System level design: six success stories in search of an industry.

Full text

Page 1

System Level Design:
Six Success Stories in Search of an Industry
Organizer:
Francine Bacchini
Thinkbold Corporate Communications
6030 Calle de Suerte
San Jose, CA 95124 California, USA
+1-408-267-6602
francine@thinkbold.com
Chair:
Grant Martin
Tensilica, Inc.
3255-6 Scott Boulevard
Santa Clara, CA 95054-3013 USA
+1-408-327-7323
gmartin@tensilica.org
Pierre Paulin
STMicroelectronics, Canada
Arie Bernstein
Intel Corp., Israel
Reinaldo A. Bergamaschi
IBM Corp., USA
Ramesh Chandra
Qualcomm, USA
Raj Pawate
Texas Instruments, India
Mohamed Ben-Romdhane
Conexant, USA
PANEL SUMMARY
System-level design is being touted as the Holy Grail that the
electronics industry has long sought, but most offers have been
disappointing because they seldom deliver results. Many
designers are fed up with the “Blah, Blah” on system-level
design as they are waiting for design facts.
Why? It seems that major breakthroughs are happening thanks
to the adoption of standard directions for modeling design at
higher than RTL level. These models are called TLM and new
languages are being adopted (SystemC, SystemVerilog). The
emergence of new standards may reshape completely the way
design industry is organized.
This panel will bring six speakers relating their success stories
about design starting at the system-level. The format is an
educational Panel aimed at informing DAC attendees of the
challenges (difficulties and pitfalls) and opportunities (sizable
benefits and lessons learned from these experiences).
1. SYSTEM-LEVEL DESIGN: PANEL
BACKGROUND
Just as Pirandello had his “six characters in search of an author”
[1], so do we have our system-level design success stories, in
search of a commercial design automation industry. As
background to the panel, let's consider the status of system-level
design today, in mid-2004. The increasing complexity of
electronic systems is aggressively pushing design, integration,
verification, and performance issues to a breaking point.
System-level design has been touted as the “once and future
hope” of the Electronic Design Automation (EDA) industry for
many years. The movement up to the “ESL” (electronic system
level) of design abstraction has been rated as a critical factor in
allowing designer productivity to keep pace with the increase in
complexity afforded by advanced IC processes. New design and
verification methods are being adopted in the dire search for a
design flow that can support sub-micron technologies.
According to Gartner Dataquest, when the initial 90-nanometer
designs came out in October of 2002, it was surprising that not
one of the designs exceeded 50 million gates – that is, none of
the designs utilized even half of the number of gates made
available by the new 90nm silicon process. System level design
and verification of electronic products is viewed as a major long
term trend that will enable the utilization of the capacity enabled
by these technology nodes.
However, as a commercial proposition, we have seen ESL
design tools come and go, burst into prominence and then fade
away, arrive in a rush of publicity and leave quietly. This has
happened in successive waves over the last decade, and
periodically the existential questions of “why is system-level
design important?”, “who is it important to?” and “is it
important enough to justify a commercial industry?” need to be
answered. The panelists will all share their thoughts about these
and related questions. But here we will sketch in a little
background along the dimensions of geography, language,
design domain and industry structure.
2. THE GEOGRAPHIC DIMENSION
Several years ago, it was possible to observe that advanced
concepts in system-level design, or ESL tools, were either
originated or adopted most widely first in Europe, then moved
westwards - rippling across North America, and finally having
impact in Japan. (Design activity in Asia-Pacific was relatively
low at the time). When we look at the geographic distribution of
interest in, and popularity of, system-level design in 2004, we
still see Europe as the pioneering originator and/or adopter of
advanced system-level design concepts as tools. But a very
interesting shift has taken place. The wave of SLD and ESL then
moves east to Japan, followed by growing interest in Asia-
Pacific, and finally hits North America.
What accounts for this geographic dimension for ESL and
indeed for the shift in the movement of the wave? There are
several possibilities. Certainly, the interest in SLD and ESL in
Europe is a partly cultural phenomenon, motivated by a greater
interest in standards and the importance of standards in creating
industrial opportunities based on interoperability. It is also a by-
product of the relative European strength in telecom-
munications, particularly wireless, in which standard setting and
conformance to standards has been vital to the growth of the
Copyright is held by the author/owner(s).
DAC 2004, June 7–11, 2004, San Diego, California, USA.
ACM 1-58113-828-8/04/0006.
349
22.1
Proceedings of the 41st Design Automation Conference (DAC’04)
1-58113-828-8/04 $ 20.00 © 2004 ACM

