Q&A with SiFive SVP Jack Kang: 5 nm chips with SiFive cores likely to be used in cars by 2025-2026

Judy Lin, DIGITIMES Asia, Taipei 0


SiFive successfully rolled out a portfolio of automotive CPUs based on the RISC-V standard, and is about to create a new ecosystem of chips, tier-1 auto component providers and OEM partners. DIGITIMES had an exclusive interview with SiFive SVP Jack Kang, who shared what triggered the RISC-V leader's entry into cartech chip IP and the new opportunity SiFive's new portfolio is creating for the supply chain.

Q: What makes the new portfolio stand out from the IP currently available on the market?

We've seen tremendous growth for RISC-V in all market segments over the last few years, and it keeps growing. Automotive is one area where there's a lot of interest now. Why? Because the automotive ecosystem is changing, it's evolving. There's a whole new software ecosystem. A lot of customers have what's called "first party software", which means they're in control of their own software stacks. That means it's easy for them to move to new architectures, especially when that architecture bring significant benefits in power performance, ecosystem choice and things of that nature.

In the RISC-V space, SiFive has the most complete portfolio of products for automotive. There are some other options by vendors who have announced certain point products, but in the RISC-V space this is the most complete portfolio of products for automotive, in terms of supporting everything from 32 bit to 64 bit to vector processing, covering all the different areas for the automotive market. The strength of SiFive is that we're able to provide the entire portfolio. There are multiple different cores within a system, and there are multiple different SOCs, which all need different versions of a CPU. Being able to provide that is very critical for large-scale adoption. Then you get the specific benefits offered by the RISC-V ecosystem combined with the benefits offered by SiFive products. From an ecosystem standpoint, by moving your software to RISC-V you now have freedom of choice. You have lots of open-source software developers working on it, and you're going to have different suppliers offering solutions.

The RISC-V ecosystem is going to grow much faster, and you're not dependent on any one vendor, which means you're going to have more diversity of choices. Then you look at the SiFive products themselves. Our products provide significant power and area advantages at the same performance level as other cores. If you compare to other competing cores from suppliers like ARM and others, SiFive typically offers 30 to 40% better power consumption efficiency, plus a smaller area. This enables more cores on a chip, a lower cost for chips, and lower power consumption.

Q: Besides Renesas, which companies have become partners with SiFive in their new products? Do your IP also have to go through years of validation with the auto makers?

On the open-source point, it's very important to understand that RISC-V is an open standard. That open standard is shared, which means lots of people can do things on it, enabled by a community of openness. And with that there's a lot of open-source software developed for RISC-V. There's a significant ecosystem of partners who develop software; high integrity, fault tolerant software for the automotive space. Now, those are not open source, right? Those partners are crucial partners for the day-to-day grind. By supporting RISC-V, then their ecosystem of hardware partners expands to not just SiFive, but others. This is why RISC-V is mutually beneficial, and there'll be large-scale adoption.

For the partners in our ecosystem, Renesas, along with some 15 other companies, announced support for our automotive launch. We have a couple of other very large customers, tier-1 chip companies, but they are not public, so you have to look for those announcements in the weeks and months to come. But we do have customers for all of these products that we are announcing today.

Our IP is very thoroughly validated on the baseline stuff, and then when we get into the automotive products, there are certain SOP/SOD type of requirements and compliance things that are required.

In many cases, either we're providing those, or we're working with our customers closely to ensure that the solutions meet safety and reliability standards. Because if we look at the automotive space, ultimately, it's about the safety and quality requirements for the final end product and for the system level. So it depends on the customer: some of it is deployed at the system level, some of its deployed in the chip level, some of it is deployed in the IP level. You'll see in the release that we talked about our products being capable and suitable for different types of applications.

SiFive has been around for seven years now. Some of our automotive products have their roots in the success of other embedded products shipping with our IP. It's hard for me to quantify how many years of verification we do for each one, but as IP providers, it is very important that we build and develop trust with our customers.

Q: What inspired SiFive to make entry to the Cartech space? How many years of R&D have you spent in it?

Actually, it was our customers that kept asking us. A couple years ago, we were a little bit hesitant. We were working in other areas, and they kept asking. There's clearly a demand here and a need for something.

Cars have been going through this very rapid technology change. With the electrification of cars and cars becoming a "data center on wheels," cars are becoming very advanced technically. But if you look at semiconductors and the CPU IP available for the space, it has not made the same progress as CPUs and chips for AI or data center chips or mobile chips. The automotive space was a little bit behind, so there was an opportunity for innovation to come in.

Our customers kept pushing until finally we decided we had to take on this market opportunity. And the more we looked into the automotive space, the more we discovered that it is a very good fit for our products and our roadmap, not just the three new solutions that we announced. And we've built a team of automotive experts to help drive this forward.

Q: Which process nodes best suit E6-A, S7-A and X280-A series of processors?

It's pretty broad because we have customers in every automotive node you can think of because RISC-V is very flexible. Now, as you get to higher and higher performing cores, and you start talking about more advanced capabilities, such as ADAS, and you get to L-3, L-4 and beyond, those chips are going into more and more advanced processes because you still need that processing capability.

That creates some additional challenges for functional safety and fault tolerance and has to be handled in other ways. For some of the other functions, maybe they will stay at some of the older nodes. As you get to the higher and higher performance ones and the more advanced nodes, you're going to be pushing into higher and higher performance cores. Some of that will probably be the cores on our roadmap that we haven't announced yet, that will be pushing towards 5 nm or lower nodes because you need that kind of compute.

Q: How long would it take for that core to go into five or below 5 nm?

The auto industry is speeding up, but it still has a longer cycle than consumer or mobile or even the datacenter. For the discussions that we're having now, our IP will be used for chips in 2024 and 2025, which means they won't make it into cars until the model years 2025 or 2026. Right now, they're all trying to accelerate that schedule. That's probably the fastest timeframe that you'd be looking at. It's not like mobile, where you announce something and you see it right away.

Q: And because cybersecurity is the issue that many people concerned about the connected vehicles, how secure is the RISC-V architecture auto processors?

Security is a system issue. One of the very things about RISC-V is from an architecture standpoint, it was developed at a time when cyber security is very important. RISC-V has a clean slate, no legacy of stuff that you're trying to attach onto it. For example, Spectre and Meltdown happened a couple of years ago, but no RISC-V cores were affected by that.

SiFive offers a security model called WorldGuard. This is an example of a security implementation our customers can choose. And there is an advantage to being built on RISC-V also. RISC-V is an open standard, that means its architecture is available for everybody to see. You get the benefit of all of the companies looking at it, and making sure there are no issues and holes, and contributing to the shared standard. At the end, you still have to do the right thing, not just in the CPU, IP, but also in the SOC and the system. Everybody has to pay attention to it. The good news is that security requirements are down to the core, and people have a much better understanding compared to even five years ago or 10 years ago. That's a big advantage that SiFive has: a clean slate for designs.

Q: Intel has this collaboration with you and also with the RISC-V community. The foundry service for all those chips will go to Intel? Or other foundry makers also have a chance.

Intel has certainly been a big proponent for RISC-V, which is great for the ecosystem. They see that this is where a lot of new chip designs are going to happen, and RISC-V will drive value for their foundry. You're going to see this across the board, all foundries are going to benefit from more RISC-V designs. This is where the foundries get to differentiate on their capabilities and features. We'll see how they compete, but it's clear that RISC-V is going to create more opportunities for all the foundries.

SiFive SVP Jack Kang

SiFive SVP Jack Kang
Photo: SiFive