Also called an "intelligent processing unit" (IPU), "neural accelerator" and "neural engine," a neural processing unit (NPU) is a set of circuits or separate chip that accelerates AI processing, especially for the user (inference). See
AI training vs. inference.
The NPU performs matrix multiplication especially for desktop and mobile devices, adding AI computing power that is normally in AI data center servers. See
GPU,
AI PC and
AI data center.
In 2016, Qualcomm introduced its Zeroth Machine Intelligence Platform and Snapdragon Neural Processing Engine that enabled mobile devices to execute AI applications on their own rather than going to a data center in the cloud. See
neural network and
neuromorphic computing.