In line with the rising demand for smarter solutions and embedded systems, Microelectromechanical systems (MEMS) have gained increasing importance for digital computing devices and Internet-of-Things (IoT) applications, most notably for mobile wearable devices. This achievement is driven by MEMS resonators' inherent properties such as simplicity, sensitivity, reliability, and low power consumption. Hence, they are being explored for ultra-low-power computing machines. Several fundamental digital logic gates, switching, and memory devices have been demonstrated based on MEMS microstructures' static and dynamic behavior. The interest of researchers in using MEMS resonators is due to seeking an alternative approach to circumvent the notable current leakage and power density problems of complementary metal-oxide-semiconductor (CMOS) technology. The continuous miniaturization of CMOS has increased the operating speed and reduces the size of the device. However, this has led to a relative increase in the leakage energy. This drawback in CMOS has renewed the interest of researchers in mechanical digital computations, which can be traced back to the work of Charles Babbage in 1822 on calculating engines. This dissertation presents axially-loaded and coupled-MEMS resonators investigations to demonstrate memory elements and different logic functions. The studies in this dissertation can be categorized majorly into three parts based on the implementation of logic functions using three techniques: electrothermal frequency tunability, electrostatic frequency modulations, and activation/deactivation of the resonant frequency. Firstly, the influence of the competing effects of initial curvature and axial loads on the mechanical behavior of MEMS resonator arches are investigated theoretically to predict the tunability of arches under axial loads. Then, the concept of electrothermal frequency tunability is used to demonstrate fundamental 2-bit logic gates. However, this concept consumes a considerable amount of energy due to the electrothermal technique. Next, the dynamic memory element and combinational logic functions are demonstrated using the concept of electrostatic frequency modulation. Though this approach is energy efficient compared to the electrothermal technique, it does not support the cascadability of MEMS resonator-based logic devices. Lastly, complex multifunctional logic gates are implemented based on selective modes activation and deactivation, resulting in significant improvement in energy efficiency and enabling cascadability of MEMS resonator-based logic devices.

Cette thèse vise à proposer des outils innovants pour le calcul de structures aéronautiques évoluant à haute température. En effet, les régimes de fonctionnement des moteurs actuels conduisent à des évolutions élasto-viscoplastiques généralisées dans les pièces métalliques et l’utilisation de modèles simplifiés (élastiques) n’est plus totalement satisfaisante en terme de précision, même en phase de préconception. De même, la géométrie complexe permettant le refroidissement continu des pièces (micro-perforations) doit être prise en compte de manière exacte. Les techniques de calcul standard pour ce genre de problème conduiraient à des simulations lentes et peu flexibles (la moindre modification entraînant une remise en œuvre complète de la chaîne de calcul). Plus précisément, cette thèse étend les méthodes de type global/local non-intrusives au cas de la viscoplasticité généralisée en utilisant deux échelles de temps et d'espace, chacune adaptée aux phénomènes locaux et globaux à capturer. La méthode est ensuite étendue au calcul de nombreux cycles complexes de chargement, par des techniques de saut de cycles. Le schéma de couplage en temps permet alors une adaptation locale du pas de temps par sous-domaine. Des techniques d’accélération de convergence sont proposées, à l’échelle d’un incrément puis à celle de la succession de cycles (sauts de cycles). Ces développements permettent d’obtenir rapidement et précisément une estimation du cycle limite qui alimente un modèle de durée de vie. Le couplage non-intrusif est réalisé dans un script de programmation pilotant un code commercial (dans notre cas le langage Python et Abaqus/Standard). La méthode a été appliquée sur des plateformes de calculs industrielles, en réutilisant directement des maillages et les mises en données issues de modèles intervenant plus tôt dans la chaîne de calcul. Un cas métier, issu d’un bureau d’études de Safran Aircraft Engines, a pu être traité. / This thesis consists in developing innovating tools destined to the simulation of aeronautical structures evolving at high temperature. Indeed, working rates of current engines lead to an elasto-viscoplastic evolution generalized in metallic parts and the use of simplified models (linear elastic) are no longer totally satisfying in term of accuracy, even in initial design process. Likewise, the complex geometry allowing the continuous cool down process of parts (micro-perforations) has to be exactly taken into account. The standard computation techniques dedicated to this kind of models would lead to slow simulations with a lack of flexibility (the slightest modifications leading to restart the whole design process of the computation chain).More precisely, this thesis extends the noninvasive global/local methods to the framework of viscoplasticity generalized to the whole structure, using two scales in time and space, each one adapted to global and local phenomena to capture. The method is then extended to the computation of high number of complex load cycles, by skipped cycles techniques. The time coupling scheme lets then a local adaptation of time steps per subdomain. Convergence acceleration techniques are also set up, first for one time step and then through several load cycles (skipped cycles). These developments conduct to obtain quickly an evaluation of the limit cycle providing data to a lifetime expectancy model.The noninvasive coupling is realized in a programming language script managing the commercial software (respectively in our case Python and Abaqus/Standard). The method has been applied on industrial computational platforms, by reusing directly meshes and data from previous engineering tasks appearing earlier in the computational chain. A genuine test case from a Safran Aircraft Engines design office, was performed successfully.

