Design of a high-performance PCI interface for an SCI network

摘要

The electronic design, test and application of a PCI-SCI bridge supporting shared memory, message passing and DMA is described. The objectives for designing a bridge between the IEEE 1596 SCI standard and the popular PCI bus were applications in high energy physics where large scale processor farms are needed to process large amounts of data in real time. The SCI's shared memory paradigm is particularly attractive for such multiprocessor applications. The functional units of the bridge, including PCI target and master are programmed into FPGAs. These functions are developed in VHDL and then simulated, synthesised, placed and routed using advanced FPGA development tools. The test results in terms of latency and bandwidth are satisfactory and meet our objectives.