Specifications
Product
Innatera Pulsar — Spiking Neural Processor
Compute Architecture
Heterogeneous: event-driven spiking neural network (SNN) fabric + CNN accelerator + RISC-V CPU
Compute Paradigm
Neuromorphic, asynchronous event-driven processing
Neural Model Type
Spiking Neural Networks (SNN) with conventional CNN and DNN support
Host CPU
Integrated RISC-V core for system management and control
Package Footprint
2.8 x 2.6 mm
Power Envelope
Milliwatt range; sensor data processing down to micro- and nano-watt ranges (vendor figure)
Energy Efficiency Claim
Up to 500x lower energy than conventional edge AI (vendor figure)
Latency Claim
Up to 100x lower latency than traditional pipelines (vendor figure)
Response Time
Sub-millisecond, event-driven inference at the sensor
SDK and Toolchain
Talamo SDK — imports TensorFlow and PyTorch models, supports native SNN model creation in Python
Deployment Flow
Train in standard frameworks, tune SNN behaviour, deploy to Pulsar
Data Privacy Model
Fully on-device processing — no cloud dependency
Target Applications
Wearables, IoT, industrial sensing — human detection, gesture recognition, environmental sensing
System Benefit
Application processors stay asleep until needed; sensing intelligence runs continuously
Price
Quote upon request
Availability
Available
Overview
Innatera's Pulsar is a neuromorphic microcontroller built for continuous, always-on sensing at the sensor itself. Rather than waking an application processor for every sample, Pulsar runs an event-driven spiking neural network fabric that only computes when input signals change — the same asynchronous behaviour that lets biological neurons stay quiet between events. That is the basis for the vendor's stated efficiency figures: up to 500x lower energy than conventional edge AI, response times up to 100x lower than traditional pipelines, and sensor processing in the micro- and nano-watt range.
Pulsar is heterogeneous by design. Alongside the spiking compute fabric it integrates a conventional CNN accelerator for standard AI workloads and an RISC-V CPU for system management and control, so end-to-end sensor data processing can happen on one chip instead of a sensor plus MCU plus accelerator chain. The whole device occupies a 2.8 x 2.6 mm footprint, which fits inside sensor housings rather than beside them.
Development uses Innatera's Talamo SDK, which imports existing TensorFlow and PyTorch models, supports building native SNN models, and deploys to the device through standard Python workflows — an important detail for teams without specialist neuromorphic expertise, since the toolchain accepts models they already know how to train.
The practical result for product designers is always-on capability without the battery penalty that normally forces duty cycling: human detection, gesture recognition and environmental sensing can run continuously on a battery-powered wearable or a remote industrial sensor node, with data never leaving the device.
Key Benefits
Event-driven spiking fabric that computes only on change. Heterogeneous single chip — SNN + CNN accelerator + RISC-V core. 2.8 x 2.6 mm package that fits inside sensor housings. Milliwatt envelope enabling always-on sensing on batteries. Talamo SDK accepts existing TensorFlow and PyTorch models.
Applications
Always-on human presence detection in wearables and smart buildings; gesture recognition for consumer and industrial interfaces; battery-powered remote sensing nodes; predictive maintenance vibration monitoring; smart home and appliance sensing; privacy-preserving on-device sensor intelligence.
Request a Quote — INNATERA PULSAR — NEUROMORPHIC SPIKING NEURAL PROCESSOR FOR THE SENSOR EDGE
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →