RISC-V CTO Mark Himelstein talks about importance of ecosystem, software and focus on IoT, automotive, HPC and cloud servers...
This week, the RISC-V community member organization, RISC-V International, announced the appointment of 16 board members to the new legal entity headquartered in Switzerland, as it transitions away from the old RISC-V Foundation entity.
The organization also hired a full-time chief technology officer, Mark Himelstein, in June 2020. We caught up with him in an online interview last month to talk about his role, challenges, if any, to widespread adoption of RISC-V, and about the effect of geo-politics.
EE Times: Tell us a little bit about how your backuground and why you decided to join RISC-V.
Mark Himelstein: Probably the three most pertinent things in my background were: 1) I was employee number 45 in MIPS doing compilers, optimizers, operating systems, and helping with ISA design; 2) from there I ran Solaris for five years for Sun Microsystems; 3) then I was founder and CTO for a massively parallel compute engine with a huge amount of Flash (Graphite Systems, acquired by EMC in 2015).
I’ve done everything from ISA to applications, and one of the reasons Calista [Redmond, the CEO of RISC-V International] chose me is because while we know that ISA is incredibly important, the ecosystem is equally important; and she wanted somebody with both the systems and the ISA experience and someone who was software-minded, because that’s the area where we probably need the most growth, and the most work in order to make the RISC-V future as successful as possible.
It was some of my colleagues at MIPS that through my name in the hat for RISC-V CTO, as I wasn’t really looking. Ultimately, I chose the job because it was technically interesting, and I like the people I interviewed with. These are the two major things at this point in my life when I choose jobs.
Given your experience in this industry over the years, what do you think are the opportunities that RISC-V offers?
Mark Himelstein: This is the first open source chip of this magnitude that started as open source. It wasn’t gifted by a big corporation who maintains oversight. It wasn’t kept closed but kind of shared by a big corporation. So, this was an opportunity for the community to treat this like they treat Linux.
In 1990 if you talked to me about Linux, you’d be talking about AIX and Solaris and so on. Now it’s like a no-brainer. We feel the same is going to be true of RISC-V in the future, it’s going to be a no-brainer, because people have learned through the open source software world that not being controlled by a big entity means having the freedom to do what you want.
The amount of intelligent people involved with it is just amazing. One of the reasons RISC-V is so flexible in going from IoT to supercomputers is because we have those people involved in the design, and they care about it. We know what’s important in those areas because we watched – everything from MIPS and SPARC, all the way up through Arm and x86.
With that benefit of history, we built an ISA that is flexible. From 16-bit instructions, floating point registers, integer registers, so that small compressed things like IoT don’t need a separate set of floating-point registers and not as much precision, but enough to do the job for an IoT situation. We’ve thought really heavily about compressing instructions and so on and how to fit into that small space.
On the other hand, because of the extension model, and the flexibility, we’re finishing off the vector spec right now. So, the Barcelona Supercomputer Center, the European initiative, are all driving that supercomputer effort. And from there and everything in between, you have things like edge networks, specific cloud server deployment.
What are the challenges for designing with RISC-V right now?
Mark Himelstein: When it comes to the question of when things roll out, it really depends heavily on what runway is required. It’s not necessarily just like Linux, where you put out an operating system, you’re not really beholden to the hardware schedules; but with software, you’re releasing the software on existing hardware, and as new hardware comes on, you build it. Here when we are building the new hardware, we have to make sure the new software catches up. But as you can imagine, a general-purpose enterprise computer, ie: multi-processor, multi-socket, requires an ecosystem that’s pretty significant.
If you go to IoT, or you go to a cloud server, or you go to high-performance computing, the application list is smaller to get off the ground. And that’s why those things are in process right now, and you’re going to see those things come out first. And then over time you’re going to see the more general-purpose stuff happen, that requires a bigger ecosystem, with all the databases and so on.
Obviously, Java, Adu, Spark all make life easier for bringing things to new platforms. But even in those cases, there are certain situations where you want to take advantage of the hardware underneath and its special capabilities.
What are the challenges to widespread adoption of RISC-V – why is it not being utilized as much as it should be, given its benefits?
Mark Himelstein: We have 600 members, and probably 120 of those who are developing their cores and hardware that’s going to end up in machines. But there’s a process to that. Hardware has a longer runway, unlike software, so it takes time. You put out prototypes, your customer goes and tries it, you may even have to some iterations. All these things are way longer than fixing a bug in Linux.
In IoT, from an architecture perspective, people are going ahead, we’ve already seen SoCs out there, we’ve already seen embedded stuff out there. There’s a whole bunch of tool vendors too. There’s a complete list that’s growing on our web site of software and core availability, and dev boards. There’s just an incredible infrastructure growing around this.
However, if you’re going to change the chip on an automobile, the turnaround time on an automobile is five years, so you have to understand there’s a long lifecycle that we have to go through. But I don’t think there’s anything at the bottom layer that’s giving problems.
Part of my job is that if anyone is having a roadblock, then I come in there and help blow them away. I have a lot of experience in systems development, and software development, and I understand who the players are, so I can help bridge the information between all the groups.
