Get email delivery of the Cadence blog featured here
MONTEREY, Calif.--One million tapeouts a year.
When Cadence Senior Vice President Martin Lund uttered those
words with a smile on his face, the small crowd at an industry event here
They chuckled because they know it would be a great number
to achieve, and they chuckled because they know the road there is not smoothly
"My vision is, let's get to a million tapeouts a
year," Lund said April 18, during a keynote address to the Electronic Design
Process Symposium. "That would be a good thing."
A good thing for the industry indeed. And it seems
achievable, especially the potential scale offered by markets such as the
Internet of Things.
But how does the industry get there? How does it get there
amid increasing design complexity? How does it get there with relentless
time-to-market pressures? How does it get there with ever-present cost
pressures and a much faster pace of innovation?
Lund (pictured, right) described a road map (link to his presentation), and his message was that the industry
gets there by rethinking old notions, considering system design more
holistically, and by delivering IP and EDA tools to get the job done.
It starts with a new outlook: a system-level mindset.
we thinking about the point problem? Point products? We need to bring to bear
system-level thinking to solve all aspects of the problem. If you don't think
about it up front, you don't design to constraints and you end up with
something that doesn't work or can't be built at a reasonable cost."
We're firmly entrenched in the era of cores-based design and,
fortunately, much of the work that once consumed design teams--designing to ever-changing
industry protocols and specs--has been assumed by third-party IP. This allows
teams to focus their design value on the core of the system.
Adding Design Value
And here is one area where system-level thinking and design
trade-offs need to be considered.
For example, designing a core in RTL is generally considered
up to 1,000 times more power efficient than other approaches, Lund said.
But there are tradeoffs:
trick is to find balance between the optimization (benefits) of RTL, which
means that it might be the best possible implementation, but it takes a long
time to verify and bring to market, and,
once you've written it, it does not change."
A team designing for the smartphone market might consider
the power demands of a system that awakens to voice command (in other words,
the system is never really asleep).
That's useful functionality for users, but if implemented in
software running on a dedicated processor, that functionality might quickly
gobble up the battery.
A programmable solution, by contrast, might offer much
better power efficiency, he added.
And then there's the ever-present time-to-market pressure.
a lot of competition, and competition drives innovation. It's the fuel for
innovation. And one of the artifacts is time to market. The guy who takes your
market window takes all your dollars. If you're third, fourth or fifth, it's not
a very good business to be in."
As teams embrace system-level thinking, they must consider
the application software that will run on top of the hardware.
"Software is what stands between us,
our chip and money. If your software doesn't run, it doesn't matter whether the
chip runs. It's only when the software is running that the value is
And that means--especially with the specter of time to
market--that software needs to be part and parcel of the design from the very
start and verified hand-in-hand with hardware. This, of course, not only speeds
design completion but helps teams optimize both hardware and software along the
a difference when people say ‘I have it passing in RTL' and ‘I have it running
on an emulator.' Once you put it close to the real hardware (in an emulator),
you're forcing all the elements to come together earlier."
On the Road
Another challenge to getting to a million tapeouts a year
is that, given the emerging market potential (especially in IoT) and almost
infinite number of possible applications, it's not going to happen on common
lot of these brave new world applications require implementations that are
innovative to fulfill the platform. To make a Fitbit (health-monitoring
bracelet) into a ring and have it be alive for 30 days, there is no standard
product you can build that's optimized for this. You'd get around to building
the ring as a standard product and then someone will want it as an earring.
Because power is king in all of these applications, you can't build common
platforms that hit all these things."
This requires an innovation ecosystem that leverages partnerships,
IP, tools and methodologies along with the system-level thinking. Fortunately,
we've begun to build that ecosystem to meet the challenge, Lund said.
SoCs require even more collaboration across the industry and the ecosystems. While
we are heavily competing in the marketplace with each other, we're also part of
an important ecosystem for the world, and it's a unique opportunity at a place
like this to form bonds and collaborate."
--Martin Lund's EDPS presentation
--Semiconductor Memory Challenges
Will Be Overcome, MemCon Keynoter Says
Mobile World Congress:
Martin Lund on IP and Electronics System Design
-- Why Cadence Exhibits at Mobile World Congress and