RISC-V is trying to launch an open-hardware revolution | Upscaled

For more background, check out our Apple and ARM video. RISC-V is a processor architecture and instruction set developed at UC Berkeley. It's attracted huge interest from everyone from startups to tech giants because it's entirely free and open source. Most current processors come with license agreements and are proprietary intellectual property, but anyone can manufacture a RISC-V chip, or design their own new processor. Big companies like Western Digital are already announcing a switch to RISC-V, and others like Google and Nvidia have partnered with them. There's a lot of ways this project could fail, but it also has the potential to make custom processor design available to a lot of people. You probably won't be getting a RISC-V PC anytime soon, and chip makers like Intel or Apple probably aren't about to switch to RISC-V but expect these chips to start showing up in all kinds of devices very soon.

Video Transcript

CHRIS SCHODT: Is RISC-V the chip of the future? For once, actually may be. These chips promise a lot, but you may never actually even see one of them. Let's take a look.

[MUSIC PLAYING]

Welcome to Upscaled, our explainer show, where we dig into all the chips running the things, you forgot even have chips in them. Today, we are talking about RISC-V, which is an open source instruction set architecture or ISA. Now, these chips promise to have a huge impact on the world of computing, but they are still super new.

There's a lot to unpack here, but if you want a more complete introduction, go check out our video on Apple and ARM. But here's a quick bit of background on instruction set architectures. The instruction set defines the most basic operations a chip can perform. Things like reading and writing values from the memory, addition, multiplication, things like that. And RISC is a type of sort of instruction set design philosophy that tries to keep those instructions short and simple.

This is the way the ARM chips in mobile and smart devices tend to work, whereas your laptops and desktops and servers tend to use a CISC design, like x86 that uses fewer, but longer and more complicated instructions. RISC chips tend to be more energy efficient, though both designs have borrowed a lot of each other's best ideas over the years. RISC-V is an instruction set based on these ideas.

And it was developed at UC Berkeley, which is the same place that David Patterson and John Hennessy developed the original RISC chips back in the 1980s. RISC-V actually grew out of an academic exercise. Another Berkeley professor, Krste Asanovic was looking for a chip that he could use to help teach students.

I talked to Asanovic and Patterson who's been involved in the RISC-V project since it started in 2010. And they told me they looked at a lot of the other conventional processor designs. Things like MIPS and SPARC and x86, but that all of them had drawbacks for teaching.

As Patterson told me even something like ARM is very complicated for a RISC chip, while x86 is just incomprehensibly baroque. More importantly, these chip designs are all proprietary. Even if x86 was the perfect tool for teaching, the Berkeley folks couldn't go out and start designing new chips based on Intel's intellectual property without getting in some serious trouble.

So Asanovic and a pair of grad students started designing their own super basic instruction set architecture. They were trying to create something that was simple, efficient, and extensible, meaning they could add to it later and their work became the basis for RISC-V.

To clarify something, RISC-V is the instruction set or ISA. It's not actually a specific chip. Some commercial ISAs can have hundreds or even thousands of instructions in them, but the 47 instructions at the core of RISC-V are still enough to build a chip that can run some basic code.

Through the next few years, this is what Asanovic and his students did, designing some basic chips based on the RISC-V ISA and crucially publishing their work along the way. Now, this may be typical for academia, but a lot of chip companies are actually pretty secretive and their work started to attract attention. Asanovic told me they really were looking at RISC-V as a purely academic project, maybe it would get picked up by some other schools to be used as a teaching tool, but by 2014, companies were starting to come to their conferences with a lot of questions about the design.

Today, dozens of companies have partnered with RISC-V International, the nonprofit that now manages the ISA, and thousands more have registered interest in using the architecture. Some of the big names include Google, Oculus, Sony, Qualcomm, Western Digital. RISC-V chips look poised to start showing up everywhere and may change the way companies build hardware.

So what makes RISC-V so special? You may have seen other videos or articles breathlessly proclaiming the amazing performance that RISC-V will enable, but well, there's nothing really all that special about these chips. According to Calista Redmond, the CEO of RISC-V International, there's nothing really in the design that we didn't already know about when folks were designing chips back in the '80s. What makes RISC-V unique is that it's an open standard.

