The explosion of data has swept the global industry and has become a new “means of production” in all walks of life. The penetration of the digital wave in the industrial field is also the only way to upgrade the global industry. However, in the data-intensive semiconductor industry, the use of big data analysis applications is relatively backward. This loop must now be filled quickly.

Born in Silicon Valley in the United States in 1991, Pudifei Semiconductor Technology Co., Ltd. (PDF Solutions, hereinafter referred to as “Pudifei Semiconductor”) has been focusing on one thing for the past 30 years – helping the entire semiconductor industry chain to mine data value. So far, the amount of data in the company’s global semiconductor database has exceeded 4,000 terabytes, which is equivalent to “watching more than 500 years of movies”.

The explosion of data has swept the global industry and has become a new “means of production” in all walks of life. The penetration of the digital wave in the industrial field is also the only way to upgrade the global industry. However, in the data-intensive semiconductor industry, the use of big data analysis applications is relatively backward. This loop must now be filled quickly.

“More and more semiconductor companies, including a large number of ‘young’ Fabless in China, are beginning to realize the importance of data analysis. This is a key step for the semiconductor industry to move from the infrastructure 1.0 era to the 2.0 era of quality improvement.” Purdy Yu Guanyuan, vice president of FeiSemiconductor, pointed out in an exclusive interview with Jiwei.com a few days ago.

How is the “explosive” growth of semiconductor big data interconnected?

According to IDC statistics, the global data volume in 2019 reached 42ZB, and it is expected to reach 163ZB in 2022, with a compound growth rate of 57%. But before these massive amounts of data can truly yield valuable insights for the industry, a lot of groundwork remains to be done.

For example, industrial data will encounter difficulties in data integrity and data quality during the collection process, which requires data preprocessing such as working condition segmentation, data cleaning, data quality inspection, data sample balance, and data segmentation. It will consume a lot of time and labor costs.

This is especially true in the data-intensive semiconductor industry. Yu Guanyuan pointed out that the biggest problem of semiconductor big data is that there are many types of data. If it cannot be well integrated, the data will be cluttered. “There are many kinds of data on a wafer, from IC design to manufacturing to packaging and testing, each process generates a large amount of data, and these big data scattered in various links are also the key to tracing problems and improving yields. “

Compared with other industries, the semiconductor industry chain is extremely long and highly subdivided, from chip design to wafer manufacturing, to packaging, testing, PCB templates, and then to products for practical application scenarios. A large number of specialized companies do it.

“This highly mature global horizontal division of labor model is an important industrial structure that has created the rapid development of the semiconductor industry in the past few decades. However, the relative independence of each link makes it difficult to connect their respective data.” Yu Guanyuan said, “The fact is that semiconductor companies To do data analysis, almost 90% of the time is spent on data cleaning and data integration in the early stage.” This makes semiconductor companies trying to introduce data analysis methods encounter practical difficulties at the specific operation level.

In addition, at the data analysis level, the lack of effective data analysis experience and means is also a difficulty faced by many domestic Fabs and Fabless. “Compared with established companies such as Qualcomm, AMD, and Intel, many domestic manufacturers are relatively young, have not been established for a long time, and lack relevant experience. After obtaining the data, what data should be integrated and what analysis can help them grasp It’s easy to get confused when you live with problems.” Yu Guanyuan pointed out.

The root cause of the above obstacles is that the underlying architecture related to the entire industry data is not well established. Yu Guanyuan compares it, this is equivalent to building a house, the ground floor, water pipes, and wires must be erected, especially the data cleaning and data integration parts, which need an open platform that can be used by the whole industry to do it well in advance. Provide effective analysis tools for companies that need to import data analysis applications.

“We have been doing a lot of work in the underlying architecture, that is, we have worked hard to lay the ‘water pipes and wires’, so that enterprises can ‘check-in’ and directly realize the value of data.” Yu Guanyuan introduced that the products provided by Pudifei The Exensio platform was born for this.

