Mecanismos de reconfiguração dinâmica aplicados ao projeto de um processador de imagens reconfigurável

As modernas aplicações em diversas áreas como multimídia e telecomunicações exigem arquiteturas que ofereçam altas taxas de processamento. Entretanto, os padrões e algoritmos mudam com incrível rapidez o que gera a necessidade de que esses sistemas digitais tenham também por característica uma grande flexibilidade. Dentro desse contexto, tem-se as arquiteturas reconfiguráveis em geral e, mais recentemente, os sistemas reconfiguráveis em um único chip como soluções adequadas que podem oferecer desempenho, sendo, ao mesmo tempo, adaptáveis a novos problemas e a classes mais amplas de algoritmos dentro de um dado escopo de aplicação. Este trabalho apresenta o estado-da-arte em relação a arquiteturas reconfiguráveis nos meios acadêmcio e industrial e descreve todas as etapas de desenvolvimento do processador de imagens reconfigurável DRIP (Dynamically Reconfigurable Image Processor), desde suas origens como um processador estático até sua última versão reconfigurável em tempo de execução. O DRIP possui um pipeline composto por 81 processadores elementares. Esses processadores constituem a chave do processo de reconfiguração e possuem a capacidade de computar um grande número de algoritmos de processamento de imagens, mais específicamente dentro do domínio da filtragem digital de imagens. Durante o projeto, foram desenvolvidos uma série de modelos em linguagem de descrição de hardware da arquitetura e também ferramentas de software para auxiliar nos processos de implementação de novos algorimos, geração automática de modelos VHDL e validação das implementações. O desenvolvimento de mecanismos com o objetivo de incluir a possibilidade de reconfiguração dinâmica, naturalmente, introduz overheads na arquitetura. Contudo, o processo de reconfiguração do DRIP-RTR é da ordem de milhões de vezes mais rápido do que na versão estaticamente reconfigurável implementada em FPGAs Altera. Finalizando este trabalho, é apresentado o projeto lógico e elétrico do processador elementar do DRIP, visando uma futura implementação do sistema diretamente como um circuito VLSI.

Implementação em hardware de uma Rede Neuronal com ummicroprocessador embebido em FPGA

O objectivo deste trabalho é a implementação em hardware de uma Rede Neuronal com um microprocessador embebido, podendo ser um recurso valioso em várias áreas científicas. A importância das implementações em hardware deve-se à flexibilidade, maior desempenho e baixo consumo de energia. Para esta implementação foi utilizado o dispositivo FPGA Virtex II Pro XC2VP30 com um MicroBlaze soft core, da Xilinx. O MicroBlaze tem vantagens como a simplicidade no design, sua reutilização e fácil integração com outras tecnologias. A primeira fase do trabalho consistiu num estudo sobre o FPGA, um sistema reconfigurável que possui características importantes como a capacidade de executar em paralelo tarefas complexas. Em seguida, desenvolveu-se o código de implementação de uma Rede Neuronal Artificial baseado numa linguagem de programação de alto nível. Na implementação da Rede Neuronal aplicou-se, na camada escondida, a função de activação tangente hiperbólica, que serve para fornecer a não linearidade à Rede Neuronal. A implementação é feita usando um tipo de Rede Neuronal que permite apenas ligações no sentido de saída, chamado Redes Neuronais sem realimentação (do Inglês Feedforward Neural Networks - FNN). Como as Redes Neuronais Artificiais são sistemas de processamento de informações, e as suas características são comuns às Redes Neuronais Biológicas, aplicaram-se testes na implementação em hardware e analisou-se a sua importância, a sua eficiência e o seu desempenho. E finalmente, diante dos resultados, fez-se uma análise de abordagem e metodologia adoptada e sua viabilidade.