Both the instruction set and some basic chip designs are available online, totally for free for anyone who wants to use them. Here's why this matters. When we think about processors, we tend to think about a chip like the Intel CPU in your laptop, or the Qualcomm SOC in your smartphone. But there are processors in everything. And I'm not just saying like, oh, your smart fridge has a chip in it.

- For this new refrigerator has an electric door.

CHRIS SCHODT: A device like your phone, or yes, your smart fridge has dozens of tiny processors in it. There are chips managing power delivery or battery life. There are storage controllers helping move data around. There are processors for sensors.

There are signal processors dealing with sound, or images from a camera, even sometimes tiny microcontrollers within the SOC or CPU itself. And all of these embedded processors are running instruction sets from different companies, and they are all proprietary intellectual property. According to Patterson, it's not uncommon for one complex chip to incorporate five or six different instruction sets.

And all of the parts in a complex device, like a laptop or a tablet, will probably require a dozen or more. These tiny embedded processors and controllers come from companies, like Texas Instruments, and Renesas, and Synopsys, companies you may not be super familiar with, but they chip billions of processors every year.

These companies have huge catalogs of chips for sale. But if you need something that they don't offer, well, if you're the size of Apple, you may be able to get them to do a custom design for you. But otherwise, you're kind of out of luck. And it doesn't get any easier if you just want to build your own chips.

Most companies do not share their instruction set designs. You can't just go and build your own x86 chip. Again, Intel or AMD will sue you. Some like ARM do license out their ISAs and some basic chip designs, but expect to pay hundreds of thousands of dollars to use them on top of royalties, plus those licenses usually have to be renewed every time the company updates or changes its ISA, which they do frequently.

Even if you are a new startup flushed with funding, where millions of dollars in either license fees or chip design is no problem to you, expect the legal end of a licensing deal to take 6 to 24 months as the lawyers hammer out all the details, which is an eternity in tech. The result of all this is relatively few companies actually have access to the resources that would be needed to design a custom chip to their exact specifications.

After all, you're not going to spend years of work and millions of dollars on a whole new ISA and architecture, just to put a custom chip in your smart pet feeder. Most companies make do with off-the-shelf parts that are pretty close to what they need and that work all right and leave it at that. What RISC-V provides is a way for companies to start designing their own chips without all this hassle.

If an engineer has an idea for a new chip, they can start playing around with a RISC-V design today, no lawyers required. Ted Marena runs RISC-V development at Western Digital, one of the biggest storage and hard drive manufacturers in the world, and he told me they are planning to transition all of their chips over to a RISC-V design they've developed in-house.

Marena told me their engineers realized that some of their products would get better performance if their storage controller could handle two threads at once, essentially issuing two instructions at the same time. However, there was no dual threaded embedded controller anywhere on the market they could buy, so some of their products just ended up having two storage controllers in them, which adds to cost and power consumption.

With RISC-V, the company designed a chip that does exactly what they need. Now, this wasn't easy. Creating a new chip meant Western Digital had to move a lot of design work in-house, and this requires time, and money, and engineers, and resources.

But for companies that don't have access to a stable of processor designers, some companies are taking the open source route and publishing their designs online. Heck, you can go look up Western Digital's new storage controller. It's called the SweRV EH1 and they put it on GitHub.

RISC-V is also designed to be flexible. Those 47 base instructions don't actually get you very far, but here's where that extensible idea comes in. The RISC-V foundation has continued to develop add-on instructions that enable things, like memory management or making the chips Linux Compatible or more complex instructions, like vector math for high performance computing. More complex instructions require more complex chips, but engineers can pick and choose which extensions they need or even design their own new ones.

This means you don't need to devote parts of the chip to instructions that you don't actually need, which can lead to smaller, more efficient, or cheaper processors. This flexibility has made RISC-V appealing to a lot of different groups, and it's got advantages for security as well. The more complex a chip, the more potential ways there are for hackers to try and exploit it. And by letting engineers only pick the extensions they need, RISC-V may end up being more secure.

