How To Build A Smartwatch: Picking A Chip
Welcome back to Tick Talk with Eric Migicovsky. The name has stuck with us for at least two episodes. We'll, we'll see how long it goes.
Today, I'd like to start off a series of podcasts and blog posts where I talk about what it's like to build. I. A smartwatch in 2025. I really wanna show that it's not that hard to do, um, especially if you have PebbleOS, which is, uh, the open source firmware that runs on Pebble watches. Um, hopefully the work that we're doing at Core Devices to build these new watches will help other folks build, um, new smart watches as well. Smart watches are not, and should not be. One size fits all devices.
I really hope that the work that we're doing in the open sourcing of PebbleOS may spark someone else's imagination and enable them to build their dream smartwatch, um, that fits their needs perfectly. A smartwatch is a system made up of three main components. You've got the watch hardware, you know the actual thing that you wear on your wrist, uh, and then you've got the software that runs on the watch. We usually call this firmware or the watch operating system. And then the third component is the mobile application, the companion app.
This is the iOS app or Android app that runs on your phone, connects to the watch, sends notifications. Downloads, watch faces, control settings, that kind of thing. So I'm gonna talk about all th uh, how you need to build each part of this system in order to build a smartwatch. But as a quick aside, I wanna show you some ancient history, some old hardware. So this is a box that actually says ancient history, and it's got some fun things that I've built over the years.
Uh, if you. We're building electronics in the early 2010s. You may recognize this. This is a Spark fund electronics box, and yeah, we all had stacks and stacks of these, So let's see what's inside a whole pile of junk. We've got some, you know. Early inPulse.
I think this is what we included. Oh, inPulse. I guess I should explain. This is the very first smartwatch that I worked on, um, way back in the day. Uh, I think we started working on this in 20 2008. Um, only worked with Blackberry.
I plugged it in. I was, I was trying to actually demonstrate it, but I, uh, I plugged it in. None of the dis, none of the displays worked. So this is. It had a transparent plastic back, kind of, kind of hilarious.
Um, one button. I can't understand why I only put one button on it. Man, that was not, that was not the best call. Oh, I've got some, these are 3D prints. Before we even came to the final, um, design for inPulse, we, you know, tested a few different designs.
This would look pretty cool. Oh. So I've got on my wrist, uh, this is a pebble, an original pebble that we launched on Kickstarter. Um, my favorite color is red. Uh, and this is, you know, this is it.
This is the one that we sold 85,000 watches. On, on, on Kickstarter. I've got my favorite.
This is my favorite sport band. It's um, it's kind of a very light, uh, silicon rubber band. Anyways, I. Take this watch to Burning Man every year. So it gets a bit dusty, but I've still got it works great.
This is, uh, yeah, like what's, let's see what the date was when I, when it was manufactured. Um, 2013. I. Yeah. Wow. Anyways, um, let's see what else we got in here. So this is the circuit board for the inPulse watch for this watch.
Um, yeah, I really wish, I really wish they would turn on, I wonder if anyone in the comments has an old, uh, oh. You could see the name. So the, the first company, uh, the company before it was called Pebble, was called Alerta. I named it that. I can't exactly remember why.
I think it was because it, well, obviously it has the word alert in it. Um, what else do we got? Oh, we've got a pebble, pebble core. Uh, one of the only pebble cores. Um, oh, this was, this was funny. This is a, uh, Kickstarter, a failed Kickstarter that me and some friends worked on where we built a iPhone. Um.
An iPhone case that would hold your AirPods. What else do we got? Oh, this is the, the first prototype for, um, inPulse that I made myself. This is the extent of my hardware engineering capabilities called the Watch Duino, which had an Arduino chip set and a screen from a Nokia thirty three ten smartphone. Um.
Yeah, some good stuff in here. Oh, this is the, this is the prototype that we used to record the first Kickstarter video. So, uh, we hadn't actually made the watch yet. Um, I could switch back to the screen.
So we hadn't actually made the watch yet. So what we did was we 3D printed a case. We didn't have a circuit board that would actually power the watch on, so we ran wires out the back and the person who was wearing it in the video, you know, if it was me or you know, one of the other people we had to run, we had to carefully run the.
The wires up our sleeve. And so if you go back and you look at all of the shots in the original Kickstarter video that had the watch actually powered on a person was always wearing long sleeves because the wires had to go up their arm to a battery pack that was kind of attached to their arm. Um, I don't actually know if I've told that story before.
So yeah, the first Kickstarter video, uh, definitely had some movie, movie magic to it. All right. Enough ancient history for now. Um, let's get back to the topic at hand.
How do you build a smartwatch? What are the steps that are needed in order to build one? As I've mentioned before, I think on a past blog post, uh, designing consumer electronics products is an exercise in constraint maximization. What do I mean by that? So first of all, when you're building a new product, you need to identify a target experience goal that you're aiming for. Like I wanna build a smartwatch with an always on daylight readable screen that lasts for 30 days. That's your target goal. Then you need to break that down into features or specifications that your watch will have. These are things like an ePaper display, Bluetooth, low energy.