Synapses are a fundamental unit of computation in the brain. Far from being passive connections between spiking neurons, synapses display striking short-term dynamics, undergo long-term changes in strength, and sculpt network-level processes in a complex manner. These synaptic dynamics, both in time and across space, may be a fundamental determinant of population-level computations and behavioral output of the brain, yet their role in neuromodulatory circuits is relatively under-explored. First, I developed and validated a set of likelihood-based inference tools to quantify the dynamics of synaptic ensemble composition throughout development. Second, I examined network computations in the serotonergic dorsal raphe nucleus through a dynamical lens, exploring the role of short-term synaptic dynamics at sparse recurrent connections, and of distinct long-range synaptic inputs, in shaping the output of spiking populations. 1. Simulation-based inference of synaptic ensembles. Functional features of synapses are typically inferred by sampling small ensembles of synapses, yet it is unclear if such subsamples exhibit biases. I developed a statistical framework to address this question, using it to demonstrate that common bulk electrical stimulation methods for characterizing the fraction of silent synapses exhibit high bias and variance, and using typical sample sizes, possess insufficient statistical power for accurate inference. I developed and validated a novel synthetic likelihood-based inference approach based on a simulator of the underlying experimental methodology. This new estimator, made available in an object-oriented Python toolbox, reduces bias and variance compared to previously reported methods, and provides a scalable method for examining synaptic dynamics throughout development. These tools were validated by targeted recording from hippocampal CA1 neurons in juvenile mice, where they reveal fundamental tradeoffs between release probability, number of synapses sampled, and statistical power. 2. Synaptic dynamics and population computations in the serotonin system. This part is comprised of two manuscripts. First, in the dorsal raphe nucleus, I uncovered slow, inhibitory recurrent interactions between serotonin neurons that are generated by local serotonin release. These connections were probabilistic, displayed striking short-term facilitation, gated the spiking output of serotonin neurons, and could be activated by long-range excitatory input from lateral habenula, representing threat signals. Targeted physiology and modeling revealed that these recurrent short-term facilitation features generated paradoxical excitation-driven inhibition in response to high-frequency habenula input. These facilitation rules additionally supported winner-take-all dynamics at the population level, providing a contrastive operation between functionally distinct serotonergic ensembles. Behaviorally, activating long-range lateral habenula input to dorsal raphe nucleus generated a transient, frequency-dependent suppression of reward anticipation consistent with these recurrent dynamics, without modulating the underlying reward association itself. These dynamics, we suggest, support sharp behavioral state transitions in changing environments. In a second manuscript, I explored the multiplexing of distinct long-range inputs in serotonergic circuits through spike synchrony. I demonstrated that a population of serotonergic neurons receives input from both lateral habenula and prefrontal cortex. These inputs produced similar subthreshold events, but prefrontal cortex triggered spikes with much higher latencies, supporting a population synchrony code for input identity. These input-specific spike timing patterns could be read out by simple linear decoders with high accuracy, suggesting they could be demultiplexed by downstream circuits receiving sparse innervation by serotonergic axons. We uncovered a novel intracellular calcium conductance in serotonergic neurons that altered the spectral characteristics of membrane voltage in a manner sufficient to generate long-latency, power law-distributed spike times, suggesting a simple dynamical origin for the production of synchronous or asynchronous spiking. This work indicates that serotonergic circuits can multiplex distinct informational streams through population spike synchrony mechanisms. Together, these investigations reveal that the dynamics of short-term facilitation and synaptic ensemble composition can act as the fundamental substrate for flexible computation by spiking networks across the brain.

No description available.

No description available.

No description available.

No / We analyze the possibility and efficiency of nonholonomic control over quantum devices with exponentially large number of Hilbert space dimensions. We show that completely controllable devices of this type can be assembled from elementary units of arbitrary physical nature, and can be employed efficiently for universal quantum computations and simulation of quantum-field dynamics. As an example we describe a toy device that can perform Toffoli-gate transformations and discrete Fourier transform on 9 qubits.

Lattice-Boltzmann method(LBM), a promising new particle-based simulation technique for complex and multiscale fluid flows, has seen tremendous adoption in recent years in computational fluid dynamics. Even with a state-of-the-art LBM solver such as Palabos, a user still has to manually write his program using the library-supplied primitives. We propose an automated code generator for a class of LBM computations with the objective to achieve high performance on modern architectures. Tiling is a very important loop transformation used to improve the performance of stencil computations by exploiting locality and parallelism. In the first part of the work, we explore diamond tiling, a new tiling technique to exploit the inherent ability of most stencils to allow tile-wise concurrent start. This enables perfect load-balance during execution and reduces the frequency of synchronization required. Few studies have looked at time tiling for LBM codes. We exploit a key similarity between stencils and LBM to enable polyhedral optimizations and in turn time tiling for LBM. Besides polyhedral transformations, we also describe a number of other complementary transformations and post processing necessary to obtain good parallel and SIMD performance on modern architectures. We also characterize the performance of LBM with the Roofline performance model. Experimental results for standard LBM simulations like Lid Driven Cavity, Flow Past Cylinder, and Poiseuille Flow show that our scheme consistently outperforms Palabos–on average by3 x while running on 16 cores of a n Intel Xeon Sandy bridge system. We also obtain a very significant improvement of 2.47 x over the native production compiler on the SPECLBM benchmark.

