Auswahl der wissenschaftlichen Literatur zum Thema „Massively parallelised micro“

The recent explosion in the speed and connectivity of the Internet has opened up the possibility of millions and possibly billions of devices connected together. Combined with the development of small, low power devices, new paradigms in the field of computing have opened up. Traditional passive electronic devices now have rudimentary computing capabilities. The resulting Internet of Things (IoT), comprised of smart interconnected devices is improving our ability to gather ambient information and make informed decisions that directly benefit humanity. However, the ubiquity of these devices also presents an interesting scenario wherein the devices can perform limited general-purpose computations when they are not performing their primary functions. A computational task divided into a large number of smaller, micro tasks, each of which take only a few CPU cycles to complete. By distributing these tasks over a large number of devices, we can achieve a substantial amount of computation with seemingly modest devices. In this work, we explore a mechanism to enable such massively parallel computations in low powered commodity hardware devices through fine-grained task parallelism.

Graphics processing units (GPUs) facilitate massive parallelism and high-capacity storage, and thus are suitable for the iterative reconstruction of ultrahigh-resolution micro computed tomography (CT) scans by on-the-fly system matrix (OTFSM) calculation using ordered subsets expectation maximization (OSEM). We propose a finite state automaton (FSA) method that facilitates iterative reconstruction using a heterogeneous multi-GPU platform through parallelizing the matrix calculations derived from a ray tracing system of ordered subsets. The FSAs perform flow control for parallel threading of the heterogeneous GPUs, which minimizes the latency of launching ordered-subsets tasks, reduces the data transfer between the main system memory and local GPU memory, and solves the memory-bound of a single GPU. In the experiments, we compared the operation efficiency of OS-MLTR for three reconstruction environments. The heterogeneous multiple GPUs with job queues for high throughput calculation speed is up to five times faster than the single GPU environment, and that speed up is nine times faster than the heterogeneous multiple GPUs with the FIFO queues of the device scheduling control. Eventually, we proposed an event-triggered FSA method for iterative reconstruction using multiple heterogeneous GPUs that solves the memory-bound issue of a single GPU at ultrahigh resolutions, and the routines of the proposed method were successfully executed on each GPU simultaneously.

Automated technologies that can perform massively parallelized and sequential fluidic operations at small length scales can resolve major bottlenecks encountered in various fields, including medical diagnostics, -omics, drug development, and chemical/material synthesis. Inspired by the transformational impact of automated guided vehicle systems on manufacturing, warehousing, and distribution industries, we devised a ferrobotic system that uses a network of individually addressable robots, each performing designated micro-/nanofluid manipulation-based tasks in cooperation with other robots toward a shared objective. The underlying robotic mechanism facilitating fluidic operations was realized by addressable electromagnetic actuation of miniature mobile magnets that exert localized magnetic body forces on aqueous droplets filled with biocompatible magnetic nanoparticles. The contactless and high-strength nature of the actuation mechanism inherently renders it rapid (~10 centimeters/second), repeatable (>10,000 cycles), and robust (>24 hours). The robustness and individual addressability of ferrobots provide a foundation for the deployment of a network of ferrobots to carry out cross-collaborative logistics efficiently. These traits, together with the reconfigurability of the system, were exploited to devise and integrate passive/active advanced functional components (e.g., droplet dispensing, generation, filtering, and merging), enabling versatile system-level functionalities. By applying this ferrobotic system within the framework of a microfluidic architecture, the ferrobots were tasked to work cross-collaboratively toward the quantification of active matrix metallopeptidases (a biomarker for cancer malignancy and inflammation) in human plasma, where various functionalities converged to achieve a fully automated assay.

The performance of current speech recognition systems is far below that of humans. Neural nets offer the potential of providing massive parallelism, adaptation, and new algorithmic approaches to problems in speech recognition. Initial studies have demonstrated that multilayer networks with time delays can provide excellent discrimination between small sets of pre-segmented difficult-to-discriminate words, consonants, and vowels. Performance for these small vocabularies has often exceeded that of more conventional approaches. Physiological front ends have provided improved recognition accuracy in noise and a cochlea filter-bank that could be used in these front ends has been implemented using micro-power analog VLSI techniques. Techniques have been developed to scale networks up in size to handle larger vocabularies, to reduce training time, and to train nets with recurrent connections. Multilayer perceptron classifiers are being integrated into conventional continuous-speech recognizers. Neural net architectures have been developed to perform the computations required by vector quantizers, static pattern classifiers, and the Viterbi decoding algorithm. Further work is necessary for large-vocabulary continuous-speech problems, to develop training algorithms that progressively build internal word models, and to develop compact VLSI neural net hardware.

