Tong Wu, Arm China: ASIC customized chips may become the future chip architecture trend

Today, the industry has entered the era of data-driven computing, also known as the fifth wave of computing, which has created a huge demand for everything from network architectures to computing architectures. Relying on ARM's world-leading ecosystem resources and technological advantages, Arm China has independently developed "Zhouyi" AIPU for the domestic market.

Wu Tong said that there are four development trends in the current AI chip market, one is the very high growth rate of the end-to-side chip market, the second is the fastest growth rate of the end-to-side reasoning market in the next 5-10 years, the third is that ASIC customized chips will become the mainstream in the future, and the fourth is that the market segment size will maintain a high growth rate.

On the other hand, the AI chip industry specific architecture (DSA) is on the rise, and mainstream AI algorithms are also showing the trend of lightweight. Under these trends, the "Zhouyi" AIPU developed by Armou China has the characteristics of complete autonomy and control and complete ecology. So far, it has launched two generations of products, namely "Zhouyi" Z1 and "Zhouyi" Z2.

Among them, "Zhouyi" Z1 is the general AI IP of edge computing, facing IOT&EDGE, based on "Zhouyi" Z1 research and development of the whole Zhi R329 intelligent voice chip is about to be large-scale commercial; "Zhouyi" Z2 is oriented to high-performance scenarios in edge computing, and the chip developed based on "Zhouyi" Z2 is about to be applied to the ground, mainly covering mid and high-end security and automatic driving/intelligent cockpit fields.

Wu also introduced two AI IP applications, including human body critical point detection, driver fatigue monitoring (DMS) and DTV super resolution applications.


1. The fifth wave of computing may drive the shipment of ARM chips to exceed one trillion

Tong Wu: First of all, I'd like to thank the organizer, Intelligent Things, for inviting me. My name is Tong Wu, AI Technology Market of Armou China. Today, I would like to share with you the theme of "Building AI Intelligent" Core Ecology ". My theme is divided into three parts. The first part is to introduce the enabling situation of the whole ARM architecture, the second part is a simple analysis of the AI chip and AIOT market, and the last part is the market and technical situation of the two generation products of AI IP "Zhouyi" developed by Armou China.

First of all, there's a concept called the fifth wave of computing, and you know back in the beginning when we had the PC, the personal computing, the Internet, and the mobile Internet, you could call it the first four waves of computing. What is the fifth wave of computing? In a word, a data-driven computing era. The main characteristic of the fifth wave of computing is that in addition to the diverse demand at the chip level, there is a huge demand for network architectures, including computing architectures. As an ecosystem company of ARM, we have been thinking about how to enable the whole ecosystem by providing more customized services with our partners in this system, including architecture and chip-based innovation.

TSMC 4nm trial production ahead of schedule! Push the automobile special new manufacturing process, the United States 5nm chip factory has started construction

This is a map of our entire ARM computing architecture from 1991 to the present as it moves toward a trillion-scale ecology. As you can see, from 1991 to 2017, it took 26 years for (ARM) to reach the first 100 billion chip shipments in the world. From 2017 to 2021, it actually took us only four years to get to the second 100 billion chips shipped, which is a very fast growth. Driven by the wave of fifth-generation computing, we hope to ship a trillion chips in the future.

From the beginning of its establishment to the present, Arm China has served more than 200 customers in China and shipped nearly 20 billion (chips) in China within a few years.

Second, the rapid growth of end-to-side AI chips and the rise of domain-specific architectures

Here's a look at AI chips and the AIOT market as a whole. Here are four charts that give an overall introduction to AI chips from the perspective of market size, scene, architecture and application layer.

1. AI chip development trend: the fastest end-to-side growth

First of all, the first chart shows the market size of the entire AI AI chip. The period from 2017 to 2025 is nearly ten years, and there is a trend. As can be seen from the red number, compared with the cloud, the market growth rate of the endside AI chip is very high, especially in the next five to 10 years.

The second chart shows the trend of AI chips divided by scene. The scene is divided into four parts, two parts. The first dimension is the whole cloud, including cloud reasoning and cloud training. The second dimension, EDGE end-to-side, includes reasoning and training. This chart concludes that in the next 5-10 years, we can see the fastest growth in the whole endside reasoning market, including cloud and Edge side reasoning.

The third diagram shows that the AI chip is divided by architecture. We can see all the mainstream architectures seen in the AI chip, including GPU, FPGA, ASIC and so on. It can also be seen that ASIC customized chips will become a mainstream in the market in the next five to ten years.

The last chart shows the scene of vertical field of artificial intelligence subdivision. From this, we can see that mobile terminal, needless to say, is also a mainstream industry of ARM, and mobile terminal still maintains a relatively high growth rate. Several market segments, such as wearables, including smart speakers, are expected to maintain relatively high growth rates in the next few years.