Not long ago, the Debon Securities Research Report compared the year-on-year growth rate of global semiconductor sales since 1976 with the GDP growth rate and found that the historical growth rate of global semiconductor annual sales showed an “M” shaped fluctuation characteristic of about every 10 years, and was mainly related to the iteration of technology-driven products.

(Source: Debon Securities Research Report)

If you look further at the development of semiconductor manufacturing technology over the years, you can also find that it has risen to a new level with each decade of change.

From the era of household appliances to the desktop era, to feature phones/laptops, 3G/4G/smartphones, 5G, AIoT/consumer electronics/automobiles. With the continuous expansion of downstream application fields,The semiconductor industry has shown continued growth in global sales, with sales increasing a full 202 times over the period from 1976 to 2022.

Here's a simple general truth revealed:The essence of a cycle is not a simple step of ups and downs, but a continuous spiral of upward development.

(Source: Debon Securities Research Report)

2023 also happens to be the first year of the AI explosion, which undoubtedly indicates that the semiconductor industry has also ushered in a new era, and is also seen as a restart of a new cycle — the explosion of AI technology places higher demands on chip demand, while also bringing unprecedented opportunities to the industry.

With the hype boom set off in the capital market, AI chips have also become popular in capital pursuit.

However, in reality, not everyone has an accurate understanding of AI chips. Next, we will thoroughly analyze the different types of AI chips to reveal their unique role, technical characteristics, and related companies involved in them.

Machine learning algorithms are generally divided into two main steps in actual application: training (training) and inference (inference). These two steps correspond to different computational requirements and chip designs, so two main types of AI chips have been derived:Training chip and inference chip.

Judging from the technical architecture, AI chips are mainly divided into four categories: graphics processing units (GPUs), field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), and central processing units (CPUs).

Furthermore, according to its location on the network, it can also be divided into categories“Cloud-edge-end” is divided to obtain cloud-based AI chips, edge, and terminal AI chips.

(Source: Public Material)

CPU: It is the core component of a computer system and is responsible for performing various computational tasks and algorithms. CPUs are widely used in AI applications. It can be used for training and reasoning tasks, including image recognition, speech recognition, natural language processing, etc. The CPU's high performance and flexibility make it an ideal choice for processing all types of data.

GPU: Also known as a display core, visual processor, and display chip, it is a type of microprocessor that specializes in image and graphics-related computation on personal computers, workstations, game consoles, and some mobile devices (such as tablets, smartphones, etc.).

FPGA: It is a product of further development based on programmable devices such as PAL (Programmable Array Logic) and GAL (General Array Logic). It appeared as a semi-customized circuit in the field of application-specific integrated circuits (ASICs). It not only solved the shortcomings of customized circuits, but also overcame the shortcomings of the limited number of gate circuits in the original programmable device.

ASIC: An integrated circuit designed and manufactured for specific user requirements and specific electronic system requirements. Currently, CPLDs (complex programmable logic devices) and FPGAs are mainly designed using CPLDs (complex programmable logic devices) and FPGAs. They all have user-field programmable features and support boundary scan technology, but the two have their own characteristics in terms of integration, speed, and programming methods.

ASIC applications include, but are not limited to, smart phones, wearable devices, security front-end cameras, smart home devices, and drones. whereasAmong domestic ASIC design and manufacturing companies, manufacturers such as Cambrian, Yuntian Lifei, Aerospace Micro, and ICG (Smart Chain Group) all occupy considerable market share in various segments. Some are even global leaders, and are worthy of in-depth exploration and continuous tracking.