Search

Most real optimization problems are defined over a mixed search space where the variables are both discrete and continuous. In engineering applications, the objective function is typically calculated with a numerically costly black-box simulation.General mixed and costly optimization problems are therefore of a great practical interest, yet their resolution remains in a large part an open scientific question. In this article, costly mixed problems are approached through Gaussian processes where the discrete variables are relaxed into continuous latent variables. The continuous space is more easily harvested by classical Bayesian optimization techniques than a mixed space would. Discrete variables are recovered either subsequently to the continuous optimization, or simultaneously with an additional continuous-discrete compatibility constraint that is handled with augmented Lagrangians. Several possible implementations of such Bayesian mixed optimizers are compared. In particular, the reformulation of the problem with continuous latent variables is put in competition with searches working directly in the mixed space. Among the algorithms involving latent variables and an augmented Lagrangian, a particular attention is devoted to the Lagrange multipliers for which a local and a global estimation techniques are studied. The comparisons are based on the repeated optimization of three analytical functions and a beam design problem., Accepted for publication in Advanced Modeling and Simulation in Engineering Sciences, march 2022

We present an optical transduction method adapted to the detection of low frequency thermal perturbations and implemented for photothermal trace gas detection. The transducer is a π-phase shifted fiber Bragg grating, stabilized and interrogated by the Pound-Drever-Hall method. The principle of detection is based on the frequency shift of the narrow optical resonance, induced by the temperature variations. In temperature measurement mode, the stabilization leads to an estimated limit of detection of 1 µK at room temperature and at a frequency of 40 Hz. When the fiber transducer is placed in a gas cell, CO

ISBN : 978-1-7281-6086-3/20; International audience; Shape-optimization techniques were quite recently applied to photonic components but to the best of our knowledge, no application to optical computing has been reported yet. Here, we present the design of a photonic integrated circuit, composed of shape optimized passive components, performing a matrix-vector product. A ≈ 2000 times gain on the overall footprint is achieved in comparison with a conventional circuit using classical Mach-Zehnder interferometers.

A highly sensitive Fabry&ndash, Perot based transduction method is proposed as an all-optical alternative for the detection of trace gas by the photoacoustic spectroscopy technique. A lumped element model is firstly devised to help design the whole system and is successfully compared to finite element method simulations. The fabricated Fabry&ndash, Perot microphone consists in a hinged cantilever based diaphragm, processed by laser cutting, and directly assembled at the tip of an optical fiber. We find a high acoustic sensitivity of 630 mV/Pa and a state-of-the-art noise equivalent pressure, as low as at resonance. For photoacoustic trace gas detection, the Fabry&ndash, Perot microphone is further embedded in a cylindrical multipass cell and shows an ultimate detection limit of 15 ppb of NO in nitrogen. The proposed optical trace gas sensor offers the advantages of high sensitivity and easy assembling, as well as the possibility of remote detection.

Photoacoustic (PA) spectroscopy is one of the most sensitive technique used to monitor chemical emission or detect gas traces. Coupled to quantum cascade lasers, this system is widely used in a large number of application fields from industrial control to health monitoring. Mass production for a large dissemination of such systems requires however further development for both decreasing their footprint and manufacturing cost. Since the last 6 years CEA-LETI has developed different versions of miniaturized photoacoustic cells. We have already demonstrated the detection of gas traces with a tiny silicon based-PA cell. Nevertheless, this first result was obtained with commercial MEMS microphones. Even if these components are reliable and enough performant they are not dedicated to photoacoustic gas detection and cannot be easily integrated into a fabrication process flow. To cope with these issues we suggest using both the M&NEMS technology and the MIR photonics. The new PAdetector termed microPA is built by stacking two 200 mm wafers: a sensor wafer, which includes the microphone (MEMS mechanical diaphragm and NEMS piezoresistive gauges), capillaries and fluidic ports, and a cap wafer, which includes the PA cell, the expansion volume, SiGe waveguides guiding the light into the PA cell, metal routing and electric contacts. Frequency response measurements as well as PA gas detection have been carried out. The system shows a mechanical resonance of the diaphragm at the frequency of 6500 Hz, in good agreement with the simulation. First CO2 and CH4 tests in laboratory condition demonstrates a limit of detection in the ppm range and a NNEA of 10-8 W.cm-1.Hz-1/2.

