Can Open-Source Semiconductors Upend the Chip Industry?

a bonus episode to our semiconductor series. It's going to be a little bit different to the other ones because we're going to be talking about a new technology, a new sort of how do I explain this? Well, I guess it's an open source hardware technology that could have a big impact on the chips industry and the way things have traditionally been done. Right. So a lot of our discussion has sort of talked about a few like

basic business models. I mean, we've talked about Intel and the sort of integrated UM chip design and fab model boundary versus fab and then the fabless semiconductor companies which have given rise to the contract manufacturer Tawan Semi, which is sort of the ultimate fab. So we've sort of like, um got this sort of like broad contours of the industry, but in terms of like interesting new technologies or different directions or totally different models of where how things could go,

still more to be discovered. Yeah, that's exactly right. So, uh, someone told us that we had to talk to this particular company because they were one of the leaders in trying to commercialize this open source hardware idea. It's called Risk five. It's a let me see if I can pull it up, it's R I S C and then

five Risk five. So what we're gonna do is actually learn about what the technology is and then dive in a little bit to what it might mean for existing chip makers and how the whole ecosystem of semiconductors actually works. I can't wait. I don't know anything about any of this stuff, so I just want to learn more. Yeah, okay, we're all learning together. So without further ado, let's bring on Chris Latner. He's the president of product and engineering

over at SCI five. He's also a long time tech guy and quite famous in the industry for reasons that we will get in. So Chris, welcome to our thoughts. Hey, thank you for having me. I'm excited to be here. Could you could you maybe give us a snapshot the elevator pitch for risk five before we begin. Sure sounds great.

Let's start with what what is risk five? So when you start talking about these companies, companies like Intel or Arm or m D things like this, they generally make processors, and so the processor in your computer it's jobs to run software. And there's this complicated dance that happens between the software and the hardware and the thing that mediates that. The thing that lives in between these two worlds is a thing called an instruction set. That instruction set is

often proprietor and so um. With Intel, for example, they'll use an instruction set called x A D six. ARM has a set of instruction sets that they named after themselves. They have the ARM Instruction Set UM. There are many, many of these things that have existed over the years. A company called MIPS remember, a company may remember ages ago called Sun had an instruction set called Spark, and many of these things existed in their designed for different

different configurations. Now these have all worked and these have been a good thing, and obviously we have a lot of successful computers today. Um, but there's a problem. The problem is that the software and the hardware lasts longer than the companies do sometimes and so MIPS for examples, in bankruptcy these days, Sun kind of went away and

Spark is in a different world. And so one of the challenges that you end up having as a device manufacturer as a software provider is that you need something that will survive. And you know, companies in general go through changes and um, you know, as an industry, it's always evolving. It's a fast paced world. Now there's another challenge with this, which is also when you have a company that's tied to a specific construction set that they create,

you only have a single provider. This can sometimes be limiting if you're building a serious advices around one company's technology, both if they go away, but also you don't have as much market pressure pushing them to innovate and pushing them to you know, have reasonable charges and rates and things like this. So RISK five is a different approach

to this. Instead of saying it's a proprietary interface between hardware and software, Risk five is an open standard and so it's you can think of it as an open source kind of instruction set which allows many different implementations of hardware to work with all the software. And so in this in this in this vein you can see to something like t c P i P, which is an open standard for networking, or you could see this LINUX, which is an open standard for for software UM in

the unex world and things like this. And so that's really what RISK five is is it's an old idea about this interface between hardware and software. But the new thing about it is that it's uh an open standard and it also has a fresh new design that really learns from what has happened in the industry and it's just technically better in most ways. So just to be clear, your company SCI five, is it manufacturing chips or is it just creating this software layer from which others could

theoretically then build ships. Great question. So, if if I look backwards, Risk five started started at the University of Berkeley, and so it's about ten years old. It started in a research lab, and you know, initially they were just building an instruction set to support academic research, support some

of the projects that they're interested in. And that that became a movement where a lot of people agreed that they needed a standardized system and that allowed people to do hardware research without having to do all the software work. Really was it was enabling new kinds of innovation coming out of the academic world. The founders of Risk five all came from that research lab in Berkeley and five years later decided to start a new company, and that

company is called SCI five. So that's where that's where we We now are leading the Risk five revolution with the founders of Risk five, and we're building a new company, but we're also building and defining and driving much of the technology the powers this open cross ecosystem standard that Risk five is, and so SCI five as a company,

we build Risk five implementations. And so you can think of as an analogy um T c P i P is an open standard for networking, but Cisco builds routers, right, and so we yeah, we're we're building implementations of Risk five and so uh we are the best place to go to if you want to build a product around

