Efficient nanophotonic devices optimization using deep neural network trained with physics-based transfer learning (PBTL) methodology

Gibaek Kim; Jungho Kim

REVIEW 4 major objections 4 minor 29 references

Efficient nanophotonic devices optimization using deep neural network trained with physics-based transfer learning (PBTL) methodology

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

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

Pith's one-line read PBTL, a physics-informed transfer-learning scheme, improves surrogate prediction accuracy by 0.69%, halves QCL training data needs, and yields 60% more physically valid optimized designs.

desk verdict New region-split transfer learning trick for QCL surrogate models, with a real but overclaimed data-efficiency gain; worth refereeing after cleaning up the 50% claim. read the letter →

arxiv 2506.10418 v1 pith:PV2UFAN6 submitted 2025-06-12 physics.optics physics.comp-ph

Gibaek Kim , Jungho Kim This is my paper · ORCID

classification physics.opticsphysics.comp-ph

keywords physics-basedtransferlearningquantumcascadelasersurrogatemodeldeepneuralnetworkgeneticalgorithmnanophotonicdesignoptimizationSchrödingerequationmodalgain

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 a neural-network surrogate that replaces numerical Schrödinger-equation solves in quantum cascade laser (QCL) optimization, together with a transfer-learning scheme called PBTL (physics-based transfer learning) to train that surrogate from less data. The central claim is that splitting the full-structure network's first hidden layer into an active-region block and an injector block, then seeding only the active block with weights from a separately trained active-region network, beats both direct learning and generic transfer learning: a 0.69% gain in prediction accuracy, a 50% reduction in required training data, and a 60% improvement in the fraction of physically valid structures among the top optimized candidates. This matters because the surrogate makes a 50-generation, 2,000-candidate genetic-algorithm optimization run in seconds instead of tens of hours, an 80,000-fold speed-up, while producing verified QCL designs that are novel relative to the training set. The authors frame PBTL as a general recipe for domains where feature importance is imbalanced and simulation data are expensive.

What carries the argument

The central mechanism is the PBTL network architecture: the first hidden layer of the DNN-TP is split into two blocks, one receiving the 8 active-region thickness inputs and the other the 16 injector inputs, and the active block is initialized with the pretrained first-hidden-layer weights of the DNN-CP. This makes the reduced latent space of the active-region model physically correspond to a sub-network of the full model, so knowledge transfer does not rely on matching full input dimensions as in generic transfer learning. The DNN-CP, trained only on active-region variations, provides the pretrained block; UMAP projections of the two training sets and weight-distribution histograms are used in the paper to argue that the shared Schrödinger-equation origin makes the transfer beneficial, while the misaligned weight transfer in generic transfer learning degrades validation performance.

What would settle it

Train DNN-TP with PBTL and with direct learning across dataset sizes up to and beyond 150,000 samples; if direct learning catches up or surpasses PBTL before the PBTL model's claimed data advantage, the transfer is a cheaper initialization rather than a physics-based generalization gain. A sharper test: permute the mapping between active-layer input positions and the pretrained active-block weights and retrain; PBTL predicts the accuracy gain disappears, while a generic initialization-benefit explanation predicts it would not.

Watch

Extended reading notes

Core claim

The discovery is that a deep neural network trained on the physically dominant active region of a QCL can be transferred to a network for the full structure by architectural correspondence rather than by generic weight copying. The authors build a DNN core predictor (DNN-CP) on the 8 active-region layer thicknesses with injection layers fixed, using 35,000 samples; then a DNN total predictor (DNN-TP) on all 24 layer thicknesses with 100,000 samples. In PBTL, the first hidden layer of the DNN-TP is divided into an active block and an injector block, and only the active block receives the pre-trained weights from the DNN-CP. The paper reports that this split transfer improves average R2 from 0.900 (direct learning) to 0.906, a 0.69% improvement, that matching that accuracy by direct learning would need about 150,000 samples, and that the PBTL-trained surrogate achieves a 60% higher pass@128 ratio during genetic-algorithm optimization. The authors take this as evidence that the reduced latent space of the active-region model encodes QCL operating principles that survive recombination with the injector block.

Load-bearing premise

The load-bearing premise is that the active-region-only network's learned representations remain valid when inserted into the full 24-feature network, because both are assumed to share the same underlying Schrödinger-equation latent space; if the first-layer features interact strongly across the active and injector blocks, the transferred weights could misalign and the accuracy gain would vanish.