International audience; A highly sensitive optical transduction system suitable for photoacoustic trace gas detection is presented. The system includes a thin deformable hinged cantilever assembled on an optical fiber to form a Fabry-Perot cavity, whose length varies according to the acoustic pressure disturbance. Consequently, the optical power of the reflected light fluctuates at acoustic frequencies around the working point, which is stabilized to prevent from environmental drift of the interference fringes. The resonant mechanical structure proposed in this study shows a spectral response in good agreement with FEM simulation, good linearity and stability, with a noise equivalent pressure of 12 $\mu$Pa/√Hz.

International audience; This article deals with the optimal design of the individual nanophotonic components of photonic integrated circuits. In the mathematical setting of the three-dimensional, time-harmonic Maxwell equations, we propose a shape and topology optimization algorithm combining Hadamard's boundary variation method with a level set representation of shapes and their evolution. A particular attention is devoted to the robustness of the optimized devices with respect to small uncertainties over the physical or geometrical data of the problem. In this respect, we rely on a simple multi-objective formulation to deal with the two main sources of uncertainties plaguing nanophotonic devices, namely uncertainties over the incoming wavelength, and geometric uncertainties entailed by the lithography and etching fabrication process. Several numerical examples are presented and discussed to assess the efficiency of our methodology.

We present several integrated technologies on Silicon, from visible to mid-infrared, for particulate matter and gas detection. We present new concepts to detect in the visible particulate matter with a high sensitivity and a discrimination of both particle sizes and refractive indices. For gas detection, mid-infrared technologies developments include on one hand, microhotplate thermal emitters, as a cheap solution for gas sensing, eventually enhanced by plasmonics, and on the other hand quantum cascade lasers-based photoacoustic sensors, for high precision measurement, and for which the integration on Silicon is pushed forward for a reduction of costs.

This paper demonstrates, for the first time, an on-chip integration of a photoacoustic (PA) detector, one of the main building blocks required for chemical sensing. This MEMS PA detector consists of a miniaturized PA cell within an acoustic transducer based on an innovative microphone architecture. Coupled with a quantum cascade laser (QCL) source, this trace gas detector can measure very low concentrations of less than one part per million.

The Mid-IR spectral range (2.5 μm up to 12 μm) has been considered as the paradigm for innovative silicon photonic devices. In less than a decade, chemical sensing has become a key application for Mid-IR silicon photonic devices because of the growing potential in spectroscopy, materials processing, chemical and biomolecular sensing, security and industry application. Measuring in this spectral range, usually called fingerprint region, allows to address a unique combination of fundamental absorption bands order of magnitude stronger than overtone and combination bands in the near IR. This feature provides highly selective, sensitive and unequivocal identification of the chemicals.

The Mid-IR spectral range (2.5 μm up to 12 μm) has been considered as the paradigm for innovative silicon photonic devices. In less than a decade, chemical sensing has become a key application for Mid-IR silicon photonic devices because of the growing potential in spectroscopy, materials processing, chemical and biomolecular sensing, security and industry applications. Measuring in this spectral range, usually called molecule fingerprint region, allows to address a unique combination of fundamental absorption bands orders of magnitude stronger than overtone and combination bands in the near IR. This feature provides highly selective, sensitive and unequivocal identification of the chemicals. Progress in Cascade Laser technology (QCL and ICL) allows to select emission wavelengths suitable to target the detection of specific chemicals. With these sources, novel spectroscopic tools allowing real-time in-situ detection of gasses down to traces are nowadays commercially available. Mid-IR Si photonics has developed a novel class of integrated components leading to the integration at chip level of the main building blocks required for chemical sensing, i.e. the source, the PICs and the detector. Three main directions of improvement can be drawn: i) extend the range of wavelengths available from a single source, ii) move beam handling and routing from discrete optics to PICs and iii) investigate detection schemes for a fully integrated on-chip sensing. This paper reviews recent key achievements in the miniaturization and the co-integration of photonics devices at chip and packaging level to address cost, size and power consumption. Perspectives on potential applications will also be presented.

