A couple of months back SparkFun did something that the company isn’t that well known for, they released a module. Built around the ultra-low-powered Ambiq Micro Apollo 3 processor, the module was intended to bridge the gap from “maker to market,” and from prototype to product. At the tail end of last week they announced full FCC/CE approval for the module.
Earning approval from both FCC and Industry Canada (IC), as well as its CE certification, means Artemis module is the first entirely US-manufactured, open source, FCC/CE/IC-certified Bluetooth LE module on the market.
It also indicates a big departure for SparkFun. Traditionally known for their breakout boards, and maker projects, aimed at hobbyists the company isn’t known for this sort of thing.
So we sat down with Nathan Seidle, the founder of SparkFun, to talk about what this means for SparkFun, the Artemis module itself, and how he sees it fitting into the maker and wider micro-controller marketplace.The module is intended to bridge the “maker to market” gap? Tell us about that?
“I have seen amazing projects built by our community of hackers and makers over the years. Every once and awhile someone would float the idea of having the thing manufactured in mass. Remember the popularity of the reverse geocache by Mikal Hart? That thing was awesome. But it took him years to productize. Manufacturing is hard, but made even harder when the inventor uses an Arduino based on the ATmega328 or the likes. Don’t get me wrong, these microcontrollers are available, but depending on where you’re manufacturing (US vs. China) and what your constraints are (cost, time to market, etc.) the ATmega series can be awful to go to market with. I heard many horror stories of people having to throw out all their code and start over again on a different micro just because the manufacturer was setup to use more modern tool sets. We chose the Apollo 3 for Artemis because it’s amazing, but we invested a ton of work into the Arduino core so that the entry level person would never know the difference between Artemis and an Uno. That is until they needed to go to scale; it’s far easier to use Artemis and our tool chains to do a commercial product.”Building something like this is a big departure for SparkFun?
“This was personally a very exciting change for SparkFun. We’ve been debating whether we could do modern manufacturing tolerances (0.5mm pitch BGA is no joke). We’ve heard stories about how hard regulatory certifications can be. But we had often felt we weren’t good enough (a bit of impostor’s syndrome). But I love a challenge! Artemis was a big lift by a lot of smart people at SparkFun but what it showed us is that we can put our big kid pants on and play. The electronics industry is getting smaller which is why SparkFun exists (making hard to access electronics easy) so it’s good to know we don’t need to slow down because ICs or tool chains are getting more complex.”How else do you see SparkFun supporting the “pro maker” market?
“We dabbled in the IoT software market for a bit with our open source data platform called Phant. What we learned is that we’re far better at hardware than online platforms. We are going to continue to focus on hardware and find ways we can help the pro market.”Do you see the module as attractive for large companies, beyond the maker market?
“Artemis absolutely needs to be in every toaster and defibrillator! No. Just no. But in all seriousness, we’re most excited to support our core, zany customers build really cool stuff. I think Artemis is perfect for a larger player but we like our friends in that prototyping phase. That said, if GE wants to build a better mouse trap, we’re happy to help them.”Why now? Why make this change?
“I had this cheeky friend[I was walking around with the prototype for the SparkFun Edge board in my laptop bag for a month —Al.] who thought it would be a great idea to try to run machine learning on a micro-controller that no one had heard of. He really wanted the hardware for his technical research and thought SparkFun could build it for them. Fast forward 6 months and we’ve had amazing success with support from Google running TensorFlow from a coin cell battery.”
“A handful of reasons. First was the Apollo 3 was the lowest power micro currently available capable of running the TensorFlow models. Second, there’s something about ‘blue sky’ design that I really enjoy. The world has piles of modules using the main stream popular ICs such as the nRF52, ESP8266, STM32, etc. No one had heard of Ambiq, but it was a phenomenal IC. Ambiq did a great job creating a industrial strength SDK to support their IC but they had no ability to bring it to the hacker market. SparkFun knew it would be a lot of work to get Arduino and TensorFlow running on a module but I really enjoy writing piles of code when I know it will help lots of people never have to think about how to setup a 2nd I2C port, or deal with software serial.”Tell us about software support for the module?
“It’s funny — we’ve tried really hard so that no one will notice how good the software is. Artemis is not a code port like the ESP8266 was. That was an amazing effort by the community to get Arduino running on Espressif’s ESP8266 but the ESP core continues to be quirky. There’s always something just a little different about using an ESP versus and Uno. We didn’t want that. We worked very hard to make sure original Uno code can run on Artemis without modification. I want users to not care what’s controlling their blender. But once they begin to hit the limits of the original Arduino they’ll learn to appreciate the extra features and flexibility of the Artemis.”Do you see this module as primarily for embedded machine learning?
“No. I love the fact that Artemis can run full TensorFlow models at less than a milliamp. But it’s a great hammer for all my general projects as well. The specs on Artemis (96MHz, 1 meg flash, BLE, lots of SPI/I2C/Serial) puts the Artemis in the middle of the pack for general microcontrollers. It’s perfect for DIY projects and a go to when the ATmega can’t quite handle it.”What’s with the open source aspect of the module?
“I had to learn a lot about how to route a four-layer board with blind vias. I had to think about how the RF shield would interact with the module. I had to think about how to layout the pads on the bottom of the device so that it would be easiest for designers to use. There are very few resources or projects where these topics are covered. It was important to me to show folks that there’s nothing magical or even hard about the design. Shoot, if I can do it…”How do you get it on tape and reel?
“This was one of the aspects of this project that had me really worried. Do we need a machine? How does it work? It turns out SparkFun has gotten loose components in the past and have had to have them inserted into tape and reel. We’ve done this already! There are companies that do it for less than $100! I just didn’t know. Once we build, program, and test the Artemis in Boulder we ship them to a company in Ohio to have them placed on tape and reel. It was just another aspect to the project that I thought would be a huge barrier but ended up being far less painful than I expected.”How rough was certification?
“That’s an entire blog post by itself. Summary: Four stars, would do again. It was less than $20k and took about seven weeks. I had to invest about a week’s worth of time dealing with paperwork and documentation but much of it overlaps with the documentation our customers need anyway. Luckily we passed first try but we had a really good experience.”
The SparkFun Artemis module is available now costing $8.95 in single unit quantities, with volume pricing available on tape-and-reel for quantities of 250 units and up. If you’re interested in picking up one of the boards built around the module; the RedBoard Artemis costs $19.95, the RedBoard Artemis Nano costs $14.95, while the RedBoard Artemis ATP costs $24.95.