Computation, Connectivity to Drive Computing over the Edge


TDK Ventures’ DX Week Experts Say Purpose-driven Innovation to Lead

As the world’s need to create, access, and mobilize data and the range of ways and devices we use to accomplish these tasks continue to increase exponentially, the criticality of high-level computing capacity at the edge cannot be overstated. Session 3 of TDK Ventures’ Digital Transformation Week (DX Week) explored the challenges, opportunities, and innovations that will drive our ability to perform computing tasks at or near the source.

Intel Capital Investment Director Elana Lian kicked off the session by noting that disparate systems and lack of standardized approaches create multiple opportunities for invention and investment.

“Monolithic software, specialized hardware, and proprietary technology limits flexibility,” she explained. “That brings us a lot of transformation opportunities to connect the unconnected, connect everything obtained in normalized data, drive access through standard consortiums, adopt IP-like system management using factory operating systems, cybersecurity, software upgradability, open API application. How to use machine learning, predictive maintenance, vision systems, digital twins to improve that efficiency, effectiveness, and responsiveness.”

With all those possibilities, Pete Warden, former Jetpac CEO and founder of the Google TensorFlow Lite Micro project, has a simple request.

“What I really want for myself is a light switch that I can just look at and say, ‘on; off,’ because I really feel like we’ve got ourselves tied in knots with a lot of the centralized hub voice interfaces,” Warden said. “I feel like there were so many improvements in the experience that we can offer through doing the processing on the edge. If you can keep things like voice and image data completely local to the device, then you really don’t have to worry about it leaking or people getting ahold of your recordings.”

“Why doesn’t your light turn on when you look at the switch, instead of your having to get off the couch and go to flip it?” asked Mike Vildibill, leader of Qualcomm’s Cloud Edge AI unit. “That would be a context-driven behavior which comes with more computation and connectivity. We have robotics today that can autonomously identify and classify objects. They can navigate a path, even on public roadways. In the future, that autonomous robot, to really be effective, is going to need a better understanding of and a better way to predict the behavior of the world around it. To integrate into the world around them, these devices at the edge must have much more processing power to establish greater context.

EdgeQ CEO Vinay Ravuri agreed that edge computing and connectivity cannot be separated.

“You can’t talk about one without the other,” he said. “We’re in the Web 3.0 world — it’s secure; it’s more sophisticated. It all comes down to the immersive, to-the-edge experiences that people want to see. That comes in many shapes and flavors, but a lot of it is going to be visual. That requires super-low latency and extremely high bandwidth delivered to the end application.”

In large cloud enterprises, compute is centralized. Moving computers closer to the data at the edge will require connectivity changes that improve latency and reliability. Once that happens, processing and computation power can be fully leveraged at the edge — where the data is generated or where the AR/VR device or the factory robot is located.

“5G is a key, but the amount of bandwidth is another piece of the puzzle. Take, for example, video. Not just for humans, but for robots as well, most decisions on the edge are made based on some form of vision sensing.”

Ravuri said machine learning, computer vision, and edge inferences will play large roles in performing those applications, putting computing at the edge or one hop away at an access point or nearby server. He said for gaming, AR/VR, and other intensive experiences, graphical computation must be accomplished.

“Those are extremely power-hungry applications,” he said. “It’s difficult to do them on a headset or even a phone. Instead, maybe there is some remote rendering that could be done and push that to the edge device itself. You would need a link that is really, really fast because you want to be able to move your head and have the scene change. The graphics would have to get rendered and presented in a millisecond.”

Vildibill said entrepreneurs, researchers, and academics must serve under the “tyranny of physics,” to overcome distance limits on latency improvements. Data movement consumes energy and continually increases the amount of computation power available at the edge.

“Technology will push an application or need with pull. These two factors will work together to really steer innovation,” he said.

Technology is definitely influencing app development, Warden said.

“There has been this explosion of the amount of compute that’s available on the edge and it seems like the hardware people have been ready with a whole bunch of computing power and looking for ways to use it,” he noted. “That’s what’s interesting about the machine learning side. We now are able to run voice recognition, person detection, computer vision algorithms, and gesture recognition locally on pretty cheap devices. It feels like the wave is starting to build because all the factors to help support it have come together.”

Dean Brenner, chair of the FCC’s Technological Advisory Council, said the convergence of the pandemic and more well-defined use cases created a perfect storm for purpose-driven innovation aimed at cloud computing.

“Covid has changed everything about this panel. Pre-Covid, I was working on the early days of 5G and the most common question I would get from policymakers, friends, neighbors, and others outside the industry was, Why do I need that?” he said. “Covid taught us that everyone wants and needs the best possible connectivity. And they want it right now. If edge computing enables faster, better capabilities, everyone’s going to want it. And the same goes for the cloud.”

He said the goal for 1G, 2G, 3G, and 4G was simply to develop the technology to try to make wireless connectivity better.

“We weren’t sure how it would be used,” he said. “With 5G it was different. There were specific use cases, specific to 5G from its inception and there still are. One of the visions for 5G was to enable better edge computing, specifically mobile edge computing. Over the next five years, as 5G matures and it enhances through the next several versions, edge computing is going to go through the roof.”

Chris Bergey, senior vice president and general manager of infrastructure at Arm, said the industry’s Covid response proved the technology’s mettle.

“The Covid dynamic and the way traffic patterns changed overnight relative to the types of bandwidth people wanted at home were working with video calls and all that kind of stuff, was darn close to flawless,” he said. “We moved to software-defined dynamic edge. Core networks were the start. Those technologies are now moving into 5G, making the network stack very portable.”

Lian said investors are looking at entrepreneurs to make some kind of industrial IoT (internet of things) at intelligent scalability. That is not easy to proliferate.

“There is still an intense cost on industrial applications, long contract sales cycles, the challenge of legacy systems,” she said. “But even so, I’m seeing entrepreneurs in the orchestration side and the device virtualization side take on these hard challenges and make scalability happens — taking scalability and consumer applications more and more toward the edge.”

Entrepreneurs and technology builders benefit from a “virtuous cycle,” which expedites development, Vildibill said.

“Performance enables new applications, which demand more requirements, which drives more innovation, which stimulates performance,” he explained. “Just over my career, we have seen a 100 million x increase in performance. That rate of growth and the need for computation at the edge is growing even faster. You can see 30 teraOPS in mobile devices today; we’re seeing 400 teraOPS in a single system on a chip SOC. The innovations in performance, connectivity including latency and bandwidth to the edge, and power efficiency are going at lightspeed. We’re going to see the applications — whether it’s ‘look at my light switch’ or more intelligent autonomous vehicles — follow suit.

He said the next five years present an opportunity to approach grand challenges by “pulling the industry together in the same direction through industry standards, interoperability, or the ability for multiple innovators to mix and match their technologies together to yield a product for consumers.”

Day 3 of TDK Ventures’ DX Week again delivered progressive visions from some of edge computing’s most insightful entrepreneurs, investors, and researchers. The discussion and ideas presented are sure to motivate a new generation of problem solvers and impact scalers who will lead the world into a new era of edge computing functionality.

"*" indicates required fields