AI has a significant impact on the design and manufacturing of the integrated circuit industry.

In today's rapidly evolving technological landscape, the rapid development of artificial intelligence (AI) technology is profoundly changing the face of various industries. Among them, the integrated circuit industry, as the core field of information technology, is on the cusp of a profound transformation driven by AI.

Today, at the 2024 International Integrated Circuit Exhibition and Symposium (IIC Shanghai), Professor Shi Longxing, a professor at the School of Integrated Circuits at Southeast University and the head of the expert group of the China Integrated Circuit Design Innovation Alliance, presented his thoughts on how AI is affecting China's integrated circuits, sparking widespread attention and lively discussion in the industry.

Professor Shi Longxing pointed out that the most notable impact of AI on integrated circuits is that with the development of large AI models, the demand for computing power is growing explosively. However, the energy efficiency improvements brought about by the miniaturization of integrated circuit devices are gradually slowing down, making it difficult to meet the increasing demand for AI computing power. This undoubtedly poses unprecedented challenges to the integrated circuit industry.

The impact of AI on China's integrated circuit industry:With the development of artificial intelligence, humanity is entering the AI era, where AI has empowered a thousand industries.

Advertisement

At present, companies worldwide that focus on AI as their main product are experiencing rapid growth. The top ten AI chip companies globally include: NVIDIA, Intel, NXP, IBM, AMD, Google, Arm, Apple, Qualcomm, and Broadcom. The top ten AI chip companies in China include Cambricon, Yunfei Litian, Sogou, Baidu, Unisound, Bitmain, Unigroup, DeepGlint, Horizon Robotics, and TianShu Zhixin. In addition, some traditional companies are also integrating AI elements.

Professor Shi Longxing pointed out that with the development of large AI models, the demand for computing power is growing exponentially. However, the energy efficiency improvements brought about by the miniaturization of integrated circuit devices are gradually slowing down, making it difficult to meet the increasing demand for artificial intelligence computing power. In this situation, in addition to general-purpose AI chips, some new types of computing architectures are being looked to with hope.

Analog Computing Architecture

Analog computing architecture is one of the new architectures that are highly regarded. Nowadays, analog computing architecture is a research hotspot at conferences such as ISSCC, IEDM, and ISCA, and it is also a focus of layout for some major semiconductor manufacturers (TSMC, Samsung).

In practical applications, the energy efficiency bottleneck of traditional digital computing AI chips based on advanced processes is improving slowly, while analog computing AI chips, although immature, have already demonstrated their significant energy efficiency potential.

The following are some issues that analog computing AI chips are still facing:

Firstly, analog computing finds it difficult to meet diverse network requirements. This is because the design concept of AI chips is complex, while the rule-based concept of analog computing is relatively strong, leading to the difficulty in satisfying diverse network demands.

Secondly, single-paradigm analog computing finds it hard to balance accuracy, energy efficiency, and area efficiency simultaneously.

Thirdly, the energy consumption of peripheral circuits such as analog-to-digital conversion is emerging as a new bottleneck for analog computing.Professor Shi Longxing stated that their team's simulation and computational research has been underway for many years, with the research content mainly including the integration of algorithmic architecture, array circuits, peripheral circuits, and demonstration systems, among others.

Computing-in-Memory (CIM)

The second promising new architecture is Computing-in-Memory (CIM). As is well known, the CIM architecture can significantly reduce the limitations of the storage wall and power wall. In-memory computing architectures can operate with high computational and transmission efficiency, making them particularly suitable for tasks that require high computational power.

In in-memory computing, there are many technological paths. Professor Shi Longxing focused on comparing non-volatile in-memory computing chips with SRAM in-memory computing chips. Compared to non-volatile devices, SRAM-based in-memory computing has strong process adaptability, high configurability, and high reliability, making it a more appropriate solution for realizing high computational power.

The research content of in-memory computing by Professor Shi Longxing's team mainly includes CIM macro units, AI accelerators, compiler design, and SOC chip development.