A specific water resistance, um, and the price, it's very important, uh, kind of feature and you need to target, so let's say $150. Then you take those features and specifications and you create a design with hardware and software components that seeks to maximize. Those values, it's very difficult to get every single specification that you want because every time you pick a new component or a new feature, you may be trading off one specification at the expense of another. More battery life, higher price, uh, less water resistance, lower price, and so you're always trying to balance that equation and maximize it to the best of your abilities.
So after almost 20 years building consumer electronics, this process almost comes naturally to me. I can do it in my sleep or as I'm walking around I'll see a new product and in my mind start thinking about, you know, what does the exploded view look like? How do the mechanical and electrical systems plug into each other? I'll estimate the manufacturing price and start thinking and imagining like, how does the software work? And it's a bit of a blessing, but a curse as well, because I'm constantly thinking about it. You can break down, watch hardware into five key systems. First you've got the microcontroller chip usually this also includes the Bluetooth radio.
Then you've got the display, uh, that's front and center on a any good smart watch. Um, the third system is sensors and outputs, like the tax switches that are inside. Touch sensor microphones, accelerometers, speakers, mics, that kind of thing. The fourth one is the other electronic components.
This could be the chips, passive components, the PCB battery. And the fifth subsystem is the mechanicals, the mechanical structure, uh, like the the case metal case, plastic case, the straps or bands. The charge cable packaging, that kind of thing.
Now picking components for the, uh, the three latter parts of this system are actually quite straight, is actually quite straightforward these days. picking components for the latter three systems is actually quite straightforward these days. There's a ton of really good options available at various price points for sensors, batteries, straps, watch cases, um, mics, you name it. So constraint maximization is relatively easy for these things, and I'll probably wrap all of these together into one blog post, um, later on. But the two most challenging decisions that you need to make when you're designing a smartwatch are selecting a microcontroller. And a Bluetooth radio, which is what I'm gonna cover today.
And selecting a display, which is extremely difficult, and I'm gonna cover that in a, in a later post. So let's dive into how you'd pick a microcontroller for your smartwatch. So back during the OG Pebble era, we used an STM 32 F two series microcontroller, also sometimes called an MCU.
How did we pick that one? Well, it's kind of a funny story. So I was hanging out back in 2011 at the Hacker Dojo in Mountain View. This was a, this was a great hacker, hacker space, hacker, you know, uh, place that we could come and work on electronics and they had a bunch of 3D printers and all kinds of kinds of stuff. But I would usually hang out in the electronics room.
And, um, one of my friends, uh, Hugo, uh, was there and he could not stop raving about this relatively new MCU that he was using at the time. I was working on inPulse, you know, the first, uh, the first generation and I. Um, the inPulse used an LPC 2103. Um, for some reason I just can never forget these microcontroller uh, names.
And so the LPC 2103 had a whopping eight K of Ram. And so Hugo thought that this was terrible and we had to improve, and he, he knew that we, uh, we should head up to, to upgrade to the S TM 32 series, which had, uh, 128 K of Ram instead of eight K. And this was, you know, the only reason why I even heard about this chip in the first place, or, um.
Ended up using it in the Pebble. And to be honest, this is like a hundred percent typical for me. I think all of the major chip selections that I've ever made have come from friends raving about a particular chip.
And I guess I have a pretty delightfully eccentric group of friends, um, who over the years have. Turned me on to all kinds of interesting chips. So this is a big thank you to my friends Hugo Trammell Hudson TL from Pine 64, Peter Barrett, um, and many others who've texted me at all hours about chips and, and why they picked them. And, you know, that's, uh, that's been super helpful. And now I, you know, I turn to Twitter and I, I. I look at Blue Sky and I
try to find new chips through there. But at, at the end of the day, you need to have, if you're building something like this, you need to have a group of friends who are just as crazy as you and can help, um, help you learn about random, interesting chips. An MCU is the heart of a smartwatch. Think of an MCU as a miniature computer.
It contains the CPU Ram, usually flash storage IO peripherals, and even radios, all on one single integrated circuit. Uh, the specs of the MCUs that we used over all of the different areas of Pebble, um, are contained in this one helpful table that I use all the time. So I'm gonna link to it in the show notes.
You could check it out and see which are the, which are the different chips and MCUs that we used over the years. But I want to talk about why picking the right MCU. Is so important for designing a smartwatch. It all goes back to constraint maximization. The MCU sits at the focal point of the most constrained governing equations, software compatibility.
I. Power consumption and cost. So on the software compatibility side, embedded software, which is the software that runs on the MCU and Powers, the watch is much more fragmented and requirement specific than computer, computer operating systems software. This is because computers have tons of hard drive space. They've got tons of ram. So there's really no need to kind of optimize things perfectly. Uh, like for example, if you run, you know, Linux or BSD or Mac or whatever, the Linux kernel contains over 17,000 device drivers.
I. PebbleOS on the other hand, as embedded software has to be effectively hard coded for the specific MCU that you're using and the drivers, um, and have custom drivers written for all of the peripherals and sensors. Um, switching to a different MCU brand like we started with St. Actually stuck with st through the entire first era of Pebble. Um, switching, switching brands actually requires writing new peripheral drivers, um, for I squared C, spy, DMA, that kind of stuff, and adopting a different SDK and sometimes even a different build system. These changes aren't necessarily risky.