Editorial extensions

If this is right

Surrogate-assisted genetic algorithm with the PBTL-trained DNN-TP completes a 50-generation, 2,000-candidate QCL optimization in seconds, an 80,000-fold speed-up over the numerical solver, making large-scale design-space exploration practical.
PBTL's accuracy gain is equivalent to roughly 50,000 fewer training samples (about 50% of the DNN-TP's 100,000-sample dataset), which directly lowers the dominant cost of surrogate training.
The 60% improvement in pass@128 means more of the top-ranked optimized structures survive physics-based validation, which is the bottleneck for trusting surrogate-guided design.
Optimized structures sit at average L1 distance ~43 from their nearest training samples (minimum ~30), indicating the surrogate plus GA generates designs beyond the training distribution rather than retrieving memorized ones.
The authors claim the same PBTL-plus-GA framework transfers to other nanophotonics or engineering domains with imbalanced feature importance and expensive data generation.

Reading between the lines

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

A testable extension the paper does not run: isolate PBTL's benefit by comparing the active-block split with randomly initialized active blocks of the same architecture; if the accuracy gain vanishes, the pretrained weights are doing the work, and if it persists, the gain is just architectural.
The paper does not vary surrogate error deliberately; a controlled study that perturbs DNN-TP predictions by increasing amounts would show how steeply pass@128 falls with surrogate error, quantifying why a 0.69% accuracy gain matters so much for optimization.
PBTL's logic suggests a general recipe for multi-region devices: train small surrogate networks on each physically dominant region and seed corresponding blocks of a full-structure network; the paper demonstrates only the active/injector split, but the same Schrödinger-equation-sharing rationale would apply to other layered quantum devices.
The reported 80,000x speed-up compares surrogate evaluation to serial numerical solves; an end-to-end comparison that includes dataset generation, training, and validation would give a more complete picture of the framework's practical advantage, which the paper does not provide.

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.

Desk Editor's Note

read the letter

Read arXiv:2506.10418. The genuinely new thing here is PBTL: splitting the first hidden layer of the 24-input total predictor into active and injector blocks, and transferring pretrained weights from an 8-input active-region-only model to just the active block. That's a physically motivated variant of transfer learning that I haven't seen in the photonics surrogate literature. The authors also do the right thing externally: they validate optimized structures with the full Schrödinger solver, report pass@128 ratios, and show L1 distances from the training set to argue novelty. The 80,000x speedup is plausible, since evaluating 2000 candidates per generation with a DNN is milliseconds versus minutes for the numerical solver.

Where it gets soft is the data-efficiency headline. The abstract and conclusion say PBTL gives a 50% reduction in training data (50k samples). But Section 3.1 says the PBTL accuracy at 100k would require ~150k samples with DL—that's a 33% reduction. Those numbers cannot both be right. And the only supporting figure, S1, shows DL learning curves only; there is no PBTL learning curve showing 50k matches 100k DL. So the 50% claim is unsupported and internally inconsistent. This is a load-bearing advertised benefit, so it matters.

Also: Table 1 shows small R2 gains (0.69% average) from a single run, no seeds, no error bars. The pass@128 improvement of 60% is striking but likewise no variance estimate. The dataset heavily samples perturbed validated designs, which helps the surrogate but weakens claims about exploring the full 10^15 design space. These are fixable with more experiments, not fatal flaws.

The math and modeling chain look solid: effective two-band Schrödinger solver, LO-phonon and IFR scattering, standard. The citation pattern is fine—they cite prior transfer learning in photonics (Qu et al., Fan et al.) and QCL GA work (Franckié & Faist, Bismuto). No code or data released, but data available on request.

Verdict: the core idea is worth taking seriously and the paper deserves a proper peer review, but the authors need to correct the data-efficiency numbers and add uncertainty quantification before publication. A careful referee should catch the 50% vs 33% inconsistency immediately.

For you: maybe worth a reading group if you're interested in surrogate modeling in photonics. I'd cite it when discussing transfer learning variants for expensive simulators, but with a caveat about the overclaim.

Referee Report

4 major / 4 minor

Summary. The manuscript proposes a deep neural network surrogate modeling framework for optimizing mid-infrared quantum cascade lasers, combining a physics-based transfer learning scheme (PBTL) with a scalarized multi-objective genetic algorithm. The authors train a core predictor (DNN-CP) on active-region-only structures and transfer its weights to the active-region block of a total predictor (DNN-TP) for full 24-layer structures, comparing against direct learning (DL) and generic transfer learning (GTL). They report a 0.69% improvement in average R², an 80,000-fold speedup, a 60% improvement in pass@128, and discovery of novel designs with an average L1 distance of about 43 from the training data.

