by Swapnil Bhartiya

Why the time is right for open source to meet hardware

Aug 03, 2016
Computers and PeripheralsCPUs and ProcessorsLinux

"Open source has proven itself to work very well in software and very successfully," says Jack Kang, VP of Product Marketing for SiFive, the newly launched open source chip startup. "We don't know why it's taken so long for it to come to hardware." rn

SiFive is a San Francisco-based fabless chip manufacturing startup that is developing the first fully open source chips. The company, which launched earlier this month, announced two chips of their flagship platform:  Freedom U500 and Freedom E300, each targeting different audience. We caught up with Kang to talk about the Freedom platform, the challenges of starting an open source hardware company, SiFive’s business model and more. Following is an edited version of that interview.

How different is your solution from previous efforts to make open source chip architecture, such as IBM’s OpenPOWER?

I think with OpenPOWER the name is a little bit deceptive. It’s not so much a free and open system as you can pay to join the membership to connect to IBM stuff. My understanding is it’s really a specification inside the system put out by IBM, and if you join OpenPOWER then you’re able to use it and able to connect to it.

RISC-V and the Freedom platforms are are completely free and open. The specifications are published under the BFD licence. Anybody can use it. Anybody can implement it.

Actually, we want other people to copy, in effect, and use those specifications because that’s what’s going to unite the ecosystem. Software developers can all develop on a common platform. In addition to the CPU, you also need to know how is the debug connected, how are interrupts connected, what does the memory map look like. All of these things are platform specific. Today if you look at different SOCs (system on chip), they all have different specifications depending on which companies are making them. They’re oftentimes closed and not open. Everything is open in our case so customers and developers get full access to our platform.

There is also a RISC-V foundation, what’s the role of the foundation?

Any company can become part of the RISC-V foundation and there are a lot of members which you can see on the member’s page. The benefits of membership is that the members decide what happens to the new revisions to the ISA, compliance tests and use of the logo. But you don’t have to be a member to utilize RISC-V. I think that’s where the free and open part comes in. Anybody can have their own implementation of RISC-V. And you are free to do your own implementation as open source or make it proprietary. It doesn’t restrict the implementation. I think that’s where the big difference is in what’s happening.

How do you plan to increase adoption of the Freedom platform?

I think a lot of it has happened very organically, and the momentum behind RISC-V as an ecosystem has grown tremendously. In January of 2016 when we organized our 3rd RISC-V workshop there were only 16 companies. In May we organized our 4th workshop and the number of companies rose to 40. It was sold out. We had to turn people away, unfortunately.

There were over 60 companies at the foundation, over 40 universities and 260 attendees. The interest is extremely high. I think it’s one of those things where it’s grown much faster than we expected.

Now since we have announced Freedom platforms, there is real platform that people can use. So it’s going to grow.

Can you talk a bit about the two platforms? What audiences do they target?

Freedom U500 Series is a high-end platform and we are seeing a lot of interest in storage, networking, machine learning … basically areas where people want to accelerate computing. A lot of these sockets are things that fit in the data center, not necessarily the main CPU, but offload and acceleration-type devices that are right for customization. We’re getting tremendous interest there.

The second platform, Freedom Everywhere family, is more of an embedded micro-controller-class device: a lot of IoTs and wearables. IoT is especially interesting because this is a market that everybody thinks will be very large. But nobody knows what it looks like. Because of that, it’s the perfect example of a very fragmented market where there’s lots of different requirements.

There’s lots of companies with ideas; there’s a lot of areas where they have special communication-type concepts or security concepts or other ideas that need to be in silicon. This is a very good example of an underserved market that the traditional companies are having a hard time servicing because their model has been going towards finding larger and larger volume sockets, and because of that they can’t customize their chips. That’s an area where we see a lot of customers coming to us because of the customization that our chip offers. It meets their unique requirements.

How do you plan to keep costs down for customers?