You know, you could, you could do them if you really wanted to, but they take time to implement and test. And so most people, when they. Pick an MCU. They actually stick with the same brand as they continue, um, through various generations. Uh, we, PebbleOS uses um, a kernel called FreeRTOS uh, and another constraint is that some chips don't even support FreeRTOS um, they require a different kernel like Zephyr.
And so more recently as we were designing, uh, the new watches, we looked at Nordics chips and we initially started designing with, um, an NRF 53 40. But we just could not figure out how to get it to work without using Zephyr. And the prospect of porting all of PebbleOS to Zephyr, uh, was too big, too big of a lift for us to really consider at the time. for Core 2 Duo, we decided to use an nRF52840 chip. This includes both the MCU as well as a Bluetooth radio all in one package. It's an older chip, but we were very familiar with it and we knew that we could Port PebbleOS to it relatively quickly.
We were initially thinking about using Nordic's soft device, BLE stack, and the 52 8 40 is the last device that supports the soft device. Uh, the 53 40 requires using Zephyr in order to, to kind of integrate the Bluetooth stack into the system. Um, so we were planning to use the soft device, uh, but then, um, thanks to the clever work of Liam, one of my past, uh, pebble colleagues and of Avid Rebble, um, contributor, we switched to using an open source BLE stack called Nimble. And funnily enough, now that we've switched to this stack, we could have.
Used the 53 40, uh, in the end, but at that point it was too late. We had already started developing the watch. Um, we were pretty far along with the schedule and we didn't feel like it was, um, it was just too risky to switch back to the 53 40. At that stage, The nRF52840 is powerful enough for the Core 2 Duo, but for the Core Time 2, we needed an MCU with more RAM and processing power. We would've really liked to stay with Nordic since we had just spent a ton of time porting PebbleOS to work with their SDK and all of their peripherals. But Nordic's roadmap in the BLE MCU space just didn't have much for us.
We looked at the nRF54L15, um, but it only has 256 K of ram, which wasn't enough. It also just recently went into mass production, so we didn't have any friends who were providing reviews or raving about it. They've also got a 54 H series with one mega of ram, but it doubles in price and we didn't actually need a mega ram. We really wanted just 512. The other consideration that we were thinking about is the bigger color display on Core Time 2 uses a special interface for memory and pixel displays. During the Pebble era with Pebble time, we used a dedicated FPGA to power and and talk to these displays, but that was an extra chip and cost more money.
So I was on the hunt for a new MCU I looked at a variety of options from Apollo BES Dialog, but I just couldn't find anything that fit our needs exactly. Um, it was actually pretty tough. I was like going in circles, trying to, you know, talking to everyone, seeing if I could find a chip. Um, one of the biggest stumbling blocks was the lack of an open source, SDK.
One chip from BES really looked pretty good, but we ran into problems as we were trying to test it. There was just no open source, SDK, no example code. Everything was locked behind an NDA. That wasn't gonna work for us because Peblo S is open source, and so we needed an SDK that would have an open source option.
Luckily, as I've seen time and time again with my MCU searches over the year, lightning struck. I randomly got an email from the CEO of a smaller Bluetooth chip company called SiFli. We exchanged a few emails over the span of probably a few hours, and it be, it became immediately clear that, uh, the CEO of this company was just extraordinarily interested in getting his chips to power an open source smartwatch. I flew out to Shenzhen a few weeks later to meet with their CEO and I think we really hit it off. Turns out SiFli chips are basically custom designed for smart watches.
They already power tens of millions of other RTOS based smart watches from brands like Red Me or APO Noise. Um, so we were in good company there. The smallest SiFli chip, the one that we're looking at using, or the one that we're gonna be using has 512 K of sram, which is exactly what we were looking at. And 16 megs of P sram, which our operating system doesn't even really use, you know, yet, but we could actually use it in the future.
And now are kind of, the wheels are turning, like how could we use this PS Ram? It also has a dedicated MIP controller designed specifically for the screen that we're using in the core time two, which means that we could eliminate this extra FPGA chip. The board gets simpler, the software gets simpler, we save a bit of money. Um, and best of all, I think is the power consumption numbers. It's wild.
I think we're looking at less than 50 microamps for the main Bluetooth controller in a connected mode. Um. Oh, and the chip's only, I think it's less than $2. So it's, it's basically perfect.
They even have a couple different options with more, um, sram if we really wanted to get to a Meg or two Megs in the future. Um, using the same SDK oh, and of course the best of all the open source SDK. Their SDKs open source, it's on GitHub and SiFli is helping us a ton to Port PebbleOS to their chips as well and publishing all of the code open source on, um, on our GitHub.
So there you have it. The chip for the core time two is gonna be an SF 32 LB five two J. I haven't got that memorized yet, but, uh, pretty soon it's gonna be etched into my, uh, memory for sure. I actually had the chance to invite Jing Ming, who's the CEO of SiFli here for, um, a chat a few weeks ago, um, which we recorded. And, um, I've got for you to enjoy. I'd love to learn more about how you started SiFli.
