Higher Performance, Lower Power Everywhere

The future of technology is all about information—not just data—at our fingertips, anywhere and at any time. But making all of this work properly will require massive improvements in both performance and power efficiency.

There are several distinct pieces to this picture. One is architectural, which is possibly the simplest to understand, the most technologically challenging to realize, and the most straightforward to solve. In the simplest of terms, bottlenecks need to be identified and fixed, signal channels need to be shortened and widened, and processing speeds need to increase for decades to come.

What’s changing is this is no longer a one-chip or one-vendor problem. The solution here appears to be a combination of higher-density CPUs, GPUs and FPGAs, more processing elements designed for very specific tasks (DSPs, eFPGAs, and new accelerator designs) and much tighter interaction between hardware and software. To make all of this work likely will require a mix of chips in a variety of packages, where some of the components are off-the-shelf and others are customized, either through programmability or with custom silicon.

This is no longer just about shrinking features, although that most likely will continue for a handful of the large processor vendors. It’s about putting together pieces in a way that serves a particular end market or a single application within an end market.

The second piece of this picture is business-related. At this point, it’s becoming clear that edge devices will have to do much more of the heavy lifting in processing data from sensors. What isn’t clear is what the hierarchy of devices at the edge ultimately will look like. Will the smart phone be the central command center, or will it be marginalized by more functionality in individual edge devices such as a car or a home network?

Apple, Google, Microsoft and Amazon, among others, are racing to provide options in each of these spaces because at this point it isn’t obvious which device(s) will move to the top of the heap. The stakes in this game are enormous, which is why these companies are building massive infrastructure behind these devices to support whichever one(s) comes out on top. The problem is that this makes it difficult to design devices for maximum efficiency and performance, because moving large amounts of data requires a different approach than processing it locally. It also makes it much easier to secure that data, which further reduces the overhead.

The third piece of this picture is probably best described as ethics. The ability to extract trends and patterns from data more quickly and turn it into something useful opens the door to misuse of that data, as well. If you think robocalls are annoying, just imagine what happens when companies can determine where you are at any point in time, which direction you’re driving (or where exactly you’re heading in your autonomous vehicle), what you’re wearing, and what you’ve been watching on TV. Some of this is already happening, but if this data is combined it will become even more invasive.

The amount of data available to marketers in the future will be astronomical. Trillions of connected things mean trillions of sources of data, and all of that can be mined with existing tools. Technology is always ahead of regulation, but the speed of technology development in areas of machine learning, deep learning and AI has widened that gap even further. It’s not clear if that will continue to widen, or whether pressure will build to slow down the pace of technological change.

Taken together, these three pieces create an interesting glimpse of how technology will develop over the next decade or so, when autonomous cars begin taking over the roadways, 5G begins to dominate the wireless spectrum, and AI and machine learning are almost ubiquitous. But at the center of this shift are still the same constants, lower power and higher performance, and that is unlikely to change anytime throughout this transition.