Significance. If the reported gains are reproducible, the framework is a useful contribution to surrogate-based photonic design: it demonstrates a physically motivated parameter-transfer strategy in a realistic QCL model, quantifies the speed advantage of replacing Schrödinger-equation solves with a DNN, and provides a clear pass@128 validation protocol. The paper is also honest about the hypothesis-like status of the transfer mechanism and about the small magnitude of the accuracy gain. However, the central data-efficiency claim is internally inconsistent and the headline numbers lack uncertainty estimates, so the significance is currently conditional.

major comments (4)

[Section 3.1 and Conclusion] The data-reduction claim is internally inconsistent and unsupported. Section 3.1 states that the PBTL accuracy level 'would require training with approximately 150,000 samples if DL is used,' which corresponds to a 33% reduction from the 100,000-sample DL baseline, while the Conclusion states 'a 50% reduction in required training data (50,000 samples).' These two statements cannot both be correct. Supplementary Figure S1 plots only the direct-learning learning curve and does not show a PBTL model trained on 50,000 samples, so the 50% reduction claim has no supporting evidence. Please reconcile the numbers and add PBTL learning-curve data.
[Table 1 and Section 3.3.2] The headline results lack uncertainty quantification. Table 1 and the pass@128 comparison are based on a single training run and a single GA run. The average R² improvement of 0.69% (from 0.900 to 0.906) is small relative to typical run-to-run variation in DNN training, and most per-feature differences are below 1%. Without multiple seeds, error bars, and a statistical test, the claim that PBTL outperforms DL is not established. Please report mean plus/minus standard deviation over at least 5-10 retraining runs and, for the GA evaluation, repeated optimization runs.
[Section 3.1 and Figure 2(d)] The PBTL architecture changes the first hidden layer of the DNN-TP by splitting it into active and injector blocks, so the comparison with DL and GTL conflates the architectural change with the transfer of physics-based knowledge. Section 3.1 supports the transfer hypothesis only with UMAP projections and weight histograms; no control experiment is provided—for example, transferring weights trained on random labels, or shuffling the correspondence between the active block and the DNN-CP weights. Please add such controls to demonstrate that the benefit comes from physical correspondence rather than from the split-layer architecture or from regularization.
[Equation (3)] Equation (3) is not a valid Gaussian objective as written. The exponent is written as −0.5(λ−λ_target)/(0.1)^2, which lacks the square on (λ−λ_target), and the prefactor 1/(0.4√(2π)) does not correspond to σ=0.1. If this formula was actually used in the GA, the wavelength objective and its relative weighting may be incorrect. Please correct the formula and confirm that the implemented objective matches the intended Gaussian form; if it was a typographical error, state so explicitly.

minor comments (4)

[Figure 6 callouts in Sections 3.3.1 and 3.3.2] The figure references appear to be swapped: Section 3.3.1 refers to 'Figure 6(a)' for the latent-space convergence and 'Figure 6(b)' for the population-size effect, but the caption labels these as (c) and (a) respectively; Section 3.3.2 refers to 'Figure 6(c)' for the pass@128 ratio, which is panel (b). Please fix the callouts.
[Abstract and Section 3.1] The abstract says '0.69 percentage increase,' while Table 1 reports a relative improvement of 0.69% in average R². These are different quantities; use consistent terminology throughout.
[Section 1 and Section 2.2] The paper mentions 'over 10¹⁵ possible configurations' and later an 'estimated 10¹⁷ fold increase in the size of the design space.' Please clarify the exact counts and how these estimates are derived.
[Data availability statement] For reproducibility, please consider making the training datasets and the trained models available in a public repository rather than 'upon reasonable request,' as the surrogate and GA results are central to the claims.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: PBTL claims are empirical comparisons against a numerical simulator, not by-construction reductions.

