Analog machine learning startup Aspinity has unveiled the first member of a new family of machine learning semiconductors designed to extend battery life of always-on devices.
The family, called analogML, is a tiny chip that reduces system power by 95%, extending the battery life of today’s devices or migrating wall-powered always-on devices to battery power. Aspinity said this potentially could create a new class of voice-first systems, home and commercial security, predictive and preventive maintenance and biomedical monitoring.
Always-on devices continuously collect large amounts of natively analog data as they monitor the environment and digitize the data immediately. This wastes system power processing data that is mostly irrelevant to the application.
The AML100, the first chip in the analogML family, delivers power savings by moving the machine learning workload to ultra-low-power analog where the chip can determine the importance of the data with a high degree of accuracy and at near-zero power, Aspinity said.
The company claims the AML100 is the only tinyML chip that can reduce data at the sensor while the data is still analog and keeps the digital components in low power mode until important data is detected. This eliminates the power penalty of digitization, digital processing and transmission of irrelevant data.
Other features of the ML chip include:
- Fully programmable configurable analog blocks.
- Reprograming through software updates or new algorithms.
- Power consumption of less than 20 µA when always sensing.
- Reduced quantity of data by up to 100 times.
- Support of four analog sensors in any combination.
