Fernando E. Ortiz

FPGA-based acceleration of the 3D finite-difference time-domain method

In order to take advantage of the significant benefits afforded by computational electromagnetic techniques, such as the finite-difference time-domain (FDTD) method, solvers capable of analyzing realistic problems in a reasonable time frame are required. Although software-based solvers are frequently used, they are often too slow to be of practical use. To speed up computations, hardware-based implementations of the FDTD method have been recently proposed. In this paper, we present our most recent progress in the area of FPGA-based 3D FDTD accelerators. Three aspects of the design are discussed, including the host-PC interface, memory hierarchy, and computational datapath. Implementation and benchmarking results are also presented, demonstrating that this accelerator is capable of at least three-fold speedups over thirty-node PC clusters.

Hardware implementation of a three-dimensional finite-difference time-domain algorithm

In order to take advantage of the significant benefits afforded by computational electromagnetic techniques, such as the finite-difference time-domain (FDTD) method, solvers capable of analyzing realistic problems in a reasonable time frame are required. Although software-based solvers are frequently used, they are often too slow to be of practical use. To speed up computations, hardware-based implementations of the FDTD method have recently been proposed. Although these designs are functionally correct, to date, they have not provided a practical and scalable solution. To this end, we have developed an architecture that not only overcomes the limitations of previous accelerators, but also represents the first three-dimensional FDTD accelerator implemented in physical hardware. We present a high-level view of the system architecture and describe the basic functionality of each module involved in the computational flow. We then present our implementation results and compare them with current PC-based FDTD solutions. These results indicate that hardware solutions will, in the near future, surpass existing PC throughputs, and will ultimately rival the performance of PC clusters.

Hardware acceleration of the 3D finite-difference time-domain method

Although the importance of fast, accurate computational electromagnetic (CEM) solvers is readily apparent, how to construct them is not. By nature, CEM algorithms are both computationally and memory intensive. Furthermore, the serial nature of most software-based implementations does not take advantage of the inherent parallelism found in many CEM algorithms. In an attempt to exploit parallelism, supercomputers and computer clusters are employed. However, these solutions can be prohibitively expensive and frequently impractical. Thus, a CEM accelerator or CEM co-processor would provide the community with much-needed processing power. This would enable iterative designs and designs that would otherwise be impractical to analyze. To this end, we are developing a full-3D, hardware-based accelerator for the finite-difference time-domain (FDTD) method (K.S. Yee, IEEE Trans. Antennas and Propag., vol. 14, pp. 302-307, 1966). This accelerator provides speedups of up to three orders of magnitude over single-PC solutions and will surpass the throughputs of the PC clusters. In this paper, we briefly summarize previous work in this area, where it has fallen short, and how our work fills the void. We then describe the current status of this project, summarizing our achievements to date and the work that remains. We conclude with the projected results of our accelerator.

Real-time Embedded Atmospheric Compensation for Long-Range Imaging Using the Average Bispectrum Speckle Method

While imaging over long distances is critical to a number of security and defense applications, such as homeland security and launch tracking, current optical systems are limited in resolving power. This is largely a result of the turbulent atmosphere in the path between the region under observation and the imaging system, which can severely degrade captured imagery. There are a variety of post-processing techniques capable of recovering this obscured image information; however, the computational complexity of such approaches has prohibited real-time deployment and hampers the usability of these technologies in many scenarios. To overcome this limitation, we have designed and manufactured an embedded image processing system based on commodity hardware which can compensate for these atmospheric disturbances in real-time. Our system consists of a reformulation of the average bispectrum speckle method coupled with a high-end FPGA processing board, and employs modular I/O capable of interfacing with most common digital and analog video transport methods (composite, component, VGA, DVI, SDI, HD-SDI, etc.). By leveraging the custom, reconfigurable nature of the FPGA, we have achieved performance twenty times faster than a modern desktop PC, in a form-factor that is compact, low-power, and field-deployable.

Modeling and simulation of nanoscale devices with a desktop supercomputer

Designing nanoscale devices presents a number of unique challenges. As device features shrink, the computational demands of the simulations necessary to accurately model them increase significantly. This is a result of not only the increasing level of detail in the device design itself, but also the need to use more accurate models. The approximations that are generally made when dealing with larger devices break down as feature sizes decrease. This can be seen in the optics field when contrasting the complexity of physical optics models with those requiring a rigorous solution to Maxwells equations. This added complexity leads to more demanding calculations, stressing computational resources and driving research to overcome these limitations. There are traditionally two means of improving simulation times as model complexity grows beyond available computational resources: modifying the underlying algorithms to maintain sufficient precision while reducing overall computations and increasing the power of the computational system. In this paper, we explore the latter. Recent advances in commodity hardware technologies, particularly field-programmable gate arrays (FPGAs) and graphics processing units (GPUs), have allowed the creation of desktop-style devices capable of outperforming PC clusters. We will describe the key hardware technologies required to build such a device and then discuss their application to the modeling and simulation of nanophotonic devices. We have found that FPGAs and GPUs can be used to significantly reduce simulation times and allow for the solution of much large problems.

