SPU-001 (Sparse Processing Unit)

SPU-001 is an ultra-efficient AI accelerator designed for real-time neural network inference on edge devices. By exploiting unstructured sparsity in both weights and activations, SPU-001 supports models up to 10× larger within the same memory footprint—while achieving up to 100× greater power efficiency than conventional accelerators. Typical workloads run at ~1mW, enabling all-day battery life for even the most demanding always-on applications. Get started immediately with our production-ready models for speech enhancement, voice commands, and more—or bring your own model and compile it with our developer-friendly SDK.

Femtosense brings intelligence to everyday devices with our Embedded AI Platform, featuring an ultra-efficient AI accelerator seamlessly integrated with an end-to-end embedded AI stack. By leveraging neuromorphic computing principles—particularly sparsity—we enable low-latency AI processing with dramatically lower energy consumption and cost. From wearables and household appliances to robotics and autonomous vehicles, Femtosense is making AI more efficient, accessible, and ubiquitous.