Like what, what, um, what caused you and your co-founders to, uh. To start a silicon company. Yeah.
Um, me and uh, and my team, we used to work on the cell, in the cell phone business. So we work on the, the application processors for cell phones, 4G and 5G. But then, you know, the cell phones, they're actually, uh, uh, uh, getting mature and not as exciting as before. So for that reason, we want to bring, uh, uh, similar experience like the cell phones to any other things that's possible. And when we look around, uh, there's not many chips that can serve that market. And so, uh, it's probably time to, you know, jump, jump off the ship, off the cell phone and do something more fun.
I think what you've identified is like a different band of technological gadgets, right? Yes. You have computers that are up here and they use heavy powered, well actually now they use almost mobile processors. Right. But cell phones use a different category of chip set than say, an embedded device.
Mm-hmm. Um, what was the first, uh, what was the first chip that you ever used? Like, did you use a pick or a basic stamp or what was the, the first micro control, I, it's 51. 80 51? Yes. Okay.
What, what did you build within 80 51? Well, I don't exactly remember. That was when I was in college. Okay. And, uh, we were building some kind of a counter or something like that.
And, um, well let me take it back. It's not 80 51, it's actually because I'm in the electrical engineering department. And, uh, we were actually building a small computer out of the discrete components.
So like Nan the Nor and the inverters, using that, using a breadboard. So you did basic digital logic? Yes. You were building flip flops and Yes.
Okay. Yes. That was, uh, way before the integrated circuits.
So we used the discrete components to put together a four bit, I think it's a four bit computer. Very, very simple. And, uh, so now people are doing that in Minecraft. Have you seen that people are building like four bit eight bit computers in Minecraft? Yeah.
Yeah. I've, I've, I've, I've seen that report before, but I haven't really tried it out. They're doing what you did on the breadboard in, uh, in a computer.
Yeah. And then, okay, so you were building, oh, I guess we're skipping around a bit, but where did you, where did you go to school? Where did you study? Uh, engineering? Uh, I, uh, my college is, uh, university in Beijing and I got my, uh, bachelor's and master's there. And I went to UCLA for my PhD. And you were studying computer engineering, electrical engineering. The whole electrical engineering. Yeah, I had the chance to visit Chinua.
It was, it was beautiful. And it was also right in the middle of like a little bit of Silicon Valley, like there were startups that were Right, right. Yeah, yeah.
Was that happening around the time that you were going to school there as well? Yes. You know, a big fund was going to the component market. There was actually a big component market in, in, and, uh, it was fun to look into those, uh, different components at the, uh, at tho at that time. All the components are imported from us and, uh, Europe.
And, uh, it was fun. And also, uh, buying the, the, the optical discs. That was a lot of fun. You mentioned that your, uh, your wife's a chemical engineer and my wife's, um, a chemist, uh, chemistry scientist. Now, you were, when, when you were at that lab, were you actually, uh, working on and, and getting down to the chemical layer of chips, like figuring out like doping and all of the um, uh, not at UCLA.
Okay. But I actually did it Inva University. We had this, uh, summer practice. So we, we went to the workshop and did all the etching and stuff to build our own board.
I think it's a, it's a charger. It's a, it's a, it's a battery charger. So you put your, uh, rechargeable battery inside and they will charge.
So you etched your own silicon at that stage? Uh, it's not silicon, it's the PCB. Okay. We etched the PCB. Cool.
Yeah, that's fun. Now, I was, one of the questions that I had, that I was interested in asking is how do you start a silicon company? I, for me, it just seems like such. And un well, it's, it's not something that I've done before.
Like I've, I've built gadgets. I've built, I, I know how to do PCBA design. I know how to do firmware, but I've always purchased chips from other people. I've ne it's never once crossed my mind of like, oh, I want to actually go and build this chip that I'm gonna use.
Mm. How did, how did you start a silicon company? And, and maybe a broader question is like, how does one start a silicon company in, you know, the modern day in 2025? You know, uh, silicon companies are mostly technology companies, so, uh, just like that you're not familiar with, uh, chips itself. We are not familiar with, uh, gadgets.
That's our problem. You know, I always, uh, you know, uh, uh, uh, uh, uh, working on the technology, but I don't see my technology at work. It's the same problem that I had, uh, when I was, uh, at school, undergraduate, and the masters. Right. You, you, you, you, you, uh, learn a lot of technologies, a lot of theories. But then how do the theories turn into a real product? Be the product, uh, a, a silicon or the product be a real gadget.
That is something I'm actually moving forward to closer and closer to the end market. But you didn't, you started a company that's just making the chips, you're making the building blocks that other people build on top of, right? Yes, uh, uh, we will do the, do the chips. We actually build the reference design as well. For example, we build a reference watch just to understand how the work, how the watch works. And, uh, then learn the experience.
What, what the, the, the customers, our potential customers will be like to, uh, have, uh, to, uh, to, uh, look into like the buttons. And also the, of course, the power consumption and the display effect, and also the hardware, uh, usability, right? So that is something we need to look into you. We don't just do the math in the silicon. We, we need to understand the real product. But take me back to the early days, like how many people, how, how big was your company when you started working on it? How much money did you need to raise in order to build the first prototype? How long did it take? Most people, you know, I, I, I think a lot of people just.