Desenvolvimento de uma arquitetura em hardware prototipada em FPGA para aplicações genéricas utilizando redes neurais artificiais embarcadas

This work proposes hardware architecture, VHDL described, developed to embedded Artificial Neural Network (ANN), Multilayer Perceptron (MLP). The present work idealizes that, in this architecture, ANN applications could easily embed several different topologies of MLP network industrial field. The MLP topology in which the architecture can be configured is defined by a simple and specifically data input (instructions) that determines the layers and Perceptron quantity of the network. In order to set several MLP topologies, many components (datapath) and a controller were developed to execute these instructions. Thus, an user defines a group of previously known instructions which determine ANN characteristics. The system will guarantee the MLP execution through the neural processors (Perceptrons), the components of datapath and the controller that were developed. In other way, the biases and the weights must be static, the ANN that will be embedded must had been trained previously, in off-line way. The knowledge of system internal characteristics and the VHDL language by the user are not needed. The reconfigurable FPGA device was used to implement, simulate and test all the system, allowing application in several real daily problems

Proposta e implementação de uma arquitetura reconfigurável híbrida para aplicações baseadas em fluxo de dados

The increase of applications complexity has demanded hardware even more flexible and able to achieve higher performance. Traditional hardware solutions have not been successful in providing these applications constraints. General purpose processors have inherent flexibility, since they perform several tasks, however, they can not reach high performance when compared to application-specific devices. Moreover, since application-specific devices perform only few tasks, they achieve high performance, although they have less flexibility. Reconfigurable architectures emerged as an alternative to traditional approaches and have become an area of rising interest over the last decades. The purpose of this new paradigm is to modify the device s behavior according to the application. Thus, it is possible to balance flexibility and performance and also to attend the applications constraints. This work presents the design and implementation of a coarse grained hybrid reconfigurable architecture to stream-based applications. The architecture, named RoSA, consists of a reconfigurable logic attached to a processor. Its goal is to exploit the instruction level parallelism from intensive data-flow applications to accelerate the application s execution on the reconfigurable logic. The instruction level parallelism extraction is done at compile time, thus, this work also presents an optimization phase to the RoSA architecture to be included in the GCC compiler. To design the architecture, this work also presents a methodology based on hardware reuse of datapaths, named RoSE. RoSE aims to visualize the reconfigurable units through reusability levels, which provides area saving and datapath simplification. The architecture presented was implemented in hardware description language (VHDL). It was validated through simulations and prototyping. To characterize performance analysis some benchmarks were used and they demonstrated a speedup of 11x on the execution of some applications

Projeto e implementação em FPGA de um processador com conjunto de instrução reconfigurável utilizando VHDL

The Reconfigurable Computing is an intermediate solution at the resolution of complex problems, making possible to combine the speed of the hardware with the flexibility of the software. An reconfigurable architecture possess some goals, among these the increase of performance. The use of reconfigurable architectures to increase the performance of systems is a well known technology, specially because of the possibility of implementing certain slow algorithms in the current processors directly in hardware. Amongst the various segments that use reconfigurable architectures the reconfigurable processors deserve a special mention. These processors combine the functions of a microprocessor with a reconfigurable logic and can be adapted after the development process. Reconfigurable Instruction Set Processors (RISP) are a subgroup of the reconfigurable processors, that have as goal the reconfiguration of the instruction set of the processor, involving issues such formats, operands and operations of the instructions. This work possess as main objective the development of a RISP processor, combining the techniques of configuration of the set of executed instructions of the processor during the development, and reconfiguration of itself in execution time. The project and implementation in VHDL of this RISP processor has as intention to prove the applicability and the efficiency of two concepts: to use more than one set of fixed instructions, with only one set active in a given time, and the possibility to create and combine new instructions, in a way that the processor pass to recognize and use them in real time as if these existed in the fixed set of instruction. The creation and combination of instructions is made through a reconfiguration unit, incorporated to the processor. This unit allows the user to send custom instructions to the processor, so that later he can use them as if they were fixed instructions of the processor. In this work can also be found simulations of applications involving fixed and custom instructions and results of the comparisons between these applications in relation to the consumption of power and the time of execution, which confirm the attainment of the goals for which the processor was developed

