Research On Configuration Data Compression Algorithm For Reconfigurable System-on-Chip

Reconfigurable System-on-Chip (RSOC) witch dynamic reconfiguration stores the configuration data in off-chip nonvolatile configuration memory. The function of the system can be changed by dynamically loading different configuration data from configuration memory to Field Programmable Gate Array (FPGA). It is widely used in domain of high performance computing. Configuration time is a main factor which has influence on system performance. Configuration compression is a common technology to shorten configuration time by reducing the size of configuration data transferred when reconfiguration. The current configuration compression technology brings limited time compression and has no reasonable evaluation criteria. This paper makes research on configuration compression technology of SOC aiming at building a scientific model to shorten configuration time and speedup configuration, and then proposes a modified algorithm to improve configuration speed.First, the evolution of RSOC and recent researches on configuration compression technology are reviewed. After that reconfiguration technology and some common compression algorithm are introduced and analyzed.Second, based on the configuration decompression architecture, double restriction to the input and output end of decompression module made by configuration memory and configuration interface is pointed out to show the shortage of using compression ratio as the evaluation index of actual configuration speed. A new model named configuration speedup ratio is designed and built to evaluate the speedup of the configuration speed introduced by configuration compression technology.Then, a compression and decompression system based on LZSS encoding is implemented. Compression ratio, configuration speedup ratio and actual configuration time are compared when using different parameters of algorithm. Experimental results show that LZSS configuration compression algorithm obtain higher configuration speed when choosing parameters with higher configuration speedup ratio. Configuration speedup ratio is much more competitive than compression ratio for evaluating a configuration compression algorithm when focusing on improving the configuration speed and shortening configuration time of RSOC.Finally, a modified Huffman configuration compression algorithm is proposed. The algorithm aims at improving configuration speedup ratio, analyzes impact of different Huffman code length and local compression ratio to configuration speed under specific difference of data rate between configuration memory and configuration interface. On one side, the algorithm increases the code length of high frequency symbol on the premise that the changed code length will not reduce the data output rate of decompression system. And on another side, algorithm shortens the code length of low frequency symbol. The configuration speedup ratio can be increased by balancing the optimal and worst local compression ratio. Experimental results show that the modified Huffman compression algorithm compared with original algorithm can achieve a higher actual configuration speed at the expense of compression ratio to some extent.