At the beginning of April, Google announced its TPU test report using custom ASIC design. The performance and power consumption are far better than the CPU/GPU combination on the market (Note 1), which is very suitable for artificial intelligence and deep learning operations. After the report was published, there were a lot of discussions on the Internet. Some friends asked me if IC design service companies can enter the AI ​​chip market. Before answering my opinion, let me briefly explain the evolution of the IC design services market.
Before the foundry model was born, IC products were mainly from IDM. At that time, IDM had its own internal standard products (ASSP), which were completed from specification, design, manufacturing, and testing. And many IDMs also provide customized chip design (ASIC), such as IBM, TI, LSI, VLSI Technology, STM, NEC, Toshiba, etc. From the business model, IDM's ASIC business is also considered A design service. After TSMC took the lead in providing foundry services, fabless IC products companies have sprung up, and related ecological chains including IP, EDA, and IC design service industries have subsequently responded. .
Because IC design service companies provide shared design resources (Note 2) to many new IC products companies that do not need to invest in their own internal teams to save money, and quickly launch products into the market, suddenly become a high demand Industry. In addition, inspired by the listing of the world's first design service company, Zhiyuan, in 1999, the newly established IC design service company established more than 30 global companies in those years, some of which focus on specific fabs, while others emphasize Support for multiple fabs can even assist in plant-to-plant conversions (porTIng). The design service industry did not take too much supply for too long, and the market competition became fierce.
With the evolution of Moore's law, the process becomes more and more complex, the chip integration degree also multiplies, and the performance and power consumption specifications of the products are also more severe. In addition to the increase of manpower/material/investment cost, the design company has technical challenges and products. The risk of failure is also relatively high. After several generations of process evolution, some of the weaker competitors were naturally eliminated. The players who stayed on the field were also roughly graded and positioned, showing a relatively healthy competitive situation. At the 16nm and below nodes (currently to 7nm), higher technical barriers not only isolate some competitors, but also increase the number of outsourced design opportunities from system companies and even IC products companies. . The position of professional and advanced IC design service company in the foundry eco-chain, the more important the higher-order process, the more obvious the market competitive advantage.
These two years of flaming applications such as AR/VR, cloud computing (big data analytics), artificial intelligence and deep learning, and even bitcoin require high performance and low power chips. Large system companies, data centers, Internet platforms and other companies are gradually eliminating the need for standard chips (ASSP) on the market, and they tend to develop their own custom-made chips (ASICs) to meet their performance and power consumption requirements. Make a difference. Google TPU is an obvious example. This trend has increased the opportunities for IC design service leaders.
The IC design service industry was born out of demand, and it was weakened and strengthened due to fierce supply competition. After 20 years, the industry is now entering a relatively healthy supply and demand market and is growing stronger.
This article is an industry trend observed by the author and has nothing to do with the company that serves as a consultant.
Note 1: Google Cloud Platform Blog: QuanTIfying the performance of the TPU, our first machine learning chip
Note 2: The sharing of resources for design services does not fall under the so-called sharing economy. Its resources are not idle, the mode is B2B, and resource allocation is not done through the trading platform, just like the technology and capacity resources of semiconductor customers sharing TSMC or ASE.
T Copper Tube Terminals,Non-Insulated Pin-Shaped Naked Terminal,Copper Cable Lugs Terminals,Insulated Fork Cable Spade Terminal
Taixing Longyi Terminals Co.,Ltd. , https://www.longyicopperlugs.com