-
Anomaly segmentation model for defects detection in electroluminescence images of heterojunction solar cells
Authors:
Alexey Korovin,
Artem Vasilyev,
Fedor Egorov,
Dmitry Saykin,
Evgeny Terukov,
Igor Shakhray,
Leonid Zhukov,
Semen Budennyy
Abstract:
Efficient defect detection in solar cell manufacturing is crucial for stable green energy technology manufacturing. This paper presents a deep-learning-based automatic detection model SeMaCNN for classification and semantic segmentation of electroluminescent images for solar cell quality evaluation and anomalies detection. The core of the model is an anomaly detection algorithm based on Mahalanobi…
▽ More
Efficient defect detection in solar cell manufacturing is crucial for stable green energy technology manufacturing. This paper presents a deep-learning-based automatic detection model SeMaCNN for classification and semantic segmentation of electroluminescent images for solar cell quality evaluation and anomalies detection. The core of the model is an anomaly detection algorithm based on Mahalanobis distance that can be trained in a semi-supervised manner on imbalanced data with small number of digital electroluminescence images with relevant defects. This is particularly valuable for prompt model integration into the industrial landscape. The model has been trained with the on-plant collected dataset consisting of 68 748 electroluminescent images of heterojunction solar cells with a busbar grid. Our model achieves the accuracy of 92.5%, F1 score 95.8%, recall 94.8%, and precision 96.9% within the validation subset consisting of 1049 manually annotated images. The model was also tested on the open ELPV dataset and demonstrates stable performance with accuracy 94.6% and F1 score 91.1%. The SeMaCNN model demonstrates a good balance between its performance and computational costs, which make it applicable for integrating into quality control systems of solar cell manufacturing.
△ Less
Submitted 1 October, 2022; v1 submitted 11 August, 2022;
originally announced August 2022.
-
Eco2AI: carbon emissions tracking of machine learning models as the first step towards sustainable AI
Authors:
Semen Budennyy,
Vladimir Lazarev,
Nikita Zakharenko,
Alexey Korovin,
Olga Plosskaya,
Denis Dimitrov,
Vladimir Arkhipkin,
Ivan Oseledets,
Ivan Barsola,
Ilya Egorov,
Aleksandra Kosterina,
Leonid Zhukov
Abstract:
The size and complexity of deep neural networks continue to grow exponentially, significantly increasing energy consumption for training and inference by these models. We introduce an open-source package eco2AI to help data scientists and researchers to track energy consumption and equivalent CO2 emissions of their models in a straightforward way. In eco2AI we put emphasis on accuracy of energy co…
▽ More
The size and complexity of deep neural networks continue to grow exponentially, significantly increasing energy consumption for training and inference by these models. We introduce an open-source package eco2AI to help data scientists and researchers to track energy consumption and equivalent CO2 emissions of their models in a straightforward way. In eco2AI we put emphasis on accuracy of energy consumption tracking and correct regional CO2 emissions accounting. We encourage research community to search for new optimal Artificial Intelligence (AI) architectures with a lower computational cost. The motivation also comes from the concept of AI-based green house gases sequestrating cycle with both Sustainable AI and Green AI pathways.
△ Less
Submitted 3 August, 2022; v1 submitted 31 July, 2022;
originally announced August 2022.
-
Boosting Heterogeneous Catalyst Discovery by Structurally Constrained Deep Learning Models
Authors:
Alexey N. Korovin,
Innokentiy S. Humonen,
Artem I. Samtsevich,
Roman A. Eremin,
Artem I. Vasilyev,
Vladimir D. Lazarev,
Semen A. Budennyy
Abstract:
The discovery of new catalysts is one of the significant topics of computational chemistry as it has the potential to accelerate the adoption of renewable energy sources. Recently developed deep learning approaches such as graph neural networks (GNNs) open new opportunity to significantly extend scope for modelling novel high-performance catalysts. Nevertheless, the graph representation of particu…
▽ More
The discovery of new catalysts is one of the significant topics of computational chemistry as it has the potential to accelerate the adoption of renewable energy sources. Recently developed deep learning approaches such as graph neural networks (GNNs) open new opportunity to significantly extend scope for modelling novel high-performance catalysts. Nevertheless, the graph representation of particular crystal structure is not a straightforward task due to the ambiguous connectivity schemes and numerous embeddings of nodes and edges. Here we present embedding improvement for GNN that has been modified by Voronoi tesselation and is able to predict the energy of catalytic systems within Open Catalyst Project dataset. Enrichment of the graph was calculated via Voronoi tessellation and the corresponding contact solid angles and types (direct or indirect) were considered as features of edges and Voronoi volumes were used as node characteristics. The auxiliary approach was enriching node representation by intrinsic atomic properties (electronegativity, period and group position). Proposed modifications allowed us to improve the mean absolute error of the original model and the final error equals to 651 meV per atom on the Open Catalyst Project dataset and 6 meV per atom on the intermetallics dataset. Also, by consideration of additional dataset, we show that a sensible choice of data can decrease the error to values above physically-based 20 meV per atom threshold.