Photoacoustic (PA) spectroscopy is among the most sensitive techniques used to monitor chemical emission or detect gas traces. In the mid-infrared, where most of gases of interest have their strongest absorption lines, this technique takes advantage of the high optical power and room temperature operation of quantum cascade lasers (QCL). We have recently demonstrated that centimeter-size PA cells can compete, with bulky commercial systems for gas sensing without any compromises on performances. We demonstrate a new step towards cost reduction, extreme integration, and mass deployment of such PA sensors with a miniaturized silicon PA-cell fabricated on standard CMOS tools. The design, fabrication and characterizations of this new sub-centimeter PA cell built on a silicon platform are presented. First, the component has been designed using a detailed physical model, accounting for viscous and thermal losses, and metamodel-based optimization techniques. Second, it has been fabricated on our 200 mm CMOS pilot line. Several wafers have been released and diced. Single chips have then been assembled with commercial capacitive microphones and finally characterized on our reference gas bench. The photoacoustic simulations and the acoustics experiments are in a good agreement. The tiny PA cell exhibits a sensitivity down to the ppm level for CO2 at 2300 cm-1, as well as for CH4 at 3057 cm-1 even in a gas flow. Taking advantage of the integration of QCLs on Si and photonic circuitry, the silicon PA cell concept is currently being extended towards a fully integrated multigas detector.

International audience; Focal cooling is a promising alternative therapy for intractable focal epilepsies, avoiding the irreversible neuronal damages induced by resection surgery. However, due to thermal conduction losses, local cooling of a deep brain region remains a challenging objective for thermoelectric or fluidic technologies. Here, we investigated the viability of an optical micro-cooler based on anti-Stokes refrigeration of ytterbium doped YLF crystals, taking into account the medical constraints for implantable device. We realized significant cooling under atmospheric pressure and developed a solution drastically reducing the harmful fluorescence heating of brain-like liquids below 2 K, thus demonstrating the relevance of this technology for biomedical applications.

International audience; We report on a novel photonic polarization rotator design obtained by multi-level shape optimization. The numerical method consists in a topological optimization scheme improving iteratively the efficiency of the component by modifying its shape on two discrete levels along the etching direction. We numerically show that, compared to the state of the art single level shape optimization, the performances can be drastically improved for a given device length. Next, the polarization conversion efficiency can be further improved up to a computed value of $\sim$98.5 % with less than 0.35 dB insertion losses on a 100 nm bandwidth for a 6 $\mu$m × 1 $\mu$m footprint.

Photoacoustic cells are a type of optical sensors than can be used to detect trace gases. The photoacoustic spectroscopic technique is based on the absorption of photons by the molecules of interest and the subsequent creation of acoustic waves. It is extremely sensitive but the sensors bulkiness and cost prevent its mass deployment in several applications, such as greenhouse gases survey or indoor air monitoring. A trend towards miniaturization is thus engaged. In this paper, two centimeter-sized photoacoustic cells are presented. After an initial design stage by the finite element method, the cells have been fabricated and characterized. The experimental limit of detection reached with a centimeter-size cell made by direct metal laser sintering and an interband cascade laser delivering an optical power of 2.2 mW at 3.36 μm wavelength, is estimated at 92 ppbv methane in nitrogen.

International audience; Shape optimization techniques were first developed in the context of mechanical engineering and, quite recently, applied to photonic components for data communication. Here, motivated by the growing application potential of mid-IR photonics driven by chemical sensing and spectroscopy, we present the design by shape optimization of passive components operating in this wavelength range.

Photoacoustic spectroscopy is one of the most sensitive techniques used to monitor chemical emission or to detect gas traces. However the sensors bulkiness, cost and complexity prevent their use in applications where mass deployment is required. In this paper, the development and characterization of centimeter sized photoacoustic sensors is presented. With a 2.2 mW optical power interband cascade laser at 3.36 μm wavelength, a limit of detection of 320 ppbv is obtained for methane.

International audience; A robust integrated power splitter based on a silicon-on-insulator adiabatic coupler configuration is demonstrated. The power separation is achieved by a slow and simultaneous change of phase and coupling constants. The geometrical parameters of the device are determined thanks to a metamodel-based iterative optimization strategy. Solving the multiple parameter problem together with a realistic bandwidth constraint provides a clear improvement of the power splitting stability. The robustness is confirmed experimentally on a single device and at the wafer scale.