Risk five technology. And we have a wide range of different implementations with different performance points, size area, power trade offs, things like this could you maybe give us a little bit of your background, So I teased it in the intro, But you are very well known within the tech and computer world, most notably well you might debate this, but for developing the Swift programming language, which I think is the thing underpinning iOS on Mac and Apple and things

like that. How did you get involved in this space and what attracted you to sci five and the risk five idea. Yeah, so, so I've been kicking around the tech industry for a number of years now. Amazing how time flies. I spent eleven plus years of Apple, for example, but did a lot of work on fundamental low level software called compilers. And compilers are the thing that takes the application that a programmer rights and gets it to work with the hardware and so that instruction set that

that that interface between hardware and software. Compilers are just on the software side of that and they handle that that making it work on the processor for you and so um. As part of that, I build a number of technologies that are all open source, including this thing called l VM and playing and a bunch of other compiler technologies that underpinn and enable a lot of that technology. Right at the hardware software boundary. There are many things

that happened in Apple. We build a lot of really great things. The Swift programming language is one of them, and so I've done a lot of work and that that space. Since then, I've worked a number of other places, including Tesla and Google, workd on machine learning, infrastructure, large scale TPUs, whole bunch of things, and it's all been

mostly this systems software working with the hardware. So when I was considering enjoining a SCI five, the really exciting thing to me about SCI five is is a number of things. Is this focus on open technologies, which you kind of have an unfair advantage when it comes to changing the world right because people love open technologies and a lot love the reach that you can have with them. Um, it's just the the idea of instead of being on the software side of the hardware software divide, be able

to straddle that. And so SCI five is both a software company and a hardware company, and so we're able to innovate by spanning that that gap and really doing things that are quite next generation because we bring this combination of knowing how the transistors work and your silicon, but also how how to get programmers to use it. And that's something that UM typically UH companies have really bifurcated on. You have software companies, you have hardware companies.

It's really rare to be able to span that gap. So give us an example of a theoretical customer, or if you want an actual customer who comes to Risk five, what are they looking for? What can they Why do they go to you as opposed to someone else. What can they get from you in terms of performance, in terms of a product that makes sense for them that from a business standpoint, UM can't get from somewhere else. Absolutely, So there's many different categories. This one example is a

replacement customer. So somebody is using a product from, for example, Myths and they're saying, hey, wow, we're building our technology on a technology staff built by a company in bankruptcy.

Maybe that's not a good idea, right. We also have many customers coming to us from arm designs, and so the arm arm company has a number of CPU designs in their portfolio, and for a variety of reasons, customers are not as happy with that as they used to be, and so a replacement of existing products typically they're looking for something that is um you know, solves the same kind of a job as what they were using before, but is better in terms of UM the area, the

size of the hardware, or the power consumes UM and things like this, and so Risk five as a technology is better than many of these things out there. Is better designed and engineered, mostly because it's much newer. And so the technology industry has invented a lot of things, and as with any any space, there are mistakes that are made along the way, and Risk five is a fresh start that allows us to get those things right and be able to draw the best ideas from across

the industry into one one system. I think that the other more interesting and exciting part of what sci Fi's business is propelled by is differentiated solutions. And so while many people come to us saying like, hey, I want to swap this thing out, that the most fun pieces are that we're able to enable products that nobody else can build. And that happens because we build our processors

and our other technologies as kind of as software. And so we build hardware with the spirit of software that's very configurable, dynamic, and you can tune it for the use case. And this allows you to say, hey, well, I want to build, you know, an AI powered thing that does voice recognition and this stuff, and so we can give you the world's best implementation of that that's really tuned to the use case. And here we are.