Don't have experience with this. This is kind of the, uh, dark, dark magic that we've never been exposed to. So shed some light share, share kind of the, the nitty gritty. How did, how did you make this happen? Well, uh, that actually, uh, mostly come, comes out of the experience in the semiconductor indu industry.
I've been working on this kind of semiconductor for more than 15 years, so I know how exactly it's, uh, being built from scratch. Then you need a lot of experienced engineers to help you. Uh, at the time that when we had the first product chip, working chip, we had about 25 engineers.
So that takes a lot of people. But, you know, nowadays. And what's the, what's the breakdown? So like there's 25 people at the company. You're there, your co-founders are there. Mm-hmm. Is it 25 engineers? Is it 25? You know.
Uh, kind of walk me through what, what does the company look like at that stage? Yeah. We had a, the chip itself is an SOC, so we had a, a combination from all different aspects. For example, there's a, a, a real circuit design designer and there's analog circuit designer to design the a DC and the pad.
And then there's digital designer to design the file so you can receive the signals from the microwave. Right. And then, um, there is a digital design engineers.
To put the, the, the arm processors together into an SOC and then you do the backend, meaning you do the, uh, you translate everything from a hardware description language to the, the real, you know, devices like transistors and then you tape out. So that part takes like 15 engineers to do, 15 engineers to do all of that. The various stages that you described. Okay.
Yes. And then there's, uh, driver engineers, right? And then there's, uh, graphics, uh, engineers, application engineers for graphics. And then there's uh, uh, uh, uh, uh, uh, Bluetooth Protocol stack engineers.
So, you know, three or four here in a year. It's a big company. It's a big startup. So like normal, like startups here, you know, are in a garage and you could.
You know, have pizza together with 25 people. You had to, you had to have some serious office space for that. Yes, we had a fairly big office, uh, there. And, uh, and we have, we, we had to have a lot of, uh, servers running. So, uh, uh, during the silicon is, uh, much more complicated than before. At least much more, much more complicated than when I was doing the etching at, uh, uh, uh, at my college.
But you're a fabulous company. So you will do the design and then work with someone like TSMC to, to go and manufacture it. Um. You worked for Marvell, correct? Yes.
Before, uh, but I, but I'm sure that when you're working at a big company, you don't really get exposed to every single part of the process, right? Mm-hmm. You, you are an expert and you are working in a particular part of the stack, or you had a team that was doing it, but when you start your own company, like. It's all you. Yeah. Like there's no one else.
Did you, was there anything unexpected? Like what were, uh, any stories that you could share of those early days where you were thinking that you knew what you were doing, but then something happened and you were like, wow, this is, uh, unexpected? Well, uh, there's a lot of things that, uh, that are, that was, uh, not expected at the beginning. For example, uh, we started with, uh, an SOC that, uh, that uses, uh, the Bluetooth low energy as the connectivity. But LA later on the market moved, uh, from, uh, uh, BLE to a dual mode Bluetooth that was not expected in the beginning.
So you were actually targeting BLE only initially. Yes. Yes. What was the product that you had, like, was there a certain class of products that you had in mind when you were starting SiFli? When we looked into this, uh, this market, uh, we saw, uh, Nordic and, uh, dialogue, all those companies, uh, behind the, the wearable gadgets. Especially dialogue. It's, uh, inside the, the, the mid band, right? The mid band.
It's, uh, pretty much the wearable in China at that point. Yeah. And that was BLE only, right? Yes, BLE only. And, uh, you know, I didn't realize, you know, the, uh, a, a phone call feature, it's important for a phone for, for, for a watch. So, uh, we thought that BLE connectivity is, uh, is good enough. Did you focus on watches exclusively at the beginning or, or wristbands? Was that like the primary thing that you were looking at? We were expanding to, expanding to o other areas, just that, uh, the watch is very typical, uh, I OT device meaning, uh, the, the AI kind of computing plus IO OT connectivity.
So, uh, we think, uh, once we have, uh, have, uh, made a footprint in a, in a, in a watch business, then we should be able to expand easily into other markets. Yeah, that is our thought. It's, uh, we are using the wearable, especially the smartwatch, as the most typical a IO OT device, you know, for the future and for the future kind of gadgets. Was there, oftentimes when you start a company, uh, there's usually one nugget of an idea or one insight that you, that you had, or one thing that you wanna push further than anyone else.
Did you see something in the market? That wasn't being worked on or like one thing that you wanted to exist to see exist? Yes. Uh, that was actually a, a, a surprise finding when I was actually looking to this market. I talked to the, the CEO of a major, you know, wearable company. He was complaining about the, the, the computing power out of the dialog chip. So the dialog was running 96 megahertz M four processor, and he was complaining it's far from enough. So I was actually curious.
I asked him, you know, how about I give you a 200 megahertz, uh, processor, which is double of what you're already having? Yeah. Is that enough? She, he said, uh, he replied, no, that's not enough. So then I got interested in what's, what's causing not enough kind of reply.
