In the era of cloud computing, do we really need new silicon? Yes we do — for a number of reasons.
You can’t turn around these days without hearing talk of our increasingly connected lives, tethered by the Internet of Everything (IoE), and cloud this, cloud that.
Don’t get me wrong, I’m not knocking this next big thing because on many levels it is inspiring creativity, as all big things do. And the ramifications are still getting worked out as far as the impact on chip design, and the ways that hardware designers will see a difference in what they do.
To this point, during a recent discussion about whether software is more of a determining factor in a system with Chris Rowen, who is a Fellow at Cadence and CTO of the IP Group there, he posed the question, “Why do people even bother to build new silicon?”
He agreed there are some domains where it’s so clearly completely software that you don’t think about it. “If somebody is saying I need to put up a new application of some kind, they’re just going to say, ‘I’m going to program it in the cloud. There’s going to be a web browser in front of somebody on their phone or an app on their phone, but all that functionality, I’m going to do everything I possibly can just using programming, and there are rich programming environments. So there is a certain class of products where the hardware is not even thought about at all, except that you know that you have certain platforms that you need to support.”
“Now,” Rowen continued, “for most of the things that we care about in the high volume silicon world that are driven a bit more by applications, then you have clearly this tradeoff to make, and there’s a much more conscious set of choices, and there’s much more awareness in that subcategory that the hardware characteristics matter a lot. It’s useful to ask the question, why do people even bother to build new silicon?”
One answer, he said is because the underlying process technology has changed and you can build a cheaper chip if you move it to 16nm instead of 28. “At a certain point in time, that’s always true. And a lot of this angst over the end of Moore’s Law is really angst over the fact that that transition process where you get kind of a free benefit in cost is slowing down.”
The other, more compelling reason why people build new silicon is that they are attempting significantly harder problems, as measured by some very raw characteristics, Rowen explained. “They have a certain amount of compute per watt or compute per dollar they need to do. They have a certain amount of memory bandwidth, they have a certain amount of memory capacity that they need, and they can’t solve that problem at all unless they provide the hardware which has those raw characteristics so that you can often distill out — regardless of the details of the applications that are being run — where is it that there’s a fundamental new horizon in compute or in bandwidth or in capacity that I only can address if i change the hardware. And that, if you look across SoCs, is the dominant driver of why people are doing new SoCs.”
This is good news, hopefully, for any of you hardware engineers concerned about being displaced.