Applying machine vision to semiconductor inspection, Jushi Technology promotes the application of AI inspection in industrial scenes
2022-8-22
In 2019, China's semiconductor sales have exceeded 40 billion U.S. dollars, and the global semiconductor industry is gradually shifting to China, and the mainland has increasingly become an emerging semiconductor base. On the other hand, my country's semiconductor products rely heavily on imports. According to data from the General Administration of Customs, in 2017, my country`s import of computer integration technology products reached US$260.1 billion (approximately 1,7561 billion yuan), accounting for 14.1% of China`s total imports that year, and the import value exceeded US$200 billion for four consecutive years.
From the external environment, the current international chip manufacturers such as Intel, Qualcomm, Xilinx, Broadcom have suspended trading, and Google mobile operating system Android and Microsoft Windows have suspended receiving new orders. The risks caused by the technology embargo have been fully spread. In front of people. To achieve the great rejuvenation of the Chinese nation, the core technology must be self-reliant. Among them, the independent control of the technology in the field of semiconductor chips has become the key to determining whether the goal can be achieved.
From the perspective of internal industry development, the current semiconductor industry is divided into front-end (wafer production), middle-end (wafer manufacturing) and back-end (wafer packaging and testing). Among them, semiconductor testing equipment is the key to chip yield control. Throughout the entire life cycle of high-end semiconductor manufacturing, it can be said to be "just needed" to some extent, and it is the core of the core.
It is no exaggeration to say that the entire semiconductor manufacturing industry itself is a quality control process.
Tens of millions of circuits are densely packed on square-inch chips. Therefore, semiconductor manufacturing has extremely high requirements for detection accuracy. Under this background, the power of artificial eyes is extremely limited, and the entire industry chain almost entirely depends on machine vision. According to gartner's data, the global wafer processing market in 2019 is about 62.7 billion U.S. dollars, accounting for nearly 20% of the global semiconductor market. It is estimated that the compound growth rate of the wafer processing market will reach 4.9% from 2018 to 2023. As the focus of global semiconductor manufacturing shifts to China, the global market share of my country's wafer processing and packaging and testing markets has increased year by year. As of October 2018, the total equipment investment of 17 wafers in China has reached 500 billion yuan. Yuan, the market space for front-end inspection equipment will reach 46 billion yuan, and the follow-up is expected to drive the demand for back-end inspection equipment of 45 billion yuan.
In such a huge market, what we have to face is the lack of independent technology.
For a long time, my country`s semiconductor testing field has almost always been monopolized by foreign equipment vendors such as Teradyne, Xcerra, and Advantest. However, this situation is being dominated by a startup company from China. break in.
"Matrixtime Technology" (www.matrixtime.com) is a local Chinese machine vision manufacturer that aims to provide industrial inspection solutions for high-end manufacturing scenarios such as semiconductors with artificial intelligence technology.
As a hard technology AI company that is deeply engaged in deep learning and computer vision, Jushi Technology is committed to using AI technology to solve difficult scene problems in the manufacturing industry. The fully deep learning-driven machine vision inspection system developed by it has been successfully applied to defect detection and quality control in the semiconductor backend field, and has been effectively verified in customer production. At present, the technology of Jushi Technology can detect the internal structure of a 14-nanometer chip, which is equivalent to looking at a nail on the earth from space.
In terms of core technology, Jushi Technology adopts the research and development of core technologies such as deep learning, computer vision, and motion planning, applies AI technology and algorithms to the field of industrial intelligent manufacturing, and develops scene-oriented industrial AI products. The solution of Jushi Technology is a highly productized and agile project development and delivery mechanism. At this stage, Jushi Technology has two major products: AI vision detection MatrixSemi® and industrial robot "eyes" MatrixRobot®. Starting from the customization business, it penetrates into vertical industries and gradually forms relatively standardized products, and based on this, it is customized for various industrial scenarios. End-to-end technical solutions.
Take the current application in the field of semiconductor back-end inspection as an example. At present, there are dozens of product defects that often occur in the semiconductor manufacturing industry such as insufficient corrosion, over corrosion, sand hole, silver deficiency, and plating offset. Jushi Technology has launched MatrixSemi equipment level solution. This solution is an AI detection platform for semiconductor defects launched by Jushi Technology specifically for the "full deep learning drive" semiconductor defects, including special lighting imaging, AI detection algorithms, underlying acceleration, ADC systems and other functions. Among them, special lighting imaging can provide micron and sub-micron special lighting imaging; semiconductor-specific deep learning models, including dozens of deep learning models, 3D/2D vision models and algorithms, specific vision algorithms and quality inspection rules analysis engine. At the same time, the solution covers the latest full-stack AI acceleration technology, supports real-time edge computing of semiconductor high-resolution vision, and reduces CT time. Provide automatic machine learning defect analysis technology ADC to realize the closed loop of semiconductor quality management. Based on this core technology, Jushi Technology independently developed a semiconductor AOI (Automatic Optical Inspection) system-Juxin 2000, which is the first set of semiconductor AI vision inspection system equipment at home and abroad. It covers more than 30 deep learning model algorithms, and the inspection is accurate. The degree of subversive improvement has been achieved, which is 10 times higher than the current foreign system. When the missed detection rate is 0, the actual false detection rate of the detection system can be maintained between 1% and 5%. Specific classification, visual positioning, and quantitative inspection of defects, and different inspection standards are set according to different defects and different customers. In addition, for different product testing, Polycore 2000 can also automatically adapt and change the testing standards. At present, it is mainly used for defect detection in advanced semiconductor manufacturing and complex 3C precision manufacturing, including front-end/back-end, wafer-related processes, lead frame stamping and etching, LED chip inspection, SiP package inspection, complex 3C manufacturing and liquid crystal panel inspection, etc. Scenes.
