Critical Choices

There’s been a lot of talk about what’s good enough. Is 10 hours of battery life enough? If the tradeoff is between a smaller battery and extra hour of battery life, which is more important?

Those kinds of discussions are at the heart of consumer electronics. Ultra-thin smart phones are more attractive than fat ones, and they’re easier to put in your pocket. But a new kind of discussion is underway, particularly at the most advanced nodes. It’s about what’s the most critical part of the device, because the reality is that there will never be time or budget to do all of the verification and testing and optimization.

The amount of data that has to be sifted through is exploding, and while tools have kept pace with much of this growth in complexity there are simply more pieces that have to be analyzed, more parts that have to be integrated, and much more to do. And all of that has to be done within the same power budget, in a design cycle that isn’t getting longer, and which has to have flexibility built in.

The key words that keep cropping up these days are flexibility and context. Tools have to be flexible enough to get really granular when necessary, or run at a higher level of abstraction for less-critical parts. The definition of good enough is a sliding scale, depending on whether a particular function really needs to be optimized or whether it can run a little slower without impacting the user experience. A camera that takes an extra second between shots is a lot different than a two-second delay on accessing the Internet or a delay in an online game. And a phone that lasts several years is probably okay, but that same reliability window doesn’t work for an automobile.

Add in software, more IP, subsystems, multiple patterning, new materials to combat quantum effects, electromigration issues, and this becomes the kind of nightmare that makes for some very tough choices. And if it keeps heading in this direction, those choices will become tougher and more critical, with the growing likelihood there will be some high-profile failures. Even worse, failures now go viral on the Internet. Witness the amount of stuff written about the iPhone antenna problem several years ago versus the video of a burning Tesla, which in the past week has received more than 3 million views.

We are living in a connected world, but in the SoC things are connected much more closely and packed in more tightly than anywhere else. SoCs are the most densely packed commercial real estate on the planet, and the number of choices about what to add, how to route signals and the number of unknowns about all of those decisions are mind-boggling. Worse, we need to understand the impacts of all of our decisions in this tightly packed and integrated space so that we can choose where to put our efforts and what to ignore. This was difficult at 90nm. At 16/14nm, depending on whose finFET you’re using, it gets a lot harder.