Soft-Core Processor Design - CiteSeer
Soft-Core Processor Design - CiteSeer
Soft-Core Processor Design - CiteSeer
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Chapter 7<br />
Conclusions and Future Work<br />
The capabilities of FPGAs have increased to the level where it is possible to implement a<br />
complete computer system on a single FPGA chip. The main component in such a system is a<br />
soft-core processor. The Nios soft-core processor is intended for implementation in Altera<br />
FPGAs. In this thesis project a Verilog implementation of the Nios architecture has been<br />
developed, called UT Nios. Performance of UT Nios has been investigated and compared to that<br />
of the Altera Nios. Performance analysis has shown that, although the Altera and UT Nios<br />
implementations are quite different, their performance is similar. <strong>Design</strong> analysis has shown that<br />
although there is little room for improvement in the current UT Nios organization, there is a lot of<br />
room for experimenting with different processor organizations and architectural parameters.<br />
The thesis offers the following contributions:<br />
• UT Nios implementation of the Nios architecture is described. The incremental design<br />
methodology used in the UT Nios design is described. This methodology resulted in<br />
an unconventional implementation with the performance comparable to the original<br />
implementation from Altera.<br />
• UT Nios benchmark set suitable for evaluation of the soft-core processor based<br />
systems is presented. The benchmark set was selected among the existing benchmarks<br />
and supplemented with benchmarks suitable for performance analysis of the Altera<br />
and UT Nios. All benchmarks were adapted for use on the Nios based systems.<br />
• A detailed analysis of the UT Nios performance and a comparison with the Altera<br />
Nios was performed and the results are presented. The results show that the<br />
application performance varies significantly with various architectural parameters. A<br />
single architectural parameter that influences performance the most is the size of the<br />
general-purpose register file. The results indicate that many applications perform<br />
worse on the system with the windowed register file of small size than on the same<br />
system that does not use register windows capabilities, but uses only a single register<br />
window. The performance comparison of the UT and Altera Nios shows that both<br />
implementations achieve similar performance.<br />
86