Du, Long Yang, Alois Knoll, and Yaodong Yang. Safe multi-agent reinforcement learning for multi-robot control.Artificial Intelligence, 319:103905, 2023

    2023

  7. [14]

    Fatemeh Hosseinpour, Ibrahim Ahmed, Piero Baraldi, Enrico Zio, Mehdi Behzad, and Horst Lewitschnig. A novel methodology based on long short-term memory stacked autoencoders for unsupervised detection of abnormal working conditions in semiconductor manufacturing systems.Proceed...

    2024

  8. [15]

    Bridging tcad and ai: Its application to semiconductor design.IEEE Transactions on Electron Devices, 68(11):5364– 5371, 2021

    Changwook Jeong, Sanghoon Myung, In Huh, Byungseon Choi, Jinwoo Kim, Hyunjae Jang, Hojoon Lee, Daeyoung Park, Kyuhun Lee, Wonik Jang, et al. Bridging tcad and ai: Its application to semiconductor design.IEEE Transactions on Electron Devices, 68(11):5364– 5371, 2021

    2021

  9. [16]

    Royal society of chemistry, 2009

    Anthony C Jones and Michael L Hitchman.Chemical vapour deposition: precursors, processes and applications. Royal society of chemistry, 2009

    2009

  10. [17]

    Highly accurate protein structure prediction with alphafold.nature, 596(7873):583–589, 2021

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ron- neberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin Žídek, Anna Potapenko, et al. Highly accurate protein structure prediction with alphafold.nature, 596(7873):583–589, 2021

    2021

  11. [19]

    Ai chips: what they are and why they matter.Center for Security and Emerging Technology, 2020

    Saif M Khan and Alexander Mann. Ai chips: what they are and why they matter.Center for Security and Emerging Technology, 2020

    2020

  12. [20]

    Prediction of silicon oxynitride plasma etching using a generalized regression neural network.Journal of applied physics, 98(3), 2005

    Byungwhan Kim and Byung Teak Lee. Prediction of silicon oxynitride plasma etching using a generalized regression neural network.Journal of applied physics, 98(3), 2005

    2005

  13. [21]

    Application of plasma information-based virtual metrology (pi-vm) for etching in c4f8/ar/o2 plasma.IEEE Transactions on Semiconductor Manufacturing, 2024

    Gwanjoong Kim, Ji-Won Kwon, Ingyu Lee, Hwiwon Seo, Jong-Bae Park, Jong-Hyun Shin, and Gon-Ho Kim. Application of plasma information-based virtual metrology (pi-vm) for etching in c4f8/ar/o2 plasma.IEEE Transactions on Semiconductor Manufacturing, 2024

    2024

  14. [22]

    A review of robot learning for manip- ulation: Challenges, representations, and algorithms.Journal of machine learning research, 22(30):1–82, 2021

    Oliver Kroemer, Scott Niekum, and George Konidaris. A review of robot learning for manip- ulation: Challenges, representations, and algorithms.Journal of machine learning research, 22(30):1–82, 2021. 13

    2021

  15. [23]

    Robohive: A unified framework for robot learning.Advances in Neural Information Processing Systems, 36, 2024

    Vikash Kumar, Rutav Shah, Gaoyue Zhou, Vincent Moens, Vittorio Caggiano, Abhishek Gupta, and Aravind Rajeswaran. Robohive: A unified framework for robot learning.Advances in Neural Information Processing Systems, 36, 2024

    2024

  16. [24]

    Deep learning.nature, 521(7553):436–444, 2015

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning.nature, 521(7553):436–444, 2015

    2015

  17. [25]

    Nett: Solving inverse problems with deep neural networks.arXiv preprint arXiv:1803.00092, 2018

    Housen Li, Johannes Schwab, Stephan Antholzer, and Markus Haltmeier. Nett: Solving inverse problems with deep neural networks.arXiv preprint arXiv:1803.00092, 2018

    2018 arXiv

  18. [26]

    Brief introduction of back propagation (bp) neural network algorithm and its improvement

    Jing Li, Ji-hang Cheng, Jing-yuan Shi, and Fei Huang. Brief introduction of back propagation (bp) neural network algorithm and its improvement. InAdvances in Computer Science and Information Engineering: Volume 2, pages 553–558. Springer, 2012

    2012

  19. [27]

    Ttt++: When does self-supervised test-time training fail or thrive?Advances in Neural Information Processing Systems, 34:21808–21820, 2021

    Yuejiang Liu, Parth Kothari, Bastien Van Delft, Baptiste Bellot-Gurlet, Taylor Mordan, and Alexandre Alahi. Ttt++: When does self-supervised test-time training fail or thrive?Advances in Neural Information Processing Systems, 34:21808–21820, 2021

    2021

  20. [28]