full rationale

The claimed derivation chain is empirical rather than definitional. The surrogate models (DNN-CP and DNN-TP) are trained on labels produced by a numerical Schrodinger-equation solver (Algorithm 1, SI Section 1), and the PBTL comparison is a head-to-head validation against DL and GTL on a held-out test set (Table 1), not a quantity fitted into the claim. The transfer-learning premise, that active-region representations transfer to the corresponding block of the 24-feature model, is explicitly stated as a hypothesis in Section 3.1 and then tested via validation loss, weight distributions, and optimization pass@128; none of these outcomes is an input by construction. The 80,000x speedup and 60% pass@128 improvement are benchmarked against the same physics simulator, which is an external ground truth for this paper, not a self-referential optimization target. The paper contains one self-citation, [14] for the effective two-band model, but the model is re-derived in the SI with standard scattering formulas, so it constitutes real evidence rather than a uniqueness theorem or an unverified ansatz. The main caveats are correctness issues rather than circularity: the '50% reduction in training data' statement is internally inconsistent (Section 3.1 says PBTL at 100k matches DL at approximately 150k, which is a 33% reduction, while the Conclusion says 50,000 samples, which is a 50% reduction from 100k), and no PBTL learning curve is shown to support either number. Additionally, PBTL's pretraining uses 35,000 extra simulations, complicating the net data-reduction accounting. These issues weaken the data-efficiency claim but do not make the derivation circular, since the surrogate is independently validated against the simulator and the PBTL accuracy gain is measured directly.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claims rest on several hand-chosen parameters (transfer layer, dataset sizes, sampling ranges, GA settings) and domain assumptions about the simulator's fidelity and the transferability of active-region features. No new physical entities are introduced. The 50% data reduction claim is not backed by a PBTL learning curve and should be treated cautiously.

free parameters (7)

Transfer layer index = third hidden layer
Selected by validating R2 on holdout data (Figure S3), not derived from physics.
PBTL first-layer split sizes = unspecified
The active/injector block sizes in the split first hidden layer are not reported; this architectural choice affects the method's effectiveness and is not reproducible.
Training dataset sizes = 35,000 (DNN-CP), 100,000 (DNN-TP)
Chosen as a trade-off between accuracy and generation cost, based on the learning curve in Figure S1; the '50% data reduction' claim depends on this choice.
GA population and generations = 2000 per generation, 50 generations, 10% mutation
Set to ensure exploration in the multi-objective landscape; not fitted to physics.
Thickness sampling ranges = Tables S1/S2
Hand-chosen ranges for each superlattice layer; they constrain the design space and affect the reported L1 novelty distances.
Perturbation parameters for validated designs = not disclosed
The strategy of perturbing prior QCL designs to create training examples has undisclosed perturbation magnitudes and seed counts; this biases the dataset and impacts generalization claims.
Network hyperparameters = learning rate 1e-4, batch size 32, layer sizes 500-500-500-500-250(-125)
Chosen by prior practice; not part of the physics claim but needed for reproduction.

assumptions (5)

domain assumption The effective two-band model with conduction band nonparabolicity accurately describes mid-IR QCL subband energies and wavefunctions.
Invoked in Section 2.1 and SI equations (1)-(2); all training labels are generated from this model.
domain assumption The LO-phonon and interface-roughness scattering rates, and the optical gain formula (Equations 7-10), are accurate enough to serve as ground truth.
Used to compute linewidth and modal gain labels; any model error propagates into surrogate training.
ad hoc to paper The DNN-CP trained on active-region-only data captures transferable knowledge for the full QCL structure.
Stated as a hypothesis in Section 3.1; the entire PBTL concept depends on this, with only empirical UMAP and weight-distribution support.
domain assumption Training samples generated by perturbing previously validated QCL designs are representative of the full design space and maintain population inversion.
Section 2.2; used to augment random sampling; affects generalization and the out-of-distribution novelty claim.
domain assumption The scalarized sum of five sigmoid/Gaussian objectives (Equations 1-6) is a faithful proxy for QCL design quality.
Used in GA fitness; target values are user-specified and not derived from first principles.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Efficient nanophotonic devices optimization using deep neural network trained with physics-based transfer learning (PBTL) methodology." pith.science (2026). https://pith.science/paper/PV2UFAN6