It is understood that Pudifei’s end-to-end analysis platform Exensio platform has big data integration, cleaning and analysis functions, and can provide services for various companies in the entire industry chain. If the entire industry chain is divided into three major links: design, wafer manufacturing and packaging and testing, Purdy has corresponding product modules for the characteristics of each link.

So far, in the IC design part, Pudifei’s Fire Engine software has analyzed 1 billion layout structures; in the wafer manufacturing process, more than 24,000 chip production equipment around the world are connected through the production monitoring software it provides; the packaging and testing part , the factory data of more than 15 major packaging and testing service providers around the world is connected to the platform, and more than 16,000 test machines and packaging equipment are connected through its software that monitors the entire operation.

More importantly, through such a data platform that connects the entire industry chain, each industry chain link that was originally independent of division of labor realizes the interconnection of the data level. This interconnection is not only between devices, but also connects semiconductor engineers in all aspects with all chip production, packaging, and testing equipment, providing an important reference for design and manufacturing, helping to reduce costs and improve performance and yield.

In the past 30 years, Pudifei has worked closely with world-class tape-out factories and design companies to help leading international semiconductor companies such as Advanced Fab, Fabless, and IDM/System to complete many advanced process mass production projects.

Open up the upstream and downstream of the industrial chain to mine the value of data

Yu Guanyuan pointed out that Pudifei is currently the only company that covers data from the front-end to the back-end semiconductor supply chain. Through decades of self-accumulation and external mergers and acquisitions, the entire supply chain has been opened up. In his view, the end-to-end full coverage of the industry chain is the core value of Pudifei.

He gave the example of Apple’s requirement for its chip suppliers that every chip they provide is traceable — which factory produced it and what each step of the factory was, including what was used in the manufacturing process. Chemical gases, on which equipment was tested, packaged, which employee was responsible, etc. Because only by tracing can we confirm which step went wrong.

In the era of big data, how does Industry 4.0 of the semiconductor industry go?

Yu Guanyuan explained that the reason why this end-to-end data analysis of the whole industry chain is becoming more and more critical is that the importance of semiconductors to various industries on a global scale is increasing year by year-semiconductors account for an increasing proportion of products. The requirements for its reliability are also getting higher and higher. “For example, in a car, the most expensive component in the past was the engine, but now, the most expensive component is the semiconductor component. And the reliability of the component is more and more important, even if it is a few dollars, if it breaks, it may be damaged. It can cause traffic accidents and even affect people’s lives.”

In other words, a few dollars of components will affect the value of hundreds of thousands of vehicles. “Therefore, the reliability of semiconductors is becoming more and more important. If there is a problem with yield and reliability, it is necessary to clearly trace where the problem is, whether it is packaging or manufacturing, or a defect in IC design.”

When it comes to advanced semiconductor process nodes, especially when it reaches 7nm and 5nm, because with the increase of process complexity, many defects are not on the surface, but buried inside. How to monitor these things during R&D and mass production becomes very difficult. Therefore, in addition to technical adjustments to improve yield, the importance of data capture capabilities is becoming more and more prominent.

Yu Guanyuan introduced that the yield improvement of advanced nodes is not just a consideration for fabs. Fabless also needs to bury the monitoring structure in the product during product design, and capture and analyze feature data quickly and accurately. To localize problems and improve product yield.

In this part, Pudifei has accumulated a large number of continuously evolving yield improvement technologies in the nearly 30 years of cooperation with world-class chip design companies and wafer fabs. For example, it can help advanced process customers to design a unique monitoring structure, which is built into the spare position of the chip, and combined with the special test equipment of PUDIFI to realize ultra-high-speed capture and analysis of a large amount of characteristic data, so that the defects of design and process can be seen at a glance.

Specifically, Exensio Platform, the semiconductor big data platform of Pudifei, includes a module for semiconductor defect detection and classification (Exensio-Control), a product test optimization module (Exensio-Test), and a semiconductor yield management module. System modules (Exensio-Yield), etc., cover the entire semiconductor industry chain from IC design, wafer manufacturing, to packaging and testing.

