|
Custom Computing Laboratory
(P. H. W. Leong and K. H. Lee)
An FPGA is an array of logic gates in which the connections can be configured by downloading a bitstream into its memory. Reconfigurable computing (RC) systems use FPGAs as a coprocessor to accelerate applications which operate on a personal computer, or as a prototype of a system which will be later implemented as an application specific integrated circuit (ASIC). Although ASICs offer potentially higher performance than reconfigurable computing, reconfigurable computing has the advantages that it has a much shorter design time; it is possible to use the same RC hardware for many different applications; Moore's law continues to offer improved silicon technology at exponential rates; the increasingly high non-recurrent engineering costs associated with ASICs are avoided; it is possible to specialize the hardware to an extent not possible in ASIC devices and finally, the reconfigurable nature makes it feasible to attempt designs employing more sophisticated algorithms which can lead to an improvement in performance.
Research in the custom computing laboratory (CCL) focuses on high performance computing using reconfigurable computing, ASIC, microprocessor and cluster computing technologies. Ongoing research projects in this laboratory include
Cryptographic hardware: Cryptographic hardware is an ideal application of reconfigurable computing and the CCL has developed high performance implementations of DES, RSA and elliptic curve cryptography. Other ciphers and public key cryptosystems are being implemented. Applications of this technology include SSL accelerators and virtual private networks (VPNs).
Improved coprocessor interfaces: Most hardware accelerated computing systems use a microprocessor together with some dedicated hardware (implemented as an ASIC or on an FPGA) to achieve a performance improvement over software based solutions. CCL is performing research on improved interfaces for such systems. A memory-slot based RC card has been developed, offering significantly improved performance over the traditional PCI based interfaces.
CAD tools: Research into ways of improving the productivity of hardware designers as well as the performance of the resulting systems is another focus of the laboratory. This includes automatic floating to fixed point conversion, hardware compilation and module generators.
Signal Processing: Biological systems use highly parallel interconnections of simple processing systems to perform feats of signal processing, which are far beyond those of even the fastest supercomputers. The laboratory is concentrating on developing both traditional and biologically inspired signal processing systems using RC and analogue VLSI technologies.
Microprocessor interfacing and cluster computing: The CCL also conducts research and consulting in the design of microprocessor based systems and device drivers. This includes studies in improving the performance of Linux based "Beowulf" clusters by using RC techniques to improve computation and communications, custom device drivers, microprocessor interfacing and parallel processing.
|
