Geof Wheelwright: Welcome back to the Arm Viewpoints podcast. I’m really excited for today’s episode, which we’re calling Raspberry Pi Empowering Millions to Innovate. I have to confess that I’m a long-time fan of the Raspberry Pi. So, I’m delighted that our guest today is Eben Upton, founder of the Raspberry Pi Foundation. He’s also a former Distinguished Engineer with fabulous semiconductor manufacturer Broadcom Inc. and founder and former CTO of mobile games middleware developer idea works 3d limited, he holds a BA in physics and engineering, a PhD in computer science, and an MBA from the University of Cambridge. Welcome, Eben.
Eben Upton: Great to be here.
Geof Wheelwright: So for those who don’t know, the Raspberry Pi is a small form factor computer that’s used for both educational purposes to learn programming and also larger professional projects. And since the company was founded 10 years ago, it’s grown from strength to strength. So maybe we can start by giving listeners a bit of background about Raspberry Pi and the company’s journey over the last decade?
Eben Upton: We had a problem I guess, in Cambridge 15 or 20 years ago, which was that nobody wanted to come here anymore to study computer science. This is it is one of the best universities in the world to study this stuff. And obviously University of Alan Turing, it’s the University of Maurice Wilks, briefly the University of Charles Babbage. I’m in the Maurice Wilks building now, and Maurice who I very fortunately got to meet when I was an undergraduate at Cambridge, ran a programme called EDSAC, which generated the first real proper scale programmable computer in the UK and one of the very first in the world back in the 1940s. So this is a great place to study computer science. And we found ourselves unable to persuade very many people, we were down to about 200. After about 2008 we were down to about 200 applicants down from an initial peak of 600 10 years before, to only 200 young people a year expressing interest in coming here to study computer science. We kind of asked ourselves as a group what had happened and the idea we came up with was, that back in the 1980s we had programmable computers. So young people like me had access to machines like the BBC Micro that we could load a program on and then we it was natural for us to go to university and study computer science. Quite simply Raspberry Pi’s is a test of the hypothesis that if we bring back the machines, if we bring out the cool tech, then maybe people become interested in computer science. And that’s really all we were trying to do solve this very simple little problem for the University here in Cambridge.
Geof Wheelwright: As you’ve been solving that problem, and in the last 10 years as you’ve developed the Raspberry Pi through a lot of different generations. There’s so many exciting and creative projects that leverage the Raspberry Pi. In fact, I recently read an article about a guitar pedal that used machine learning for effects, which was fascinating. So which ones have you found the most interesting over the years?
Eben Upton: I’ve always found the space related or space adjacent, Raspberry Pi applications a lot of fun. Big culture of high altitude balloon with Raspberry Pi, putting Raspberry Pi’s under a weather balloon get up to about 40 kilometres and taking pictures and you can see the curve of the earth blackness of space. It seems fun as a geek, but it’s also fun from an educational perspective because it really dramatizes to kids exactly how close spaces. So I really enjoyed that one, we have our own space related programme as well, called Astro Pi, where we have a couple of first generation Raspberry Pi’s, so fairly antique hardware, now on the International Space Station, about 15,000 groups of schoolchildren now have run their code in space on a Raspberry Pi, which is kind of fun.
Geof Wheelwright: Well, you’ve clearly harnessed the desire for a lot of people to make exciting projects, and it’s created an enormous developer and maker community working with Raspberry Pi. So maybe you can tell us a bit more about your involvement with them, and why you think Raspberry Pi has been so successful and popular?
