Designing hardware to output pixels for light field displays or multi-projector systems is challenging owing to the memory bandwidth and speed of the application. A new technique of hardware that implements ‗anywhere pixel routing‘ was designed earlier at the University of Kentucky. This technique uses hardware to route pixels from input to output based upon a Look up Table (LUT). The initial design suffered from high memory latency due to random accesses to the DDR SDRAM input buffer. This thesis presents a cache design that alleviates the memory latency issue by reducing the number of random SDRAM accesses. The cache is implemented in the block RAM of a field programmable gate array (FPGA). A number of simulations are conducted to find an efficient cache. It is found that the cache takes only a few kilobits, about 7% of the block RAM and on an average speeds up the memory accesses by 20-30%.

Thesis--Auburn University, 2009. / Abstract. Vita. Includes bibliographic resources (p.88-89).

Classical Molecular Dynamics simulation (MD) models the interactions of thousands to millions of particles through the iterative application of basic Physics. MD is one of the core methods in High Performance Computing (HPC). While MD is critical to many high-profile applications, e.g. drug discovery and design, it suffers from the strong scaling problem, that is, while large computer systems can efficiently model large ensembles of particles, it is extremely challenging for {\it any} computer system to increase the timescale, even for small ensembles. This strong scaling problem can be mitigated with low-latency, direct communication. Of all Commercial Off the Shelf (COTS) Integrated Circuits (ICs), Field Programmable Gate Arrays (FPGAs) are the computational component uniquely applicable here: they have unmatched parallel communication capability both within the chip and externally to couple clusters of FPGAs. This thesis focuses on the acceleration of the long range (LR) force, the part of MD most difficult to scale, by using FPGAs. This thesis first optimizes LR acceleration on a single-FPGA to eliminate the amount of on-chip communication required to complete a single LR computation iteration while maintaining as much parallelism as possible. This is achieved by designing around application specific memory architectures. Doing so introduces data movement issues overcome by pipelined, toroidal-shift multiplexing (MUXing) and pipelined staggering of memory access subsets. This design is then evaluated comprehensively and comparatively, deriving equations for performance and resource consumption and drawing metrics from previously developed LR hardware designs. Using this single-FPGA LR architecture as a base, FPGA network strategies to compute the LR portion of larger sized MD problems are then theorized and analyzed.

This master's thesis describes the design of effective working artificial neural network in FPGA Virtex-5 series with the maximum use of the possibility of parallelization. The theoretical part contains basic information on artificial neural networks, FPGA and VHDL. The practical part describes the used format of the variables, creating non-linear function, the principle of calculation the single layers, or the possibility of parameter settings generated artificial neural networks.

This dissertation studies the effects of radiation-induced single-event upsets (SEUs) on field-programmable gate array(FPGA)-based circuits. It analyzes and quantifies a special case in data communication, that is, the synchronization issue of signals when they are sent across clock domains in triple modular redundancy (TMR) circuits with the presence of SEUs. After demonstrating that synchronizing errors cannot be eliminated in such case, this dissertation continues to present novel synchronizer designs that can guarantee reliable synchronization of triplicated signals. Fault injection tests then show that the proposed synchronizers provide between 6 and 10 orders of magnitude longer mean time to failure (MTTF) than unmitigated synchronizers. This dissertation also studies the reliability of block random access memory (BRAM) on FPGAs. By investigating several previous reliability models for single-error correction/double-error detection (SEC/DED) memory with scrubbing, this dissertation proposes two novel MTTF models that are suitable for FPGA applications. The first one considers non-uniform write rates for probabilistic write scrubbing, and the second one combines deterministic scrubbing and probabilistic scrubbing into a single model. The proposed models reveal the impact of memory access patterns on the reliability of BRAMs. Monte Carlo simulations then demonstrate the correctness of the proposed models. At last, the memory access patterns of a type of FPGA application, digital signal processing (DSP) is studied, and mitigation mechanisms for DSP applications are discussed.