New breakthrough in the semiconductor industry: emerging industries are expected to usher in explosive growth

Highlights

Automotive semiconductors and artificial intelligence chips have brought new development opportunities to the semiconductor market.

Currently, the technologies behind many cutting-edge digital devices we use rely on semiconductors. Due to the development of emerging technologies such as driverless cars, artificial intelligence, 5G and the Internet of Things, as well as continued investment in technology research and development and fierce competition among major market players, the global semiconductor industry is expected to continue to grow steadily in the next decade.

Mergers and acquisitions in the semiconductor industry have reached their peak, and professional vertical integration has gradually become the focus of the industry. Japan and South Korea are trying to revitalize their semiconductor industries through acquisitions, while at the same time, the ongoing trade war and intellectual property disputes will hinder China's large-scale investment around the world. As demand for consumer electronics saturates, the growth of the semiconductor industry will slow down. However, many emerging fields will bring ample opportunities to the semiconductor industry, especially semiconductor applications in automobiles and artificial intelligence.

The semiconductor industry landscape continues to evolve

In the past few years, the growth of the global semiconductor industry has mainly relied on the demand for electronic devices such as smartphones, as well as the expansion of technology applications such as the Internet of Things and cloud computing. It is expected that the total revenue of the global semiconductor industry will increase from US$481 billion in 2018 to US$515 billion in 2019, and the growth trend is expected to continue into the next decade.

▲ Global semiconductor industry sales revenue (2016-2022, unit: billion US dollars)

Automotive electronics and industrial electronics will become the two fastest growing areas in the semiconductor industry, and revenue from consumer electronics, data processing and communications electronics will grow steadily.

▲ Growth rate of semiconductor revenue for each category of electronic equipment (2017-2022)

Asia Pacific will remain the world's largest semiconductor consumer market. The increase in the share of Chinese products is stimulating the growth of the entire Asia Pacific market and will provide a major impetus. In addition, the increase in merger and acquisition activities will be beneficial to the future development of the semiconductor industry.

▲ Regional analysis of semiconductor industry sales (2018)

In terms of growth, the US market grew fastest in 2018, mainly due to the rise of dynamic random access memory and high demand for microcontroller units, especially in the storage device market. As memory prices rise and contribute huge profits, the memory market is developing rapidly, and the Asia-Pacific region has benefited from this. The mainland's integrated circuit industry grew by 24.8%, which has strongly promoted the development of the Asia-Pacific regional market.

However, although the competitiveness of Chinese semiconductor manufacturers has been significantly improved in recent years, key components still need to be imported from Western countries in large quantities, and the self-sufficiency rate is less than 20%. The Chinese government is very concerned about this issue and has formulated a number of favorable policies to support the development of the semiconductor industry.

▲ Major players in China’s semiconductor industry

In general, there are four types of companies in China's semiconductor industry: the "national team", the "local team", private equity/venture capital funds and multinational companies, competing to make China the driving force of the global semiconductor industry.

Breakthrough in automotive semiconductors

The automotive industry has gone through a long period of development to achieve the front-end installation of automotive electronics with safety and comfort as the core. As early as 2004, only a quarter of the vehicles shipped had built-in airbags, and less than 50% of vehicles were equipped with front-mounted power seats. However, driven by government regulation and consumer demand, safety-related electronic systems have quickly become popular. Today, most innovations in the automotive industry occur at the electronic system rather than the mechanical level. Between 2007 and 2017, the proportion of automotive electronics costs increased from about 20% to about 40%.

▲ Automotive front-end electronics

▲ The automotive industry has gone through a long period of development before it has realized the pre-installation of automotive electronics with safety and comfort as the core.

▲ The proportion of electronic systems in the total cost of a car (%)

Semiconductor costs (i.e., the cost of electronic system components) have increased from $312 per vehicle in 2013 to about $400 today. Automotive semiconductor suppliers are benefiting from the sharp increase in demand for various semiconductor devices such as microcontroller units, sensors, and memory. By 2022, semiconductor costs are expected to reach nearly $600 per vehicle.