While additive manufacturing based on multi photon polymerization is currently considered to be a very promising technique for the fabrication of 3D micro and nano structures, long fabrication times are a major limitation of this approach. Parallelization of the fabrication process is an important technique to overcome this issue. The fabrication process is parallelized by imaging a 1920x1080 pixel Spatial Light Modulator (SLM) into an ultra‐sensitive Triplet‐Triplet Annihilation (TTA) resist. However, proximity effects between close pixels generate uncontrolled polymerization and make the controlled fabrication of 3D structures difficult. This work models light propagation and chemical interactions in our system to predict fabricated structures with a view to precompensating plot data and improving 3D resolution by performing optical and chemical proximity correction. Our simple model gives reasonable predictions of fabricated structures helping us fabricate fully 3D structures in parallel.This article is protected by copyright. All rights reserved.

Les structures de taille submicroniques en 3D sont utiles dans de nombreux domaines (photonique,optique, biologie...). La fabrication de telles structures est difficile. La polymérisation multiphotonique est une technique adaptée, mais les temps de fabrication actuels sont longs (une journée pour fabriquer une structure d’un mm3), rendant la production industrielle couteuse et limitant le développement de ces structures. Nous présentons notre contribution au développement et à l’optimisation d’un procédé de fabrication rapide de ces structures par polymérisation multiphotonique massivement parallélisée. Deux techniques de parallélisation sont étudiées à IMT Atlantique : une avec un élément optique diffractif, et l’autre, plus étudiée dans cette thèse, avec un modulateur spatial de lumière en configuration imagerie et une résine ultrasensible TTA (annihilation triplet-triplet), permettant d’écrire avec 1920×1080 faisceaux en parallèle. L’utilisation de multiples faisceaux d’écriture peut entraîner des effets de proximité qui limitent la résolution. Nous présentons notre simulation numérique du processus photochimique pour comprendre, prédire et corriger ces effets. Ensuite, nous présentons des améliorations effectuées, identifiées grâce aux simulations et à une meilleure compréhension du système optique. La méthode de fabrication développée permet de fabriquer des structures avec une résolution d’environ un micromètre en X,Y et de plusieurs dizaines de micromètres de hauteur sur des surfaces de l’ordre du cm2 en quelques minutes. Enfin, des exemples d’applications en biologie et en ophtalmologie, adaptés à ces performances, sont présentés
Submicron 3D structures are required in many fields (photonics, optics, biology, etc.). Fabricating such structures is difficult. Multiphoton polymerization is a suitable technique, but current fabrication times are long (one day to fabricate a mm3 structure), making industrial production costly and limiting the development of these structures. We present our contribution to the development and optimization of a massively parallelised multiphoton polymerization fabrication process for these structures. Two parallelization techniques are investigated at IMT Atlantique: one using a diffractive optical element and another, studied in this thesis, using a spatial light modulator in an imaging configuration and an ultra-sensitive TTA resist (Triplet-Triplet Annihilation), enabling writing with 1920 × 1080 beams in parallel. The use of multiple write beams can lead to resolution limiting proximity effects. We present our numerical simulation model of the photochemical process to understand, predict and correct these effects. We present possible improvements based on these simulations and the improved understanding of the optical system. The fabrication method we have developed enables us to fabricate structures with a resolution of around one micrometer in X,Y and several tens of micrometers in height on surfaces of the order of cm2 in just a few minutes. Finally, examples of applications in biology and ophthalmology, adapted to the photoplotter performance are presented

Il est reconnu aujourd'hui que pour un grand nombre d'applications les performances globales des systèmes sont fortement limitées faute d'un transfert suffisament rapide entre les unités de calcul et les dispositifs de stockage. L'idée développée au long de cette thèse est qu'il est possible de réaliser un système d'E/S universel et performant dans un environnement extensible si l'on respecte quelques principes dans sa conception. Pour ce faire, il est nécessaire d'y faire participer le matériel, le système d'exploitation, le système de fichiers et les utilisateurs, chacun au niveau approprié. Notre travail intègre toutes les composantes d'un sous-système d'E/S. En premier lieu, nous choisissons une architecture matérielle adéquate aux divers types de demandes d'E/S observés dans les applications parallèles. Nous présentons une architecture universelle et extensible qui permet de maximiser l'exploitation du parallélisme. En deuxième lieu, nous utilisons ParX, un micro-noyau parallèle conçu à l'intérieur de notre équipe, pour fournir les mécanismes de base à l'exécution d'un système de fichiers parallèle. Nous concrétisons d'abord certaines extensions indispensables pour mieux adapter ParX aux besoins des E/S parallèles, et ensuite, afin d'exploiter la projection des fichiers dans l'espace d'adressage, nous développons des mécanismes originaux, nécessaires à l'implémentation d'un espace d'adressage commun dans une architecture extensible à mémoire distribuée. En troisième lieu, nous introduisons les principes de base qui doivent être respectés afin de concilier la généralité et les hautes performances dans la conception d'un système de fichiers parallèle extensible. L'architecture du système de fichiers proposée à la fin du rapport est le résultat de l'application de ces principes.