One of the macro phenomena that's going on today is the thing that many people would call the end of Moore's law. And so the process technology, the fabs are not not not moving at the speed that they used to, and the economics are really shifting. And what this means is this means that there's an increased desire and increased

need for custom solutions. And custom solutions can be much more power efficient, they can be more cost efficient, they can be just better for the you know, being part of your life as a human and they fit in as IoT light bulbs are like all these other things. And so that's really driving customization way that we haven't seen before. And this is really what SCI five strength is. So I want to go back to that replacement idea because a big theme running through this entire semiconductors series

is the politicization of chips. And of course we've seen the US put restrictions on China and things like that. I think one of your customers was reported to be Ali Baba, So I mean, is this a way for companies too, I guess accelerate a reduction in their dependency on things like TSMC chips or maybe Western chips from Intel. Also, I think that's a complicated topic and I'm certainly not an expert on that. I think there's many factors that

are going on here. Risk five is an open standard, allows collaboration, and that collaboration happens across political boundaries, and so the US and sci FI, for example, is investing a lot in Risk five. China is also investing a tremendous amount into Risk five software in particular, and they're starting to look at Risk five hardware and so um. From a technology perspective, this just kind of flats all boats like this just makes things go faster, the technology

curve go that way. In terms of the plutotas politicization of fabs and things like this, I think it really comes down to the business model of the individual companies. And from sci Fi's perspective, we're very happy to work with UH, Samsung and TSMC and multiple different fabs and UH for that. For US it's really about what our customers are looking for and what the products they're trying to build are and trying to find the best possible

answer for their their outcome. So the big FABS that everybody knows when it comes to the annual manufacturer, your their client. UM. Generally, the way it works is that we provide technology to other hardware companies and so UM let me give you an example of this. UM. We we recently launched. UH there's an FPGA company called Microchip, and they launched a product that incorporates our technology into their product. UM, they package it up, they work with

the FABS, they do all of that work. We we are what's called a fabulist semiconductor company, and so we provide technology to other companies that then build build a larger product pulled together documentation and software and things like this,

and then we enable that product to be built. UM. There are other cases where we actually are the the the device manufacturer, and so we do have developer boards, for example, and so we'll make little boards that developers can use to write software and work with risk five and that's been very popular. I mean, all of those sounds really great. UM. And obviously you're making a good

case for the technology, but what's what's adoption like? Do you find that anyone is resistant to this new idea or is everyone um sort of jumping on board with the customization pitch. Well, So I think that there's different

aspects of this. One is risk five as a industry trend, and there we see just a complete explosion in adoption from kind of the who's who of companies including Google and Video, Qualcom, Samsung, UM, Western Digital, also Ali Baba and all kinds of all kinds of companies are are working within doing things with Risk five, and so I think there's a question of when, when and if does Risk five take over everything, which I think is an entirely possible future where you know, you fast forward a

hundred years from now, it could be risk five and um and hopefully much sooner than a hundred years um. With SCIE five we're in we're in effectively all of the big semiconductors products, and we don't talk about our customers specifically, but the adoption and the drive we're seeing is quite intense. It's very exciting time right now. So do you have any competitors in the Risk five space or is it too early for that and would you expect them to emerge at some point? Yeah, So, so

there are other companies that build risk five processors. UM Risk five as an open standard really enables this, and we encourage that. It depends on what you mean by competitors. So, for example, UM risk five is why is widely adopted within the academic context, and so there's many university projects where people are building risk five course. UM. I don't see these as competition. I see these as you know, the next generation of engineers and we'd like to hire UM.

But you could say that they're playing in the same space. UM. Many companies are also building small risk five cores themselves, and I think that they're playing with the technology in many cases and they're starting to understand it. UM. There are a relatively small number of companies that are commercial productizing and commercializing risk five course the way we are, and I think that is also great. That's that's one of the great things about this standard is that you

can get different implementations from different companies. UM. In comparison, SCI five has been at it for you know, our our team has been at for ten years. UH. SCI five as a company has been at it for five years, and so we're just much further ahead and than the pack in terms of the breadth for a portfolio, the maturity of the technology, um the established customer base. And so I'm very thrilled that our customers love our products,

and that's really what I am We focus on. So one of the things that's come up a few times in some of our discussions has been, you know, you sort of hinted at it, which is the sort of death of Moore's law, you know, the sort of the the increasing difficulty of the arms race, how much it costs to improve speed size. And when we've heard from a couple of people, they say things like you asked the question like, okay, what could theoretically up end the

dominance of a company like Taiwan Semiconductor. And the answer that a couple of people have given up is, well, maybe customers don't necessarily need the very fastest chip anymore. Maybe like the sort of traditional metrics of performance that we sort of think about with chips, certain measures of speed and so forth do not become the key things

that people absolutely need. I'm trying to understand a little bit more like does risk five, does your model sort of create the avenue for a sort of I get maybe break in the arms race of chips such that the industry can go in a more direction, or in a direction that's not just about more expensive, faster chips and offer different value propositions. But I think that it's it's hard to paint the entire industry with one brush.

