# Neural processing unit

Source: [https://docs.qualcomm.com/doc/80-PV086-5P/topic/neural-processing-unit.html](https://docs.qualcomm.com/doc/80-PV086-5P/topic/neural-processing-unit.html)

The NPU230 is the dedicated neural processing unit for performance and always-on neural network (NN) use cases. The network execution engine consists of:
      
- 4096 8-bit and 1024 16-bit fixed point match – scalable for AI compute
- Generates complex address patterns to fetch data for MACs
- Native support for SNPE, Future Android NN, and Caffe 2 ONNX

      It has a tightly coupled memory (256 KB) with built-in support for NPU-centric load/store operations and synchronization with the DMA. The NPU230 has a 3D DMA controller for DNN, which supports 3D data structures movement, streaming DMA, gather/scatter, padding/cropping in any dimension.

**Parent Topic:** [SoC subsystems](https://docs.qualcomm.com/doc/80-PV086-5P/topic/SoC-subsystem.html)

Last Published: Jul 07, 2023

[Previous Topic
Compute DSP with Quad HVX and HCP 1.1](https://docs.qualcomm.com/bundle/publicresource/80-PV086-5P/topics/compute-dsp-with-quad-hvx-and-hcp.md) [Next Topic
Connectivity](https://docs.qualcomm.com/bundle/publicresource/80-PV086-5P/topics/connectivity.md)