In addition, AI has also had a significant impact on the manufacturing of the integrated circuit industry. For example, virtual manufacturing can greatly reduce the dependence of integrated circuit research and development on design tape-out testing, leading to shortened time and reduced costs. Finally, AI has also had a profound impact on the development of EDA tools. Professor Shi Longxing stated that the next generation of EDA needs to be intelligent and agile, and AI-EDA may become the key to breaking through bottlenecks.

In addition to the above, Professor Shi Longxing also took everyone through a review of several main characteristics of China's integrated circuit industry.

Industry Characteristics of Integrated Circuits

The industry characteristics of integrated circuits mainly include the following points:Firstly, the development of integrated circuits is a history of entrepreneurship. The industry characteristics of integrated circuits have shown a pattern of division and integration. Since the founding of Fairchild Semiconductor by Robert Noyce, Gordon Moore, and other "eight greats" in 1957, the integrated circuit industry has set off a fission trend. Today, there are more than three thousand chip design companies in China, and there will inevitably be a process of mergers and integrations in the future, leading to the birth of more leading enterprises.

Secondly, the integrated circuit industry is centered around products. As we all know, industry collaboration is aimed at producing products. If we liken integrated circuits to a fruit tree, the roots of this tree are the current important semiconductor equipment, materials, EDA, and other tools, the trunk is the manufacturing of integrated circuits, and the final fruit is the final integrated circuit product. For a fruit tree, only with deep roots can there be lush leaves. However, the foundation of China's integrated circuit industry is not yet solid, and lush leaves can also nourish deep roots. Products and markets can also promote a more solid foundation for the integrated circuit industry.

Thirdly, the integrated circuit industry is driven by system applications. Professor Shi Longxing stated that the application of mobile phones to various AI devices today is driving the integrated circuit industry forward, which is essentially a matter of ecology.

Fourthly, the development of integrated circuits is essentially an economic issue. According to Moore's Law, the feature size of transistors is reduced by half and performance is doubled every 18 months. While power consumption is reduced, the manufacturing cost of integrated circuits is also decreasing. In addition, the integrated circuit industry itself is a capital-intensive and investment-driven industry.

Fifthly, the global division of labor in the integrated circuit industry chain. The United States holds a significant advantage in many sub-industries of integrated circuit support and manufacturing, especially in EDA/IP, logic chip design, and manufacturing equipment, with a share of more than 40%. Looking at other countries and regions globally, Japan has an advantage in integrated circuit materials, while Taiwan and mainland China are leading in wafer manufacturing and packaging testing. The global integrated circuit industry chain has formed a deep division of labor and collaboration pattern, with integrated circuit enterprises in related countries and regions having a high degree of specialization, forming a complementary relationship in the design and manufacturing of integrated circuits.

Technical characteristics of integrated circuits

Professor Shi Longxing also summarized the technical characteristics of the integrated circuit industry.

Firstly, the history of integrated circuit development is a history of innovation. From the birth of the world's first transistor in 1947, to the birth of the world's first germanium integrated circuit in 1958, and then to the proposal of Moore's Law, the invention of dynamic random access memory, the emergence of EDA tools, etc., it can be seen that the entire history of integrated circuit development is a history of innovation.

Secondly, the trend of technological innovation: high integration, high energy efficiency, and agility. With the diversification of computing power needs, the demand for energy efficiency is increasingly high, and high energy efficiency is the eternal pursuit of integrated circuit design. Agility is also an important trend in the technological innovation of integrated circuits, and "EDA + design methodology" is becoming an important means to achieve design agility in the post-Moore era.Thirdly, computing power is a necessity. Professor Shi Longxing stated that the primary issue to be addressed at present is the problem of computing power. The development of integrated circuit product applications cannot be separated from the support of computing power, and the enhancement of computing power relies on the innovation of devices, materials, and architectures.

Finally, Professor Shi Longxing pointed out that the integrated circuit industry has another very obvious characteristic - it is people-oriented. Integrated circuits are an interdisciplinary field, with connections to more than ten primary disciplines, hence the industry has a great demand for talent. The integrated circuit industry needs not only practical talents but also more compound and innovative talents.