Analyze-First Architecture for Ultra-low-power Always-on Sensing
Over the next five years, billions of hands-free, battery-operated, always-on sensing devices in consumer, IoT, biomedical and industrial markets will assist us in our daily lives at home and work. As users become more dependent on such devices, they want smaller always-on products with longer battery lifetimes. Device manufacturers who can deliver more power-efficient solutions in ever-smaller form factors will gain a competitive edge. MEMS and sensors suppliers can help achieve incremental improvements in system power with lower-power and multi-mode components, but to effect great change, a system-level approach is needed to achieve significant power- and data efficiency. The problem is that the current “digitize-first” system architecture digitizes all the incoming sensor data, which is mostly irrelevant data, early in the signal chain — before sending it to the cloud for processing. A new alternative, the “analyze-first” edge system architecture, uses ultra-low-power analog processing and analog neural networks to enable the detection of events — such as voice, specific acoustic events or a change in vibrational frequency — from raw analog sensor data before the data is digitized. This “analyze-first” architecture reduces the volume of sensor data that is processed through higher-power system components (e.g., digital processors and ADCs) by up to 100x and reduces always-on system power by 10x for the next generation of portable, always-on sensing devices. MEMS and sensor components can integrate easily into an “analyze-first” architecture to offer battery-powered always-on sensing designs that enable smart portable products that run for months or a year instead of days or weeks.