A HYBRID APPROACH FOR PERFORMANCE ASSESSMENT
IN EMBEDDED SYSTEMS DESIGN
C. Talarico
Dept. of Electrical Engineering
University of Hawai‘i
Honolulu, HI 96822, USA
claudio@eng.hawaii.edu
J.W. Rozenblit
Dept. of Electrical and Computer Engineering
The University of Arizona
Tucson, AZ 85721, USA
jr@ece.arizona.edu
A. Stritter
Design Automation
Infineon Technologies AG
Balan str. 73, 81541 Munich, GERMANY
albert.stritter@infineon.com
ABSTRACT
In recent years, embedded computing systems have
become so complex, and time-to-market
requirements so stringent, that a need for new
methodologies and computer aided design (CAD)
tools has emerged. The key to cope with complexity
is to tackle designs at the highest level of abstraction
- the system level. Toward that objective, we pursue
an implementation independent system level
methodology using executable discrete event models.
This paper introduces a hybrid approach for
performance assessment of hardware/software
designs. We propose both analytical and simulation
methods for early evaluation of design models. The
approach is illustrated through an application
example.
The originality of this work is to provide a
method for deriving system performance from the
execution of high level models rather than gate level
or transistor level pre-characterizations.
1. INTRODUCTION
The design of embedded systems has traditionally
been carried out by decomposing and allocating the
system to hardware and software, then allowing
separate hardware and software design teams to
design their respective parts, and finally integrating
hardware and software. This separation of design
tasks leads to the potential for initial design mistakes
to be carried until the integration phase where they
are much more difficult and costly to correct. We
address this problem using a unified framework for
describing both hardware and software. The idea it to
keep the hardware and software design activities
tightly coupled. An effective paradigm to pursue this
strategy is the concept of “platform” (Sangiovanni-
Vincentelli 2001).
A platform is simply an abstract view, i.e., an
executable model, of the system and its components.
Thus, designing an embedded system can be regarded
as the process of stacking up layers of platforms.
Depending on the abstraction layer (i.e., the level of
details used to describe the system) different
concerns can be addressed and solved. The key is to
model the system at each abstraction layer only with
the minimum number of details needed. At each layer
the goal is to collect quality/performance indexes to
assist the design team in making sound engineering
decisions.
In our approach, we use stepwise refinements of
simulatable models and offer the opportunity to
abstract system components at multiple levels of
representation. Model components are specified at a
high level of abstraction to remain implementation
independent. We endeavor a satisfying level of
confidence in the correctness of models through
computer simulation. A simulation test setup is called
an experimental frame. Such frames specify
conditions under which the model of the system is
observed. Simulation is then executed according to
the conditions prescribed by the frames. At the end of
the simulation process a virtual prototype is obtained.
The design can be then partitioned into hardware
software and the corresponding interfaces using a
process that we call “model mapping” (Schulz 1998;
Schulz 2002). A simplistic illustration of the modelbased
design process is given in Figure 1. The
process begins with the definition of the functional
requirements. The output of this step feeds both the
design and test process flows.
.....