Semiconductor design engineers live and die by Moore’s Law – every 18 months, your device should shrink by 2 and be twice as powerful. That law has served consumers well for many years, demonstrated by the fact that our smartphones today are significantly more powerful than the desktop PCs we worked on even last decade. However, should it really apply to everything semiconductor? That question sat in the back of my head when my team and I examined the sensor hub market. And, as it turns out, I don’t think that it should. Here’s why…
The sensor hub market today is served mainly by Application Processors (AP’s) and Microcontroller Units (MCU’s). During the research phase, OEMs were very clear that they don’t envision enabling always-on sensing unless the total sensor sub-system power is less than 2% of battery life. No existing solution provides that. While both of these types of products are fully capable of performing as sensor hubs, neither of them perform sensor tasks in a processing and power-optimized way. Let’s look at AP’s – today’s mobile device AP’s operate in the GHz processing range, which is great for applications, 4G web browsing, etc… However, the sensors in mobile devices operate in the Hz range – does it really make sense to purpose a GHz processor for Hz functions? I think not, and this is demonstrated in that even the most power efficient 4+1 core processor, in low power mode, still requires more than 6% of system power to operate as a sensor sub-system (sensor hub + 9-axis sensing, which includes the accelerometer, gyroscope, and magnetometer). Moving to microcontrollers, the devices are capable of MHz processing in a more power-efficient architecture than AP’s. However, microcontrollers still consume above the 2% power threshold, and that’s without factoring in the 9-axis sensing power consumption.
So, today’s existing product offerings are more than capable in terms of processing capabilities, but consume way too much power. Let’s go back to Moore’s Law – do we really need something more powerful? In this case, the processing capabilities are actually well in advance of what is needed. Why not rethink the equation? Why not use the existing AP’s for GHz processor-efficient tasks, and off-load lesser tasks to a separate, appropriate device?
Rather than figure out how to build something more powerful, we actually took a step back and built something processing-appropriate in the ArcticLink 3 S1. We didn’t build the latest GHz processing device – we built a right-sized processing device. The ArcticLink 3 S1 features a Finite State Machine plus a CISC-based Arithmetic Logic Unit that processes all sensor data. What that enables us to do is to process all sensor data (including collection, processing, change detection, and AP notification) at 1/30 the power consumption of some our competitors with no detriment to performance. Mostly though, it allows us to run always-on sensing at less than 2% of system battery. This allows OEMs to keep the sensor subsystem active at all times, enabling a whole range of sensor-based applications, from today’s fitness and indoor navigation apps to tomorrow’s environmental, health, crowdsourced, and consumer applications.