Eben Upton: I think Raspberry Pi was drawn to what we call the category defining product. It is a product that you couldn’t have focus grouped into existence, hard to ask somebody, do you want this thing before it exists? But it turns out there was latent demand. So just as with tablets, people actually turn out we did one tablets, but the only way to find that out was to go make them and see if discovered they were incredibly popular. And people had a latent desire. Lots of people in the hobbyist community have a latent desire for something which occupies the role of glue between the physical world and the computing world and the networked world. And then we, you know, we put it in the market, and it just tapped into this latent demand. Unfortunately, of course, we were able to, because the way our business model works, of course, very similar business model to Arm where we’re less primarily a licencing company. Because of the way our business model worked, we were able to scale the business very quickly into this demand, you know, we went from nothing to 100,000 orders on the first day to a million units shipped in the first year, it was very hard to take off, and we were able to kind of keep up with a nurture that demand all the way through, and where that demand initially came from the hobbyist sector. Over time, what we’ve seen is those hobbyists saw the people who have taken Raspberry Pi into their workplace. You know, a lot of our industrial designs come from individual hobbyist engineers, bringing the things they love into the workplace. They’re also the people who took it into education. So I’ll kind of attack on our interface to education, the educational world has been a two step process of engaging enthusiastic teachers, enthusiastic volunteers, and then those teachers and those volunteers then going and engaging the children.
Geof Wheelwright: So that kind of leads me to my next question, and it’s around the Arm Cortex-M0+ processors, and your use of them. So what was the team’s experience of designing Raspberry Pi’s first silicon device in house? and what led the team to develop an MCU and to select the Cortex-M0+?
Eben Upton: Prior to Raspberry Pi Pico, which is the product that launched in January of this year, all Raspberry Pi’s have been basically the same. There’s enormous difference in performance between a Raspberry Pi 1 and the Raspberry Pi 4; About a 40 fold increase in performance but they are fundamentally the same thing. They’re fundamentally Linux computers. What they aren’t is microcontroller platforms. And of course, they come with it that that that distinction comes with lots of advantages, you know, we have lots of interfacing opportunities, USB 3 Gigabit Ethernet, dual band WiFi, enormous massive memory, you know, gigabytes of memory, gigahertz or processor, and, but there are things you can’t do. So it’s not a platform, which is well suited to very fine deterministic control of IO, in the way that microcontroller platforms are. It’s not a platform that will scale down to the power consumption of the same way that microcontroller based platforms. There was never really a way to build a device, which was recognisably a Raspberry Pi but was also a microcontroller platform and so the team, I think, as I understand it, there was a sort of a process of going out and looking at what was available in the market, seeing dissatisfaction which was what was available on the microcontroller market, and then the determination to build something in house. What that led to was, you know, a selection is a core selection, which is effectively based on efficiency. What’s good about Cortex-M zero plus, it’s a small core, it’s a power efficient core. If what you want is integer, lots of integer performance, it’s a great core to great core for that. It has a feature called SIO, it has a single cycle IO bus that lets you connect to an supervise external piece of hardware piece of hardware elsewhere. And the SoC lets you control those with very low latency. So it’s kind of a good choice for for a chip that was focusing on flexible IO. Flexible IO really is one of the one of the key selling points of this device. Yeah, I think it’s worked out pretty well.
Geof Wheelwright: Yeah, it looks like it has, were there any other key architecture decisions behind this choice that you wanted to highlight?
Eben Upton: Let’s have a think about what we think is good about this chip. It has a substantial amount of additional performance. Because it’s implemented on a modern processors node, it’s been possible to take those Cortex-M0+ cores and implement them at a relatively high frequency. So they come out of the box running at 133 megahertz. Chips got an enormous amount of overclocking headroom actually over that point, but even at 133 megahertz, with two cores, you’re getting a lot of integer throughput. It has a little memory on there. So yeah, certainly for a Cortex-M0+ basis, it has over a quarter a megabyte of single cycle SRAM on the chip. That’s extremely useful in terms of, I guess, programmer productivity, you can add obviously, you can do more with a chip that has more memory on but why do I find more memory exciting, I find it exciting because it means programmers can focus on implementing features, rather than trying to squeeze their code down into eight or 16 or 32k of RAM. I guess the last thing is IO. So it has this it has the usual collection of low speed serial interfaces you URTs, SPIs I2Cs. But it also has a thing called PIO, which is a programmable IO subsystem, which really automates the processes often called bit banging, the process of taking a novel, a novel wire protocol that doesn’t have native support in the chip where there’s no native control of that protocol. You know, the example people often give is neopixels, these little LED bulbs that you can put on a ribbon that have a little, a little piece of logic inside them that makes them addressable. Very few microcontrollers have a peripheral that can control that. But PIO lets you build a very efficient software based peripheral that can control those LEDs. What we found with Pio is it’s been incredibly flexible, you can use the same piece of the same piece of silicon, to drive a DPI display, a DVI display, read an SD card drive some of these LED driver IDs, audio codec, all of these things can be bit banged with PIO. So it’s that combination of high energy performance, lots of memory, and very, very flexible IO. I can see that going in a whole bunch of different directions in terms of what people could do with it.
