Sub-threshold design for exponentially low MCU power consumption
May 08, 2015
With processor manufacturers clambering over one another screaming of their "low power" achievements, naturally it's a term that that's become over us...
With processor manufacturers clambering over one another screaming of their “low power” achievements, naturally it’s a term that that’s become over used and thus is under-valued. The term “low power” is of course relative (and what it is low relative to is rarely clearly defined!) and it’s that aspect that has permitted it to be ubiquitous in (almost) every embedded component’s marketing material for many years now – since the focus first began switching from performance to power consumption.
It’s easy to become weary of claims of low power in the embedded marketplace, and it’s precisely that reason I could have missed a genuine innovation in low power technology that really is relative – promising power consumption reduced by factors of 10, not a reduction of a mere few mW. This innovation employs the exciting, though complex area of sub-threshold circuit design techniques.
In standard circuit board designs, transistor’s “on” state is typically trigged by a voltage in the region of 1.8 V, which from now I’ll refer to as “super-threshold”, whereas sub-threshold design can derive that same on signal from a fraction of that. As energy consumption is the square of voltage applied, reducing this 1.8 V to, say, 0.5 V would offer a 13x improvement in consumption, dropping to 0.3 V offers a 36x improvement.
This all sounds very logical and, dare I say, simple – but the complex considerations involved in reliably operating transistors in this fashion are far from simple. When considering the obstacles involved in sub-threshold design, it is a staggering achievement that this technology has now been fully commercialised and is inherently reliable.
For example, in a traditional super-threshold design, an ambient temperature difference of 25 °C translates to a 20 percent difference in current at 1.8 V, operating sub-threshold at 0.25 V equates to a difference of 660 percent!
For the technology to be commercially viable, not only did it need to address those fundamental challenges of operating sub-threshold, such as poor transistor modelling at these voltage ranges, logic swings, and noise. It must be employable (as closely as possible) using the same processes used to design traditional super-threshold circuitry, thus not be entirely alien to electronic designers.
Ambiq Micro‘s Sub-threshold Power Optimised Technology (SPOT) approach was developed through years of intensive testing at sub-threshold levels, modelling using test chips and wafer shuttles to exhaustively test the processes, and satisfy themselves that the end product is infallible. This in itself represented significant challenges as industry standard test equipment doesn’t support picoAmp and nanoAmp levels generates; special test fixtures were carefully adapted to address this, of course still utilising standard industry testers – that compatibility to existing technology essential to Ambiq Micro.
The industry challenged Ambiq Micro to demonstrate real world usage of this technology beyond the RTC they successfully implemented some years ago. Open to a challenge, Ambiq set about to create an MCU with an ARM Cortex-M4F core that offered lower power consumption than any ARM Cortex M0+ offering from the competition – introducing the Apollo Ultra-Low Power MCU. They achieved an industry leading 30 uA/MHz and a mere 100 nA in standby. The M4F wasn’t chosen purely out of vanity, the IoT and wearable markets they target are increasingly reliant on large numbers of sensors and complex algorithms that rule out M0+ with raw processing requirement.
It’s sometimes easy for companies pioneering new technology to develop tunnel-vision and suggest that every circuit (and all of it) blindly would perform better with sub-threshold technology. The reality is that parts of the circuit that are only selectively active (perhaps on start-up) are actually better left in the traditional super-threshold domain as they do not affect true run-time consumption.
It is this approach to me that defines Ambiq Micro as a technology company with credibility and integrity. Their technology is likely to be met with skepticism given (what appeared to be) past terminal obstacles thus they’ve made it their mission to validate, substantiate and educate the electronics industry to a technology that genuinely can be described as revolutionary.