Just as this year started, Raspberry Pi flipped 2021 on its head with two surprising announcements: they had begun producing their own chip, called the RP2040, and they were using it to power their first entry into the world of microcontroller dev boards, the Raspberry Pi Pico. To coincide with the launch, four other major board manufacturers released their own RP2040-powered products, followed soon after by a slew of RP2040-based products from the maker community.
We wanted to hear how this new endeavor developed for the organization, so we connected with Raspberry Pi CEO Eben Upton to learn about making modules, silicon, and one last surprise product announcement to close out the year.
Make: Raspberry Pi is now nine years old. Give me a quick overview of where things are at.
Eben Upton: So in nine years we’ve sold about 42 million Raspberry Pis. We did about 7 million in the last year, so quite high sales right now.
We’re about two years into Raspberry Pi 4, which is the biggest step change. Or, I guess technically Raspberry Pi 1 to Raspberry Pi 2 was, because it went from single core to quad core. But in terms of user experience, really 3+ to 4 is the biggest jump that I’ve seen. That’s been really well received by people.
We have the Pi itself and the Compute Module 4. Then we’ve added a third form factor, which is the Pi 400, a kind of consumer PC version of Raspberry Pi 4. And those are all going pretty well.
Make: The CM4 has caught my attention more than any other module offering that I’ve seen previously.
Upton: Compute Module 4 has taken off a lot faster than previous Compute Modules. One, it has wireless, you have an option to have wireless on board so you can just leave wireless to us and we take care of it. And then the other one’s the power design. With previous Compute Modules, you had to provide them with several power rails yourself, all of which sequenced properly coming up and coming down. With this one, you just put 5 volts in and it makes its own rails. I think those two things together have made it much easier to work with.
And so we’re selling tens of thousands of CM 4s a month at a point where, with previous Compute Modules, we probably would have been doing hundreds to thousands. We’re seeing volume orders. So yeah, it’s busy.
Plus silicon of course. That’s the other thing that’s going on. Pico is continuing to do really well. We’re seeing lots and lots of RP2040 boards, third party boards built on our silicon. Before we launched, we spoke to SparkFun, Adafruit, Pimoroni, and Arduino. Then after launch, even though we’re still in a backlog on Pico, we’ve been diverting about 15% of chip production to third parties. That’s been working out well.
Make: The silicon conversation itself is huge. It’s a whole new thing.
Upton: We call it Franchise Two, because it is a new thing. Everything we’ve done, we do lots of things, but everything we’ve done up until then is Franchise One — the single board computer and all of the stuff around it, the accessories. And this is the first thing we’ve done that isn’t Franchise One.
Make: It seems like this could end up surpassing Franchise One.
Upton: I think it could. The microcontroller market is a 28 billion-unit-per-year business, and it’s an 18 billion-dollar-per-year business. Those two numbers are interesting, because they imply that your average microcontroller only costs 60 to 70 cents.
So you can really see that a microcontroller business is dominated by the ultra low end. The interesting thing about the RP2040 is it is a sub-dollar product, but it has quite a lot of processing power, a lot of memory, and a flexible approach to IO. It’s never going to be a dust microcontroller. You can buy a 10 cent microcontroller if you only need a few tens of bytes of RAM and a few hundred bytes of program storage. And it’s never going to be one of these things at the other end of the bell curve, which has got Cortex M7s, these things they call crossover microcontrollers. Things that are almost application processors. I think the Microchip ones have A5s, right? It’s never going to be those ends of the bell curve. And of course it’s not a connected product as it doesn’t have a radio on it. But there’s an awful lot of stuff in the middle of that bell curve. I think it’s a really competitive product in that space. And they exist. That’s the other thing of course for this year. By the middle of next year we will have produced 10, 20 million of these. And so people who want microcontrollers before 2023 may end up looking at it independently of whether it’s a good product, because it will exist.
Make: How have you guys managed these chip shortages?
Upton: We were lucky that we came into the year in a reasonable stock position, in finished goods and in chips. And we’ve spent this year depleting those reserves. We’ve been doing anywhere between 600,000 and 800,000 Raspberry Pis a month this year. In terms of difficulty. I think this year is probably more of a story about having more demand than we were expecting, rather than having less supply. We have about as much supply as we were expecting, but we have to eek that supply out to service probably 20 to 30% more demand. Last year we did 7 million, this year if things go well in the second half, maybe seven and a half million units. But the really galling thing is there was a 9 million unit year there for us. And that’s what hurts, that we won’t get there. Next year, we spent a lot of time early this year making sure that we had our ducks in a row, so I think 2022 shouldn’t be too bad for us.
