Professional Documents
Culture Documents
Problem Taxonomy. To Be Realistic: N and Concurrent Engineering
Problem Taxonomy. To Be Realistic: N and Concurrent Engineering
Problem Taxonomy. To Be Realistic: N and Concurrent Engineering
January 1993 85
compare them to estimates for full plications. Much attention must be fo-
hardware or full software implementa- cused on partitioning methods and
tions. All this boils down to the most tools for industrial use.
famous question in hardware-software Partitioning starts with There is plenty of room for research
codesign: How well is the system parti- and experimentation. The field is
tioned into hardware and software? the system-design evolving rapidly. A book based on the
Partitioning actually starts with the modeling step. First International Workshop for
system-design modeling step, in which Hardware-Software Codesign in Gras-
the designer expresses the system’s sau, Germany, held in the spring of
behavior formally with, for example, 1992, and the follow-up in Estes Park,
parallel random-access machines that Colorado, in late September, is sched-
encapsulate state diagrams and time- labs are working in this area, we are uled for publication early in 1993. In-
continuous transfer functions. De- far from a standardized representa- terest is increasing among industrial
pending on the underlying theoretical tion. labs and production sites in the US,
model, abstraction level, and integra- Another issue is that few tools are Europe, and Asia. The Computer-
tion strategy, several estimation and designed for hardware-software cross Aided Software-Hardware Engineer-
analysis methods are available. Deter- specification, development, simula- ing Codesign Workshop, scheduled
ministic estimation, for example, re- tion, integration, and test. Most de- for May 1993, and special sessions of
quires a fully specified model with all sign tools apply to a single domain, the Design Automation Conference
data dependencies removed and all making it difficult to observe the se- and the European Design Automation
component costs known. This method quence of operations or connect such Conference will increase the commu-
leads to very good partitions, but fails tools to framework environments. De- nity interested in the topic as well.
whenever data items are unavailable. sign flow management, operation
Statistical estimation, based on the monitoring, or task cross-checking be- References
analysis of similar systems and certain comes very hard if not impossible.
design parameters, is then required. Tool vendors must support hardware- 1. Collected papers from the First Inter-
Profiling techniques rely on the exam- software codesign efforts by supplying national Workshop on Hardware-Soft-
ination of control flow and data flow mechanisms for framework interfac- ware Codesign, sponsored by IFIP WG
within an architecture to determine ing. Frameworks should also be im- 10.5 in cooperation with WG 10.2,
computationally “expensive” parts proved with respect to communica- Grassau, Germany, May 19-21,1992.
prone for realization in hardware. tion, cosimulation, auditing, and
2. Workshop handouts from the First In-
In the strategic area, the issue is checkpointing capabilities between ternational Workshop on Hardware-
early versus late binding. Early bind- physically distant development sites Software Codesign, sponsored by ACM
ing is the preferred strategy in the in- linked via networks. and IEEE, Estes Park, Colo., 1992.
dustry because it supports complete Because estimation and analysis
planning of the development cycle to tools are in their infancy, automatic Klaus Buchenriederis research manager at
guide design decisions. In contrast, system partitioning is not yet possible. Siemens AG, Corporate R&D, Base Tech-
late binding can help in finding better At best, users get some hints from es- nologies, Software and Engineering, ZFE
solutions to product performance is- timations to guide their intuitions. BT SE 52, Otto-Hahn-Ring 6, W-8000 Mu-
sues and in addressing “moving tar- This is not sufficient for industrial ap- nich 83, Germany.
get” problems such as change requests
from consumers.
After parallel implementation of as-
signed team modules, all design Extending CAD (continued from p. 85)
threads are joined for system integra-
tion. Clearly, successful integration de- thus a hardware-software codesign Embedded systems often fall into the
pends on the quality of the partitions. problem with possibly multiple solu- class of reactive systems. They are
Since both steps complement each oth- tions. CAD tools can explore the meant to react to the environment -
er, interfaces must be specified during trade-off and suggest a convenient im- executing functions in response to spe-
system partitioning and prepared for plementation. The Piper synthesis cific stimuli. In some cases, their func-
automatic synthesis and system test, program is an example of a codesign tions must execute within predefined
tool that addresses this problem.’ It time frames. Hence, they are called
State of the art. Codesign is slowly provides pipe stage partitioning and real-time systems. Examples of reac-
gaining acceptance in the industry - pipeline scheduling, and also deter- tive real-time systems are pervasive in
partly because meeting the design-to- mines the instruction reorder that the the automotive field (for example, en-
cost bounds on the development of in- corresponding back-end compiler gine combustion control), the manufac-
creasingly complex systems under rigid should use to avoid hazards. turing industry (robot controllers), and
time-to-market constraints requires Embedded systems are computing the consumer and telecommunication
new approaches. Confidence that hard- systems dedicated to an application. industries (portable telephones).
ware-software codesign can solve some The most restrictive view of an embed-
of these problems stems from success- ded system is a microcontroller or a Embedded systems and ASICs. Em-
ful automotive and telecommuncations processor running a fixed program. bedded systems can be thought of as a
applications. In addition, many small- This model can be broadened to a gen- generalization of application-specific
scale codesigned systems do exist.’%2 eral-purpose processor, assisted by ap- integrated circuits (ASICs), where a
Critics bring forward the issue of a plication-specific hardware and memo- processor coupled with its software
missing common model or standard ry, that performs a dedicated function. program can be viewed as a system re-
for unified, exchangeable design rep- Sensors and actuators allow the system source. The similarities between em-
resentation. Even though research to communicate with the environment. bedded systems and ASIC design
86 COMPUTER