This paper describes a design flow for data-dominated embedded systems. We use The Mathworks' Simulink\trademark environment for functional specification and algorithmic analysis. We developed a library of Simulink blocks, each parameterized by design choices such as implementation (software, analog or digital hardware, \ldots) and numerical accuracy (resolution, S/N ratio). Each block is equipped with empirical models for cost (code size, chip area) and performance (timing, energy), based on surface fitting from actual measurements. We also developed an analysis toolbox that quickly evaluates algorithm and parameter choices performed by the designer, and presents the results for fast feedback. The chosen block netlist is then ready for implementation, by using a customization of The Mathworks' Real Time Workshop\trademark to generate a VHDL netlist for FPGA implementation, as well as embedded software for DSP implementation.
[1]
Luciano Lavagno,et al.
Simulink-Based HW/SW Codesign of Embedded Neuro-Fuzzy Systems
,
2000,
Int. J. Neural Syst..
[2]
Luciano Lavagno,et al.
Hardware-software co-design of embedded systems: the POLIS approach
,
1997
.
[3]
Marcello Chiaberge,et al.
A pulse stream system for low-power neuro-fuzzy computation
,
1995
.
[4]
Sharad Malik,et al.
Performance Analysis of Embedded Software Using Implicit Path Enumeration
,
1995,
32nd Design Automation Conference.
[5]
Sharad Malik,et al.
Power analysis of embedded software: a first step towards software power minimization
,
1994,
IEEE Trans. Very Large Scale Integr. Syst..
[6]
Edward A. Lee,et al.
Dataflow process networks
,
1995,
Proc. IEEE.