I think there’s two areas where we’re looking primarily at how we can reduce cost in the market. On the hardware side, building chips is expensive. You still need to fab the chip. There’s still a lot of tools involved; there’s still third-party IP. All of that are areas that we are working on. We think RISC-V, of course, helps, but also by utilizing the Freedom platform, we’re also focusing the changes on the areas that make sense. The rest of the platform, the common part of the platform, is verified, is tested, and we don’t have to reinvent that every time and redo a lot of development and effort work.

The much larger savings actually comes from the software side. Today, there’s a tremendous amount of software work required to take your chip to production. This software work isn’t done by the customer. A huge portion of it is done by the chip company because a large part of the chips are proprietary to the chip companies, and only they can fix the problem. We are looking at open source software — we are utilizing it, relying on it. Since both software and hardware is open source it means people can solve bugs and fix things on their own. They’re not held hostage by the chip supplier who is the only one that can fix the problem because, for example, that bug is in some binary blob or it’s using some black box piece of hardware that they can’t look into.

We think that’s actually an area where if you deep dive into the production costs, tremendous amounts of savings are going to come from the software side.

Open source has been extremely successful on the software side. But open source hardware is still far behind. Why do you think that is?

Open source has proven itself to work very well in software and very successfully. I think when it first came out, people questioned how the business model would work and why people would do it and how it would make money. And I think it’s proven to be a tremendous model for innovation and business. We don’t know why it’s taken so long for it to come to hardware. We think open source hardware makes absolute sense. We think the economic factors of Moore’s Law are effectively stopping. The cost per transistor is no longer going down. The cost of design has grown exponentially. We think the time is really right for open source to meet hardware.

What about patents? Should people be concerned about patents when using the RISC-V platform? Will customers need to pay royalties to some chip maker?

I think there’s a long genealogy and I think the RISC-V Foundation actually has a list that you can look at that show that many of the ideas have been around for a long time. I think specific to your question, it would be hard to answer in more detail.

It depends on what they would like inside their chip. Certainly we don’t believe that there’s any royalties that need to be paid for the RISC-V CPU.

What has been the biggest challenge for you so far?

I think the biggest challenge was getting a hardware company started. There’s so many software companies being started, but the open source hardware company has fallen out of favor. You don’t see very many hardware startups. I think that this is something that’s very unique to [Silicon] Valley, or I guess as a whole, just in terms of hardware startups. Maybe open source hardware will encourage more companies to start.

What about Raspberry Pi?

I think Pi is a great board. It enabled a lot of makers to play around with stuff. I think the big thing there was [that] even though the board was open source and the hardware was open source, the chip was not. We think open source is the next step. Then we think we can help people graduate to where they’re not hacking around with the board, but can go build their own devices with their own custom silicon.

As a company will there be any secret sauce in your open hardware?

Philosophically we want to open source as much as we can. There will be, for example, third-party IPs that we cannot open source. Again, it’s about making sure the software works, so things that we can open source will be very well-documented from a software perspective.

What’s going to be your business model?

As a company we are going to make chips based on the Freedom platform, the specification of it itself is also open so others can use it. Even though the design is open source, somebody still needs to put that together, assemble it and deliver sand, as I call it. We’ll do that. That’s not easy. That’s really hard even if everything’s open source. That’s a part that we want to take on and help the ecosystem. Then we will sell finished silicon.

When will the first chip be released?

We’re working with our lead customers so it depends on a little bit of their schedule and how our alignment works. It could be as early as the end of the year or it could be next year. It depends on our customers.

Who is backing you financially?

We’re backed by Sutter Hill Ventures. They’re a pretty well-known VC with a track record of success. They incubated Pure Storage, NVIDIA and a whole bunch of other companies. We’re not disclosing at this time how much funding we’ve raised, but from a commercial aspect, we will commercialize RISC-V silicon, and our fundamental business model is to sell chips.