International audience; Multiple models of a physical phenomenon are sometimes available with different levels of approximation. The high fidelity model is more computation-ally demanding than the coarse approximation. In this context, including information from the lower fidelity model to build a surrogate model is desirable. Here, the study focuses on the design of a miniaturized photoa-coustic gas sensor which involves two numerical models. First, a multifidelity metamodeling method based on Radial Basis Function, the co-RBF, is proposed. This surrogate model is compared with the classical co-kriging method on two analytical benchmarks and on the photoacoustic gas sensor. Then an extension to the multifidelity framework of an already existing RBF-based optimization algorithm is applied to optimize the sensor efficiency. The co-RBF method brings promising results on a problem in larger dimension and can be considered as an alternative to co-kriging for multifi-delity metamodeling.

In photoacoustic (PA) spectroscopy, the signal is inversely proportional to the resonant cell volume. This favorable scaling behavior has provoked in recent years a growing interest in the miniaturization of PA cells. Due to dimension downscaling, technological constraints, and preliminary design choices, the modeling hypotheses used at the macro-scale are no longer valid. Here, a new model adapted to miniaturized and integrated PA ( $$\upmu $$ -PA) sensors is presented. The coupled model takes advantage of commercial software to address, respectively, (i) the electromagnetic mode propagation in the waveguide, (ii) the mid-IR light illumination of the chamber and its interaction with the molecules of interest, and (iii) the creation and propagation of acoustic waves in the cell. The model has been used to confirm the validity of the choice of the differential Helmholtz resonator principle and to provide a prototype $$\upmu $$ -PA cell geometry. It is shown that, in spite of the specific issues inherent to miniaturization and integration, in particular, the strong divergence of the light beam and the crucial influence of viscothermal dissipation processes, the $$\upmu $$ -PA sensor produces a pressure signal compatible with the next generation of resonant microphone technology. The model will be further improved when measurements performed on the prototype currently in fabrication are available.

We present here the development and the characterization, in laboratory conditions, of our photoacoustic cell associated with QCL’s and specific readout circuit for gas traces sensing. Assessment has been done on CO2.

A new Boundary Integral Equation (BIE) formulation for Stokes flow is presented for three-dimensional and axisymmetrical problems using non-primitive variables, assuming velocity field is prescribed on the boundary. The formulation involves the vector potential, instead of the classical stream function, and all three components of the vorticity are implied. Furthermore, following the Helmholtz decomposition, a scalar potential is added to represent the solenoidal velocity field. Firstly, the BIEs for three-dimensional flows are formulated for the vector potential and the vorticity by employing the fundamental solutions in free space of vector Laplace and biharmonic equations. The equations for axisymmetric flows are then derived from the three-dimensional formulation in a second step. The outcome is a domain integral free BIE formulation for both three-dimensional and axisymmetric Stokes flows with prescribed velocity boundary condition. Numerical results are included to validate and show the efficiency of the proposed axisymmetric formulation.

The multipole expansion is an approximation technique used to evaluate the potential field due to sources located in the far field. Based on the multipole expansion, we describe a new technique to calculate the far potential field due to ring sources which are encountered in the boundary element method (BEM) formulation of axisymmetric problems. As the sources in the near field are processed by the slower conventional BEM, it is important to maximize the amount of multipole calculations taking advantage of both interior and exterior multipole expansions. Numerical results are presented for an axisymmetric potential test problem with Neumann and Dirichlet boundary conditions. The complexity of the proposed method remains O ( N 2 ), which is equal to that of the conventional BEM. However, the proposed technique coupled with an iterative solver speeds up the solution procedure. The technique is significantly advantageous when medium and large numbers of elements are present in the domain.

This work reports numerical and experimental investigations of capillary stop valves, which simply rely on the sudden expansion of a microchannel to control the liquid position. The simulation approach is based on the solution of the free surface equilibrium problem by the Finite Element Method. In contrast to analytical solutions, it deals rigorously with the complex geometries found in planar microfluidic systems. In contrast to energy minimization approach, it leaves the way open to the future introduction of dynamic effects. Microfluidic systems, including a new design of capillary stop valves expanding along both channel width and depth, have been fabricated by a two-level Deep Reactive Ion Etching technology. The silicon substrate is limited by a bonded PolyDimethlylSiloxane (PDMS) cover. The microvalves with cross-section dimensions down to 15 μm × 15 μm, which are able to stop biological buffers and burst at pressures in the 1–10 kPa range, are suitable for Lab-on-a-Chip applications. The computed burst pressure agrees well with the measurements when the liquid is water. However, when liquids containing surfactants are used a deviation is ascertained. It may be explained by the transient variation of surface tension due to the dynamics of meniscus expansion.

