Calif.—The path to tackling system design complexity runs straight through
system-on-chip (SoC) IP design and methodology, because getting differentiated
systems to market in a timely manner can no longer be done from scratch.
That was the message from Krishna Yarlagadda,
President of Imagination Technologies Inc.,
who keynoted March 11 at Cadence's CDNLive Silicon Valley user conference. Yarlagadda has an appropriate perspective: He
began his career designing SPARC workstation processors, and his career arc has
tracked with the evolution of processor architectures from hard-core computing
to graphics to mobility. His company Hellosoft (V.VOIP and VoLTE technologies)
was purchased by Imagination three years ago.
He defined today's big market opportunity, the
Internet of Things (IoT) as largely your computer, phone, tablet, and car. But
this landscape will change rapidly in the coming years as systems, devices, and
applications emerge for healthcare, the home, and other areas.
As this happens, engineers must confront some
traditional design challenges in power, area, cost, software, and security.
"You can't do these things from scratch or
just pick up a CPU core or do your own core," Yarlagadda told a
standing-room only audience at the Santa Clara Convention Center. "There are a lot of smart people out there who do cores. We all know
how to design them. But is that the right thing to do?"
key is to take the existing IP that's best in class and bring it together in a
way that you can deliver your solution faster along with your core intellectual
At Imagination, they
call the approach a heterogeneous SoC platform that combines CPUs, GPUs, and video
and imaging processing blocks as well as communications processor blocks
called RPUs (radio processing units).
"You can bring all that together with your
own IP and others' IP and you can create these systems much more
efficiently," Yarlagadda said.
By way of example, Yarlagadda shared a slide of
a PowerVR-based heterogenous vision platform (pictured, left):
"You take the image from the image
sensor and go through the ISP and process it in various ways. You can do
compute algorithms on various CPUs; you can run some GPU algorithms. You might
have your own custom block that you want to run various things on. By the way,
software is key."
And as design teams
are bringing all these blocks and software together, they must be mindful of
power and thermal constraints: "This is a serious problem and a lot of
people ignore this. It already dominates at 28nm design."
In an afternoon session with
members of the technical press, Yarlagadda expanded a little on the design
challenges ahead. Clearly, an IP/platform-based approach to design will be a
logical way to tackle complexity, he noted. But the economics of the IoT market
(small devices, low cost, thin margins) loom as another major challenge.
"There are architectural
issues. Systems must be simpler, both hardware and software must be simpler
because these (devices) can't be too expensive," he said. Right now, "they have
a function but they're expensive."
Imagination's focus is to
enable the design and make things simpler, he said. "That's why we have silicon
IP, graphics, MIPS CPU, RPU for communications and we have the cloud offering (FlowCloud) that
allows the IoT (devices) to work together in a secure, authenticated
Related Blog Post