It is imperative to embrace AI and move data to the cloud. Predictive analysis is the trend

Yu Guanyuan introduced that at present, semiconductor companies are most concerned about yield diagnosis and analysis in big data analysis, that is, after a problem occurs, the source of the problem can be quickly traced through a systematic method, so as to carry out diagnosis analysis and improve yield.

However, with the deepening of the intelligence of the whole industry, this passive response obviously cannot meet the demand in the long run. Predictive analytics before problems occur are the focus of more industry leaders.

In the era of big data, how does Industry 4.0 of the semiconductor industry go?

A year or two ago, TSMC stated that it has begun to use artificial intelligence in its factories. With the help of artificial intelligence, it can produce 20%-30% more wafers without adding machines. For example, in some key processes, artificial intelligence is used to dynamically adjust the maintenance time of the machine to improve production efficiency. In addition, artificial intelligence can also integrate the experience and professional skills of many experts, so that the experience of an expert can be used in a large area without the presence of experts, so as to achieve better experience inheritance.

“The next thing to do is to foresee, through AI, machine learning and other means, to propose value from the data to prevent possible problems.” Yu Guanyuan said.

For example, taking a fab as an example, part of the function of Purdial’s Process Control software is to monitor machine equipment, and to analyze and predict how many hours of operation the machine may have problems through the collected past data, which may lead to production problems. The wafer is scrapped. This allows the machine to be stopped in advance for maintenance. In the past, the equipment was stopped until the machine started producing scrapped wafers. But often when it stops, some of the scrapped wafers may have been produced.

With the advancement of technologies such as AI, machine learning, and 5G, the amount of data is growing exponentially, and the demand for on-demand computing is getting higher and higher. Going to the cloud to obtain more flexible computing resources and storage advantages has become an inevitable choice for the industry to go deeper into Industry 4.0.

Yu Guanyuan took a wafer as an example. With the introduction of advanced technology, the amount of data generated by a wafer may be 10 times or even 100 times the original. At the same time, the introduction of AI and machine learning means a huge amount of computing, and the requirements for data storage space and computing power are increasing day by day. It is obviously unrealistic to rely on the internal structure of a company.

Migrating to the cloud has become one of the cores of the advanced manufacturing industry, but the domestic semiconductor industry still has reservations about it. Among them, the security of data is the biggest concern of fabs and design companies. Yu Guanyuan believes that there is a misunderstanding, “Everyone thinks that data is safer in their own company than in the cloud. In fact, a semiconductor company is not a pure IT company, and its human investment in security is far less than that of cloud service providers. These cloud services Service providers will provide multiple security architectures, which are enough to build a strong barrier for customer data.”

It is reported that the DEX network launched by Pudifei has helped major global packaging and testing plants to achieve cloud deployment and edge computing, including rapid prediction and predictive response through edge analysis, reducing data loss and improving data quality.

According to reports, in addition to close cooperation with world-renowned cloud service providers such as AWS, Pudifei has also taken a lot of measures to ensure data security in the software, and has hired many security companies to conduct simulated intrusion tests.

It is worth mentioning that Pudifei, which entered the Chinese market in 2006, has gradually upgraded its business model with the development of China’s semiconductor industry in the past 15 years. Yu Guanyuan introduced that the domestic semiconductor industry has been upgrading in terms of technology, and chip design companies are the fastest growing group. According to the characteristics of chip design companies, Pudifei previously launched the Exensio-Hosted semiconductor data analysis platform based on cloud deployment. It is an enterprise-level cloud data analysis system that does not require any IT maintenance, allowing customers to access data anytime, anywhere. Complete customized data analysis to quickly find the root cause of problems. The platform is compatible with a variety of data, supports efficient data extraction and loading, provides a highly interactive data visualization module, as well as fast alarms, deep tracking, and powerful data mining capabilities.

The Links:   ZUW252412 LCBA7Z211M14 COMPONENTS