Monolithic CdZnTe gamma-ray detectors are used to build gamma cameras for nuclear medicine imaging but their energy resolution is currently limited by charge sharing between adjacent anodes. For this work, charge sharing is simulated using Ulysse, a numerical model that takes into account the physical processes of charge creation by ionizing radiation and charge transport within the semiconductor detector. The charge carrier cloud size, following the gamma-ray photon interaction, is computed by the Monte Carlo method. Electron cloud diffusion and charge induction on the electrodes are computed by the finite element method. This study shows that the electron diffusion strongly influences the final electron cloud diameter.

A new transient finite element model of extracellular action potentials recording with a microelectrode in a tissue slice is presented. The neuron model is based on the Hodgkin-Huxley equations implemented with a thin film approximation of the neuron membrane. The computations of the membrane potential and currents, as well as that of the intra and extracellular potential fields, are performed at the same time, within a single finite element software. The neuron membrane model is validated by comparison with a NEURON simulation. It is shown that the finite element model is able to properly represent the neuron behavior in terms of membrane currents and potential. Moreover, it is demonstrated that an ideal measurement system model can be used, provided that the electronic recording system is adapted to the electrode-tissue interface impedance. A brief study of the influence of the relative position of the neuron and recording microelectrode on the recorded signal is presented. It is shown that the maximum recorded extracellular action potential is obtained when the electrode is placed below the neuron soma and hillock-initial segment areas.

A combination of the multilevel fast multipole method (MLFMM) and boundary element method (BEM) can solve large scale photonics problems of arbitrary geometry. Here, MLFMM-BEM algorithm based on a scalar and vector potential formulation, instead of the more conventional electric and magnetic field formulations, is described. The method can deal with multiple lossy or lossless dielectric objects of arbitrary geometry, be they nested, in contact, or dispersed. Several examples are used to demonstrate that this method is able to efficiently handle 3D photonic scatterers involving large numbers of unknowns. Absorption, scattering, and extinction efficiencies of gold nanoparticle spheres, calculated by the MLFMM, are compared with Mie's theory. MLFMM calculations of the bistatic radar cross section (RCS) of a gold sphere near the plasmon resonance and of a silica coated gold sphere are also compared with Mie theory predictions. Finally, the bistatic RCS of a nanoparticle gold-silver heterodimer calculated with MLFMM is compared with unmodified BEM calculations.

In the framework of research program “MIRIADE”, GSMA and CEA-LETI report simulations and developments of Si-integrated resonant photoacoustic cells developed for the optical sensing of the atmosphere when associated with mid-infrared sources such as QCLs.

International audience; Shape optimization techniques were first developed in the context of mechanical engineering and, more recently, applied to photonic components for data communication. Here, motivated by the growing application potential of mid-infrared photonics driven by chemical sensing and spectroscopy, we present the design by shape optimization of passive components operating in this wavelength range. A focus is placed on the creation of designs that are fabricable and robust to manufacturing uncertainties.