@misc{pith2026250610418,
  author       = {Pith},
  title        = {Pith review of: Efficient nanophotonic devices optimization using deep neural network trained with physics-based transfer learning (PBTL) methodology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PV2UFAN6}},
  note         = {Machine review of arXiv:2506.10418}
}

read the original abstract

We propose a neural network(NN)-based surrogate modeling framework for photonic device optimization, especially in domains with imbalanced feature importance and high data generation costs. Our framework, which comprises physics-based transfer learning (PBTL)-enhanced surrogate modeling and scalarized multi-objective genetic algorithms (GAs), offers a generalizable solution for photonic design automation with minimal data resources.To validate the framework, we optimize mid-infrared quantum cascade laser (QCL) structures consisting of two regions, active and injection, which have different levels of feature importance. The optimization targets include five key QCL performance metrics such as modal gain, emission wavelength, linewidth, and effective injection, extraction energies. To address the challenge of multiple local optima in the output latent space, we integrate a deep neural network total predictor (DNN-TP) with a GA, enabling scalable and nature-inspired optimization. By replacing computationally expensive numerical simulations with the DNN-TP surrogate model, the optimization achieves a speed-up of over 80,000 times, allowing large-scale exploration of the QCL design space.To improve model generalization with limited data, we introduce PBTL, which transfers knowledge from a DNN core predictor (DNN-CP) trained on active-region structures. This approach yields a 0.69 percentage increase in prediction accuracy, equivalent to a 50 percentage reduction in training data requirements, and leads to generate more feasible device structure with 60 percentage improvement in evaluation metric during optimization.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 27 canonical work pages

[2]

Empowering nanophotonic applications via artificial intelligence: pathways, progress, and prospects,

W. Chen, S. Yang, Y. Yan, Y. Gao, J. Zhu, and Z. Dong, “Empowering nanophotonic applications via artificial intelligence: pathways, progress, and prospects,” Nanophotonics, vol. 14, no. 4, pp. 429–447, 2025

work page 2025
[3]

Deep learning enabled inverse design in nanophotonics,

S. So, T. Badloe, J. Noh, J. Bravo-Abad, and J. Rho, “Deep learning enabled inverse design in nanophotonics,” Nanophotonics, vol. 9, no. 5, pp. 1041–1057, 2020

2020
[4]

Quantum cascade laser,

J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, and A. Y. Cho, “Quantum cascade laser,” Science, vol. 264, no. 5158, pp. 553–556, 1994

work page 1994
[5]

Quantum cascade unipolar intersubband light emitting diodes in the 8–13 μm wavelength region,

C. Sirtori, F. Capasso, J. Faist, D. L. Sivco, A. L. Hutchinson, and A. Y. Cho, “Quantum cascade unipolar intersubband light emitting diodes in the 8–13 μm wavelength region,” Applied physics letters, vol. 66, no. 1, pp. 4–6, 1995

work page 1995
[6]

Evaluation of some scattering times for electrons in unbiased and biased single- and multiple-quantum-well structures,

R. Ferreira and G. Bastard, “Evaluation of some scattering times for electrons in unbiased and biased single- and multiple-quantum-well structures,” Phys. Rev. B, vol. 40, no. 2, pp. 1074–1086, 1989

work page 1989
[7]

Femtosecond luminescence measurements of the intersubband scattering rate in AlxGa1−xAs/GaAs quantum wells under selective excitation,

M. Hartig, S. Haacke, B. Deveaud, and L. Rota, “Femtosecond luminescence measurements of the intersubband scattering rate in AlxGa1−xAs/GaAs quantum wells under selective excitation,” Phys. Rev. B, vol. 54, no. 20, pp. R14269–R14272, 1996

work page 1996
[8]

Development of a multi-objective evolutionary algorithm for strain-enhanced quantum cascade lasers,

D. Mueller and G. Triplett, “Development of a multi-objective evolutionary algorithm for strain-enhanced quantum cascade lasers,” Photon- ics, vol. 3, no. 3, p. 44, 2016

work page 2016
[9]

Particle swarm optimization approach to identify optimum electrical pulse characteristics for efficient gain switching in dual wavelength quantum cascade lasers,

P. Ashok, “Particle swarm optimization approach to identify optimum electrical pulse characteristics for efficient gain switching in dual wavelength quantum cascade lasers,” Optik, vol. 171, pp. 786–797, 2018

work page 2018

Show all 29 references

[10]

QCL design engineering: automatization vs. classical approaches,

A. S. Dashkov and L. I. Goray, “QCL design engineering: automatization vs. classical approaches,” Semiconductors, vol. 54, no. 14, pp. 1823– 1825, 2020

2020
[11]

Bayesian optimization of terahertz quantum cascade lasers,

M. Franckié and J. Faist, “Bayesian optimization of terahertz quantum cascade lasers,” Phys. Rev. Applied, vol. 13, no. 3, p. 034025, 2020

