# Hardware Accelerator

For higher-end devices like drones, service robots, industrial cobots, the Idle Network stack can include the Edge-AI Accelerator IP:

* IP Core licensed to OEMs → integrated alongside their main SoC.
* Provides 140 FPS inference throughput @ 50 W, nearly 2× faster than equivalent GPU at 60% less power.
* Optimized for matrix multiplications (GEMM ops) common in CNNs (Convolutional Neural Nets).
* Interfaces with the Idle Compiler to accept quantized model binaries.

This gives OEM devices a dual purpose: they perform their primary function and generate financial yield through Idle tasks when idle.

<br>