This thesis focuses on the mathematical field of shape optimization and explores two topics:one concerns the systematic determination of the design of nanophotonic components and the other one the optimal shape and location of boundary conditions defining partial differential equations (PDE).• In the mathematical setting of the three-dimensional, time-harmonic Maxwell equations, we proposea shape and topology optimization algorithm combining Hadamard’s boundary variation methodwith a level set representation of shapes and their evolution. A particular attention is devotedto the robustness of the optimized devices with respect to small uncertainties over the physical orgeometrical data of the problem. In this respect, we rely on a simple multi-objective formulation todeal with the two main sources of uncertainties plaguing nanophotonic devices, namely uncertaintiesover the incoming wavelength, and geometric uncertainties entailed by the lithography and etchingfabrication process. Several numerical examples are presented and discussed to assess the efficiencyof our methodology.• The second application concern the optimization of the shape of the regions assigned to different typesof boundary conditions in the definition of a “physical” PDE. This problem proves to be difficult inthe case of a Dirichlet-Neumann transition since it requires a precise study of the singular nature ofPDE solutions at the transition between two regions supporting these boundary conditions. On theone hand a full mathematical study is carried out on this theoretical problem and on the other handa numerical method based on a regularization of the boundary conditions is proposed to optimizethese regions. Various numerical examples are eventually presented in order to appraise the efficiencyof the proposed process.; Cette thèse contribue au domaine mathématique de l’optimisation de forme et explore deux sujets:l’une concerne la détermination automatique du design de composant nanophotonique et l’autre l’obtention de la forme optimale des conditions aux limites permettant de définir une équation au dérivées partielles (EDP).• Dans le cadre mathématique des équations de Maxwell tridimensionnelles et harmoniques dans letemps, nous proposons un algorithme d’optimisation de forme combinant la méthode d’Hadamard etune représentation des formes par la méthode level-set. Une attention particulière est accordée à larobustesse des dispositifs optimisés par rapport à des petites incertitudes sur les données physiques ougéométriques du problème. À cet égard, nous nous appuyons sur un algorithme provenant du domainede l’optimisation multi-objectifs pour traiter les deux principales sources d’incertitudes affectant desdispositifs nanophotoniques, à savoir les incertitudes sur la longueur d’onde de la lumière injectée ainsique les incertitudes géométriques liées au procédé de fabrication par lithographie-gravure. Plusieursexemples numériques sont présentés permettant d’évaluer l’efficacité de la méthode proposé.• La deuxième application concerne l’optimisation de la forme des régions affectées à différentes condi-tions limites dans la définition d’une EDP d’un problème “physique”. L’étude de ce problème s’avèreêtre délicate dans le cas d’une transition Dirichlet-Neumman car elle nécessite une analyse précise dela singularité des solutions d’une EDP à la transition entre deux régions supportant ces conditionslimites. Nous proposons d’une part une étude mathématique complète de ce problème dans le casde l’équation du Laplacien et d’autre part une méthode numérique basée sur une régularisation desconditions aux limites pour optimiser la forme de ces régions. Différents exemples numériques sontprésentés afin d’évaluer l’efficacité de notre méthode.

Numerical simulation is widely employed in engineering to study the behavior of a device and optimize its design. Nevertheless, each computation is often time consuming and, during an optimization sequence, the simulation code is evaluated a large number of times. An interesting way to reduce the computational burden is to build a metamodel (or surrogate model) of the simulation code. Adaptive strategies are then set up for the optimization of the component using the metamodel prediction. In the context of this thesis, three representative devices are identified for applications that can be encountered within the CEA LETI optics and photonics department. The study of these cases resulted in two problems to be treated. The first one concerns multifidelity metamodeling, which consists of constructing a metamodel from two simulations of the same component that can be hierarchically ranked in accuracy. The simulations are obtained from different approximations of the physical phenomenon. The work on this method for the case of the photoacoustic cell has generated the development of a new multifidelity surrogate model based on radial basis function. The second problem relate to the consideration of manufacturing uncertainties in the design of photonic devices. Taking into account the differences observed between the desired geometry and the geometry obtained in manufacturing for the optimization of the component efficiency requires the development of a particular method for the case of the adiabatic coupler. The entire work of this thesis is capitalized in a software toolbox.; La simulation numérique est couramment utilisée pour étudier le comportement d’un composant et optimiser sa conception. Pour autant, chaque calcul est souvent coûteux en termes de temps et l’optimisation nécessite de résoudre un grand nombre de fois le modèle numérique pour différentes configurations du composant. Une solution actuelle pour réduire le temps de calcul consiste à remplacer la simulation coûteuse par un métamodèle. Des stratégies sont ensuite mises en place pour réaliser l’optimisation du composant à partir du métamodèle. Dans le cadre de cette thèse, trois dispositifs représentatifs des applications pouvant être traitées au sein du CEA LETI sont identifiés. L’étude de ces cas permet d’établir deux problématiques à résoudre. La première concerne la métamodélisation multi-fidélité, qui consiste à construire un métamodèle à partir de deux simulations du même composant ayant une précision différente. Les simulations sont obtenues à partir de différentes approximations du phénomène physique et aboutissent à un modèle appelé haute-fidélité (précis et coûteux) et un modèle basse fidélité (grossier et rapide à évaluer). Le travail sur cette méthode pour le cas de la cellule photoacoustique a amené au développement d’un nouveau métamodèle multifidélité basé sur les fonctions à base radiale. La deuxième problématique concerne la prise en compte des incertitudes de fabrication dans la conception de dispositifs photoniques. L’optimisation des performances de composants en tenant compte des écarts observés entre la géométrie désirée et la géométrie obtenue en fabrication a nécessité le développement d’une méthode spécifique pour le cas du coupleur adiabatique.