△ Less
Submitted 2 October, 2022; v1 submitted 11 July, 2022;
originally announced July 2022.
-
Proton-induced activation cross sections in the energy range below 1 GeV
Authors:
I. V. Moskalenko,
A. A. Andrianov,
A. V. Bytenko,
T. A. Frolova,
R. S. Khalikov,
A. Yu. Konobeev,
Yu. A. Korovin,
T. V. Kulevoy,
I. S. Kuptsov,
K. V. Pavlov,
A. Yu. Stankovskiy,
E. M. Syresin,
Yu. E. Titarenko,
V. D. Vu
Abstract:
(Abridged) Modern studies and industrial applications related to the design, radiation protection, and reliability of nuclear facilities, medical applications, as well as space research and exploration are relying on extensive simulations and modeling. Computer codes realizing semi-classical and quantum molecular dynamics (QMD) approaches are often employed to make up for the lack of accelerator d…
▽ More
(Abridged) Modern studies and industrial applications related to the design, radiation protection, and reliability of nuclear facilities, medical applications, as well as space research and exploration are relying on extensive simulations and modeling. Computer codes realizing semi-classical and quantum molecular dynamics (QMD) approaches are often employed to make up for the lack of accelerator data on many nuclear reactions at intermediate and high energies (>10s of MeV/n) and are in high demand. This contribution focuses on the methodology of generating reliable proton-induced cross sections in the energy range below 1 GeV. We developed a problem-oriented computer framework based on MCNPX and CASCADE/INPE codes to calculate activation cross section data at intermediate and high energies. Goodness of the fits of nucleon-nucleus interaction models to the existing data is evaluated based on elaborated algorithms. The method is based on the analysis of a large set of data and calculated cross sections for different targets and residual nuclei in a wide range of proton energies using numerous criteria. In practice, this could be done by tuning the model parameters and algorithms to fit required experimental data subset, or through achieving unification and consistency of fundamental parameters for all considered reactions. The presented framework is pursuing the latter approach. We use proton-induced reactions on Si and Fe nuclei to illustrate the application of the developed multi-criteria algorithm, where we use all data below 1 GeV available from the EXFOR data library and the IAEA CRP "Benchmark of Spallation Models." We show that the analysis of the predictive power of various intermediate and high-energy models based on the multi-criteria algorithm allows more sophisticated selection of appropriate models in a given energy range and residual nuclei domain.
△ Less
Submitted 1 March, 2021;
originally announced March 2021.
-
Vertical engineering for large stimulated Brillouin scattering in unreleased silicon-based waveguides
Authors:
Laura Mercadé,
Alexander V. Korovin,
Yan Pennec,
Bahram Djafari Rouhani,
Alejandro Martínez
Abstract:
Strong acousto-optic interaction in silicon-based waveguides generally requires releasing of the silicon core to avoid mechanical leakage into the underlying silica substrate. This complicates fabrication, limits thermalization, reduces the mechanical robustness and hinders large area optomechanical devices on a single chip. Here, we overcome this limitation by employing vertical photonic-phononic…
▽ More
Strong acousto-optic interaction in silicon-based waveguides generally requires releasing of the silicon core to avoid mechanical leakage into the underlying silica substrate. This complicates fabrication, limits thermalization, reduces the mechanical robustness and hinders large area optomechanical devices on a single chip. Here, we overcome this limitation by employing vertical photonic-phononic engineering. Specifically, the insertion of a thick silicon nitride layer between the silicon guiding core and the silica substrate contributes to reduce GHz-frequencies phonon leakage enabling large values of the Brillouin gain in an unreleased platform. We get values of the Brillouin gain around 300 (Wm)$^{-1}$ for different configurations, which could be further increased by operation at cryogenic temperatures. These values should enable to observe Brillouin-related phenomena in cm-scale waveguides or in more compact ring resonators. This finding could pave the way towards large-area unreleased cavity and waveguide optomechanics on silicon chips.
△ Less
Submitted 4 November, 2020;
originally announced November 2020.
