Making Things Better

The challenge of improving performance and energy efficiency is going for less visible purposes.


For the better part of the past decade the focus in semiconductor design has been on improving energy efficiency—making batteries last longer and lowering the energy bills for data centers—and continuing to boost performance. The benefits of future engineering efforts may be less obvious.

In fact, progress in the future may be measured more against improving what’s already there today rather than increasing battery life or lowering electric bills. Consider voice recognition, for example. While it works better now than in the past, there isn’t anyone who would argue it’s good enough. It’s still obvious, even with advanced automated answering systems, that you’re talking with a computer.

Getting to the next level of interaction, which is based increasingly on artificial intelligence—or at least some semblance of AI, where enough scenarios are programmed in and respond quickly enough not to frustrate users—requires changes in the following areas:

  1. Massive improvements in bandwidth;
  2. More memory with faster access times and lower latency;
  3. An array of processors, including some with very narrowly defined functions;
  4. More energy efficiency in the design, which requires hardware-software co-design as well as architectural changes;
  5. New use models that take advantage of near-threshold computing, DVFS and other energy-saving techniques;
  6. Less current leakage, either through new transistor designs such as finFETs or silicon on insulator, or both, and
  7. A widespread recognition that all of this has to be well thought out ahead of time.

The challenge isn’t that this can’t be done. There have been big advances in all of these areas over the past few years. It’s that this stuff is less visible to the consumer, which means they may not be so willing to pay for incremental changes that ultimately will bring about an entirely different user experience. It’s easy to see the difference in smart phone performance moving from 3G to 4G LTE, but it’s less obvious when a voice recognition system such as the iPhone’s Siri understands a particular word or concept and responds more accurately.

Still, this is the direction that low-power and high-performance engineering will need to take over the next decade. These incremental changes will pave the way for massive changes in electronics—ones that will make today’s devices look primitive and dumb. The only question now is who’s going to market it to customers, how they’re going to pitch it, and whether they’re going to reap enough of a return on what ultimately will be a very large investment.

The fact that this will borne out across things as well as people will help, of course. The Internet of Things doesn’t have accents, pronunciation or language issues, even though it may have some protocols to standardize. But will people really see a difference, and can companies convince them that this is a direction and a journey, not a single device upgrade? That’s a big question, and it’s one that will have to play out in many markets over the next few years.