2020
[12]

Migrating knowledge between physical scenarios based on artificial neural networks,

Y. Qu, L. Jing, Y. Shen, M. Qiu, and M. Soljačić, “Migrating knowledge between physical scenarios based on artificial neural networks,” ACS Photonics, vol. 6, no. 5, pp. 1168–1174, 2019

2019
[13]

Transfer-learning-assisted inverse metasurface design for 30% data savings,

Z. Fan et al., “Transfer-learning-assisted inverse metasurface design for 30% data savings,” Phys. Rev. Applied, vol. 18, no. 2, p. 024022, 2022

2022
[14]

Effect of conduction band non-parabolicity on the optical gain of quantum cascade lasers based on the effective two- band finite difference method,

G. Cho and J. Kim, “Effect of conduction band non-parabolicity on the optical gain of quantum cascade lasers based on the effective two- band finite difference method,” Semicond. Sci. Technol., vol. 32, no. 9, p. 095002, 2017. 11

2017
[15]

Theoretical and experimental study of optical gain and linewidth enhancement factor of type-I quantum-cascade lasers,

J. Kim et al., “Theoretical and experimental study of optical gain and linewidth enhancement factor of type-I quantum-cascade lasers,” IEEE J. Quantum Electron., vol. 40, no. 12, pp. 1663–1674, 2004

2004
[16]

Intersubband absorption linewidth in GaAs quantum wells due to scattering by interface roughness, phonons, alloy disorder, and impurities,

T. Unuma, M. Yoshita, T. Noda, H. Sakaki, and H. Akiyama, “Intersubband absorption linewidth in GaAs quantum wells due to scattering by interface roughness, phonons, alloy disorder, and impurities,” J. Appl. Phys., vol. 93, no. 3, pp. 1586–1597, 2003

2003
[17]

Interface-roughness-induced broadening of intersubband electroluminescence in p-SiGe and n-GaInAs/ AlInAs quantum- cascade structures,

S. Tsujino et al., “Interface-roughness-induced broadening of intersubband electroluminescence in p-SiGe and n-GaInAs/ AlInAs quantum- cascade structures,” Appl. Phys. Lett., vol. 86, no. 6, p. 062113, 2005

2005
[18]

Intersubband linewidths in quantum cascade laser designs,

A. Wittmann, Y. Bonetti, J. Faist, E. Gini, and M. Giovannini, “Intersubband linewidths in quantum cascade laser designs,” Appl. Phys. Lett., vol. 93, no. 14, p. 141103, 2008

2008
[19]

A tutorial on Bayesian optimization,

P. I. Frazier, “A tutorial on Bayesian optimization,” Jul. 08, 2018, arXiv: arXiv:1807.02811. doi: 10.48550/arXiv.1807.02811

2018 doi arXiv
[20]

Free-form optimization of nanophotonic devices: from classical methods to deep learning,

J. Park, S. Kim, D. W. Nam, H. Chung, C. Y. Park, and M. S. Jang, “Free-form optimization of nanophotonic devices: from classical methods to deep learning,” Nanophotonics, vol. 11, no. 9, pp. 1809–1845, 2022

2022
[21]

Fully automatized quantum cascade laser design by genetic optimization,

A. Bismuto, R. Terazzi, B. Hinkov, M. Beck, and J. Faist, “Fully automatized quantum cascade laser design by genetic optimization,” Appl. Phys. Lett., vol. 101, no. 2, p. 021103, 2012

2012
[22]

Géron, Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: Concepts, tools, and techniques to build intelligent systems

A. Géron, Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: Concepts, tools, and techniques to build intelligent systems. O’Reilly Media, Inc., 2022

2022
[23]

Inverse design of nanophotonic devices enabled by optimization algorithms and deep learning: recent achievements and fu- ture prospects,

J. Kim et al., “Inverse design of nanophotonic devices enabled by optimization algorithms and deep learning: recent achievements and fu- ture prospects,” Nanophotonics, vol. 14, no. 2, pp. 121–151, 2025

2025
[24]

Large-scale photonic inverse design: computational challenges and break- throughs,

C. Kang, C. Park, M. Lee, J. Kang, M. S. Jang, and H. Chung, “Large-scale photonic inverse design: computational challenges and break- throughs,” Nanophotonics, vol. 13, no. 20, pp. 3765–3792, 2024

2024
[25]