Make: What led you to even start thinking about doing a chip?
Upton: I think that was driven by just a dissatisfaction with what was out there already. I mean there’s lots of cool chips, but maybe nothing so exciting as to motivate you to go out and make a whole new sort of product. It was great that the team here were able to think of something a little bit different. I was super pleased with what they’ve done. This is amazing. It’s just an amazing team of engineers here, and they like a challenge.
Make: You must have been working on this for a long time!
Upton: All of these things take a long time, right? That’s the big lesson of Raspberry Pi, everything takes a long time. A lot of it’s driven by the desire to make things that can be made at scale. There’s a huge difference between making things that could be made at prototype scale, or even at small manufacturer scale, 10,000-unit scale. There’s a huge difference between 10,000-unit scale and million-unit scale in terms of how much attention to detail you need to put upfront into the design, because otherwise you just drown in manufacturing issues, support issues, product quality issues. If you cut corners, you spend more time in the end.
So all these things take a long time. But it’s good. The results are worthwhile. And it’s always the case at Raspberry Pi that the things people can see that launched were always done by a subset of the team that exists. Very many of the people here are working on things that will appear in the future. Something like Raspberry Pi 4, in terms of the hardware engineering, the board-level hardware engineering, was only a very small handful of people, it was only two or three people. Quite a much larger software team, obviously. But yeah, some of the biggest runners have the smallest teams. Compute Module 4 was one person.
Make: Wow. I wouldn’t have guessed that.
Upton: That’s a guy named Dominic Plunkett, a fairly recent joiner, and it’s his first product for us. The other one is Simon Martin, who joined us quite a long time ago and worked on a number of very long duration projects that all of a sudden all came to fruition at once. The two big ones being the high-quality camera last year and then Pi 400; they were both Simon Martin products. And because of the way that the timescales work, they all end up bunched up together and suddenly every product we launch is the Simon Martin product for about six months.
Make: (Holds up board) I’ve got a Pico right here with the reel packaging that you guys set up and it’s brilliant.
Upton: I love computers on a reel. We have them in the shop. You come in, “I want two yards of computers, please.”
Make: (Holds up a different board) Now there’s a new thing. And nobody’s really seen this before, but this is pretty fancy. Tell me about it.
Upton: So that is a Raspberry Pi Zero 2 W. It’s a successor to the Zero W, which was itself the successor to the Zero. Zero was our $5 Raspberry Pi, and Zero is a fun product. I still love the Zero, more I think than anything else we’ve ever done. Because it is sort of us challenging ourselves to keep being aggressive after we had some success. What Raspberry Pi has done is it’s used Moore’s law in a different way from the way people normally use Moore’s law. People normally pick a price point and they use the declining, specific cost of computing power to fill that price point up with progressively more computing every year. What Raspberry Pi wanted was to say can we take a PC from 10 years ago and then use Moore’s law to squash that PC down to a much lower price point. And we did that. And then we launched Raspberry Pi 2 in 2015, three years later. Then lo and behold, what do we do? We picked our price, $35, and we filled it up with a bunch of computing. Because that’s what everyone always tells you to do. That’s what your customers always say they want.
Raspberry Pi 2 was a great product, but there was a little bit of a feeling after we launched it that we’d fallen into this trap of being very conventional again. And then Zero is the output of then saying can we challenge ourselves again to squeeze the same performance into a low cost structure. So your $35 Raspberry Pi 1 gets stripped down in a variety of ways and repackaged as a $5 computer. Then Zero W comes along about 18 months later. And that was very popular. And of course now it is the most popular Zero product, at $10. A lot of fun for five extra bucks, right? So 90% of Zeros that sell are Zero Ws. But still people ask for more performance. And Zero is a Raspberry Pi 1, right? So compared to Raspberry Pi 4 it’s got about a 40th of the performance of a Raspberry Pi 4 and there are things you can’t do with it.
