Next: 8 Acronyms
Up: Case Study Index
Previous:6 Conclusion
The benchmark demonstrated a top-down design process that moved from executable
specification, to performance model, to abstract behavior model, to detail behavior model
and synthesizable, register-transfer level using executable models written in VHSIC
Hardware Description Language. The use of hierarchical virtual prototyping minimized
design errors and design time by verifying hardware/ software components before
manufacturing. The physical prototype of the SAR processor with a mix of custom and
commercial, off-the-shelf (COTS) hardware/ software components was built and quickly
integrated with first-pass design success that passed all performance requirements.
Performance models were used to quickly simulate various mixtures of custom and COTS
hardware with numerous software mappings of the SAR algorithm during architecture
trade-offs. Originally, an ADSP21060-based COTS board was chosen for the design;
however, delay in the delivery of a fully functional digital-signal-processor component
forced a late change to i860-based COTS boards. Due to the previously completed
performance modeling during the architecture trade-offs, the new mapping configuration of
the software and amount of hardware and memory configurations were already known, so
a quick change was possible.
The SAR benchmark demonstrated a new technique using data -flow graphs of the
algorithm for quick development of efficient, real-time, embedded, application code. The
overall development time for the real-time application software was reduced by a factor of
seven and the development cost was decreased by a factor of four. The processing
efficiency of the autocoded software was only 10 percent less than manually-optimized
code.
7.0 Conclusions
The Synthetic Aperture Radar (SAR) benchmark demonstrated independently-verified
reductions in cycle time of 40 percent and reductions in software development time by a
factor of seven.
Next: 8 Acronyms
Up: Case Study Index
Previous:6 Conclusion