Abstract
|
|
---|---|
In this paper, the interdependency of the design tasks of fixed-point optimization and architectural synthesis is analyzed, aiming at FPGA implementations. Fixed-point arithmetic is widely adopted for low-cost FPGA implementations of DSP algorithm. Despite his success, there are still many open issues regarding the automatic synthesis of fixed-point circuits. Normally, fixed-point optimization is performed without considering any architectural issues or assuming a fully parallel implementation. Thus, any architectural optimization made after may produce non-optimal results. The effect of using different fixed-point optimization methods on the final architecture is studied. Several methods to perform fixed-point optimization, architectural synthesis, and the combination of both, are presented. The analysis considers both LUT-based resources and DSP blocks. The results yield that when the architectural model used during the two design tasks, does not match, then, it is not necessary to perform a thorough -- and time-consuming -- fixed-point optimization (i.e. simulated annealing). Also, the combined synthesis techniques produces area reductions up to 20\%. | |
International
|
No |
Congress
|
XI Jornadas de Computación Reconfigurable y Aplicaciones, JCRA'11 |
|
960 |
Place
|
Tenerife (España= |
Reviewers
|
Si |
ISBN/ISSN
|
978-84-614881-4-8 |
|
|
Start Date
|
07/09/2011 |
End Date
|
09/09/2011 |
From page
|
269 |
To page
|
278 |
|
Actas de las XI Jornadas de Computación Reconfigurable y Aplicaciones |