So people ask us “can we have another Zero?” The problem that we have with the Zero is that design is very dependent on that SoC that we use, the 2835 SOC that we use, which has a PoP package. You have the SoC and then you have the DRAM stacked on top of it. And we don’t have any other PoP chips. So if you look at a Raspberry Pi 2, 3, 3+, 4, they all have external memory, which is mounted on the PCB. There is no room on that PCB. It used to be, if you want to stay single-sided or really actually I think even a double design would be quite challenging to achieve.
And so we kind of backed into a corner and so it took us a long time to figure out how to get out. And Zero 2 is the product that is enabled by RP3A0, which you’ll see is written on top of the chip. That’s not silicon innovation, that’s packaging innovation. So effectively what that is is a BCM271081 die, the same piece of silicon that’s used in Raspberry Pi 3, put inside a package with an LPDDR2 memory die to make a finished object. And then on the outside it has our logo and RP3, and that’s what it is.
So what you’re holding is a slightly downclocked 1GHz versus 1.2GHz Raspberry Pi 3 with half as much memory and a smaller set of accessories around it on the board. And it’s 15 bucks, we have actually had to grow the price in order to do that. But it’s about 10 times the performance, for a multithreaded benchmark, of a Zero.
Make: Pi 3 was where people started looking at Raspberry Pi as not just maybe an educational product, but something that could do serious work. You could turn this potentially into a desktop device that you use daily. And to go from that to this size, for 15 bucks…
Upton: Yes! It’s little, it’s a lot of fun.
We’ve always called ourselves a PC company, we’ve always aspired to be a PC company. And they’ve always been viable PCs for some subset of the things people do with PCs. Raspberry Pi 4 is the one that finally delivers for all but the most hardcore users. But it’s a PC, so you can run a web browser on it very comfortably. Raspberry Pi 3 is probably the one that delivers the same thing, but without the web browser. So pretty much anything else that you want to do, as long as it isn’t running that one, horrible, memory-hungry, processor-hungry application. Pretty much anything else you want to do, you can do all the Raspberry Pi 3. And that’s why it’s exciting to bring that level of performance to the Zero family.
Make: Absolutely. And so you’re saying, just to look at it closely, here’s the Zero 2 W, and this is the original Zero (holding up two boards). And looking at them side by side here, you can see, this is the the RP3A0, that’s this guy right here.
Upton: That’s what we call a SIP. So it’s a system-in-package. It’s multiple silicon die, stacked in, or sometimes stacked in package.
Make: And then one of the big differences that I’m seeing here, we’ve got on this one, this is the original chip here that you were saying, is it a stacked design that is actually two chips?
Upton: So what you see there is two packages. I’m not sure if you turn it on this side, you might be able to see, it’s a little bit hard to see. There are actually two, it’s a little sandwich. There are two outside different colors. So there’s a top package which has got Elpida, I think, written on it, that was effectively Micron. And then you’ve got the bottom, underneath is the bottom package, which is the process package, the BCM 2835 package. And so those, when you’re assembling that, those are two separate pick and place steps. And it’s actually a slightly specialized, POP is a slightly specialized assembly step in that.
When you have a single layer of components, what you can do is screen print solder paste onto the board. Obviously the question is how do you get solder paste onto the top of the package? You pick and place that onto the board. How do you get solder paste on top? The answer is you take the top package, you pick it up and you have a little tray with solder paste and this little round tray with solder paste, and have a very finely calibrated depth.
And then you dip it in. And a little solder paste adheres to each of the balls on the bottom of the top package, and then you put it down and there’s a little kind of arm that goes radially across this disc. And every time you do a dip, the thing rotates through 360 degrees and the little arm smooths the solder paste down to this very finely defined depth.
So there’s some subtlety there. There’s not much subtlety in the packaging design. It’s just two chips. The bottom package obviously has pants on top of it, which is a bit unusual. But then there’s some subtlety in the pick and place. What we’ve done really here is we’ve moved the subtlety into the packaging. So you’ve now have two die inside one package.
Make: Yeah. Got it. And so one of the things that I thought was interesting, again, back to the Pico, when you launched this a few months back, one of the things that really caught everyone’s attention was putting the state machines, the PIO functionality into it, something that most people really weren’t familiar with, were excited to see, especially at this size and also at a $4 price point, which is pretty amazing. (Holds up Zero 2 W) But when this showed up and I really got to see it and take a good look at the Zero 2 W first thing that popped in my head is what kind of surprises does this have baked into it?
