“Edge AI is fundamental to bringing intelligence into the physical world, enabling new applications and technologies that will improve our society. Neuronova contributes to this shift with a processor that consumes less than one microwatt, bringing always-on intelligence where until recently it was simply unthinkable. But no company is an island: joining the Edge AI Foundation gives us the opportunity to collaborate with an exceptional ecosystem, contribute our perspective on neuromorphic and ultra-low-power computing, and help shape the future of edge computing.” – Alessandro Milozzi, CEO & Co-Founder of Neuronova
Based in Milan, Neuronova is developing fully analog neuromorphic processors for ultra-low-power, always-on AI. Rather than continuously digitizing raw sensor signals and sending them through conventional processing pipelines, Neuronova’s technology uses physical implementations of spiking neurons and synapses to process temporal information directly in the analog domain.
Rethinking Always-On Intelligence
As intelligence moves from the cloud into billions of physical devices, power consumption becomes one of the defining constraints of Edge AI. Neuronova is tackling that problem at the beginning of the compute pipeline.
Its Analog-to-Information™ architecture is designed to act as an intelligent analog front end between sensors and digital processors. Instead of keeping an ADC, MCU, DSP or application processor continuously active, the system can interpret signals at the source and wake the digital system only when meaningful events occur.
The company’s H1 processor is fabricated and measured on silicon and is designed for always-on inference at below 1 µW. The chip incorporates three analog cores in a footprint smaller than 2 × 2 mm, with standard SPI/GPIO interfaces for integration into existing systems.
Starting With Audio, Scaling Across Sensors
Neuronova is initially applying its architecture to always-on audio applications, including:
- Wake-word and keyword detection
- Voice activity detection
- Acoustic event detection
- Acoustic scene classification
- Contextual assistance for denoising and downstream audio processing
These capabilities can run continuously at the sensor front end while allowing higher-power digital processors to remain asleep until they are actually needed.
But audio is only the starting point.
Neuronova envisions the same architecture extending across motion, vibration, biosignals, pressure, environmental sensing and other temporal signals-ultimately embedding intelligence directly into sensors rather than treating sensors simply as sources of raw data.
From Silicon to the Developer Ecosystem
An important part of bringing new computing architectures into the Edge AI ecosystem is making them accessible to developers. Alongside H1, Neuronova has developed the Silk 1 evaluation board and a silicon-calibrated SDK. The SDK provides a PyTorch-compatible workflow for training, hardware-aware optimization, power and accuracy estimation, and deployment to H1 silicon. Developers can use the evaluation platform to benchmark inference power, latency and event-driven MCU wake-up behavior on real hardware.
This combination of new compute architecture, real silicon and developer tooling makes Neuronova particularly relevant to the Edge AI Foundation community as we explore how AI systems move beyond conventional CPU/GPU/NPU architectures and deeper into the physical world.
Neuronova brings an important perspective to that conversation: what if intelligence begins before the signal ever becomes conventional digital data?
We’re excited to welcome Neuronova to the foundation and look forward to connecting their work in analog neuromorphic computing with our community of semiconductor companies, researchers, developers, startups and organizations building the next generation of intelligent physical systems.
Welcome to the EDGE AI FOUNDATION, Neuronova!