Desenvolvimento de uma arquitetura reconfigurável para o processamento de modelos no ambiente ABACUS

Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)

Mecanismo de tolerância a falhas através de escalonamento para uma arquitetura reconfigurável de grão grosso

The continuous evolution of integrated circuit technology has allowed integrating thousands of transistors on a single chip. This is due to the miniaturization process, which reduces the diameter of wires and transistors. One drawback of this process is that the circuit becomes more fragile and susceptible to break, making the circuit more susceptible to permanent faults during the manufacturing process as well as during their lifetime. Coarse Grained Reconfigurable Architectures (CGRAs) have been used as an alternative to traditional architectures in an attempt to tolerate such faults due to its intrinsic hardware redundancy and high performance. This work proposes a fault tolerance mechanism in a CGRA in order to increase the architecture fault tolerance even considering a high fault rate. The proposed mechanism was added to the scheduler, which is the mechanism responsible for mapping instructions onto the architecture. The instruction mapping occurs at runtime, translating binary code without the need for recompilation. Furthermore, to allow faster implementation, instruction mapping is performed using a greedy module scheduling algorithm, which consists of a software pipeline technique for loop acceleration. The results show that, even with the proposed mechanism, the time for mapping instructions is still in order of microseconds. This result allows that instruction mapping process remains at runtime. In addition, a study was also carried out mapping scheduler rate. The results demonstrate that even at fault rates over 50% in functional units and interconnection components, the scheduler was able to map instructions onto the architecture in most of the tested applications.

Implementação de uma matriz de neurônios dinamicamente reconfigurável para descrição de topologias de redes neurais artificiais multilayer perceptrons

The Artificial Neural Networks (ANN), which is one of the branches of Artificial Intelligence (AI), are being employed as a solution to many complex problems existing in several areas. To solve these problems, it is essential that its implementation is done in hardware. Among the strategies to be adopted and met during the design phase and implementation of RNAs in hardware, connections between neurons are the ones that need more attention. Recently, are RNAs implemented both in application specific integrated circuits's (Application Specific Integrated Circuits - ASIC) and in integrated circuits configured by the user, like the Field Programmable Gate Array (FPGA), which have the ability to be partially rewritten, at runtime, forming thus a system Partially Reconfigurable (SPR), the use of which provides several advantages, such as flexibility in implementation and cost reduction. It has been noted a considerable increase in the use of FPGAs for implementing ANNs. Given the above, it is proposed to implement an array of reconfigurable neurons for topologies Description of artificial neural network multilayer perceptrons (MLPs) in FPGA, in order to encourage feedback and reuse of neural processors (perceptrons) used in the same area of the circuit. It is further proposed, a communication network capable of performing the reuse of artificial neurons. The architecture of the proposed system will configure various topologies MLPs networks through partial reconfiguration of the FPGA. To allow this flexibility RNAs settings, a set of digital components (datapath), and a controller were developed to execute instructions that define each topology for MLP neural network.

Implementação de uma matriz de neurônios dinamicamente reconfigurável para descrição de topologias de redes neurais artificiais multilayer perceptrons