Upton: (Laughing) The surprise is there’s no surprises, the surprise is that is BCM 271081 in that really. So that’s, I think it’s an important surprise. It sounds mundane really. But it is a Raspberry Pi. It runs all the Raspberry Pi software, the software is almost indistinguishable to the system software from a Raspberry Pi 3.
And so yeah, upsides and downsides… The upside: total software compatibility, which is something we care about enormously. Downsides, yeah. I wasn’t able to squeeze a PIO state machines into the GPi subsystem. There is no analog input, there isn’t a PIO. None of this, none of that fancy stuff, there’s no cool real time stuff that the Pico has.
I guess the other thing that you’ll notice when you look at it is there’s a can over the wireless. This is silver can over the wireless. And I think this is probably this reflects another aspect of our evolution since 2016.
So Raspberry Pi 3 was the first product that we shipped, in 2016, with wireless. And we’ve evolved from, “Hey, we put wireless on the board.” What you realize is you put wireless on the board and then if someone wants to take your product and integrate it into another product, they have to basically, with some other circuitry, they have to basically start from scratch, particularly with FCC, in North America, they have to start from scratch and completely reconform that product. Do a complete conformance run. One of the reasons why people use Wi-Fi modules, they go out and buy a Marotta Wi-Fi module or something, one of the reasons why people use Wi-Fi modules is that they hide that from you. So you can just treat them as a modular component and you only need to worry about the unintentional radiation and susceptibility of the rest of your design, and you just say “and that box does radio and that’s nothing to do with me” and the function of the can is actually not to hide the the outside world. Emissions from the radio is to hide the radio from interference from the outside world from emissions from the radio, it’s to hide the radio from interference from the outside world, so that you can make that modular claim,”What happens inside this box is unaffected by what other changes I’ve made in the system.” Now, the fun thing about Raspberry Pi Plus and Raspberry Pi 4 and Raspberry Pi Compute Module is that they are modular products. So they’ll have cans over their Wi-Fi. So from an FCC, all of our products post Raspberry Pi 3+ are FCC radio modules. But they also do other things, like have a whole computer on them. But they have the same integration rules. And so that was the thing where you do get people integrating Zero W into products, but they have to do, as with Raspberry Pi 3, more work, more expensive work at the FCC level. And so the aspiration with this product was “look, let’s as we do with everything now, let’s make it a radio module.” And then it’s very easy to integrate.
Make: Speaking of these integrations, and obviously you’re accommodating the people using Raspberry Pi in commercial grade products, especially by having the Compute Module line, but… Thinking back to 2012 when the first Pi came out, versus the way that they’ve been adopted by everyone, are you surprised?
Upton: Yeah. We are surprised. And it happened very early on. So I think you can see the story for Raspberry Pi, like we had these educational aspirations, but if you look at the people who bought them in 2012, they weren’t kids and they weren’t engineering companies, they were hobbyists, enthusiasts, your readers.
But the interesting thing about that demographic is a lot of them are educators. And a lot of them are professional engineers. And so it kind of likes being the root of the tree for Raspberry Pi that by, because we were popular with those guys, those are the people who took us into schools. Those are the people who took us into their workplace. And when that boss said, “Hey, go build me a thing,” they’re like, “Hey, I know how to do this.” Exactly the same story with Arduino,, right? It’s exactly the way that Arduino has been successful. And that’s been one of the great things about this year is having an ability to work really closely with Arduino on the 2040-based products.
So yeah, it’s surprising. And last year, 7 million Raspberry Pis. Well over half of those were industrial applications. And that’s ranging from big giant companies, to this maker pro world that we’ve always been interested in, I guess that we’re investing a lot of effort this year and trying to sustain, back to a franchise two, we’ve been targeting what spare RP2040s we have outside of our own needs, and before close-in guys, really to the kind of maker pro-type businesses who can’t make anything this year.
So it’s an interesting mix. It’s an interesting volume mix. We have hundred-thousand-unit customers, we have a handful of those. We probably have fewer than 10 people who are buying a hundred thousand units off us a year. But then we’ll have thousands that are buying a thousand off us a year. So it’s a very broad base.
Make: So do you imagine with a new Zero what is this slot in with what you see?