Geof Wheelwright: But what sort of projects do you see this being used for in compared to other Raspberry Pi products?
Eben Upton: I think it will be used, I think it is being used in combination with the big Raspberry Pi products and a lot of applications. Because you often need both a lot of computing capability and ability to access the network and high speed peripherals along with deterministic control of the real world. So I think in a lot of cases, it’s part of a pairing of the two tiers of details and product. On its own, really, I think one of the big places that shines is power consumption. It isn’t the lowest power microcontroller in the world but compared to a big Raspberry Pi platform it is extremely low power consumption. So for battery powered operations where you have days of operation out of the battery, it certainly shines in that area.
Geof Wheelwright: So will Raspberry Pi designing more silicon chips in the future, and if so what areas do you see it addressing?
Eben Upton: Obviously, there is silicon design capability and Raspberry Pi, and it would be ashamed not to use it. Right now though I think we are very focused on this one design. You know, there is there is a lot of work to do and a lot of work for the team to do in bringing this design to scale. You know, we have it’s been in the market for four or five months and it’s sold hundreds of thousands of units, we have the of the Pico product, so we’re kind of unusual in that the organisation is both making silicon and they’re making a product which uses that silicon. So there’s a lot of effort, actually at the moment on production engineering, and then the same people who do the production engineering. So you know, obviously, I’m sure we’ll come back to this in a few years time, I would think you know, maybe different core choices, different IO choices, conceivably, in a different packaging choice. I mean, there’s all the usual dimensions, and now that you can look at the world of microcontrollers, and you can say, well, we’re in one corner of that world, you know which bits of it, could we go and address with some subsequent product. For now, I think we probably got to, we frequently find ourselves maybe on a three year cadence where you probably have a year of learning, a year of thinking about what to do next, a year sort of thinking about what changes you’ll make in response to those lessons, and then a year of implementation and the launch. So I think we’re probably about three years away, we’re a year away from having any idea of what we want to do. We’re probably three years or two and a half to three years away from actually doing it.
Geof Wheelwright: Whether that’s that’s a great sketch of kind of the medium to long term. I wanted to switch now and talk a bit about kind of how you get there and what the journey is like, particularly in working with Arm. I know that you join the Arm’s Flexible Access Program to help design your own custom systems on a chip. Maybe you can tell us a bit about that experience some of the challenges, what you see the benefits being and how it’s all worked?
Eben Upton: Well, one of the one of the problems with traditional IP procurement for silicon, is you have to know what you want before you get it. And that’s fine a lot of the time you know, if you if you want a PCI Express controller, you know, you want a PCI Express controller, you can go to the companies that they PCI Express controllers and say give me a PCI Express controller, give me a quote for a PCI Express controller. And you can get a bunch of quotes and then you can procure some IP. The challenge in a space certainly like processes is there’s quite a broad spectrum of IP that you could choose even within one vendor like Arm. And it can be a little bit hard to know which thing you want until you’ve got it. And that’s pretty inefficient. And, I guess promotes, promotes all sorts of suboptimal behaviours. What is very attractive about Arm Flexible Access was that it gives you both access to a very broad menu of IP, a really surprisingly broad menu of IP, I mean, there’s, there’s an awful lot of stuff in the on the menu and in Arm Flexible Access. And it’s a very broad menu that gives you predictable pricing. So you know what something is going to cost before you engage with a particular piece of IP, and you know what it would cost if you decided to take IP all the way through to production. So those two are very, those two are actually very attractive things and lets you play around with the experiment. That was very helpful. The contractual complexity is obviously very simple as well, there isn’t really a contractual negotiation process is just the Arm Flexible Access licence and it’s a very easy to sign piece of paper. And really, it does take a lot of the friction out. I think that’s the thing, all of these things I’m describing are friction, you know, either a real friction, or they’re kind of a fear of missing, there’s kind of a fear of missing out, right? There’s a kind of this this fear of missing out, which which imposes a psychological friction on you. And so having access to all but the very highest end elements of the Arm IP offering has been very helpful.