▲ The proportion of automotive electronics and semiconductors in each vehicle cost

Semiconductor suppliers play a vital role in the automotive industry supply chain. In the traditional automotive industry ecosystem, semiconductor suppliers sell their products to Tier 1 electronic system suppliers, who then integrate the technology into modules and hand them over to vehicle manufacturers for assembly. In recent years, the automotive industry has undergone tremendous changes, and the ecosystem will be completely transformed in the next few years.

The automotive industry will be reshaped by the development of technologies such as artificial intelligence, electric vehicles, autonomous driving, energy storage and cybersecurity; social awareness of topics such as safety and shared mobility; concerns about environmental issues such as pollution; economic considerations such as infrastructure spending; and the growth of the Asian market.

▲ The role of semiconductors in the automotive ecosystem

Automation, electrification, digital connectivity and safety: These four major trends will drive the continued growth of semiconductor components in automotive electronics and subsystems over the next decade.

▲Main trends in automotive semiconductors

Although mobile phones are the largest market for semiconductor companies now and in the future, the growth of this field has been very saturated for many years. The automotive semiconductor market is an exception. As electronic components such as advanced driver assistance systems and in-vehicle infotainment are increasingly used in automobiles, this field has strong demand and has become an important growth market for semiconductor companies.

▲ Automotive semiconductor revenue and output in various regions around the world

▲ Automotive semiconductor applications and equipment growth forecast

The race for artificial intelligence chips begins

The AI ​​framework can be roughly divided into three levels. The infrastructure level includes core AI chips and big data, which are the basis of sensing and cognitive computing capabilities at the technical level. The application level is at the top level, providing services such as unmanned driving, intelligent robots, smart security and virtual assistants. AI chips are the core of the AI ​​technology chain and are crucial to AI algorithm processing, especially deep neural networks.

"Depth" refers to the number of layers and nodes in a neural network model. In recent years, the complexity of the layers and the number of nodes have grown exponentially, which has posed a great challenge to computing power. Although traditional central processing units are relatively good at handling general workloads, especially those based on certain rules, they are now unable to meet the parallel computing requirements of artificial intelligence algorithms.

▲ The role of AI chips at different levels of AI

There are two main ways to solve the problem of parallel computing: first, add dedicated accelerators to the existing computing architecture; second, completely redevelop and create a new architecture that simulates the human brain neural network. The second method is still in the early stages of development and is not suitable for commercial applications. Therefore, the main method currently used is to add artificial intelligence accelerators. Various types of artificial intelligence chips can achieve acceleration. Mainstream accelerators include graphics processors, field programmable gate arrays, and application-specific integrated circuits, which include variants such as tensor processors, neural network processors, neural network processors, vector processors, and brain processors. Each artificial intelligence chip has its own advantages and disadvantages.

Deep learning has two completely different ways of deploying AI: training and inference. AI “trains” neural network models based on big data, using the training data set to obtain newly trained models. These newly trained models are then given new capabilities to “infer” conclusions based on new data sets.

Because a huge data set needs to be applied to the neural network model, a lot of computing power is required in the training phase. This requires high-end servers with advanced parallel computing capabilities to process a large number of highly parallel data sets of various types. Therefore, this phase of work is usually completed using cloud hardware devices. The inference phase can be completed in the cloud or with the help of edge devices (products). Compared with training chips, inference chips need to consider power consumption, latency, and cost more comprehensively.

▲ Two major stages of deep learning

AI chip innovation is just getting started, and vendors are taking different approaches to chip acceleration. For example, Google has chosen the ASIC route, while Microsoft has demonstrated that field-programmable gate arrays can achieve comparable or better results. Meanwhile, Xilinx, Baidu, and Amazon are all working to reduce the traditional barriers to using ASICs.