Artificial intelligence belongs to one of the advance technologies determined in the present time. It enables the machines to learn, reasoning, think, act, and solve the situations as just same as humans do. Artificial intelligence spread to vast area of fields like marketing, trainings, modern sports education, brain inspired cognitive AI, AI in medical (prediction of post mortem Interval, etc.), Artificial Intelligence of things wearable systems for cardiac detection and many more .
So, AI is defined as intelligence demonstrated by machines, unlike the natural Intelligence displayed by humans and animals. Generally, if we talk about Artificial Intelligence, it can be classified in two ways- Based on capabilities and based on functionality.
Table of Contents
Classification of Artificial Intelligence on the Bases of Capability and Functionality
Capability Bases | Functionality Bases |
Narrow or Weak AI (mimic like a humans) | Reactive Machines |
General AI (Just like a humans) | Limited Memory |
Strong or Superior AI (Superior than Humans) | Theory of Mind |
Self-Awareness |
As for the implementation, there are some resources on which a technology build. If we talk about Artificial intelligence, the prime resource is a hardware on which AI get stable after it the next thing is ANN (Artificial Neural Network) and another resource that makes it advancement for the upcoming future is Deep Learning which is a subset of ANN or machine learning.
Neural Processing Unit (NPU)
The hardware on which AI resides is NPU (Neural Processing Unit) or Neural Processor. It is a neural processing engine which is a specialized circuit that implements all the necessary control and arithmetic logic necessary to execute machine learning algorithms, typically by operating on predictive models such as Artificial Neural Networks (ANNs) or Random Forests (RFs). However, NPU is classified in two categories that is training and inference.
Processors Equipped with Integrated or dedicated NPUs
Brand | Processor name | N.P.U. (Neural Processing Unit) | Process technology | Launching dates |
Qualcomm | Snapdragon 865+ | Dedicated Quadcore Hexagonal tensor NPU accelerator | 7nm | 8 July 2020 |
Snapdragon 865 | Dedicated Quadcore Hexagonal tensor NPU accelerator | 7nm | March 2020 | |
Snapdragon 765 | Hexagon 696 HVX + Tensor 5.4TOPS AI | 7nm | 4 December 2019 | |
Mediatek | Dimesity 800 | Integrated quadcore NPU | 7nm | March, 2020 |
Dimesity 820 | Integrated mediated APU 3.0(tri core N.P.U.) | 7nm | 2nd quarter of 2020 | |
Dimesity 1000 | Integrated APU 30 hexacore NPU | 7nm | 23 April 2019 | |
Samsung | Exynos 990 | Dual core NPU | 7nm | 2020 |
Exynos 9825 | Integrated NPU | 7nm | 2019 | |
Exynos 980 | Integrated NPU | 8nm finFET | End of 2019 | |
Apple | A13 Bionic Chip | Octa core Neural engine | 7nm | 10 Sept 2019 |
A12 Bionic chip | Octa core Neural engine | 7nm FinFET | 12 Sept 2018 | |
A11 Bionic chip | Dedicated neural network | 10 nm FinFET | 12 Sept 2017 | |
Huawei | Hisilicon kirin 990 | Dedicated dual core daVinci NPU | 7nm | 6 Sept 2019 |
Hisilicon kirin 980 | Dedicated dual core NPU | 7nm | 31 August 2018 | |
Hisilicon kirin 970 | Dedicated NPU | 10nm | 1 Sept 2017 |
Training NPUs
Training NPUs are used for the training or preparing purpose for the new models. It works by putting the values in the algorithms by guessing the correct values. It guess the different values or data until unless it gets perfectly fits for the convenient, more correct and smooth working of the algorithms designed for accelerating the training of new models.
Inference NPUs
Inference NPUs are utilized to implement inference operations on complete trained models. Infact, these are used to accelerate the input of a piece of data (a picture or image), after which it gets processed by other trained models in order to get desired results.
Artificial Neural Networks (ANNs)
Now one of the subsets of AI is the Artificial Neural Network (ANNs) sometimes also called as machine learning (ML). It is related to a part of NPU and works as an application software for it. Basically, ANN is a subset of Artificial intelligence which is programmed in NPU circuit as an application which has various interconnected nodes to pass the logical inputs and manipulates them according to the situation with the help of mathematical algorithms.

These algorithms are designed to work just like the neurons working in a brain of a human or animal. In short ANN is defined as computing systems vaguely inspired by the biological neural networks that constitute animal brains. Now if we go in advance ANN then it is known by a term that is Deep Learning (DL).
Deep Learning (DL)
DL is a subset of machine learning or ANN in artificial intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured and un-labelled. It sometimes also known as Deep neural learning or deep neural network.
DL is an advanced neural processing of ANN in which random data inputs are arranged in a hierarchical format to predict the possible situations just like the natural predictive brain of humans. It uses the hierarchical pattern to detect speech recognition, face detection, finding anomalous and suspicious activities by fraudulent, etc.

According to this study, deep learning is a subset or collaborated to artificial neural network (ANN) and ANN or machine learning is a subset of artificial intelligence (AI) and AI resides (collaborated) on NPU as a physical existence. All these things are related to each other to mimic and solve the real-world situations and problems just like a human brain.