EDA is Multi-Core,
EDA is Parallelism, EDA is Software
Abstract
Gary Smith’s annual review of technology trends in electronic
design - the pre-DAC EDA Forecast - has a long tradition. What are
the trends shaping design in the next 5 years? What is the EDA outlook
for the upcoming years? What are the major design issues of today?
What is the multicore problem? Will software engineers ever pay
top dollar for EDA tools? Questions like these were covered in this
traditional ‘must-see’ event.
Panelists
• G. Smith, Gary Smith EDA
• M. Olsson, Gary Smith EDA
• D. Nadamuni, Gary Smith EDA
and as a guest speaker
• T. Starnes, Objective Analysis
Compilers and Moore’s Law
Gary Smith’s tradition of giving a pre-DAC EDA Forecast is
nineteen years old. He used to present as chief analyst of Gartner
Dataquest; this year, he presented as President and CEO of his own
company. The meeting took place directly after the EDAC reception.
More finger food and very drinkable California wine were served,
followed by presentations from Gary and his colleagues, Daya Nadamuni,
Mary Olsson and Tom Starnes. Initially, the most difficult task
was to calm down the large and lively audience.
Tom Starnes
Tom Starnes, Director of Processors at Objective Analysis, started
the presentations with a discussion entitled “Multi-Core Processor
Considerations in Modern Day SoC Designs”. Starnes outlined
some of the commonly accepted approaches to SoC design. Processor-based
approaches have already made great strides in the implementation
of the modern “single-chip”-system-on-achip. Today’s
SoC comprise a number of processors of different classes, such as
MCUs, DSPs, specialized processors and accelerators. Starnes sees
three major steps in selecting, building and programming the processors.
“Multiple processors
are here. Deal with it. You’re going to have to work with
the entire system design team.”
Tom Starnes from Objective Analysis
He noted that design challenges include task partitioning, memory
hierarchy, bus structure, power domain partitioning, process technology,
verification, simulation, and power dynamics. One of the most difficult
things to simulate is dynamic power. Starnes concluded by repeating
Gary’s recent message: “Software, software and software!”
Mary Olsson
Mary Olsson, Chief Analyst at Gary Smith EDA, then gave a presentation
entitled “What EDA and Chip Designers Don’t Think About”.
Facing the
changing landscape, she asked, “Semiconductor production:
Where in 2007?” and posed the provocative question:
“Do EDA software and chip designers really know how
to communicate with package and board designers and contract manufacturers?”
Mary Olsson from Gary Smith EDA
Her up-to-date statistics showed that by far the most designs are
targeted at 90 nm, which will probably not change in the near future.
Also, the shares of different product types are very stable.
Olsson observed that MEMS (Micro-Electro-Mechanical Systems) are
emerging for 3D IC integration, and that applications drive new
package designs and “real estate squeeze”.
Furthermore, lots of expensive software does not work with new packages
because of incompatibility or other reasons. Other important points
for new package designs are that MEMS are driving system-in-package
(SIP) and that mobile applications in general are very important
drivers. As most software and IC designers do not yet see problems
and opportunities around stacked dies and do not envision the variability
of assembly options, Olsson called for more collaboration of independent
parties. She summarized, people should maintain R&D on next
generation process nodes, but ignore the hype as well. And she reminded
us that there are EDA revenue opportunities with both packaging
and silicon manufacturers.
Daya Nadamuni
Daya Nadamuni, Chief Analyst at Gary Smith EDA, talked about “The
Crisis in Software Programming: How To Prepare for Multi-Core”
and
started by asking: “Where are we today (with software)?”
And she answered:
“Parallel programming is understood in theory but
underused in praxis.”
Daya Nadamuni from Gary Smith EDA
So, she asked, is there a crisis? The attitude that hardware and
software development are two different and separate worlds is unsustainable,
and will lead to serious losses. Due to the increasing demand for
processing power, Moore’s law will be substituted by multi-core,
which means parallelism (or throughput) instead of speed, she predicted.
Of course this will create new challenges, especially in the programming
of multiple processors. The semiconductor industry already understands
symmetric multiprocessing (SMP), and she predicted that asymmetric
multiprocessing (AMP) will be the next step.
