BM110: Always-On Analog AI Audio Inferencing SoC

Redefining Edge AI with Ultra-Low Power, All-Analog Compute

Overview

The BM110 is a groundbreaking, all-analog AI inferencing System-on-Chip (SoC) engineered to bring continuous, real-time intelligence to the absolute edge. Designed specifically for always-on audio and time-series data processing, the BM110 shatters the power barriers of traditional digital AI processors.

By consuming a fraction of the power required by competitive digital solutions, the BM110 extends the battery life of edge devices from hours to weeks or months for wearables and years for security systems. It makes "always-on" AI a reality for even the most power-constrained environments, unlocking a future of truly ubiquitous and intelligent listening devices. Our system level approach to maximize batter life and simplify system level integration for always on audio inferencing manifest itself in multiple ways in the BM110 (figure 1)

System MCU is in DEEP SLEEP for >>99% of the time – BM110 generates an interrupt on a trigger event. Simple SPI interface to BM110 to query trigger and access audio buffer.
All analog processing dramatically reduces audio inference power while providing low latency key word spotting.
Voice, keyword and up to 10 additional commands in a single low-cost device.ost sensitive 2-to-4-layer PCBs solutions.
Small, low-cost analog microphone input to the BM110 saves significant additional system level power in always-on applications further extending battery life.

Integrated audio buffer stores ~2 seconds of audio which is optionally triggered on the keyword or voice detection. By enabling post trigger validation and audio capture the integrated audio buffer dramatically simplifies system design and further reduces system level power considerations.
High fidelity audio pass through support from the analog microphone to a digital output to feed the system MCU if required.
Tiny ~4.5mm² wafer scale CSP (~2.25mm x ~2.0mm) for space constrained systems supports c

Frictionless Development Workflow:

Train your neural networks in the cloud using standard frameworks like PyTorch, TensorFlow, or Caffe. Blumind's proprietary translator software automatically handles the quantization, compression, and mapping directly onto the BM110.

Core Technology: The Analog Advantage

At the heart of the BM110 is Blumind's patented AMPL™ (Analog Machine Learning Processor) architecture. While traditional AI chips rely on digital systems that inefficiently shuttle data back and forth between memory and processing units, the BM110 takes a brain-inspired approach. Read more about the Blumind technology