Neuromorphic processors are designed to work like the human brain, firing signals only when needed. This makes them far more energy-efficient than traditional chips. Dutch company Innatera has introduced Pulsar, the first commercially available neuromorphic microcontroller, aiming to bring this technology into mainstream AI applications.
Pulsar dramatically reduces latency up to 100× faster and consumes as little as 1/500th the power of conventional AI processors. It combines hybrid analog-digital spiking cores, allowing developers to choose between maximum energy efficiency or higher flexibility. The chip also includes accelerators for convolutional neural networks and fast Fourier transforms, plus a 32-bit RISC-V CPU, all packed into a 2.8 × 2.6 mm chip.
What makes Pulsar unique is its complete system integration. Unlike other neuromorphic solutions that lose efficiency when interacting with external components, Pulsar handles sensing and processing within a single chip. This enables “always-on” AI for extremely low-power devices like smart doorbells, industrial sensors, and IoT gadgets.
For example, radar-based presence detection can run on just 600 µW, significantly extending battery life and improving accuracy. Innatera also released a PyTorch-based SDK, Talamo, to make neuromorphic development more accessible and encourage wider adoption of this emerging technology.
Read more-https://spectrum.ieee.org/innatera-neuromorphic-chip