Implementation of three-dimensional FPGA-based FDTD solvers: an architectural overview

Maxwells equations, which govern electromagnetic propagation, are a system of coupled, differential equations. As such, they can be represented in difference form, thus allowing their numerical solution. By implementing both the temporal and spatial derivatives of Maxwells equations in difference form, we arrive at one of the most common computational electromagnetic algorithms, the Finite-Difference Time-Domain (FDTD) method (Yee, 1966). In this technique, the region of interest is sampled to generate a grid of points, hereafter referred to as a mesh. The discretized form of Maxwells equations is then solved at each point in the mesh to determine the associated electromagnetic fields. In this extended abstract, we present an architecture that overcomes the previous limitations. We begin with a high-level description of the computational flow of this architecture.

Comparing FPGAs and GPUs for high-performance image processing applications

Modern image enhancement techniques have been shown to be effective in improving the quality of imagery. However, the computational requirements of applying such algorithms to streams of video in real-time often cannot be satisfied by standard microprocessor-based systems. While a scaled solution involving clusters of microprocessors may provide the necessary arithmetic capacity, deployment is limited to data-center scenarios. What is needed is a way to perform these techniques in real time on embedded platforms. A new paradigm of computing utilizing special-purpose commodity hardware including Field-Programmable Gate Arrays (FPGAs) and Graphics Processing Units (GPU) has recently emerged as an alternative to parallel computing using clusters of traditional CPUs. Recent research has shown that for many applications, such as image processing techniques requiring intense computations and large memory spaces, these hardware platforms significantly outperform microprocessors. Furthermore, while microprocessor technology has begun to stagnate, GPUs and FPGAs have continued to improve exponentially. FPGAs, flexible and powerful, are best targeted at embedded, low-power systems and specific applications. GPUs, cheap and readily available, are available to most users through their standard desktop machines. Additionally, as fabrication scale continues to shrink, heat and power consumption issues typically limiting GPU deployment to high-end desktop workstations are becoming less of a factor. The ability to include these devices in embedded environments opens up entire new application domains. In this paper, we investigate two state-of-the-art image processing techniques, super-resolution and the average-bispectrum speckle method, and compare FPGA and GPU implementations in terms of performance, development effort, cost, deployment options, and platform flexibility.

An embedded processor for real-time atmoshperic compensation

Imaging over long distances is crucial to a number of defense and security applications, such as homeland security and launch tracking. However, the image quality obtained from current long-range optical systems can be severely degraded by the turbulent atmosphere in the path between the region under observation and the imager. While this obscured image information can be recovered using post-processing techniques, the computational complexity of such approaches has prohibited deployment in real-time scenarios. To overcome this limitation, we have coupled a state-of-the-art atmospheric compensation algorithm, the average-bispectrum speckle method, with a powerful FPGA-based embedded processing board. The end result is a light-weight, lower-power image processing system that improves the quality of long-range imagery in real-time, and uses modular video I/O to provide a flexible interface to most common digital and analog video transport methods. By leveraging the custom, reconfigurable nature of the FPGA, a 20x speed increase over a modern desktop PC was achieved in a form-factor that is compact, low-power, and field-deployable.

Reconfigurable device for enhancement of long-range imagery

In this paper, we discuss the real-time compensation of air turbulence in imaging through long atmospheric paths. We propose the use of a reconfigurable hardware platform, specifically field-programmable gate arrays (FPGAs), to reduce costs and development time, as well as increase flexibility and reusability. We present the results of our acceleration efforts to date (40x speedup) and our strategy to achieve a real-time, atmospheric compensation solver for highdefinition video signals.

Accelerated modeling and simulation with a desktop supercomputer

The performance of modeling and simulation tools is inherently tied to the platform on which they are implemented. In most cases, this platform is a microprocessor, either in a desktop PC, PC cluster, or supercomputer. Microprocessors are used because of their familiarity to developers, not necessarily their applicability to the problems of interest. We have developed the underlying techniques and technologies to produce supercomputer performance from a standard desktop workstation for modeling and simulation applications. This is accomplished through the combined use of graphics processing units (GPUs), field-programmable gate arrays (FPGAs), and standard microprocessors. Each of these platforms has unique strengths and weaknesses but, when used in concert, can rival the computational power of a high-performance computer (HPC). By adding a powerful GPU and our custom designed FPGA card to a commodity desktop PC, we have created simulation tools capable of replacing massive computer clusters with a single workstation. We present this work in its initial embodiment: simulators for electromagnetic wave propagation and interaction. We discuss the trade-offs of each independent technology, GPUs, FPGAs, and microprocessors, and how we efficiently partition algorithms to take advantage of the strengths of each while masking their weaknesses. We conclude by discussing enhancing the computational performance of the underlying desktop supercomputer and extending it to other application areas.