As part of DARPA’s Domain System on Chip (DSSoC) program, Arizona State University (ASU) assembled a team of academics and industry partners to pioneer a new approach to software-enabled radio frequency (RF). Here, ASU professor Dan Bliss explains Arm’s vital role in the project, as provider of a range of IP, along with constant close support…
“I'm the director of the Center for WirelessC Information Systems Computational Architectures at Arizona State University. We build experimental stuff that actually goes out and functions in the world, as well as doing theoretical work. I am doing a lot of interesting work now on radio frequency (RF) convergence and spectrum sharing, working on techniques to run multiple radio functions on the same spectrum, simultaneously.
We have partnered with Arm as part of the DARPA-funded Electronics Resurgence Initiative program, Domain-Specific System on Chip (DSSoC), on a specific project called Domain-Focused Advanced Software-Reconfigurable Heterogeneous SoC (DASH-SoC). The aim is to try to enable software-defined RF, which has lots of different functions around communications radar and positioning navigation and timing.
ASU is the prime, leading the project; and Arm is working as part of the project, along with General Dynamics, the universities of Arizona, Michigan, Texas (Austin) and Wisconsin. Each player has its own area of expertise, and we all collaborated on the proposal.
A lot of the interesting stuff in this space is computationally expensive, which naturally pushes me to think about processors. You build a chip to solve a problem and it might be the right answer in the moment – but it will be the wrong answer tomorrow. The obvious response, which many people have tried over the years, is to build custom accelerators for those particular pieces. That works, but as you start getting to a reasonable number of accelerators, and you are trying to figure out how to thread information through them all, it becomes really problematic. And requiring a person to do this process-threading does not scale with having more applications. So we are building a sophisticated, intelligent scheduler that takes this out of the hands of the programmer.
“You build a chip to solve a problem and it might be the right answer in the moment – but it will be the wrong answer tomorrow.”
There are a number of problems we need to solve to make it work. First, we need a fast interconnect between all the pieces. Arm’s CoreLink CMN-600 Coherent Mesh Network is critically valuable for this. It gives us lots of tools and knobs we can turn, so it is not only fast but flexible too, which means we can be power-efficient.
We are also using the Arm Cortex-A53 processor and the Cortex-A72 processor. Arm’s processors work so well that it is no longer a question whether or not you would use one. Arm is just so good, it is such a clear winner, and we would never think of doing it any other way.
The other issue is that if every little accelerator has a different interface and different problems with things like debugging, it can quickly become a nightmare. So it is incredibly valuable to have a common way to interact with these accelerators. Arm’s Revere-AMU architecture prototype is helping us do that, serving as the interface between the various accelerators on our chip and the rest of the chip.
That on-chip network was a scary part for us. DARPA’s requirements required us to really push the boundaries, and we were all concerned about this from the start. But because of Arm and how well the Revere-AMU architecture works, it went from a high-risk part of the project to low risk. So for the people running the project it is like: ‘Ahhh, that is nice.’
Most people have no idea how the world of tech development works. Every interesting bit of tech we have now represents the end point of a 40-year arc. Look at your phone, this truly beautiful piece of work: it is astounding where each of the many subsystems that are integrated into it came from. There have been many sources of funding for the development of computers and the development of processors. But as well as playing a key role here, DARPA funding also enabled early internet development and voice technologies which are now widespread.
“DARPA’s requirements for the on-chip network were rather extreme, but because of Arm and how well the Revere-AMU architecture works, it went from a high-risk part of the project to low risk.”
We understand how the process works. For our project on this processor, we have built our chip so it is ready to be mixed up and then put into someone else's product, in whatever form they need. They probably will not take exactly what we have, but it is easy for them to modify it and put it in their product.
But this is not a scenario where we are given access to some IP and Arm thanks us and offers to check back in next year. Arm brings other sorts of help. They are an active part of the team. They are on our bi-weekly calls, supporting us. And there are constant little tweaks we are looking for. First, understanding, and getting the CMN-600 to work the way we want it to. Then there is Revere-AMU and how it is working for us. We have regular interactions on multiple levels, and working with Arm on it all has been really positive. We face several program development risks, but Arm has helped us navigate and mitigate these to keep moving the project forward. It is hard to convey how important that is.
“I do not have a clear-cut sense of the line where academia ends and industry starts. The idea that you would build something that stayed in academia and never left seems a horrible waste of time.”
Personally, I do not have a clear-cut sense of the line where academia ends and industry starts. The idea that you would build something that stayed in academia and never left seems a horrible waste of time. It would be a bit of performance art that does not actually do anything. If I really want to make sure I am working on problems that help society more broadly, I need to have some connection with industry.
In the case of our DSSoC project, we are driving some tech development and pushing it out to industry. Meanwhile industry helps us make sure we are focused on the issues that are of actual real value to them. That makes us more efficient as researchers.
Arm has reached out and is willing to team with us, and to expend its energy and funds in support. So if anyone asked me whether I’d recommend working with Arm, clearly the answer is yes. Do it. We would not have been successful without our collaboration with Arm. There is no two ways about it. We have had a wonderful experience.”
Dan Bliss, professor in the School of Electrical, Computer and Energy Engineering at Arizona State University and director of the Center for Wireless Information Systems and Computational Architecture.
Arm Research is committed to supporting our academic partners. How have we aided their research, and how can we support you?
Learn more