    A fast and manufacture-friendly optical proximity correction based on machine learning.Microelectronic Engineering, 168:15–26, 2017

    Xu Ma, Shangliang Jiang, Jie Wang, Bingliang Wu, Zhiyang Song, and Yanqiu Li. A fast and manufacture-friendly optical proximity correction based on machine learning.Microelectronic Engineering, 168:15–26, 2017

    2017

  21. [29]

    Fast pixel-based optical proximity correction based on nonparametric kernel regression.Journal of Micro/Nanolithogra- phy, MEMS, and MOEMS, 13(4):043007–043007, 2014

    Xu Ma, Bingliang Wu, Zhiyang Song, Shangliang Jiang, and Yanqiu Li. Fast pixel-based optical proximity correction based on nonparametric kernel regression.Journal of Micro/Nanolithogra- phy, MEMS, and MOEMS, 13(4):043007–043007, 2014

    2014

  22. [30]

    John Wiley & Sons, 2006

    Gary S May and Costas J Spanos.Fundamentals of semiconductor manufacturing and process control. John Wiley & Sons, 2006

    2006

  23. [31]

    Enabling dfm and apc strategies at the 32 nm technology node

    Kevin M Monahan. Enabling dfm and apc strategies at the 32 nm technology node. InISSM 2005, IEEE International Symposium on Semiconductor Manufacturing, 2005., pages 398–401. IEEE, 2005

    2005

  24. [32]

    A novel approach for semiconductor etching process with inductive biases.arXiv preprint arXiv:2104.02468, 2021

    Sanghoon Myung, Hyunjae Jang, Byungseon Choi, Jisu Ryu, Hyuk Kim, Sang Wuk Park, Changwook Jeong, and Dae Sin Kim. A novel approach for semiconductor etching process with inductive biases.arXiv preprint arXiv:2104.02468, 2021

    2021 arXiv

  25. [33]

    Modeling simplification for thermal mechanical analysis of high density chip-to-substrate connections.Journal of electronic packaging, 133(4), 2011

    Ping Nicole An and Paul A Kohl. Modeling simplification for thermal mechanical analysis of high density chip-to-substrate connections.Journal of electronic packaging, 133(4), 2011

    2011

  26. [34]

    Future of plasma etching for microelectronics: Challenges and opportunities.Journal of Vacuum Science & Technology B, 42(4), 2024

    Gottlieb S Oehrlein, Stephan M Brandstadter, Robert L Bruce, Jane P Chang, Jessica C DeMott, Vincent M Donnelly, Rémi Dussart, Andreas Fischer, Richard A Gottscho, Satoshi Hamaguchi, et al. Future of plasma etching for microelectronics: Challenges and opportunities.Journal of ...

    2024

  27. [35]

    Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730...

    2022

  28. [36]

    Damage-free plasma source for atomic-scale processing

    Junyoung Park, Jiwon Jung, Min-Seok Kim, Chang-Min Lim, Jung-Eun Choi, Nayeon Kim, Ju-Ho Kim, and Chin-Wook Chung. Damage-free plasma source for atomic-scale processing. Nano Letters, 24(37):11462–11468, 2024

    2024

  29. [37]

    Plasma heating characterization of the large area inductively coupled plasma etchers with the plasma information for managing the mass production.Physics of Plasmas, 31(7), 2024

    Seolhye Park, Yoona Park, Jaegu Seong, Haneul Lee, Namjae Bae, Ki-baek Roh, Rabul Seo, Bongsub Song, and Gon-Ho Kim. Plasma heating characterization of the large area inductively coupled plasma etchers with the plasma information for managing the mass production.Physics of Pla...

    2024

  30. [38]

    Perspectives on artificial intelligence for plasma-assisted manufacturing in semiconductor industry

    Kapil Sawlani and Ali Mesbah. Perspectives on artificial intelligence for plasma-assisted manufacturing in semiconductor industry. InArtificial Intelligence in Manufacturing, pages 97–138. Elsevier, 2024. 14

    2024

  31. [39]

    Advanced plasma processing: etching, deposition, and wafer bonding techniques for semiconductor applications.Semiconductor technologies, 27:81–105, 2010

    Michael Shearn, Xiankai Sun, M David Henry, Amnon Yariv, and Axel Scherer. Advanced plasma processing: etching, deposition, and wafer bonding techniques for semiconductor applications.Semiconductor technologies, 27:81–105, 2010

    2010

  32. [40]

    Three-dimensional integrated circuit (3d ic) key technol- ogy: Through-silicon via (tsv).Nanoscale research letters, 12:1–9, 2017

    Wen-Wei Shen and Kuan-Neng Chen. Three-dimensional integrated circuit (3d ic) key technol- ogy: Through-silicon via (tsv).Nanoscale research letters, 12:1–9, 2017

