Brainchip offers the Akida neuromorphic AI platform, a sensor‑agnostic, host‑flexible processor IP that performs event‑driven inference at milliwatt power levels. By leveraging sparsity and Temporal Event‑Based Neural Networks, Akida enables real‑time detection, classification, and on‑chip learning for battery‑powered edge devices such as wearables, IoT sensors, radar and LiDAR without requiring constant cloud connectivity.
Funding
$21.5M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
Edge devices and sensors often require continuous AI inference, but traditional neural network processors consume watts of power, generate heat, and need constant cloud connectivity, making always‑on intelligence impractical for wearables, IoT, radar, and other battery‑powered applications.
Solution
BrainChip’s Akida platform provides a sensor‑agnostic, host‑flexible neuromorphic processor IP that leverages sparsity at the data, weight, and activation levels to execute AI only when meaningful events occur. By converting streaming inputs to events and pruning unnecessary parameters, Akida runs convolutional, recurrent and spatio‑temporal models in the milliwatt range, enabling real‑time detection, classification, and adaptation directly on the device. The solution includes a full development stack—model conversion tools, pre‑trained model zoo, on‑chip learning, and cloud‑based Akida Cloud for remote benchmarking—so developers can deploy always‑on AI without redesigning hardware or relying on external servers.
Target Audience
Primary customers are OEMs and system integrators building battery‑powered edge products such as wearables, smart sensors, radar/EFW systems, LiDAR, audio processors, and industrial IoT devices that require low‑latency, always‑on AI.
Features
- Digital neuromorphic NPU core optimized for sparse data, weights, and activations, reducing compute and memory by up to 10×
- Event‑driven processing that fires only when input thresholds are crossed, achieving continuous inference at milliwatt power levels
- Supports CNNs, DNNs, RNNs, spatio‑temporal CNNs, State‑Space Models and Temporal Event‑Based Neural Networks (TENNs) for motion, audio, vision and radar workloads
- On‑chip learning capability for personalization and adaptation without cloud connectivity
- Integrated development ecosystem: MetaTF conversion tools, pre‑trained model library, SDK with simple API, and Akida Cloud for remote model evaluation and benchmarking
- Flexible integration options: ASIC IP, FPGA prototypes, and reference development boards (AkidaTag, Akida Pico) for rapid prototyping
- Built‑in privacy: all inference runs locally, only model weights are stored, eliminating data transmission requirements