Postcards From The Edge (Of The Cloud)

The view from the edge of the cloud is pretty spectacular. Out here, there’s endless possibility. But out here on the edge, there’s turbulence, the cold buffeting swirl of today’s engineering challenges.

On the edge, some see the Internet of Things (IoT) stalled. I wrote about this last month. The first wave of IoT devices — especially wearables — has quickly commoditized, driven by low-cost fitness trackers, early smart watch designs and remote camera devices.

As consumers, we’ve bought them but the experience has left us with that sort-of “ummmm that’s nice but…” feeling.

Slow evolution
I chatted recently with a technical marketing engineer I’ve known since the early 1990s. Some six to seven years ago he built himself a state-of-the-art house in the Silicon Valley and went all-in on home-automation systems, from lighting to watering.

Today, he’s not impressed. The technology teamwork he expected from his new systems and networks never really materialized. Now, in his current role as investor and board member, he looks at IoT business plans the way hikers look at thunderclouds.

Going forward, though, we know what’s needed. We need to make this IoT world, its devices and applications, much more intuitive—much more automated—if IoT is to live up to its hype. This requires a considerable rethinking and re-engineering of what we do today. We need as much of a cultural shift in systems engineering as we do technological innovation.

Much of this will be accomplished in the application software (you don’t want your modern refrigerator simply to text you when you’re out of milk and eggs; you want it to add it to your next Google Express delivery order).

As my Cadence colleague Seow Yin Lim likes to say, IoT is something that has blend into human lifestyles. The interface must become more natural. This is going to take more processing power than we have conventionally allotted to edge devices. SoC design will get more complex because in order to achieve—from a power-optimization standpoint—what you want to achieve, engineering teams will need to think about offloading more work from the main applications processor into DSPs and other programmable devices.

Think about remote camera systems like Dropcam. Now, they identify movement but soon they will recognize faces and alert you that Jane has arrived home or that all your team members have assembled in the remote office for a meeting. That requires not only clever algorithms but also beefed-up hardware.

That beefed-up hardware at the edge will have to get simultaneously much smarter and more powerful—without busting the power budget. That’s no small engineering feat.

Cultural shift
To achieve that IoT potential the design of edge devices also requires a change in mindset; here’s where the cultural shift comes in.

Mike Gianfagna, vice president of marketing at eSilicon, hit the nail on the head in an interview with Semiconductor Engineering Editor-in-Chief Ed Sperling. He said silicon designers don’t necessarily understand system design, where the emphasis is on fast, cheap and easy designs at established process nodes.

So designing those edge devices requires understanding what data is being computed and how the larger system functions. This notion of “follow the data” is something Cadence IP Group CTO Chris Rowen has been talking about for some time.

Also in Sperling’s piece, Drew Wingard, chief technology officer at Sonics, took Gianfagna’s point a step further. Silicon vendors in IoT struggle because “they don’t know exactly what to integrate until the actual device requirements are understood.” Conversely, systems companies often don’t possess the full skill set needed to develop SoC designs.

The last element of this need for mindset change comes with respect to power design itself. Elad Alon, an engineering professor at U.C. Berkeley, spoke in November at the Cadence Low-Power Technology Summit.

During a presentation, he said:
“Unlike performance where you are looking for the max—meaning the worst-case path—power is set by the sum of everything. From a design standpoint, this changes things in a dramatic way. Every small piece can matter quite substantially.”

He later said: “What this really means is you’re going after an energy efficient (design)… you have to think about entire system as a whole.”

Getting to IoT 2.0 from the edge
So where does this leave us? We have an enormous opportunity to really kick-start the growth of IoT market. Right now, we’re at IoT 1.0 or 1.2 and we’re looking up ahead to 2.0. But to get there, we need to reconsider the edge, reconsider those devices’ role in the overall system (whatever it is), reconsider their processing power (and by extension the architectures and functional partitioning we do) and their power.

That requires a little bit of thinking on our part while we gaze up at the cloud, daydreaming of the future.