Then, uh, when I, you mean like what software does he want to run that's not currently being served by the dialogue chip sets? Basically he wants a bigger display. That's the, the a clear trend. Yeah. And also there's, uh, uh, as you need more CPU to be able to drive the pixels and Yes. Yeah. And there is, uh, sensor algorithms that are becoming more complicated and also, uh, uh, uh, neur network based.
So when you're talking about the neural network, then the, the existing, uh, computing power is far from enough. And then the same, similar thing for the graphics. When you talk about graphics, the, they, they, they, they pretty much use the CPU to do all the graphics. So that was the early stage of, uh, feature phone. Then, uh, you know, we could do that. We, we realized that the nesting is, uh, the, so the, the park consumption.
We need you, we need to, uh, cut the park consumption to be a much lower profile than the cell phone. That's exactly what we did for this company. Interesting. So. CPU more power, more, more power to be able to run algorithms and pump pixels out to the screen. GPU.
Yes. Uh, which hadn't really been done in the microcontroller. Class of processors. And then power.
Yeah. Uh, cool. Actually, the, all these elements are already, uh, there in a, in a cell phone. Silicon. Yeah. But, uh, but the power is not, the power is much higher at the different profile.
Right. And so, uh, uh, for the, uh, a IOT devices that we foresee when you need a fair, um, fair amount of, uh, computing power, including, uh, graphics, including the AI type of computing power and the regular, uh, uh, CPU kind of power on the other hand, then you need to have, um, a much lower power consumption profile so that, uh, you can, the device can last for a much longer period with a small battery. That is something we see as a combination. So the two extreme one on one side, uh, yeah. Computing power. People want it All right.
People want it all. That's why we started with, uh, this multi, multi-core architecture. And that is actually directly borrowed from, uh, uh, a, a, a mobile device, uh, mobile CPU, having multiple cores.
Yeah. Yes. one of the things that, um, well, we actually, why don't we talk about how. You and I met, uh, so, uh, you know, we were uh, kind of relaunching PebbleOS as an open source project a couple months ago. Yes.
And, um, we were initially using Nordic chip sets, the nRF52840. Mm-hmm. And I think I had already published that.
So it was on the blog that we were using Nordic. And then one day I checked my email and. There was an email from Jing Ming.
Mm. How did, how did you get the idea to email me about this? You know, uh, we are, uh, uh, we're actually working on the open source community, uh, project, uh, since, uh, sometime last year. And, uh, I'm actually, uh, looking into all kinds of news, anything open source. But then again, I still realize I'm actually late. To that, uh, PEB os open source news.
So, um, once I, I saw it and, uh, I, I, I read the news that, uh, you are relaunching the, the devices that was actually a, a lot of fun and, uh, and exactly what we are working on. So that's a perfect match. It's rare. There's not too many chip companies that do open source SDKs.
Why do you think that is? Well, that's the post, uh, uh, cell phone area that I'm foreseeing because, uh, you know, for the other ELT devices. The thing can get fairly complicated, but again, they're not coming with high volume as cell phones. For the higher volume devices, you can invest a lot of r and d into the device, so you don't really need a, an open source. SDK, you have plenty of engineers. You're saying like the, the companies that use the chip have a lot of people, have a lot of engineers, so they don't need the open source kind of community.
Right. But you're targeting products that aren't doing that, that are. Yeah. So, uh, we, I haven't seen anyone, any, uh, new gadgets that that can be compared to a cell phone in volume and in the, in the market value.
Right? And, uh, in the, in that sense, I, I see people like their own. Choices, for example, a smartwatch. I haven't had the, my favorite smartwatch yet. That's What's your favorite smartwatch. What's your dream smartwatch look like? Uh, firstly it has to be streaming low power.
Okay. Uh, uh, it's better to be a, a solar powered solar power. Okay, cool. Never have to recharge it.
Uh, that would be my first choice. Right. Okay. And the second I want to have a fairly big display.
So that I can see it because, you know, I'm in my fifties. I need to have display to see it. It's, it's funny. Yeah, like the first pebble I put together when I was 22. It needs a bigger screen now.
Yeah. Okay. Long battery life. Mm-hmm. Ideally, solar powered, bigger screen. What else? What is your, uh, a very sleek look.
Okay. That is very personal because I'm a a, a, a tech guy. I, I need to have something that looks techy, right? Mm-hmm. And, uh, there's probably many people like me who are looking for that kind of gadgets.
And maybe there is also, and also a, a female, female users, that's an underserved market from the smartwatch perspective. But now you're thinking as a business, I want you to think as a person, just as Xing Ming, because like there's a difference between what you think is a good product that's gonna sell a lot. And what's a product that you, yourself, you wanna put on your wrist? Yeah. The, the thing is, uh, I think there's many people like me who don't, haven't found their favorite gadgets yet. So, uh, it's probably not, not in a very high volume.
Like, maybe you have 20 K enough. Yeah. A maximum to, uh, to serve that market. How do we help them? That is, uh, help me, is to help other people.