According to Helena Handschuh, a security fellow at Rambus, it's also just newer and doesn't come with the baggage of the past. By contrast, Intel's x86 still supports instructions going all the way back to 1978. For now at least, RISC-V also seems immune to the side-channel attacks, like Spectre and Meltdown, that caused such headaches for Intel and AMD. And the RISC-V foundation is working on extensions for enhanced security, like cryptography and trusted execution environments.

Lastly, by being open source, anyone can dig around in RISC-V's design and look for bugs. Software companies have understood the value of this approach for years, but it's much less common in hardware. RISC-V's biggest use is almost certainly going to be in low-powered embedded chips in smart devices and IoT gadgets. But some groups, like the Barcelona Supercomputing Center and the European Processor Initiative, are looking at building RISC-V accelerators for uses in supercomputers.

Alibaba is working on a 16-core RISC-V chip for servers and data centers. And India's Shakti initiative is funding six different risk core designs with multiple universities collaborating on the new processors. I personally still think it'll be years before you see RISC-V in a high performance device like a laptop or smartphone. But you can already find RISC-V designs in Huami's Amazfit smartwatches, and SiFive, a company that builds custom RISC-V chips, has raised $300 million and is planning on debuting a RISC-V based PC dev kit this month.

RISC-V has been of particular interest for folks outside of the US. Up until this point, a lot of processor IP has been held by a few US companies, who license or sell their designs to the rest of the world. This has worked OK, and it has made a lot of money for US companies.

But in recent years, a lot of other countries are looking around and starting to think, maybe, their trade deals with the US are not as secure as they thought they were. If you live outside the US, it's starting to look like any new president or trade war could shut you off from access to a lot of critical chip technology. So everyone from the EU to Pakistan are looking at ways to develop and design their own homegrown chips.

So what can go wrong here? Well, if these chips end up taking longer than planned to get to market, or they underperform, or for a host of other possible reasons, support for the RISC-V foundation can dry up, and this whole project could just wither away. There have been other attempts at open source chips before including from big companies, like Oracle and IBM, and none of those designs have really caught on.

So RISC-V might just suffer the same fate. Though to be fair, most of those designs were not quite as open as RISC-V until recently. It's also possible that an exploit like Spectre or Meltdown gets found for RISC-V.

In fact, with enough time, it's probably inevitable. Now, if a big hack gets blamed on the RISC-V architecture, that could push engineers away from working with the design in the future. RISC-V International does have security researchers working on this, and they have a plan in place for distributing patches to partners.

But with the decentralized nature of RISC-V, it would be up to manufacturers to try to implement the fixes on their own, kind of the situation Google is in with Android. Even with an open instruction set, it's also just really hard to design chips. The microarchitecture, the physical implementation, and the verification, which is the process of making sure the chip not only works, but doesn't have huge security flaws-- all of this can take teams years of work, hundreds of millions of dollars, and an incredible amount of processing power.

It also requires very specialized tools and software. And while folks are working on some open source solutions, it's possible that companies will develop these tools on their own, but then make them proprietary and not share them. Or they might do the same thing with their designs and their instruction sets, and choose not to publish them and keep all the information for themselves.

If this happens, the end result could be pretty much the situation we have now, where a few companies have the resources to actually manufacture and design new chips, and everyone else just has to buy off the menu and take what's available. Lastly, as an engineer friend of mine said, academic projects are cool, but I just don't trust the big tech companies to let people go and do things for free. A lot of the big players in this space are currently supporting RISC-V.

But if they change their minds or maybe it starts to enrich their competitors, they could easily find ways to slow this whole process down from Penn lawsuits to pressuring suppliers. 2020 was the year that RISC-V at least started to feel real. And it's got the potential to make custom hardware accessible to a lot more companies, from the big tech firms all the way down to small startups, who would have just had to content themselves with off-the-shelf parts.

It might change business more than it changes what's powering your phone, but RISC-V really does have the potential to shake up the world of computing. Just don't expect your RISC-V PC to arrive anytime soon.

Thank you, folks, for watching. This was actually an audience suggestion topic from the comments in our Apple and ARM video. And again if you haven't watched that one yet, I would go check it out. If there are any other topics you want us to dive into, let us know in the comments below and we'll catch you next time.

[MUSIC PLAYING]