Power Intelligence for Sustainable Always-on AI


An explosion of always-on intelligent devices is almost upon us. Fueled by a vast array of sensors that continuously sense the environment for data, the billions of smart products in our lives are on a collision course with a sustainable future. If industry experts are on target, we may have 45 trillion sensors generating an astounding >1 million zettabytes of data per year by 2032i.  Moving and processing such a heavy load of data consumes a phenomenal amount of energy, and that’s true for both battery-powered and wall-powered devices.

What’s to be done?

To achieve true power intelligence we need to think beyond individual components and develop near-zero power always-on AI systems. We can do that today by using analog processing. Sophisticated, inherently low-power, and now programmable using the same development environments that are already familiar to digital designers, Aspinity’s pure analog machine learning processor solves the power-consumption problems that have bloated energy consumption in always-on devices for far too long.

Aspinity’s AML100 enables the lowest-power always-on AI solutions on the market. Acting like an intelligent gatekeeper, it pulls machine learning into the ultra-low-power analog domain to determine data relevance at the earliest point in the signal chain keeping the downstream system off unless triggered by relevant events, data or anomalies.

Eliminate the wasteful processing and transmission of irrelevant data, and you’ve reduced always-on AI system power by an order of magnitude. Power-intelligent battery-powered devices reduce the burden on landfills placed by tons of old batteries. Power-efficient wall-powered devices relieve the drain on the power grid.

At scale, adopting Aspinity power-intelligence into always-on devices can save 1000+TWh in annual energy consumption and avoid 200+ million metric tons of CO2 from being generated every year from wasted always-on energy.

iSemiconductor Research Corporation (SRC), The Decadal Plan for Semiconductors