That's exactly the reason why I'm looking to the open source way. Maybe, uh, we can come up with a, a new generation of software like pipes we're talking about to, uh, to as, uh, the, the, the, the Linux for a LT devices. That is something I'm looking for. What are the open source projects that you are looking towards as good examples? Like things that you and your company will emulate? Are there some companies or some chips or some SDKs that you prefer or think are doing a good job? Uh, yes.
There was one point when I was, uh, doing the research on smart watches. I saw this watch watching. Yeah, that's, uh, by S-Q-F-M-I. Yes. Yeah. And, uh, that was a very interesting device.
Uh, unfortunately it's not really a low park consumption one. No. Which is my number one because it uses an ESP 32. Yes, yes. But on the other hand, the SP 32 has done a good job to enable the, the open source community. That's also the reason why, you know, for smartwatch for, even though the SB 32 is not a perfect match, people can still use it, can make it happen.
Yeah. I I, I don't know exactly how S-Q-F-M-I chose SB 32, but I imagine that part of it was because the SDKs are public, there's no NDA that you have to sign, and there's a lot of good examples on GitHub. Exactly, yeah. And there, there was, uh, some customers complaining in front of me about the original NDA stuff that we had, uh, like five years back. And they were frustrated because, you know, in our NDA there was a, a, a penalty, if you know Yeah. It's, well that's any NDA any contract has that.
Yeah. They didn't like it. So you decided, so your current SDK has no NDA you can just download it. No, no nda. It's already on GitHub publicly. Uh, GitHub.
Yeah. And that was one of the reasons why after you emailed me and, and we started chatting more, uh, one of the reasons why we decided to pick SiFli, um, for the new set of watches, because I. Yeah. You were, you were publicly kind of stating that you wanted to be part of this community. Yes.
Which is great. Um, let me take you back, uh, one question I like to ask is, what, what's a gadget? That had, that had an impact on you, maybe an early, an early gadget that you owned or one that you really wanted to own, but you never got to buy? Uh, what do you remember? Uh, that I can, it can take me a full day to talk about it. Good. I want to hear it.
It's actually a true six 50. Six 50. Okay. So it's exactly the Trio six 50. Yes. I had one of those.
We'll, we'll put a picture up. Okay. That was, uh, in, uh, 2004 when I first got it. That was a big assignment because I was, uh, gonna get my PhD degree in the, but in the, in the end of 2004. So I bought it as a gift for myself. That was a, a, a major fund for me.
And, uh, did, did you have a Palm pilot before that? No. Okay. It was actually too expensive for, for a, a PhD student. So you had, but you had a, you had a cell phone before, uh, basic cell phone, yes. Okay. So this was a pretty big upgrade if you're upgrading from a cell phone to your first Palm Pilot, PDA.
Right. Okay. And what did you like about the six 50? Uh, first of all, it's a, it's a color screen.
Yeah, color. This display. That is a very good, I like it. The second it's, uh, I can hack it to it.
That is a lot of fun. Okay. I actually installed a dictionary app in the, in the device because you know, I'm an international student, sometimes I run into new words. Right.
So I learned from that dictionary a lot. And then I actually installed A-A-A-G-P-S navigation app. I think it was a hack from the internet. I don't remember.
It's probably Tom. Tom or, yeah, back then 2004 it would've been Tom. Tom, yeah, exactly. Yeah. And uh, that helped a lot. Because in those days when we go go for road trip, we actually print the, the, the Google Maps.
Uh, not Google Maps. It's, it was maps. MapQuest, yeah, exactly. Yahoo Maps. And then we put all the, all the, so the trio had a GPS chip in it.
It doesn't, so we actually did, bought a GPS receiver with Bluetooth connectivity. Okay. So we connect the Bluetooth device to, uh, the GPS device, to our, uh, did it have wifi as well? No. Okay.
Uh, no wifi. Have you ever gone war driving? Do you remember that? Where people would drive around to find open wifi networks? Uh, yes. I think I've seen one before. I wonder if the, uh, the trio could do that. You know, wifi was actually a new thing during my PhD period.
My PhD was about wifi. Okay. It's, uh, the eight by eight wifi, eight antenna, uh, wifi, mi o mi o. Yeah. You don't see it very often still in a, in the consumer markets. But then in the commercial markets, you see there's a, there's eight by eight.
Interesting. Um, cool. Uh, last question. So I was.
Um, I wanted to learn from you, where do you see, so we were talking about these different layers, right? Like the, the computer layer, the laptop layer, the cell phone layer, and then the microcontroller embedded layer. Mm-hmm. Where do you see that layer going over the next few years? Are there new technologies that you think are gonna start appearing on this layer? Or what's, what's next? Like, where do you see MCUs embedded radios in the next three to five years? Uh, first of all, I think the m cs is, uh, going through a major upgrade because of all the connectivity. So we call a LT or sometimes MC plus. So it's a connected M Cs, then it's a, it's a, it's a, it's a game changer. It's like the changing the game in a, in the cell phone business.
It's changing the business of MCU as well to add radios to add connectivity. Right. And second, it's, uh, the, the two chain.