Stencil computations are iterative kernels often used to simulate the change in a discretized spatial domain overtime (e.g., computational fluid dynamics) or to solve for unknowns in a discretized space by converging to a steady state (i.e., partial differential equations).They are commonly found in many scientific and engineering applications. Most stencil computations allow tile-wise concurrent start ,i.e., there exists a face of the iteration space and a set of tiling hyper planes such that all tiles along that face can be started concurrently. This provides load balance and maximizes parallelism. Loop tiling is a key transformation used to exploit both data locality and parallelism from stencils simultaneously. Numerous works exist that target improving locality, controlling frequency of synchronization, and volume of communication wherever applicable. But, concurrent start-up of tiles that evidently translates into perfect load balance and often reduction in frequency of synchronization is completely ignored. Existing automatic tiling frameworks often choose hyperplanes that lead to pipelined start-up and load imbalance. We address this issue with a new tiling technique that ensures concurrent start-up as well as perfect load balance whenever possible. We first provide necessary and sufficient conditions on tiling hyperplanes to enable concurrent start for programs with affine data accesses. We then discuss an iterative approach to find such hyperplanes. It is not possible to directly apply automatic tiling techniques to periodic stencils because of the wrap-around dependences in them. To overcome this, we use iteration space folding techniques as a pre-processing stage after which our technique can be applied without any further change. We have implemented our techniques on top of Pluto-a source-level automatic parallelizer. Experimental evaluation on a 12-core Intel Westmere shows that our code is able to outperform a tuned domain-specific stencil code generator by 4% to2 x, and previous compiler techniques by a factor of 1.5x to 15x. For the swim benchmark from SPECFP2000, we achieve an .improvement of 5.12 x on a 12-core Intel Westmere and 2.5x on a 16-core AMD Magny-Cours machines, over the auto-parallelizer of Intel C Compiler.

La simulation moléculaire est un outil indispensable pour comprendre le comportement de systèmes complexes pour lesquels les expériences s'avèrent coûteuses ou irréalisables à l'heure actuelle. Cette thèse est dédiée aux aspects méthodologiques de la simulation moléculaire et comprend deux volets. Le premier volet porte sur la recherche de chemins de réaction et de points col d'une surface d'énergie potentielle. Nous proposons, dans le chaptire 3, une amélioration d'une des méthodes de cette classe, appelée '"Activation Relaxation Technique"(ART). Nous donnons également une preuve de convergence pour un algorithme prototype. Le deuxieme volet porte sur le calcul d'énergie libre pour les transitions caractérisées par une coordonnée de réaction. Nous nous plaçons dans le cadre d'une méthode d'échantillonnage d'importance adaptative, appelée 'Adaptive Biasing Force' (ABF). Ce volet comprend en soi deux sous-parties. La première partie (chapitre 5) s'attache à montrer l'applicabilité à un système biomoléculaire, d'une nouvelle mise en oeuvre parallèle d'ABF, nommée 'multiple-walker ABF' (MW-ABF), consistant à utiliser plusieurs répliques. Cette mise en oeuvre s'est avérée utile pour surmonter des problèmes liés à un mauvais choix de coordonnée de réaction. Nous confirmons ensuite ces résultats numériques en étudiant la convergence théorique d'un algorithme d'ABF adapté. Le chapitre 6 comprend une étude de convergence en temps long utilisant les méthodes d'entropie relative et les inégalités de Sobolev logarithmiques / Molecular simulation is an essential tool in understanding complex chemical and biochemical processes as real-life experiments prove increasingly costly or infeasible in practice . This thesis is devoted to methodological aspects of molecular simulation, with a particular focus on computing transition paths and their associated free energy profiles. The first part is dedicated to computational methods for reaction path and transition state searches on a potential energy surface. In Chapter 3 we propose an improvement to a widely-used transition state search method, the Activation Relaxation Technique (ART). We also present a local convergence study of a prototypical algorithm. The second part is dedicated to free energy computations. We focus in particular on an adaptive importance sampling technique, the Adaptive Biasing Force (ABF) method. The first contribution to this field, presented in Chapter 5, consists in showing the applicability to a large molecular system of a new parallel implementation, named multiple-walker ABF (MW-ABF). Numerical experiments demonstrated the robustness of MW-ABF against artefacts arising due to poorly chosen or oversimplified reaction coordinates. These numerical findings inspired a new study of the longtime convergence of the ABF method, as presented in Chapter 6. By studying a slightly modified model, we back our numerical results by showing a faster theoretical rate of convergence of ABF than was previously shown