Upton: With the can, it’s designed to be integrated. I think a probably one observation has been that when we talk about, we sold something like half a million Compute Modules last year, but we did about 4 million industrial units. So 7/8s of our industrial business was SBCs, not SOMs. And I guess we’ve learned that people are very keen, people go a long way to avoid making a baseboard.
People think about the internet of things and they think about thermostats and light switches and stuff. But actually a lot of the things that people connect these days are air conditioners, HVAC units, and big pieces of factory equipment. And those objects actually, an SBC may well be the right platform for that, because you don’t have to cut a baseboard, you basically have a hat or just a wiring loom. Sometimes you just have wires going down, soldered down onto a board. Zero 2 W is going to be a platform for, on the industrial side, it’s a platform for those kinds of applications where people want kind of more performance than they get from Zero, they don’t need the peripheral sets, big USB, multi USB ports that you get from a 3 or a 3+. They don’t need the performance of a 4 and they want a lower cost point and they want something that they can integrate into a smaller space.
It’s probably our smallest integrated, measured by board area, it’s probably fractionally smaller than CM4 and CM3. So it’s probably about our smallest integral, easily interoperable product.
Make: How do you decide which boards to make? You’re talking about Dominic and Simon. What’s the decision tree like?
Upton: This is a Simon product, by the way. Actually both the SiP and the board itself are a Simon product. So you talk about what sorts of people do you need actually, people who can do lots of stuff. And Simon is probably one extreme in terms of his ability to do lots of things.
How do we decide what to make? We make the things we want.
That’s weird, and it sounds flippant, but there’s a sort of a traditional marketing way of deciding what products to make, and I don’t like it. Cause you go in to speak to customers and you ask them what they want. And of course we listen to our customers, but it’s important to not overemphasize what customers want, because otherwise you can start to look at all your customers and say these people want this thing and these people want this thing and I’ll make a product. And these people want this thing. And if I make a product, I can cover all of those sets of functionality with one product. That’s great. I’ll do that.
And what you end up doing is you can make a product which is in the middle that interpolates between the needs of a bunch of different people and actually pleases nobody.
You either make a thing which doesn’t have the function, doesn’t actually have the functionality people want, or you end up making the thing which has loads of stuff piled onto the design, so it’s expensive. So everyone’s sad because they’re paying for functionality they don’t need. The nice thing about making products that you want is guaranteed at least one person wants the thing that you have made.
And I think the other fruit company, there’s a theory about the other fruit company. In that classic era, they basically made what Steve wanted. And it turned out a lot of people want what Steve wanted.
And so they were successful. And so I think, we’re geeky engineers and we want particular things. We make those things and “Hey, a lot of other geeky engineers like the same stuff that we like.” So that’s how that’s literally how we do it.
Make: It’s funny that you bring up Apple because it’s hard to not start to see some similarities in the journeys that you guys are going on. How do you look at what they’ve done and think about where Raspberry Pi is now?
Upton: Obviously they’re an amazing company and I’m using a piece of their equipment to talk to you right now. They do great stuff. They have a very admirable sensibility in terms of engineering. They are very, very good at engineering. And I think probably, to call out one similarity, is this kind of full-stack approach. Obviously they have now famously a full-stack approach that goes all the way through to the very advanced M1. Very, very advanced semiconductors all the way up to product design, not worrying about, we call it NAND to Tetris sometimes, there are courses in, called NAND to Tetris, which tries to teach you a little bit about everything that’s in the stack between an amp gate and writing Tetris in JavaScript. And, being able to worry about the NAND gates on the chips and the color of the icons on the desktop is a wonderful experience. One that I very much enjoyed.
I think what you find is Raspberry Pi tends to do this somewhat, sometimes a little bit more through partnership rather than the kind of brute force approach that you can take if you’re a trillion dollar company. So Raspberry Pi works with partners to do these things. The silicon, the, there were two pieces of Silicon in that SIP on that board, and they come from two different partners with whom Raspberry Pi has an incredible relationship. And then Raspberry Pi has to bring to the table the thing that it’s best at, and often the thing it’s best at is good engineering — but also, having relationships with people. And so that SIP is the product of two relationships and a good engineer.
And so, you can do the full stack thing several ways. And I think we are finding interesting new ways to do that.