So I think that if you look at UM, for example, the Internet of Things, IoT space for example, there are time to market is worth a lot, right because a lot of it is finding product market fit, and so spending three to five years to build a chip is really kind of an untenable place because five years from now the industry will be changing, and you know it

doesn't it doesn't really make sense to do that. And so I think that that that is a pretty big shift and how people think about things from back in the Windows PC days, where until it would come out with a chip every every so often and it was

just the steady, say, march up and performance. On the other hand, I think you look in the server space, for example, and there there you're kind of limited by power, and so each each socket in a large scale cloud data center can dissipate roughly two hundred fifty watts, and so the question is how much how do you get

how much can you get out of that? Right? And so to me, uh, we we as an industry have solved that problem by pushing, pushing process technology and chasing from one fab node to the next, which is very expensive and as you say, it's it's not a particularly scalable approach. The other way to tackle this is to

innovate in the hardware. And so what we're seeing now is we're seeing a rise of GPUs, and we're seeing a rise of machine learning accelerators, and we're seeing a rise of other other accelerators, and to date they're primarily purpose built. So I want to accelerate video and cooding, you know. And so if I want to accelerate video and cooding, I can build a box that does video encoding, and I can build a chip for video encoding. The far extreme of that is I want a fully general

box that can do anything. And that's what you see with typical CPUs. There's a big space in the middle where you say, if you're a cloud provider, for example, I want to be able to run a wide range of different specialized workloads. And so I want to do a little bit of video encoding, I want to do a little bit of signal processing. I want to do a little bit of machine learning. I want to do a little bit of graphics. I want to do a little bit of this. And so what you can do

is you can build what are called heterogeneous computers. And these heterogeneous computers can have accelerators for different domains and assembling those together into one package is something that you know, Risk five is standard and a lot of the ecosystem that we're helping drive well, I think really changes the nature of compute. And so I think that we as an industry or shifting and it's kind of hard to see the end state from you know, we're halfway through

this transition, but it's a really exciting time. So going forward, how do we know if people are using Risk five? So my understanding is that you know, if you're using something from ARM, you have to sign like some sort of licensee agreement and we can kind of figure out who signed those. But I don't think using Risk five you necessarily have to disclose what you're doing because it's open source. How how do we know what adoption actually looks like and how how much this gets embedded in

the actual industry. Well, so it's it's hard to get precise metrics. So for example, in video has proactively said we use RISK five and our g v us, right, and so if they if they had not said that, then nobody would know. It's just an internal implementation detail and it's not exposed. And so, UM, I don't think you'll ever get a true number. But what we can see is the number of members of the open Risk

five Foundation, for example, is just exploding, right. We know that the number of design wins coming out of companies like sci Fi for example, is just going exponential. We see we see, uh, you see the metrics and the the secondary indicators that that you can tell that there's there's a lot of momentum and a lot a lot

driving this. And so while it's very difficult to know the precise number of design winds or the precise number of transistors configured for Risk five, we know it's it's incredible. So I have a weird question. But um, we've talked a little bit about this on the podcast the sort of culture around open source technology on Silicon Valley. We've spoken about it with the Camille Fournier and people like that. But is it weird too commercialize and open source technology

like this. I've seen some people like in Risk five to Linux, like does it does it feel weird to be doing that or does it feel like this sort of commercialization is what's needed to get that technology rolled out to a wider group. Yeah, So, I mean, I there are many different ways of looking at this. I've I've worked on open source and open technologies for a couple of decades. Now I can I can share some

examples of that. So at Apple, for example, we built these compiler tools and these technologies that that application developers used to build their apps, right, which is a very

meta thing. Um. And so Apple is a company that believes in producing value and keeping it proprietary, of course, but even in that context, there was incredible value in making that open and sharing sharing contributions that they were doing with the world, because they then were able to benefit from the work that everybody else in the industry was doing. And so that open standard, the open technology, the open source code that was being developed, was not