The Artificial Neural Networks (ANN), which is one of the branches of Artificial Intelligence (AI), are being employed as a solution to many complex problems existing in several areas. To solve these problems, it is essential that its implementation is done in hardware. Among the strategies to be adopted and met during the design phase and implementation of RNAs in hardware, connections between neurons are the ones that need more attention. Recently, are RNAs implemented both in application specific integrated circuits's (Application Specific Integrated Circuits - ASIC) and in integrated circuits configured by the user, like the Field Programmable Gate Array (FPGA), which have the ability to be partially rewritten, at runtime, forming thus a system Partially Reconfigurable (SPR), the use of which provides several advantages, such as flexibility in implementation and cost reduction. It has been noted a considerable increase in the use of FPGAs for implementing ANNs. Given the above, it is proposed to implement an array of reconfigurable neurons for topologies Description of artificial neural network multilayer perceptrons (MLPs) in FPGA, in order to encourage feedback and reuse of neural processors (perceptrons) used in the same area of the circuit. It is further proposed, a communication network capable of performing the reuse of artificial neurons. The architecture of the proposed system will configure various topologies MLPs networks through partial reconfiguration of the FPGA. To allow this flexibility RNAs settings, a set of digital components (datapath), and a controller were developed to execute instructions that define each topology for MLP neural network.

Integrating hardware and software information flow analyses

Security-critical communications devices must be evaluated to the highest possible standards before they can be deployed. This process includes tracing potential information flow through the device's electronic circuitry, for each of the device's operating modes. Increasingly, however, security functionality is being entrusted to embedded software running on microprocessors within such devices, so new strategies are needed for integrating information flow analyses of embedded program code with hardware analyses. Here we show how standard compiler principles can augment high-integrity security evaluations to allow seamless tracing of information flow through both the hardware and software of embedded systems. This is done by unifying input/output statements in embedded program execution paths with the hardware pins they access, and by associating significant software states with corresponding operating modes of the surrounding electronic circuitry.

Power network in loop : a paradigm for real-time simulation and hardware testing

This paper discusses a new paradigm of real-time simulation of power systems in which equipment can be interfaced with a real-time digital simulator. In this scheme, one part of a power system can be simulated by using a real-time simulator; while the other part is implemeneted as a physical system. The only interface of the physical system with the computer-based simulator is through data-acquisition system. The physical system is driven by a voltage-source converter (VSC)that mimics the power system simulated in the real-time simulator. In this papar, the VSC operates in a voltage-control mode to track the point of common coupling voltage signal supplied by the digital simulator. This type of splitting a network in two parts and running a real-time simulation with a physical system in parallel is called a power network in loop here. this opens up the possibility of study of interconnection o f one or several distributed generators to a complex power network. The proposed implementation is verified through simulation studies using PSCAD/EMTDC and through hardware implementation on a TMS320G2812 DSP.

Low-cost hardware speech enhancement for improved speech recognition in automotive environments

Voice recognition is one of the key enablers to reduce driver distraction as in-vehicle systems become more and more complex. With the integration of voice recognition in vehicles, safety and usability are improved as the driver’s eyes and hands are not required to operate system controls. Whilst speaker independent voice recognition is well developed, performance in high noise environments (e.g. vehicles) is still limited. La Trobe University and Queensland University of Technology have developed a low-cost hardware-based speech enhancement system for automotive environments based on spectral subtraction and delay–sum beamforming techniques. The enhancement algorithms have been optimised using authentic Australian English collected under typical driving conditions. Performance tests conducted using speech data collected under variety of vehicle noise conditions demonstrate a word recognition rate improvement in the order of 10% or more under the noisiest conditions. Currently developed to a proof of concept stage there is potential for even greater performance improvement.

A synthesizable hardware evolutionary algorithm design for unmanned aerial system real-time path planning

The main objective of this paper is to detail the development of a feasible hardware design based on Evolutionary Algorithms (EAs) to determine flight path planning for Unmanned Aerial Vehicles (UAVs) navigating terrain with obstacle boundaries. The design architecture includes the hardware implementation of Light Detection And Ranging (LiDAR) terrain and EA population memories within the hardware, as well as the EA search and evaluation algorithms used in the optimizing stage of path planning. A synthesisable Very-high-speed integrated circuit Hardware Description Language (VHDL) implementation of the design was developed, for realisation on a Field Programmable Gate Array (FPGA) platform. Simulation results show significant speedup compared with an equivalent software implementation written in C++, suggesting that the present approach is well suited for UAV real-time path planning applications.