Page 2

industry. An emphasis on system-level thinking, modelling and
analysis, reflects a more formal design approach.
The recent growth of interest in Japan is due to the prominence
of consumer electronics industries, as well as a convergence
phenomenon: increasingly all devices are taking on multiple
roles – computing, multimedia and wireless communications.
The growth of the design industry in the Asia-Pacific region, the
complexity of today's multi-functional products, and the desire
to catch-up and leapfrog well-established European and North
American rivals by adopting the most advanced methods, are
reasons for the growth in interest in that region.
What may be hardest to explain is the relative lack of interest in
North America. Certainly the military-aerospace industries have
emphasized a certain level of formalism and analysis, and the
wired telecommunications industry has been standards-based.
However, cultural forces may be at work here: design in North
America has been more “component-based” than “system-
based”, and the relative weakness of both mil-aerospace and
wired telecom in recent years may account for some lack of
take-up of the most advanced ESL and SLD concepts. In
addition, the computing industry, as opposed to wireless and
consumer products, has tended to be the trendsetter for design
practices in North America. Its focus on components, dedicated
system architectures, and a more ad-hoc design style, has tended
to emphasize the “RTL to GDS II” design flows over any system
modeling and analysis. Automotive is also a major industry in
the U.S., but electronics is only recently reaching major
importance via electronic control units and telematics. This
contrasts dramatically with Europe where ESL has already been
adopted by the European automotive suppliers.
Will geography and local design cultures play an important role
in the creation, take-up and commercialization of SLD and ESL
tools? And, if so, to what should the industry pay its greatest
attention?
3. THE LANGUAGE DIMENSION
One key problem in adopting a new abstraction level and set of
modelling languages is the problem of ensuring payback and
reuse of all the design effort spent in building models. If models
are orphaned or divorced from implementation flows, then the
effort in building them is often perceived as wasted. If models
are captured in proprietary languages attached only to one
vendor's tools, then the incentive to create them is low and their
fate is tied too closely to a single vendor. Lastly, even models
built in open languages - if developed using different concepts
of interfaces - have a very low chance of interoperating when
provided by different IP vendors. Each of these characteristics
can be observed in previous generations of ESL tools.
In 2004, however, we are seeing some profound differences.
Standards-based ESL languages, such as SystemC, are seeing
fairly wide adoption. Design methods and flows that link models
to implementation are beginning to emerge. Modelling
languages such as SystemC, and implementation languages such
as SystemVerilog (and thus eventually IEEE 1364 Verilog as it
evolves) are defining notions such as “transaction-level models”
(TLM) that will provide an opportunity for flow between
modelling and implementation, and also provide interoperability
between models sourced from different IP vendors. Thus the
major hurdles to seeing widespread IP and design model
creation and use seem to have been successfully jumped. Will
the current language efforts reduce or eliminate the barriers to
the creation, adoption, interoperability, reuse and flow of SLD
models? What additional work is required?
4. THE DESIGN DOMAIN DIMENSION
A key aspect of SLD is the importance of particular design
domains and use models. SLD does not share a common “design
ideology”, as does digital design from RTL to GDSII (the
synthesis-place and route flow). Niche-specific flows –for
example, the modelling and implementation of dataflow
algorithms – and specific use models – for example, creating
architectural models of SoC 'platforms' for performance
analysis, verification, and providing fast-executing 'system
virtual prototypes' for embedded software development and
verification – seem more the case in SLD and ESL. Which of
these specific niches and/or use models will be able to foster a
commercial EDA or tools industry – and which will not? What
can SLD success stories tell us about the most important user
communities and their needs? Which markets are viable for
ESL tools?
5. INDUSTRY STRUCTURE DIMENSION
From this last point, we can raise several interesting questions
about the structure of an SLD and ESL tools industry. It may be
that SLD is so closely associated with particular designs, and the
architectural and system design teams are so small, that the most
important ESL tools will be created and used within large
system and semiconductor companies.
As a corollary to this hypothesis, the most important ESL tools
and SLD concepts may be so intertwined with particular design
IP characteristics, that the commercial IP industry will need to
invest in the creation, proliferation and support of SLD tools
tied closely to their offerings. In both these cases, some generic
and low-level ESL tools (such as SystemC capture, simulation
and debug environments) may be sourced by the commercial
EDA industry, but all the most valuable tools will come from IP
suppliers and/or be created within large design companies.
In contrast to this, it may be possible that the commercial EDA
industry will identify many truly viable niches for advanced ESL
tools, as well as sufficiently large user communities, to spur a
real ESL industry as forecast by Gartner Dataquest. This may be
triggered by the rapid growth of IP-based design – IP integration
complexity will lead to new tool opportunities for the generic
EDA industry. Power consumption may also be another trigger
for generic ESL tools. Even in this scenario, however, one point
for debate is whether it will be small startups and medium sized
EDA companies which will specialize in ESL tools, or whether
the larger EDA companies, who to some extent have reduced
their interest in ESL and SLD, will find this area attractive
(again). And if they do find it attractive, will it be via acquisition
only, or also via indigenous research and development of new
tools?
6. REFERENCES
[1] Pirandello, Luigi. Six Characters in Search of an Author.
Translated by Eric Bentley. Signet Publishers, 1998.
350
Proceedings of the 41st Design Automation Conference (DAC’04)
1-58113-828-8/04 $ 20.00 © 2004 ACM