Will the Future of the Chip Industry be Dominated by System Companies Such As Apple and Google?
Since the iPhone 4 was equipped with Apple’s self-developed A4 processor in 2010, Apple has rushed along the road of self-developed processors. In 2020, it replaced the Intel processors on the Mac series with self-developed M1 chips. Apple’s success has increased the confidence of system companies in choosing self-developed chips.
In 2016, Google launched the first-generation self-developed tensor processor unit TPU, which has now been iterated to the fourth generation.
Since then, Amazon, Alibaba, and Baidu have also chosen self-developed chips. The self-developed chips of these system companies with huge demand for chips will inevitably reduce the demand for buying chips from chip giants such as Intel, Nvidia, and AMD. With this, the voice of traditional chip giants will also weaken.
Will the Chip Industry be Dominated by System Companies in the Future?
It is foreseeable that system applications will be the core driving force of chip design, which means that EDA (Electronic Design Automation), a key tool for chip design and chip manufacturing, needs to be innovated in order to support the diversified and customized requirements of system applications.
It is expected that Global EDA Tools Market to Reach $14.9 Billion by 2026. In the past, every breakthrough of EDA has brought revolutionary changes to the chip industry. Does this mean that 2026 is the starting point for system companies to dominate the chip industry?
System companies are “forced” to develop their own chips
The global chip industry is still developing at a high speed and has already entered a mature period. From design to production, to packaging and testing, the complex chip industry chain has achieved a global division of labor and collaboration.
Starting in 2015, Moore’s Law began to slow down, and the performance improvement of general-purpose chips became slower and slower.
However, the new round of AI craze demanded more and more computing power for chips, and the demand for chips for IoT applications became more and more diverse.
System companies like Apple and Google that purchase chips in large quantities are increasingly dissatisfied with the improvement of general chip performance, so they have begun to develop their own custom chips.
In the past few decades, general-purpose chips have had advantages in terms of performance improvements from process and architecture improvements. Customized chips are difficult to compete with general-purpose chips because of their small usage.
However, Moore’s Law, which has dominated the chip industry in the past 40 years, has gradually approached its limit, allowing the industry to gradually relax the performance and cost requirements of custom chips. At the same time, after most of the application requirements are met by general-purpose chips, the purpose of custom chips is no longer to simply pursue high performance, but to achieve differentiation in function, power consumption, safety, etc., and achieve innovation through system-level collaborative optimization and unique competitiveness.
Now it is software that determines everything. Since system companies have mastered the software, they will have many ideas. They know the best implementation path for the system, but chip companies don’t know, and EDA companies do the same.
Apple is a typical example. In the early stage of self-research, the performance of A-series processors was significantly different from that of chip companies. However, through Apple’s system-level optimization, there is no significant disadvantage in the experience. Through continuous iteration, Apple’s A-series processors have become industry benchmarks. Not only that, through innovations such as unified memory architecture, Apple’s more powerful M1 processor has achieved a huge improvement in performance and energy efficiency, which can replace Intel’s mature Core processor.
Google’s self-developed TPU also satisfies its own map, photo album, search, and other businesses, and achieves more differentiated functions and better experience at a lower cost. There is quantitative data showing the advantages of Google’s self-developed processor. Its self-developed video coding unit VCU is used to accelerate YouTube’s video coding and decoding.
The mature chip industry and the failure of Moore’s Law have allowed system companies to embark on the path of self-developed chips. However, the chip industry has a very high technical threshold. Even with a mature division of labor in the chip industry chain, it is a huge challenge to shorten the chip design cycle.
Therefore, Google has begun to use AI to reduce the difficulty of chip design.
Google recently published a paper entitled “A graph placement methodology for fast chip design” on Nature, which uses deep learning to optimize the layout planning method of chips to automatically generate floor plans. Optimize the performance, power consumption, and area of the chip. Human engineers need months to complete the work, Google can achieve the same effect with AI in only 6 hours.
In fact, Google’s experiment is showing that the current EDA needs more innovation.
Moore’s Law has made many general-purpose chip companies, but as Moore’s Law is approaching its limit, the chip industry is entering a new era, and the trend of heterogeneity and customization is becoming more and more obvious. More and more system companies are starting to develop their own chips. On the one hand, they are dissatisfied with the performance and cost of general-purpose chips. On the other hand, it is the best choice for system companies to maintain their competitiveness.
But for system companies to customize chips with higher efficiency and lower cost, breakthroughs in the top-level EDA tools are needed.
The chip industry is brewing a new revolution but we have to see who will become the leader of the chip industry in the new era?