Nadamuni mused that EDA vendors are seeking the next ‘big
thing’, but are not sure how to deal with the software - which
is REALLY the next big
thing! And as embedded software development tools and RTOS vendors
are themselves not sure how to deal with multi-core, confusion reigns
in both the EDA and embedded system design worlds.
Nadamuni pointed out that mainstream software developers do sequential
programming, while object oriented programming has yet to be understood
or adopted by the mainstream. Multicore programming remains a largely
unsolved issue. This leads to the question of whether we have 10
years to train everyone in concurrent programming before the crisis
hits. Nadamuni proclaimed:
“We have a programming crisis now!”
Daya Nadamuni from Gary Smith EDA
She claimed multi-core is here to stay and that software concurrency
issues will create roadblocks.
Finally, she asked where we go from here, and announced that Gary
Smith would give the answer in his talk entitled “Alive and
well but software challenged”.
Gary Smith
The statement “alive and well” is derived from Smith’s
latest EDA market statistics. This year’s EDA revenue forecast
is optimistic. It predicts $4.795 billion in 2007, up from $4.423
billion in 2006. Smith predicted $6.585 billion revenues in 2011.
While Smith claimed that the market size is OK, he added a big BUT
that points to a dramatically changing world.
Commenting on the great semiconductor consolidation, Smith said
that the “R” of “R&D” is responsible
for the high costs. For him it is the process research that is consolidating,
not the semiconductor industry as a whole.
Semiconductor vendors want EDA vendors to provide tools for all
of the emerging design challenges, and the most important design
problem will be embedded software. “If you want to stay providing
the stuff you’ve always provided, you probably won’t
be around in five years,” he said. “If you want to take
advantage of the opportunity, it’s a nice market.”
Smith showed a second chart in which embedded software development
tools market is combined with the EDA market. The combined market
totals over $8 billion in 2011. The problem is that “only
a couple of EDA vendors really understand embedded software.”
This statement was similar to his statement last year that embedded
software development is the biggest problem in systemon- chip (SoC)
design.
By providing tools for system IC vendors, the EDA market is growing
to include all design tasks, he stated. However, Smith named parallel
processing as a big challenge for EDA vendors. Without the use of
parallel processing (when developing their tools) most EDA tools
will not be able to handle designs with new semiconductor processing
technologies. Thus in 65/45 nm, many tools have to be re-written,
and much existing code has to be completely replaced. Compared to
that, the move to 64 bit processing was child’s play, Smith
said. Nevertheless it is both a huge challenge and a huge opportunity
for the EDA industry.
Meeting this challenge will result in the accumulation of experience
in parallel programming, which will give the industry a competitive
edge in multi-core processor design and programming, hence the challenge
is connected to an opportunity. In fact, an important topic at this
year’s DAC is support for multi-processing and multicore architectures,
Smith said. Several companies introduced EDA products written from
the ground up to take advantage of parallel processing. Finally,
Smith pointed to another problem - most EDA vendors do not understand
embedded software. Nevertheless, his conclusion remained optimistic:
“That isn’t the end, it’s just the beginning.”
Gary Smith from Gary Smith EDA
Summary
As always, Gary’s pre-DAC EDA Forecast was worth listening
to. Again, the analyst and his team highlighted some new trends
and their view on
the scene, which was a good start to DAC. But as the importance
of software and multi-core has already been discussed for more than
a year, and was stressed at DATE 2007, it was no real surprise.
In our view, Gary’s observation that the semiconductor industry
consolidation was really a process research consolidation is extremely
important. Consolidation at the fab end of the flow brings economies
of scale (good), whereas consolidation at the front end would imply
fewer designers and less EDA revenue (bad). Of course, this is not
the case. The number of parties concerned with front end design
has actually increased for some years, as evidenced by the proliferation
of fabless semiconductor companies. This dynamic is likely to continue
so the prospect for EDA revenue growth is probably quite good.
It is quite positive, that Gary sees more opportunities than show
stoppers in EDA. Meanwhile it should be considered that the most
relevant parts of these opportunities consist in embedded software
which leads to a lot of imponderability as the given examples showed.
Hence a certain skepticism is appropriate – so be curious
about exciting times in EDA.
Links
• Video
interview with EDA analyst
Gary Smith
performed by Richard Goering from EETimes
• Gary Smith: What
to see @ DAC
Author
Ralf Popp, edacentrum
|