The Intel Curie story: why it really created the button-size chip

Make no mistake: Intel is taking wearables seriously. Last year at CES it introduced Edison, an SD-card computer designed for miniature devices, and it’s followed this up just a year later with the button-sized Curie. And let’s not forget that in March 2014 it bought Basis, manufacturer of arguably the most accurate fitness tracker.

In a briefing behind closed doors, I heard directly from Intel’s vice president Mike Bell on the company’s motivations, the technical challenges, and how wearables need to improve. Plus, what he thinks of the “voodoo science” behind some fitness trackers.

First, why did Intel introduce Curie at all. Wasn’t Edison small enough?

“You can think of Curie as Edison for wearables… Edison was designed specifically for the internet of things, and it is very powerful in ways that may be overkill for wearables.

“After working with wearables for a year and a half now we said, ‘Wow, to make a good wearable you would really want to fundamentally rearchitect some of the ways the pieces inside the chip play together to work the ways wearables work.’ So the Quark SE chip inside the Curie board is actually the first chip we’ve designed specifically for the wearable use case.”

So why is Curie round rather than square?

“Because it turns out that a lot of people building round wearables today are building round wearables. It could be any size. And in fact, if people don’t like the round board, we’re going to sell the chip separately and help people to build their own if they want, so it’s not like ‘this is it or nothing’.

“We showed it [at the Intel keynote] built into a button – it’s as small as we can make it essentially at the moment. Could we make it square? Sure. But why make it bigger if we don’t have to? So we took the corners off.”

Why is Intel spending so much time and money on wearables?

“The good thing for us is, a lot of the more complicated things [to do with wearables] require a lot of compute power in a very low power envelope, and that’s kind of what we do.

“So I can go back to the teams working on the process technology in the chip and say, I need a specific accelerator block that does this one thing in a really low-power way because I’m going to be using it all the time.

“In fact, in the Quark SE chip that’s within Curie we have a pattern-matching accelerator, and the thing it’s really good for is being able to feed it activity data and say, ‘If you see stuff like this, this person’s running. If you see stuff like this, the person’s playing tennis.’ And they can do this in a really low-power way so we can build these specific blocks to allow us to do very specific things with wearables, whether it be input, whether it be activity classification, that sort of thing.”

That brings us to the tricky subject of accuracy. Why will Intel be any better than the rest?

“Before we bought Basis last year, I bought one of everything in the market, tried them. The test for me is that I have a sports car with really tight suspension; if I drove to San Francisco some of those fitness bands would give me, like, 2,000 steps. Basis was the only one on the market that I found was accurate and had sleep stuff based on scientific fact and testing, as opposed to just, ‘Well, if you’re moving you’re probably turning over,’ which I call voodoo science.

“The Basis folks have done some testing to prove that what they have is a pretty accurate representation of a sleep profile. The other stuff on the market – well, I think it’s kind of natural selection. The ones that are based on voodoo science will be weeded out.

“We’re doing our part by saying, ‘We believe in this, hey, we believe in this so much we bought Basis, we think it works really well, we’re committed to improving it over time. 

“As we find edge cases – because everyone’s different, we’re going to find somebody out there who this doesn’t work for – we’re going to put fixes in there to make it better and better and better.”

So how will Intel’s chip adapt to individuals?

“The pattern-matching engine we built into the Quark SE is so good that if you feed it a bunch of data it can figure out, yeah, it’s different, but it’s not so different – because it’s almost like a neural network capability to be able to learn and figure out what it looks like when you run.

“And I think that these wearables have to adapt to people over time because not everybody is the same, and speech recognition is a great example of that – it works for a population of the public and there are people who it just doesn’t work for.

“These devices have to be adaptive and smarter and this is part of what we’re looking to build into our chipsets going forward. Not only in the hardware but in software.”

What specifically about Curie makes it work for wearables?

“Curie is lower power with internal architecture that makes it perfect for an always-sensing application like a cellphone.

“With a wearable it’s on all the time. It’s sensing stuff, it’s watching you walk, it’s listening, so fundamentally when you architect the guts of the chip it has to account for the fact that you want part of the chip lit up all the time but not all of it, else your battery will last an hour. 

“We’ve studied it a lot, we’ve gone through a lot of prototypes, it’s our first chip that says, hey, this is really for wearables. 

“Edison is great… but Curie lets you build stuff that’s even smaller. So you could build a fitness tracker into a button that doesn’t just track steps, it can figure out what you’re doing, or figure out when you’re sleeping, which is taking [the Basis Peak watch] and making it so small that everybody can have one.”