-
Stabilization of $\varepsilon$-Fe$_2$O$_3$ epitaxial layer on MgO(111)/GaN via an intermediate $γ$-Fe$_2$O$_3$-phase
Authors:
Victor Ukleev,
Mikhail Volkov,
Alexander Korovin,
Thomas Saerbeck,
Nikolai Sokolov,
Sergey Suturin
Abstract:
In the present study we have demonstrated epitaxial stabilization of the metastable magnetically-hard $\varepsilon$-Fe$_2$O$_3$ phase on top of a thin MgO(111) buffer layer grown onto the GaN (0001) surface. The primary purpose to introduce a 4\,nm-thick buffer layer of MgO in between Fe$_2$O$_3$ and GaN was to stop thermal migration of Ga into the iron oxide layer. Though such migration and succe…
▽ More
In the present study we have demonstrated epitaxial stabilization of the metastable magnetically-hard $\varepsilon$-Fe$_2$O$_3$ phase on top of a thin MgO(111) buffer layer grown onto the GaN (0001) surface. The primary purpose to introduce a 4\,nm-thick buffer layer of MgO in between Fe$_2$O$_3$ and GaN was to stop thermal migration of Ga into the iron oxide layer. Though such migration and successive formation of the orthorhombic GaFeO$_3$ was supposed earlier to be a potential trigger of the nucleation of the isostructural $\varepsilon$-Fe$_2$O$_3$, the present work demonstrates that the growth of single crystalline uniform films of epsilon ferrite by pulsed laser deposition is possible even on the MgO capped GaN. The structural properties of the 60\,nm thick Fe$_2$O$_3$ layer on MgO / GaN were probed by electron and x-ray diffraction, both suggesting that the growth of $\varepsilon$-Fe$_2$O$_3$ is preceded by formation of a thin layer of $γ$-Fe$_2$O$_3$. The presence of the magnetically hard epsilon ferrite was independently confirmed by temperature dependent magnetometry measurements. The depth-resolved x-ray and polarized neutron reflectometry reveal that the 10\,nm iron oxide layer at the interface has a lower density and a higher magnetization than the main volume of the $\varepsilon$-Fe$_2$O$_3$ film. The density and magnetic moment depth profiles derived from fitting the reflectometry data are in a good agreement with the presence of the magnetically degraded $γ$-Fe$_2$O$_3$ transition layer between MgO and $\varepsilon$-Fe$_2$O$_3$. The natural occurrence of the interface between magnetoelectric $\varepsilon$- and spin caloritronic $γ$- iron oxide phases can enable further opportunities to design novel all-oxide-on-semiconductor devices.
△ Less
Submitted 4 August, 2019; v1 submitted 26 July, 2019;
originally announced July 2019.
-
Substrate induced magnetic anisotropies and magneto-optical response in YIG nanosized epitaxial films on NdGG(111)
Authors:
B. B. Krichevtsov,
V. E. Bursian,
S. V. Gastev,
A. M. Korovin,
L. V. Lutsev,
S. M. Suturin,
K. V. Mashkov,
M. P. Volkov,
N. S. Sokolov
Abstract:
Nanosized Y3Fe5O12 epitaxial films have been grown on Nd3Ga5O12 substrates using laser molecular beam epitaxy method. Magneto-optical polar Kerr effect, ferromagnetic resonance and spin wave propagation measurements show that the stress-related anisotropy field has an opposite sign, compared to that in the YIG/GGG systems. This leads to a considerable decrease of the effective magnetization that o…
▽ More
Nanosized Y3Fe5O12 epitaxial films have been grown on Nd3Ga5O12 substrates using laser molecular beam epitaxy method. Magneto-optical polar Kerr effect, ferromagnetic resonance and spin wave propagation measurements show that the stress-related anisotropy field has an opposite sign, compared to that in the YIG/GGG systems. This leads to a considerable decrease of the effective magnetization that opens a perspective to get YIG films with perpendicular magnetization for utilizing forward volume spin waves. Longitudinal magnetooptical Kerr effect magnetometry reveals a large contribution of quadratic in magnetization terms into dielectric permittivity tensor at optical frequencies. This effect strongly increases with temperature decrease and is explained by magnetization of the interface Nd3+ ions that are exchange coupled to the Fe3+ ions.
△ Less
Submitted 12 December, 2018;
originally announced January 2019.
-
Role of gallium diffusion in the formation of a magnetically dead layer at Y3Fe5O12 / Gd3Ga5O12 epitaxial interface
Authors:
S. M. Suturin,
A. M. Korovin,
V. E. Bursian,
L. V. Lutsev,
V. Bourobina,
N. L. Yakovlev,
M. Montecchi,
L. Pasquali,
V. Ukleev,
A. Vorobiev,
A. Devishvili,
N. S. Sokolov
Abstract:
We have clarified the origin of magnetically dead interface layer formed in yttrium iron garnet (YIG) films grown at above 700°C onto gadolinium gallium garnet (GGG) substrate by means of laser molecular beam epitaxy. The diffusion-assisted formation of a Ga-rich region at the YIG / GGG interface is demonstrated by means of composition depth profiling performed by X-ray photoelectron spectroscopy,…
▽ More
We have clarified the origin of magnetically dead interface layer formed in yttrium iron garnet (YIG) films grown at above 700°C onto gadolinium gallium garnet (GGG) substrate by means of laser molecular beam epitaxy. The diffusion-assisted formation of a Ga-rich region at the YIG / GGG interface is demonstrated by means of composition depth profiling performed by X-ray photoelectron spectroscopy, secondary ion mass spectroscopy and X-ray and neutron reflectometry. Our finding is in sharp contrast to the earlier expressed assumption that Gd acts as a migrant element in the YIG/GGG system. We further correlate the presence of Ga-rich transition layer with considerable quenching of ferromagnetic resonance and spin wave propagation in thin YIG films. Finally, we clarify the origin of the enigmatic low-density overlayer that is often observed in neutron and X-ray reflectometry studies of the YIG / GGG epitaxial system.