Multi-objective optimization using genetic algorithms: A tutorial,

A. Konak, D. W. Coit, and A. E. Smith, “Multi-objective optimization using genetic algorithms: A tutorial,” Reliab. Eng. Syst. Saf., vol. 91, no. 9, pp. 992–1007, 2006

2006
[26]

Optuna: A next-generation hyperparameter optimization framework,

T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, “Optuna: A next-generation hyperparameter optimization framework,” ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2019, pp. 2623–2631

2019
[27]

Physics-enhanced deep surrogates for partial differential equations,

R. Pestourie, Y. Mroueh, C. Rackauckas, P. Das, and S. G. Johnson, “Physics-enhanced deep surrogates for partial differential equations,” Nat. Mach. Intell., vol. 5, no. 12, pp. 1458–1465, 2023

2023
[28]

Lee, Python machine learning, New Jersey, John Wiley & Sons, 2019

W.-M. Lee, Python machine learning, New Jersey, John Wiley & Sons, 2019

2019
[29]

Harrison and A

P. Harrison and A. Valavanis, Quantum wells, wires and dots: theoretical and computational physics of semiconductor nanostructures, New Jersey, John Wiley & Sons, 2016

2016
[30]

Simulating 500 million years of evolution with a language model,

T. Hayes et al., “Simulating 500 million years of evolution with a language model,” Science, vol. 387, no. 6736, pp. 850–858, 2025. 12 Supplementary Material 1 Numerical Modeling of QCLs Quality of deep neural network ( DNN) surrogate model is highly determined by a quality of...

2025

Pith tools

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

[1] [2]

Empowering nanophotonic applications via artificial intelligence: pathways, progress, and prospects,

W. Chen, S. Yang, Y. Yan, Y. Gao, J. Zhu, and Z. Dong, “Empowering nanophotonic applications via artificial intelligence: pathways, progress, and prospects,” Nanophotonics, vol. 14, no. 4, pp. 429–447, 2025

work page 2025

[2] [3]

Deep learning enabled inverse design in nanophotonics,

S. So, T. Badloe, J. Noh, J. Bravo-Abad, and J. Rho, “Deep learning enabled inverse design in nanophotonics,” Nanophotonics, vol. 9, no. 5, pp. 1041–1057, 2020

2020

[3] [4]

Quantum cascade laser,

J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, and A. Y. Cho, “Quantum cascade laser,” Science, vol. 264, no. 5158, pp. 553–556, 1994

work page 1994

[4] [5]

Quantum cascade unipolar intersubband light emitting diodes in the 8–13 μm wavelength region,

C. Sirtori, F. Capasso, J. Faist, D. L. Sivco, A. L. Hutchinson, and A. Y. Cho, “Quantum cascade unipolar intersubband light emitting diodes in the 8–13 μm wavelength region,” Applied physics letters, vol. 66, no. 1, pp. 4–6, 1995

work page 1995

[5] [6]

Evaluation of some scattering times for electrons in unbiased and biased single- and multiple-quantum-well structures,

R. Ferreira and G. Bastard, “Evaluation of some scattering times for electrons in unbiased and biased single- and multiple-quantum-well structures,” Phys. Rev. B, vol. 40, no. 2, pp. 1074–1086, 1989

work page 1989

[6] [7]

Femtosecond luminescence measurements of the intersubband scattering rate in AlxGa1−xAs/GaAs quantum wells under selective excitation,

M. Hartig, S. Haacke, B. Deveaud, and L. Rota, “Femtosecond luminescence measurements of the intersubband scattering rate in AlxGa1−xAs/GaAs quantum wells under selective excitation,” Phys. Rev. B, vol. 54, no. 20, pp. R14269–R14272, 1996

work page 1996

[7] [8]

Development of a multi-objective evolutionary algorithm for strain-enhanced quantum cascade lasers,

D. Mueller and G. Triplett, “Development of a multi-objective evolutionary algorithm for strain-enhanced quantum cascade lasers,” Photon- ics, vol. 3, no. 3, p. 44, 2016

work page 2016

[8] [9]

Particle swarm optimization approach to identify optimum electrical pulse characteristics for efficient gain switching in dual wavelength quantum cascade lasers,

P. Ashok, “Particle swarm optimization approach to identify optimum electrical pulse characteristics for efficient gain switching in dual wavelength quantum cascade lasers,” Optik, vol. 171, pp. 786–797, 2018

work page 2018