Make: I am curious, next year will be 10 years for Raspberry Pi…
Upton: Wonderfully it can’t quite be 10 years because we launched on the 29th of February. So it can never be quite, it can be eight years or 12 years, but it can never quite be 10, but I think we’re going to declare it, we’ve thought about this and we’re going to declare it on the 28th. So we decided to round down rather than rounding up.
Make: 10 years tends to be a number of significance for everyone, a decade it’s a big thing, but when you talk about the second part of the Raspberry Pi name being a mathematical constant that is its own thing, I don’t know if 10 years really has that same significance or if you’re thinking about that.
Upton: I would love to have a product announcement for 10 years, and I’m not going to. And that’s sad. I wanted it badly enough. I almost suggested we slip that one to be it. But it would’ve been synthetic actually. We try never to be synthetic in that way.
So we launch as soon as we can. We launch as soon as we got something. I don’t know what it’s going to feel like because it was a very strange day, launch day. It was a very very odd day. You mentioned Pi… of course we normally don’t ship any product for the first couple of weeks, and if we delayed exactly two weeks, then we could have announced on Pi day. So maybe that was a missed opportunity. But, the launch day, obviously knocking over on those websites, doing a hundred thousand, we sold a hundred thousand Raspberry Pis on the day, and then spent six months making Raspberry Pis to send to the first, I think the partners started sending out t-shirts after three months to all the people to say, “hang on. Look guys, we’re doing our best,” and even though by that point, there was enough volume, tens of thousands of units, there was enough volume that they were showing up on Twitter, it was still only at a point where people who had ordered before 11:00 AM — which is basically no Americans because no one was awake —people had ordered before 11:00 AM, UK time had got them.
It was still quite wearing for people who’d ordered them at two in the afternoon or got up first thing in California and ordered them, and that was 5:00 PM UK time. So I don’t know what it’s going to feel like having an actual solid decade under our belt. It’s still fun. That’s the thing, it’s fun. And if it wasn’t fun, I’d stop doing it.
And I’m amazed because I have no real patience for anything. And I’d never done anything for more than four years. And to have a thing that I’ve done continuously for a decade is, and of course that we were doing for several years before we launched. It is just outrageous, really.
Make: It’s significant!
Upton: I’ll get some champagne. There’ll be a party, there’ll be a party and there’ll be champagne. But sadly, there won’t be a Raspberry Pi 5.
Make: We’ll be looking for it at whatever it does come out. One of the great things is just how much enthusiasm and excitement the community has for anything that you guys are putting out. And I think it’s just a testament to what good work you’ve been doing.
Upton: It’s a testament to that there was this community there. What’s the surprise of Raspberry Pi, the surprise is that there was a community there waiting and there was demand waiting.
People talk about category defining products and the iPad is of course a great example of a category defining product. It’s the other good thing about this weird approach to marketing, which is you couldn’t have focus-grouped the iPad into existence, because you can’t go out and ask people if they want a product that doesn’t exist yet, you can just make it and then see if people go for it.
And Raspberry Pi is like that. We’ve made it. And sometimes you just throw a dart and you hit a pool of demand. You hit a kind of a place where there was a lot of demand, latent demand, sat there, and nobody knew about it because there was no products there tapping it. And then you throw your dart and your dart hits the public demand and bang a hundred thousand units.
And of course, you know that this maker community, the community that’s around… I remember going to Maker Faire, my first Maker Faire in New York in the autumn of 2011 with a Raspberry Pi prototype and a monitor. And I got a trestle table. And we just set it up running Quake 3, which is one of the things that it could do very well. And just this crowd of people gathered around, but next to us, the trestle table next to us was this kid who had an Arduino and he’d automated a doll’s house. Lights, light switches in this doll’s house. And that was the first time that I really felt that we would, that there was a community out there waiting for it, that there were people that it might — and also on the educational side that it might work, that they were young and not just a community, but a community of children who wanted this.
And so yeah, it’s been a wild ride, 10 years. And it’s at least another 10 years of fun, I think. Maybe I’ll just wake up one day and it won’t be fun anymore and I’ll stop and do something else. But it’s very hard to see more than 10 years in our industry at the moment because there’s a lot of discontinuous, particularly with the end of Moore’s law, of discontinuous change going on. But I can see a decade. I can see 2030 from here. And that’s kinda fun.
ADVERTISEMENT