△ Less
Submitted 6 November, 2018; v1 submitted 4 November, 2018;
originally announced November 2018.
-
Tunable polymorphism of epitaxial iron oxides in the four-in-one ferroic-on-GaN system with magnetically ordered α-, γ-, ε-Fe2O3 and Fe3O4 layers
Authors:
Sergey Suturin,
Alexander Korovin,
Sergey Gastev,
Mikhail Volkov,
Masao Tabuchi,
Nikolai Sokolov
Abstract:
Hybridization of semiconducting and magnetic materials into a single heterostructure is believed to be potentially applicable to the design of novel functional spintronic devices. In the present work we report epitaxial stabilization of four magnetically ordered iron oxide phases (Fe3O4, γ-Fe2O3, α-Fe2O3 and most exotic metastable ε-Fe2O3) in the form of nanometer sized single crystalline films on…
▽ More
Hybridization of semiconducting and magnetic materials into a single heterostructure is believed to be potentially applicable to the design of novel functional spintronic devices. In the present work we report epitaxial stabilization of four magnetically ordered iron oxide phases (Fe3O4, γ-Fe2O3, α-Fe2O3 and most exotic metastable ε-Fe2O3) in the form of nanometer sized single crystalline films on GaN(0001) surface. The epitaxial growth of as many as four distinctly different iron oxide phases is demonstrated within the same single-target Laser MBE technological process on a GaN semiconductor substrate widely used for electronic device fabrication. The discussed iron oxides belong to a family of simple formula magnetic materials exhibiting a rich variety of outstanding physical properties including peculiar Verwey and Morin phase transitions in Fe3O4 and α-Fe2O3 and multiferroic behavior in metastable magnetically hard ε-Fe2O3 ferrite. The physical reasons standing behind the nucleation of a particular phase in an epitaxial growth process deserve interest from the fundamental point of view. The practical side of the presented study is to exploit the tunable polymorphism of iron oxides for creation of ferroic-on-semiconductor heterostructures usable in novel spintronic devices. By application of a wide range of experimental techniques the surface morphology, crystalline structure, electronic and magnetic properties of the single phase iron oxide epitaxial films on GaN have been studied. A comprehensive comparison has been made to the properties of the same ferrite materials in the bulk and nanostructured form reported by other research groups.
△ Less
Submitted 15 December, 2017;
originally announced December 2017.
-
High Energy Activation Data Library (HEAD-2009)
Authors:
Yury A. Korovin,
Anatoly A. Natalenko,
Alexander Yu. Konobeyev,
Alexey Yu. Stankovskiy,
Stepan G. Mashnik
Abstract:
A proton activation data library for 682 nuclides from 1-H to 210-Po in the energy range from 150 MeV up to 1 GeV was developed. To calculate proton activation data, the MCNPX 2.6.0 and CASCADE/INPE codes were chosen. Different intranuclear cascade, preequilibrium, and equilibrium nuclear reaction models and their combinations were used. The optimum calculation models have been chosen on the basis…
▽ More
A proton activation data library for 682 nuclides from 1-H to 210-Po in the energy range from 150 MeV up to 1 GeV was developed. To calculate proton activation data, the MCNPX 2.6.0 and CASCADE/INPE codes were chosen. Different intranuclear cascade, preequilibrium, and equilibrium nuclear reaction models and their combinations were used. The optimum calculation models have been chosen on the basis of statistical correlations for calculated and experimental proton data taken from the EXFOR library of experimental nuclear data. All the data are written in ENDF-6 format. The library is called HEPAD-2008 (High-Energy Proton Activation Data). A revision of IEAF-2005 neutron activation data library has been performed: A set of nuclides for which the cross-section data can be (and were) updated using more modern and improved models is specified, and the corresponding calculations have been made in the present work. The new version of the library is called IEAF-2009. The HEPAD-2008 and IEAF-2009 are merged to the final HEAD-2009 library.
△ Less
Submitted 10 March, 2010;
originally announced March 2010.
