The co-processing core is called the ‘flexible fusion engine’ (FFE, diagram below).
“We worked out the power needed to do a 32bit read from memory was 10x the power needed for multiplication, and RISC processors tend to have a load-store architecture with a lot of memory access,” QuickLogic marketing director Andy Green told Electronics Weekly. “For the fusion engine, we came up with long isntructions: input, multiply and store is one instruction. It’s a CISC-type processing engine. If a sensor wakes up, you have one instruction.”
Green spoke to Electronics Weekly at the Wearable Technology Show in London this week, where QuickLogic’s EOS S3 sensor processing reference design (see below) got its first public airing.
That is not to say QuickLogic is against RISC processors – alongside the fusion core on the same EOS chip there is an ARM Cortex-M4F.
The Runtastic fitnness watch (right) is an EOS design win. The watch uses sensors to classify what activity the wearer is engaged in – sitting, walking, running – and provides estimates on: number of steps in a day, distance travelled, calories burned, sleep duration and sleep cycles.
Battery life is typically six months, compared with three months in the predecessor.
“This comes from shutting the [ARM Cortex-Mx] processor down longer,”explained QuickLogic wearable specialist Robert Dawson. “It was running the pedometer and classification in a Cortex-M0. We added an EOS device with an FFE, and the FFE took over the pedometer and classification which left the M0 turned off.”
Tell me more about wearables
Having been involves with multiple design-ins, Dawson knows about wearables.
According to him, activity trackers can be split into a number of types:
GPS is power hungry, but gives a smart watch accurate autonomous positional information. If you need position, you need GPS.
Some position-concious watches save power by not having their own GPS receiver, but linking with a GPS-enabled smartphone which has to be carried at the same time.
If activity needs to be monitored, but position is not important, then accelerometers can be used to implement sophisticated pedometers which estimate distance by the number of steps taken and also estimate activity type.
Accelerometers take very little power and, in non-GPS watches, the use of Bluetooth can make a significant difference to battery life. While a battery might last six months in a watch with always-on Bluetooth LE, said Dawson, triggering Bluetooth LE manually with a push-button can stretch this to a year.
Which strategy is used depends on the watch company’s plans: is the watch intended to offer value-added services by constantly interacting with a phone, or does it only need to be able to download post-run information, for example.
Gyros take more power than accelerometers – mA compared with µA – and are therefore used less, but a recent 200µA gyro+accelerometer combo from Kionix ( KXG07 /08) could see gyros being used more widely.
There are no GPS-less watches that use gyros and accelerometers to implement full inertial navigation like a nuclear submarine or Boeing 747. However, for places with poor GPS reception, in cities for example, magnetic compass and step-counting may be combined for brief periods of dead-reckoning.
The S3 has the fusion engine, running at 20MHz, plus enough FPGA to implement a second fusion engine or a customer-specific function. Its 80MHz Cortex-M4F (F=floating point) is there to run the operating system when required, and has 512kbyte of sram.
On Monday, QuickLogic announced a reference design to go with the EOS S3, which supports always-on voice triggering and voice recognition.
It incorporates environmental sensors, inertial sensors, light sensors and microphones, and connects to a Google Nexus 5 phone via USB, supporting Android Lollipop and Marshmallow.
QuickLogic’s Celeris integrated development environment can be used with the reference design, and there is an Eclipse plug-in for Cortex-M4F support.
“OEMs will have the flexibility to evaluate always-on listening applications,” said marketing director Mao Wang. “They can either utilise QuickLogic’s SenseMe library of motion, context and gesture algorithms, or employ their own proprietary software stack.”