Know About Neural Processing Unit (NPU)




How much do you know about NPU? Let's introduce you with NPU and it's benefits and use of NPU in the mobile phone.

NPU stands for Neural Network Processing Unit. "Neural network" may sound a bit confusing, but it seems to be more familiar with "Artificial Intelligence". NPU is the hardware on the phone that specialises in deep learning of artificial intelligence. Reading this, many of us may think of the first NPE on the Qualcomm Snapdragon 820. Indeed, NPE and NPU have similarities. NPE stands for “Neural Processing Engine”. Both NPE and NPU can handle similar tasks. But NPE is the software layer and NPU is the hardware level. If we choose a computer as an example, the graphics card is its image processor, which is hardware and frost, illusion, EGO, etc. are all game engines still software. Of course, Qualcomm's NPE is still done with the help of CPU and GPU. Because neural network algorithms and machine learning need to involve a huge amount of information processing, and current CPU / GPU cannot achieve such efficient processing capabilities, it is necessary to have a separate processing chip to do this.




ARM released the Cortex-A75 and A55 CPUs and the Mali-G72 GPU design earlier this year, all with more advanced machine learning algorithms, but not through the NPU. The key to ARM's ability to enhance SoC machine learning is the Computer Library, which includes a full set of capabilities for imaging and visual projects, as well as machine learning frameworks such as Google TensorFlow, which developers can use to recompile their needs. version.
In addition to ARM, Qualcomm also has its own Hexagon SDK, which also includes a general matrix algorithm for machine learning, which enables more efficient DSP operation efficiency. In addition, the Qualcomm Symphony System Manager SDK provides an API that optimises computer vision, image and data processing, and low-level algorithms, as well as the common processing needs of smartphones.

Benefits -


The NPU can help the phone to recognise the shooting scene more accurately and quickly. Let the phone choose the most appropriate image processing algorithm, make the phone more accurate to the edge blur when the double camera background is blurred, and the recently very hot AR camera. NPU can significantly improve the speed of rendering and reduce power consumption. For example, with the NPU the mobile phone can transfer part of the voice and semantic recognition work to the mobile phone local, improve the performance of voice and semantic recognition and also improve the experience of voice interaction applications (such as voice input method, intelligent voice assistant etc) in mobile phones.

NPU Usually Used in Samrtphones for -


User Behaviour Learning: Analyse the user's usage habits by knowing where users are often doing things, in order to automatically implement certain functions for users in certain scenarios after a period of learning. Thereby optimising the memory storage of the mobile phone and making system resource optimization in time.
Image recognition: Image recognition is perhaps a term that has often been heard recently. In the past, our mobile phone couldn't know what was in the picture, except for our face. Now, with the help of NPU, the phone can know where you took a photo? After a long period of learning the habit of taking pictures, the mobile phone can analyse the shooting scene in real time during the shooting process, and set the camera parameters for different scenes, so as to achieve good photos.