Peter Greenhalgh is the director of technology and fellow in ARM’s CPU group, responsible for all micro-architecture developments across the CPU group, technology innovation and engineering aspects of the ARM CPU roadmap. He was the lead architect of Cortex-A7 and Cortex-A53 and led the development and deployment of ARM’s big.LITTLE technology. I booked some time with Peter, in advanced of his ARM TechCon keynote address, to get his thoughts on processor design challenges in the years to come.
Q: Peter, you’ll be giving your presentation on a perennially pressing issue: power. What’s your vision for the future of technology as we scale to finer line widths and power considerations become more crucial?
Greenhalgh: Well, today it’s not just improving power efficiency. It’s increasingly about delivering low power and high performance at the same time. It’s the marriage of multiple factors ranging from process geometry, to micro-architecture evolution, to tools to per-core thermal budgets. Thermal budgets, in general, don’t change whether you’re looking at a mobile phone or an ADAS system in your car’s rear-view mirror or high core counts enterprise systems.
Q: No pressure there! Well, we have a much more robust design ecosystem today. Is that giving SoC designers better visibility earlier?
Greenhalgh: Certainly the ARM ecosystem has become a model in some ways to continue sharing information and driving design forward efficiently. We have great relationships with EDA tools vendors, the foundries, silicon partners. ARM’s PDG team has excellent relationships with foundries, and they give us strong guidance about what’s coming in the next process node that extends into what Greg Yeric—my fellow keynoter—and his team in R&D are looking at on the bleeding edge. Then our long relationships with silicon partners and OEMs go back decades. We’ve been able to have conversations that are confidential and we’ll take their info and advice and fuse it with other information to build the most efficient and optimal-performing product for that generation.
Q: Let’s look to the recent past. You’ve been involved in a lot of key ARM CPU architecture directions. Where did you have anxiety at one point during development and then happily overcame it?
Greenhalgh: The speed of product development that’s required now is extremely high. It’s the time to develop the hardware and the relationship between hardware and software development that can be challenging. Trying to get all of that right is tough on everybody.
Q: What drew you to CPUs versus other engineering paths you could have explored?
Greenhalgh: It’s the fast-moving industry that CPU design is. There’s also the fact there’s a phrase “an engineer can do for a cent what any fool can do for a dollar.” I think of the phrase as you have to make decisions when you don’t have all the information available to you. You have to decide when nodes aren’t set, tools aren’t set. We have to make informed judgments based on what we’ve seen in the past and trust our silicon EDA and foundry partners. You’re making a highly informed guess on what’s going to happen in the future and do it in a cost effective way. That’s my analogy. If you were to try to control and research every piece of information, you’d be there forever and a day or you’d be beaten to market. It’s that ability to make decision with only some of the pieces of the puzzle available.