What does “bionic chip” in Apple’s new iPhone mean?
Bionics is a scientific method of simulating the function and behaviour of biological systems to construct technical systems. It breaks the boundaries between biology and machines and communicates different systems.
From this perspective, artificial intelligence chips can be counted as bionics. I think that the current artificial intelligence is still only in the artificial stage, the real intelligence should be self-learning ability (do not need to rely on big data training, will learn independently). I think the new brain-like chip is a true bionic chip.
First of all, Apple A12 is not a bionic chip (sounds a bit higher), it is actually the AI operation acceleration kernel added to the mobile phone SoC. Apple A11/A12 integrates independent processing units (independent cores) for neural network computing, which is the NPU that everyone often hears (different vendors have different methods).
Generally, in the SoC of a mobile phone, the CPU is mainly responsible for some general-purpose calculations, can handle various complicated tasks, and has strong adaptability. For example, it can also handle graphics tasks and perform artificial intelligence calculations, but it is less efficient than GPU and NPU.
The GPU is mainly responsible for processing graphics tasks. Its advantage is that it can support large-scale parallel computing architecture, which is very suitable for some simple and repetitive computing tasks.
It is currently used by many manufacturers for deep learning and data model training. Although large-scale deployment is powerful, the power consumption is also very large. Therefore, it is mainly used in the field of artificial intelligence computing where power consumption requirements are not very high.
The AI chip that everyone is talking about is actually a dedicated ASIC chip. Its AI computing power and computational efficiency are directly customised according to the needs of specific AI algorithms, so it can achieve small size, low power consumption, high reliability, high confidentiality, high computing performance, and calculation with respect to CPU and GPU.
High efficiency and other advantages. Therefore, at present on the terminal side (especially mobile terminals) AI calculation, many manufacturers have begun to choose to use a dedicated AI chip, or integrate the NPU core in the SoC.
Secondly, looking back at the integrated NPU kernel in Apple A11/12, this is actually the hardware of some specific neural network types, namely AI algorithms (such as voice/picture recognition, face recognition, etc.). Acceleration for these AI algorithm operations. For example, the original image recognition can only recognise one hundred images in one minute, and the NPU acceleration can realise thousands of images in one minute, which is faster and consumes less power.
However, NPU also has its shortcomings. Once the design is completed, the corresponding AI algorithm cannot be changed and cannot be upgraded. But it can work with the CPU/GPU/DSP in the mobile phone SoC to perform AI calculations (the AI neural network engine of Qualcomm Snapdragon 845 is to coordinate the AI calculation by deploying CPU/GPU/DSP), so it can make up to some extent. The lack of flexibility.
At present, the AI capabilities that can be provided on mobile phone chips are mainly for some applications such as voice, picture, face, and AR. In addition, the neural network engine also has certain machine learning capabilities. For example, you can unlock your face every day, so AI also learns some changes in your face (the face of a person will change due to aging).
In general, we should look at the A12 chip as a whole, and its AI capabilities are not only provided by the NPU. In some scenarios, other CPUs such as CPU/GPU will also participate in the AI operation. The role of the NPU is also primarily accelerated for specific AI algorithms.
The 1 minute Knowledge Kit is a 36-day, more question-and-answer forum. It is designed to be limited to one minute per day. Ask quickly to answer an important business question. Today we are answering the chip related issues. If you have any questions about the recent business world, please leave a message in the comments section, and the kit is responsible for finding the answer for you.