Geof Wheelwright: So, I was going to ask you, if you could pick one thing, that the Arm Flexible Access initiative has enabled Raspberry Pi to do? What would it be? And I mean, it sounds like you that flexibility is part of it. Do you have other aspects that you don’t want to highlight?
Eben Upton: I would say pick the right core. And that’s, that’s a that’s an instance of flexibility. It lets you pick the right core. And I think we did, I think the core choice in RP2040 is the right core and it’s easy to make that right choice in the context of flexible access.
Geof Wheelwright: Maybe we can talk more broadly about the ecosystem and what synergies you see between Arm and Raspberry Pi for supporting and growing that ecosystem?
Eben Upton: I think Arm and Raspberry Pi have very similar goals in a lot of ways. We both have an interest in the engineers of the future. You know, Arm has run a very impressive education programme for a very long time and And so we kind of both have an interest, probably complementary interest, Arm’s educational efforts are often focused at the tertiary level, they focused at the university level, where we’re really focused as an organisation. So there’s the charity of getting people to the tertiary level, where I would focus on getting people through school, and nurturing some enthusiasm for computing during their lives. So that’s the thing. It’s also the kind of commercial side of me, I grew up with a BBC Micro and I grew up thinking about, I grew up thinking about wanting of Acorn Archimedes, which of course, was the very first machine that had Arm in it. And then I think we went through a period where a PC was synonymous with another instruction set architecture. And that was very sad for me. And so there is certainly there’s a great deal of satisfaction for me, in being one of the world’s largest, is one of the world’s largest PC companies. But we’re certainly one of the world’s largest Arm-based PC companies and sort of bringing on bringing the ARM architecture has been enormously successful in embedded and mobile applications, and is increasingly successful in the PC space, being part of bringing this architecture back, really, because that’s where it started bringing this architecture back into the PC space, certainly, personally very satisfying for me.
Geof Wheelwright: I could see that. And, you know, as you said, that’s kind of harken back to where we kind of started, that there was this excitement, energy and enthusiasm that you’ve helped nurture and inspire in people. And I wonder where that takes us over over the next five years, and particularly, what’s next for Raspberry Pi over the next five years? And then we talked a little bit about that earlier. But maybe you could kind of cast your mind out over a five year span. And tell me which
Eben Upton: This is an interesting question. I think, you know, you can break it down into I guess the products and the and the mission. In terms of the products still kind of in the we’re coming up actually on the second anniversary of Raspberry Pi 4, and Raspberry Pi 4 was a transformational product for us because it’s the first piece really genuinely PC-class Raspberry Pi. It’s also interestingly a product that came along in a very relevant time for the pandemic. We’ve done an enormous amount of work over the last year in deploying Raspberry Pi based PCs to young people primarily in the UK, but also some of the US who have who have needed hardware in order to be able to work from home, but we’re still in my mind early in the Life of Pi 4, I think it’s because it was such a revolutionary uplift in performance for us, it’s still a platform that has a lot of legs, it has a lot of optimization yet to do here, we can eke more performance out of it in, in software, we may be able to eke more performance out of it and other ways as well. So you probably get to see for most of that five years actually us being very focused on Raspberry Pi 4, but probably more on the software side than the hardware side. That’s the that’s the platform as the platform in terms of the mission, I think we’ve been very successful at reversing this decline that I talked about, in the number of people interested in computing, something at Cambridge, we went, as I said, from 600, in 1999, to 200, this is number of applicants for about 100, places, 600 to 200 to 1400 last year. So computer science has gone from being one of the easiest courses to get into at Cambridge to almost the hardest course. And that uplift in interest uplift in interest is not entirely due to Raspberry Pi, but we’ve been an important participant in the ecosystem that has driven that increase in interest. So I think we’ve and that’s been mirrored across the rest of the higher education system in the UK. I think the opportunity really, in terms of mission is to recapitulate that effort in the in the context of other developed world countries. And in the context