Built on a decade of research, Pulsar delivers up to 100X lower latency and 500X lower energy consumption than conventional AI processors and is being described as a new class of ultra-efficient device, that brings brain-inspired intelligence directly to the sensor edge.
With sensors embedded almost everywhere there is a need for real-time, secure, energy-efficient data processing at the edge. Pulsar looks to tackle this challenge by processing data locally and intelligently, at the sensor level, eliminating the need to rely on brute-force compute in power-hungry edge processors or data centres to make sense of sensor data.
“Pulsar is not just another AI chip – it represents a fundamental shift in how we bring intelligence to the edge,” claimed Sumeet Kumar, co-founder and CEO of Innatera. “This launch marks the moment that our brain-inspired technology becomes ready for mass-market deployment. As demand for real-time, power-efficient intelligence in edge devices continues to grow, Pulsar delivers the capabilities that traditional AI hardware simply can’t – ultra-low latency, minimal power draw, and on-device decision-making. More importantly, it lays the foundation for a new class of intelligent systems that are adaptive, autonomous, and scalable.”
Pulsar’s compute architecture is based on Spiking Neural Networks (SNNs), a generational leap in AI hardware that processes data the way the brain does, focusing only on changes in input.
This event-driven model reduces energy use and latency while delivering precise, real-time decision-making. Pulsar also combines neuromorphic compute with traditional signal processing in a new architecture. Integrating a high-performance RISC-V CPU and dedicated accelerators for Convolutional Neural Networks (CNNs) and Fast Fourier Transform (FFT), this architecture provides increased versatility on a single chip.
“Innatera’s Pulsar chip has the potential to redefine what’s possible at the edge,” said David Harold, senior analyst, Jon Peddie Research. “By using brain-inspired Spiking Neural Networks, it brings real-time processing to ultra-low-power devices without leaning on the cloud. That means sensors that can think for themselves – faster responses, lower energy use, and smarter performance across everything from wearables to industrial systems.”
Pulsar gives product teams a shortcut to smarter features that were previously off-limits due to size, power, or complexity. Filtering and interpreting sensor data locally keeps the main application processor asleep until needed, in some cases, eliminating the need for a main application processor or cloud computing, extending battery life by orders of magnitude.
With sub-milliwatt power consumption, Pulsar now makes always-on intelligence viable, enabling everything from sub-millisecond gesture recognition in wearables to energy-efficient object detection in smart home systems. For example, it achieves real-time responsiveness with power budgets as low as 600 µW for radar-based presence detection and 400 µW for audio scene classification.
According to the company, Pulsar transforms traditional sensors into self-contained intelligent systems. With its small memory footprint and efficient neural models, it can fit into tight form factors while eliminating the need for heavy external compute and reducing reliance on complex, custom DSP pipelines. Sensor manufacturers can now deliver plug-and-play smart sensor modules, accelerating development and time to market.
Innatera’s Talamo SDK also makes neuromorphic development easier. Developers can build spiking models from scratch, in a PyTorch-based environment, simulate, optimise, and then deploy.
To further support this ecosystem, Innatera is launching its developer program, now open to early adopters. An upcoming open-source PyTorch frontend and marketplace will create a collaborative ecosystem for neuromorphic AI.
