Uncertainty, But Not Over Power

While industry contemplates next steps in architectures, materials and process nodes, reducing power remains a fixed target.

popularity

The semiconductor industry has reached a crossroads. Lithography has stalled out, NRE is rising, and chipmakers are torn between choices of when and whether to jump to the next process node—and even more daunting, the next one after that—or whether to take half steps with fan outs, 2.5D, or fully depleted SOI.

While chips do continue to tape out, the number of critical choices that need to be made over the next few years is daunting even for those with the clearest path forward—the memory and processor makers and high-volume mobile SoCs. As most of them concede, it’s not just the next stop on the Moore’s Law road map that needs to be considered. It’s how companies will need to be reconfigured to move forward, including changes in methodologies, tool flows, and in some cases their internal business structures.

One thing hasn’t budged in all of this, however. There is an almost universal recognition that power needs to be reduced, no matter which direction chipmakers decide to go. Until several years ago, the PPA equation usually favored performance and area/cost over power. Depending on the application, performance and area now seem to be more elastic, while reducing power is a fixed priority.

The impetus comes from many different directions. In the cloud and data center, the energy costs are often measured in seven figures. In mobile devices, it’s measured in time between charges. In cars, it’s miles per gallon and/or per charge. And in industry and home markets, it’s electricity bills and reliability. And across all of these markets, there is an overriding concern about global warming.

What’s also changed is that the drivers for reducing power are present at all process nodes. It’s no longer just the threat of cooking a chip’s circuitry, although that is still a concern at the most advanced nodes.. It’s a marketing-driven decision that now approaches battery life the same way computers used to talk about MIPS. Even software teams are beginning to focus their efforts on efficiently accessing hardware rather than focusing entirely on functionality and performance.

What a difference a few years makes.