the differentiating value for their products. It was the enabling technology that they benefited from. You know, Google used the same technology in their data centers and like that was actually great for Apple because you know, all the work that Google put into it directly benefited Apple and Apple's products. So um, I think that there's a long, long balance of trying to figure out what to open, how best

to work in foster communities. With five, I think I'm seeing very similar results that I saw I've seen in other domains where open has a lot of advantages that are hard to quantify. So for example, there are you know, we're all humans, and humans like to work on things with large scale impact, and so um, open technologies arguably are it's easier to draw the best minds into working on open technologies because they end up having larger impact

on the world. The way that risk five is defined and standardized and driven in the ecosystem is by an open committee development standardization process, and it's kind of the who's who's of gurus in the industry. They are coming together to make sure that the best ideas win and you get you get a really good result, and it's kind of hard hard to match that with a company off in the corner that even if they're a massive semiconductor company, they still have only a slice of the bright,

bright minds in the industry. And so there's there's a lot of there's a lot of nuanced pieces of this. But then also you there has to be a business model, right we have sci fi or for profit company we are you know, we're we're here to build a scalable business that is an amazing an amazing feed by itself.

And so the question is how do you fund and propel and make the open technologies thrive while also having sustainable businesses that you can you know, pay your engineers to work on the open technology and that that divide is often tricky. But if you look you look to the industry, Red Hat for example, as a company that was built on Linux and they put tons of engineering at the Linux and they build a great services business

and there are now a huge part of IBM. Right. Um. So there's many, many of these companies that have found different business models. Google, for example, is apparently a profitable company, you could say, um and and they are built and

contribute a tremendous amount of open technologies. And so it's it's all about figuring out the divide and there's different divides that makes sense for different industry sectors and different technologies and different companies, but I think it's a pretty proven model by this point, and and for it's worth, I think that open software is much further ahead than open hardware, and so I think that we're really on the leading edge of that for hardware, and I think

that ten years from now it's going to be a completely different world. Well, Chris, that was really really interesting, and I think we're all going to be um watching what happens with Risk five and with SCI five for years to come. So thank you so much. We really appreciate that. Yeah, thank you for having me. Thanks Chris. That was awesome. So Joe, I'm glad we had Chris on. Definitely a little bit different to what we spoke about on some of our other semiconductor podcasts, But I think

there are two things that probably stand out. One is that idea of Risk five as a cheat for the end of Moore's law, Like if you can't eke out any more efficiencies in terms of the actual technology, then maybe you could do it through customization and the way that technology works with the rest of the system. And then I guess the second one is this idea that Risk five might actually be a way of increasing independence

away from some of the big manufacturers. Yeah. No, I thought that was super interesting and just this like this idea that's like, Okay, you have all these companies that could build the chips, you have these end customers that have specific needs, you know, you could I think, like he said, I don't know the exact word, but the idea of like mindshare, like people like working on open source products, people like being part of something that's potentially

world changing. So it seems like potentially one of these things where it's like sure, like the sort of dominant players in the industry, the dominant platforms X say, the six arm et cetera, aren't going anywhere, but you can very easily, as he describes it, imagine it just continuing to gather steam over the long term. Yeah. Um, definitely

an interesting one to watch. And I guess the issue is who comes out and announces that they're actually using it, so we get a good idea of what adoption actually looks like. All right, well but no, definitely, UM going to have it on my radar. Yeah, um, this is our gift to add thoughts listeners demanding more semiconductor content. UM, hope you enjoyed it, and I'm sure we'll be recording more semiconductor episodes eventually, but for now, we're going to

move on to some other things. Right. I think there's some other stuff going on that we should talk about in markets No way, Okay, all right, this has been another episode of the All Thoughts Podcast. I'm Tracy Alloway. You can follow me on Twitter at Tracy Alloway and I'm Joe Why Isn't All? You can follow me on Twitter at the Stalwart. You can follow our guest Chris Latner on Twitter. He's at Sea Latner Underscore l l v M. Follow our producer Laura Carlson. She's at Laura M. Carlson.

Follow the Bloomberg head of podcast, Francesco Levie at Francesco Today, and check out all of our podcasts under the handle AD Podcasts. Thanks for listening.