Aspinity, the leader in analog machine learning processing, has announced the commercial availability of the only glass break detection solution to produce a five-year battery life while also eliminating false alarms to common household sounds. Integrating the AML100 with the company’s purpose-built glass break algorithms, Aspinity’s always-listening solution is the first to implement sensing, processing, and decision-making completely within the ultra-low-power analog domain, eliminating the digitization of irrelevant data and its associated wasted power. The Sensors Converge team recently recognized this first-of-its kind chip as a finalist for the 2022 Best of Sensors Awards.
Aspinity provides algorithms that speed the evaluation of its AML100 solutions for an array of security and home automation applications that require an extended battery lifetime, including the detection of T3/T4 alarm tones (smoke and carbon monoxide, respectively).
“Consumers want home security systems that both run for years on battery and don’t trigger false alarms,” said Tom Doyle, founder and CEO, Aspinity. “Unfortunately, existing glass break sensors only deliver one or the other, which frustrates consumers and diminishes brand satisfaction. False alarms can also prove expensive if law enforcement charges a penalty when called to the scene for no reason. While today’s glass break detection solutions force designers to trade off battery life or accuracy, Aspinity’s solution frees designers from having to make that undesirable choice,” Doyle said. “Our analog machine learning processor, on the other hand, enables an extremely low-power always-on machine learning architecture with algorithms that are trained to detect window glass break. And this uniquely delivers the extended battery life and performance accuracy that gives consumers confidence in their home security systems.”
Two types of traditional glass break sensors have dominated the market for years. The first type employs a traditional DSP architecture that continuously analyzes ambient sound for some combination of the frequency, amplitude, and duration of the sounds that might represent glass break. This method—which doesn’t include machine learning—may deliver an extended battery life but is prone to triggering false alarms on common, loud household sounds, e.g., pots and pans clanking, dogs barking, and items dropping.
The second type, which also uses a digital architecture, includes a digital tinyML chip that improves glass break detection accuracy. But these systems consume a whopping 3000µW or more to continuously digitize and analyze all sound data. The upshot is fewer false alarms but with a scant one- to three-year battery life.