I don’t really see any issues at the lower end, but as you get towards the higher end, there are some challenges, but that’s just because of where we are in the lifecycle. For example, in the history of the world, you can see that other architectures have let too many variations go out, and then there ends up being too much for them to do, they are overwhelmed. We intend not to go there.
What is RISC-V International working on to help the ecosystem?
Mark Himelstein: We have this concept of profiles we are putting together. Then we can say, “Go to these profiles and we will guarantee you things like application compatibility.” Which is important. We believe we have a path to do that, but that work needs to be done, which is the reason they brought me in to ensure that is in place.
We have to make sure we have a plan in place that allows people to do innovation, add proprietary value, and allow them to share with the community stuff that’s common denominator. We are in the process of doing a gap analysis.
But I will tell you, user stories are good. You have to actually tell somebody who is going to use it, why they are going to use it, and how they are going to benefit from it. So there’s a whole bunch of verticals that we care about and have a complete list for you to succeed, whether it be automotive, telco, finance or oil and gas and anything that’s a good target for RISC-V.
How are you working with other organizations in this plan?
Mark Himelstein: All the other organizations out there helping with RISC-V is just phenomenal: OpenHW Group, Chips Alliance, OpenTitan. It just brings home the point that people want open source, they want to control these things, they want influence on these things.
We pick up the base ISA, and the stuff around it, the ecosystem; one-third of the work we do is the ISA, and two-thirds of the work is the ecosystem around it, whether it be the boot loaders or configuration files, or Linux ports, or compilers, or security guidance. All that is ecosystem.
But it’s clear there is more beyond it, when we start talking about, for example, sharing SoCs or sharing dev boards or sharing security stuff or whatever, there’s a lot of people working on these things, and they really like RISC-V. Everybody’s got to decide, how much do they do of their own, and how much do they use somebody else’s.
People will start with a core that someone else created – either one of these outside groups or from one of the 35 or so cores on the RISC-V web site that we helped to roll out. And then people have to decide what kind of innovation they want to do.
What do you say to some perceptions that RISC-V is still in the academic domain?
Mark Himelstein: Let’s be very honest, we’re still working on some of the specs. But there are solution providers out there that offer help. For example, I won’t go into names, but there’s one group working with an automotive company and providing them with the solution. The automotive company’s not going to do that [the design]. They hire somebody to do that. The Walmart’s of this world will go to a solution provider and might say, we want RISC-V and here’s why.
How are you making people aware of how RISC-V can benefit their designs?
Mark Himelstein: I am the chief tech evangelist for RISC-V. My number one goal is the deployment of products with real RISC-V cores. In the end, that’s the only thing that counts.
What next for RISC-V, not that it has you in place?
Mark Himelstein: I see my role as having three parts. The first is operational. Get the specs done. Do the gap analysis. The second part is planning. Going to these verticals and making sure we have a list of everything they need to succeed, so we enable them, and they don’t have any barriers. The third part is outbound. Speaking to media, going to conferences and evangelizing the product.
Because I bridge the gap from ISA all the way up to apps, I have a more holistic view than a lot of people. I can sit in a meeting and say, ‘hey guys, have you talked to the software guys about this?’ and then force a conversation to occur.
You said you can’t go for every vertical and need to focus on a few key areas. What are those immediate focus areas?
Mark Himelstein: The guys who have the shorter runways are where we need to spend time. So, we’re taking a look at IoT, but not just IoT, but where is it going into and what do they need. We are similarly doing the same in automotive, high-performance computing (HPC) and cloud servers. Those are the four areas I see moving very quickly. And I want to make sure we get ahead of that game, and make sure they have everything there that they need.
One of the issues in general with these types of projects is that there’s the potential for people redundantly doing work that isn’t of proprietary value. And part of my job is to do synthesis and attraction, figure out what could be shared and encourage conversations where people can leverage the community.
Tell us what you think about some of the geo-political challenges, given that you established RISC-V International as a Swiss entity.
Mark Himelstein: Let me tell you, this is no different to export controls we had in the 1990s. Things were very strict, and I ran Solaris when we open sourced PKI basically so that we could sell security into China. Because you couldn’t do it as part of a proprietary operating system, but you could do it as part of open source.
I think over time, everything will find its own level and you figure out how to do business and so on. I’ve seen no aggressiveness from any country or any participant with respect to RISC-V. In fact, it’s just the opposite. They want to fuel the community because they want to take advantage of the community work. The nice part is, it’s all open source, so it’s not as if we are hiding anything. It’s all out in the open, just like Linux, just like Adu. And just like them, we have people using them in China, and Pakistan, and other places, and it’s OK.
Obviously, there are geo-political forces that heighten things in this period of time, more than they’ve been since the 1980s and 1990s. We have to pay respect for that, we have to make sure we honor that, and not do anything that is considered to be violating any country’s restrictions.
The reason we moved to Switzerland is because Switzerland is a neutral place. So, it gives people comfort. Even though there’s nothing in place right now that would stop us doing business in the United States, it gave our members comfort being in Switzerland. That they have less exposure to any kind of changes as we move forward. We’ve had tremendous acceptance of the move to Switzerland, and the truth is, it’s still a worldwide organization with everyone contributing. It hasn’t changed the operations of how we do things.