of other developing countries as well, obviously, computing is both at the individual level, computing, education is at the individual level, a really powerful driver of economic mobility, teaching young people to do STEM subjects, STEM subjects are hard, you know, we talked about talking about this thing where you say, you know, success in STEM subjects is in some to some extent, it’s a very objective thing, right? You know, your computer programme doesn’t care who your dad is, you know, you can’t, you know, if dad gets you a job as a computer programmer, and you suck at computer programming, you’re not going to be out in the street, right? So he really cares about what you’re good at, rather than who you know. So at an individual level, enormously, it’s a fantasy. I know so many people who have come from very challenging backgrounds, and have had very successful life on the basis of having had some exposure to STEM education, where they were young. So on an individual level, it’s very powerful, on a societal level is incredibly powerful as well. Because you know, a society that trains a lot of people to do interesting STEM stuff as a society is going to be able to compete in the modern era. It’s obviously great for the UK that we’ve been doing that here. But obviously, we want to recapitulate that ever else. And of course, it can be a powerful driver for lifting some low middle income countries really up into the high income bracket. So very excited about that just hired our first employee and in our first employee in Africa, based out of Legos, which is, which is super exciting for us. So I think there are there are some wonderful opportunities to do good work there, on the commercial, mechanical side, and also on the charitable side.
Geof Wheelwright: Yeah, it seems that there’s a real democratising aspect to what you’ve done with Raspberry Pi. I mean, just in terms of the enthusiasm and the breadth that it generates ,and also the fact that it is, you know, very, very cost effective. And there are a few platforms that you could look to, that will provide the same level of capability at the kind of price point that that you’re able to achieve. Can you just talk a little bit about that democratisation and what you see the impact of that being?
Eben Upton: It’s interesting, if you go back to there’s a thing called the whole earth catalogue, you may have come across.
Geof Wheelwright: I remember indeed
Eben Upton: and it’s that was all about access to tools, right? So this is this is from the late 1960s or early 70s. In the United States, big catalogue of all the stuff you need, if you want to go a lot of credit would be emphasis on self sufficiency. And if you want to go and live in a, in a Buckminster Fuller dome out in the wilderness in California somewhere and you want to be able to purify your water you get from the stream, where do you actually in the pre-internet world? Where do you find out how to build your Buckminster Fuller home? And where do you find out how to purify water? So this idea of access to tools has always been at the heart of what we do at Raspberry Pi. You know, there are people it’s back to, you know, why is Raspberry Pi successful? Because there were people who wanted tools and we made it we made tools. So it’s always been about providing people with missing tools in their lives. And of course a big part of that’s always going to be about cost. We think about being kind of the new BBC Micro, but in a lot of ways, we owe a lot of debt to the Sinclair machines, you know, there’s an exciting one on the Sinclair Spectrum, because while the BBC Micro was a platform that proceeded from a feature set to a price, the simpler machines were more machines that proceeded from a price to a feature set. Now, people ask me how do you decide what to put into a Raspberry Pi? And the answer is really whatever you can fit in to $35 in any given year. And over time, the operation of Moore’s Law outside our organisation, and the kind of accumulation of operational experience and efficiency inside the organisation has driven this enormous increase in what you can do for $35. But then always still proceeds from $35 in the way Sinclair did.
Geof Wheelwright: You know, that that does remind me of Sir Clive and, and his kind of designing to a price point, and then having the ecosystem kind of fill in the gaps. I’m inspired by the work you’re doing and your insights on working with Arm. Thanks so so much for bringing them to us. Don’t forget that Arm DevSummit 2021 is coming to your screens October 19-21. The three-day virtual conference where hardware and software join forces will serve up insights into the latest technology trends and provide engineers, developers and tech enthusiasts with technical sessions, hands-on workshops and networking opportunities with like-minded software developers and hardware designers. Sign up to stay updated by visiting http://devsummit.arm.com/