Because, you know, before we were programming like CC plus plus, at most nowadays, people are programming using Python or JavaScript. That makes the programming a lot more easier and, uh, and quicker. So we need to have, uh, uh, uh, the two chain prepared for that kind of change. The second is the silicon itself, because, uh, Python, the Java programming needs a lot of computing power and a much more memory.
That's why, you know, for our, all of, all of our chips. They are equipped with, uh, much more higher, uh, memory capacity. So you don't have to worry about the memory. Uh, uh, ti tighten constrained by those like 30 2K, 256 k, uh, sram. So what are we, what are we gonna see on chip, right? Right now your chips go up to 32 megs of PS Ram, uh, up to 64.
64. So what are we gonna see? Are we gonna see like a. We are thinking about a gig? No AI models running on the MCU.
Okay. Yeah. And so, but, but where does the, like, there must be some, as you graph the power consumption of an MCU against the capability, there must be some point where the trade off is such that you may go to an arm eight or like to a higher power. No, it's possible.
Is there like a certain, like, will, will, that will the kind of phone level m uh. Um, uh, microprocessors, will they come down in price and power consumption to meet MCUs, or how will, how will that work? Uh, the biggest power consumption of the application processors is actually the operating system. Hmm. It's the Linux because the Linux, uh, uh, is, uh, requiring that much of footprint and a lot of data throughput. So that, uh, is something that you cannot really, uh, change. But again, if we can come up with a much more powerful, uh, RO system and it solves your, uh, uh, per consumption problem, and it solves the, the, the, the hardware cost problem that actually make, uh, make life a lot easier for the iot devices.
Sorry, I, yeah. Uh, eyes are just watering too much. Uh, too light, too bright.
Yeah. I haven't done that, um, before, sorry. I'm not, I'm not in tears from your, your, uh, I'm not crying about the future.
You see your, your life is harder 'cause it's, I know I've got a, I've got a few more. Yeah. Yeah.
Um, cool. Uh. Oh, this was one, one last one. If, if, um, if I could ask one more question, uh, what's the cleverest hack that you've seen someone do in the embedded world? Like, you know, when you're, you, you, you hear about someone using a chip in a very interesting way or someone doing something with software. Do you have any, anything off the top of your head that you remember about someone doing something, kind of like a crazy hack? Yes.
There's, there's two things that I can come up with. The first thing is with our own engineers. Uh, we had this, uh, uh, AI engine in our chip, so it's uh, basically an accelerator, hardware accelerator for the, uh, uh, uh, uh, matrix competition. And the, uh, convolution, there's a math called, uh, convolution.
And uh, at some point there's a customer talking to us about the goin effect Authing blur, which we see on the iPhones and Apple watches. And they were asking us if this is doable, is it hardware accelerated or. It, it is hardware accelerated, but never before. I'm a smartwatch. Okay. So, um, so I looked into, I told that idea to our, uh, engineers and the two of our engineers once a hardware engineer.
The other is a software engineer. They came up with the i this idea that we can probably use our AI accelerator to do this. Go simpler. But your AI accelerator, is that running on the GPU or it, it's a, it's separate from the GU.
It's an NPU. Okay. Yeah. So we, we, I didn't realize this, uh, kind of, uh, AI accelerator can be used for graphics. That was a major surprise to me.
And, uh, they did a very good job because actually basically the fundamental math, it's very similar. It's always matrix, uh, convolution. So that's, that's, that's why they actually, so they trained a, did they train a model to do Gagosian blurs or No, they didn't do that. Okay. It's just a, a, a plain filtering. Okay.
Yeah, that's one thing. Cool. That's a nice hack. Yeah. The, the other surprise, it's actually happening in our current open source community, there's, uh, there's, uh, one, uh, uh, uh, developer, his name. It's, uh, well, his, uh, web name is Simon.
Yeah. He actually, uh, ported the Unix to our, uh, American controller. What, which is a total surprise. And, uh, he was actually running a game from the, the, the, the nineties on, on that Unix.
That was fun. Maybe I can show you. That's pretty cool. Okay. Uh, so that's people in your, in your, like on your GitHub posting interesting hacks that they're doing with SiFli.
It's not exactly GitHub, it's actually a qq uh, uh, group. QQ group. Okay. Yeah, we, we have a big, like 300 something.
Uh, engineer QQ group. Nice. And they were actually posting their fun stuff in, in the, in the group. In my past life, I did a lot of chat interconnectivity. Maybe we need to build a QQ to GitHub Bridge so people can see on both sides. Yeah.
What's happening? Yeah. I actually, we actually just did a bridge between the, the GitHub and the GI Gate because, you know, there was a glitch a few days ago on GitHub. They blocked all the ips. From, uh, China and Hong Kong. Really? Okay.
Yeah. Oh yeah. I think I remember seeing that. Yeah.
So you now have to build that bridge so people can communicate on both sides. Yeah. Yeah. Very cool. Cool.
Well, thank you very much for the Ming. Okay. Thank you, Eric. Thank you for having me. Well, thank you Jing Ming for being the first uh guest on my podcast. And thank you everyone for joining me on another episode of Tick Talk with Eric Tchaikovsky.
We'll see you soon.
2025-05-20 03:26