    2017

  33. [41]

    Machine learning-guided etch proximity correction.IEEE Transactions on Semiconductor Manufacturing, 30(1):1–7, 2016

    Seongbo Shim and Youngsoo Shin. Machine learning-guided etch proximity correction.IEEE Transactions on Semiconductor Manufacturing, 30(1):1–7, 2016

    2016

  34. [42]

    Mas- tering the game of go with deep neural networks and tree search.nature, 529(7587):484–489, 2016

    David Silver, Aja Huang, Chris J Maddison, Arthur Guez, Laurent Sifre, George Van Den Driess- che, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lanctot, et al. Mas- tering the game of go with deep neural networks and tree search.nature, 529(7587):484–489, 2016

    2016

  35. [43]

    Test: Test-time self- training under distribution shift

    Samarth Sinha, Peter Gehler, Francesco Locatello, and Bernt Schiele. Test: Test-time self- training under distribution shift. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 2759–2769, 2023

    2023

  36. [44]

    Chemical vapour deposition.Nature Reviews Methods Primers, 1(1):5, 2021

    Luzhao Sun, Guowen Yuan, Libo Gao, Jieun Yang, Manish Chhowalla, Meysam Heydari Gharahcheshmeh, Karen K Gleason, Yong Seok Choi, Byung Hee Hong, and Zhongfan Liu. Chemical vapour deposition.Nature Reviews Methods Primers, 1(1):5, 2021

    2021

  37. [45]

    Test-time training with self-supervision for generalization under distribution shifts

    Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei Efros, and Moritz Hardt. Test-time training with self-supervision for generalization under distribution shifts. InInternational conference on machine learning, pages 9229–9248. PMLR, 2020

    2020

  38. [46]

    Modeling of layout-dependent stress effect in cmos design

    Chi-Chao Wang, Wei Zhao, Frank Liu, Min Chen, and Yu Cao. Modeling of layout-dependent stress effect in cmos design. InProceedings of the 2009 International Conference on Computer- Aided Design, pages 513–520, 2009

    2009

  39. [47]

    Tent: Fully test-time adaptation by entropy minimization

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. InInternational Conference on Learning Representations, 2021

    2021

  40. [48]

    Data-driven machine learning predictor model for optimal operation of a thermal atomic layer etching reactor.Industrial & Engineering Chemistry Research, 63(45):19693–19706, 2024

    Henrik Wang, Feiyang Ou, Julius Suherman, Matthew Tom, Gerassimos Orkoulas, and Pana- giotis D Christofides. Data-driven machine learning predictor model for optimal operation of a thermal atomic layer etching reactor.Industrial & Engineering Chemistry Research, 63(45):19693–1...

    2024

  41. [49]

    Attention is all you need

    A Waswani, N Shazeer, N Parmar, J Uszkoreit, L Jones, A Gomez, L Kaiser, and I Polosukhin. Attention is all you need. InNIPS, 2017

    2017

  42. [50]

    SPIE press, 2009

    Yayi Wei and Robert L Brainard.Advanced processes for 193-nm immersion lithography, volume 189. SPIE press, 2009

    2009

  43. [51]

    Multiscale modeling and recurrent neural network based optimization of a plasma etch process.Processes, 9(1):151, 2021

    Tianqi Xiao and Dong Ni. Multiscale modeling and recurrent neural network based optimization of a plasma etch process.Processes, 9(1):151, 2021

    2021

  44. [52]

    Etching process prediction based on cascade recurrent neural network.Engineering Applications of Artificial Intelligence, 139:109590, 2025

    Zhenjie Yao, Ziyi Hu, Panpan Lai, Fengling Qin, Wenrui Wang, Zhicheng Wu, Lingfei Wang, Hua Shao, Yongfu Li, Zhiqiang Li, et al. Etching process prediction based on cascade recurrent neural network.Engineering Applications of Artificial Intelligence, 139:109590, 2025

    2025

  45. [53]

    Autonomous driving system: A comprehensive survey.Expert Systems with Applications, 242:122836, 2024

    Jingyuan Zhao, Wenyi Zhao, Bo Deng, Zhenghong Wang, Feng Zhang, Wenxiang Zheng, Wanke Cao, Jinrui Nan, Yubo Lian, and Andrew F Burke. Autonomous driving system: A comprehensive survey.Expert Systems with Applications, 242:122836, 2024

    2024

  46. [54]

    SE" represents senior engineers, 16 and “JE

    ZW Zhong. Overview of wire bonding using copper wire or insulated wire.Microelectronics Reliability, 51(1):4–12, 2011. 15 Appendix A More Details on Plasma Etching Process in Semiconductor Manufacturing Modern computer chips feature complex circuit patterns that conduct electr...

    2011

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.