"The use of deep learning and complex machine vision to improve the level of quality inspection and control is of great significance to high-end semiconductor manufacturing," said Zheng Jun, founder and CEO of Jushi Technology. "The more complex the scene, the more obvious our AI inspection advantage. In the field of industrial machine vision, we have surpassed traditional monopolies such as Cognex, Keyence, Germany and other traditional monopolies in complex scenarios, and won a large number of orders from well-known customers, which proves that Polycore 2000 has global competitiveness. ."
In addition, compared with the leading technology level, the advantage of Jushi Technology is that its products have extremely high adaptability and convenience.
As semiconductor design and manufacturing become more and more specialized, almost every fab needs a dozen different large-scale equipment to complete hundreds of steps in the wafer production process, which makes any additional process flow It has become very difficult, especially for general-purpose equipment in the testing process, which often requires complex transformation of the production line and long-term training for employees.
The Juxin 2000 developed by Jushi Technology can realize one-step "dumb" operation through the AI system-workers or engineers can quickly get started and use directly without training. On the other hand, the unique embedded design of Jushi Technology enables the finished and defective products to be clearly separated at the discharge end once the Juxin 2000 is connected to the normal production line, without the need to send the products to the inspection room separately. And the laboratory for testing. This alone can reduce inspection costs by more than 10 times for enterprises and integrators.
This ability to commercialize advanced technologies through engineering capabilities and product thinking is also the core competitive barrier of Jushi Technology. At present, Polycore 2000 has launched the production line of Ningbo Kangqiang Electronics, China's largest semiconductor lead frame manufacturer, in September 2019. The operation effect is good and it has been widely praised by the enterprise.
From the perspective of the team, Jushi Technology's member configuration can also be called "hard core". The company does not even have direct sales personnel. More than 90% of the team are product R&D and technical personnel, including 22 PhDs, and there are no shortage of technical experts from specific production lines such as Siemens and Foxconn. Founder and CEO Zheng Jun once worked at Bell Labs and has contact experience with Foxconn, SAIC and other B-end production lines.
At present, many startups in the market have applied machine vision to industrial inspection, but most startups will choose to cut into machine vision inspection from the easier 3C and auto parts industries. Only Jushi Technology has chosen the most difficult AI application. , With high-precision semiconductor testing as a breakthrough.
Everything seems to be just as the founder Zheng Jun said, "Use the most interesting technology to do the most meaningful things, and the most difficult things are often the most valuable."
At present, Jushi Technology has deployed more than 40 patents in deep learning and machine learning, computer 2D/3D vision, computer graphics, lighting imaging and intelligent mechanical control.
Through technology advancement and independent research and development, Jushi Technology has successfully opened an innovative hole in the field of chip testing that foreign giants are waiting for, and has made useful explorations for domestic semiconductor high-end manufacturing. Zheng Jun said that this year the company will continue to complete the product research and development of algorithms, while carrying out market replication and business development. In 2020, Jushi Technology will focus on the three tracks of semiconductor, photovoltaic and automotive precision manufacturing.
(Source: Provided by Jushi Technology)
From the external environment, the current international chip manufacturers such as Intel, Qualcomm, Xilinx, Broadcom have suspended trading, and Google mobile operating system Android and Microsoft Windows have suspended receiving new orders. The risks caused by the technology embargo have been fully spread. In front of people. To achieve the great rejuvenation of the Chinese nation, the core technology must be self-reliant. Among them, the independent control of the technology in the field of semiconductor chips has become the key to determining whether the goal can be achieved.
From the perspective of internal industry development, the current semiconductor industry is divided into front-end (wafer production), middle-end (wafer manufacturing) and back-end (wafer packaging and testing). Among them, semiconductor testing equipment is the key to chip yield control. Throughout the entire life cycle of high-end semiconductor manufacturing, it can be said to be "just needed" to some extent, and it is the core of the core.
It is no exaggeration to say that the entire semiconductor manufacturing industry itself is a quality control process.
Tens of millions of circuits are densely packed on square-inch chips. Therefore, semiconductor manufacturing has extremely high requirements for detection accuracy. Under this background, the power of artificial eyes is extremely limited, and the entire industry chain almost entirely depends on machine vision. According to gartner's data, the global wafer processing market in 2019 is about 62.7 billion U.S. dollars, accounting for nearly 20% of the global semiconductor market. It is estimated that the compound growth rate of the wafer processing market will reach 4.9% from 2018 to 2023. As the focus of global semiconductor manufacturing shifts to China, the global market share of my country's wafer processing and packaging and testing markets has increased year by year. As of October 2018, the total equipment investment of 17 wafers in China has reached 500 billion yuan. Yuan, the market space for front-end inspection equipment will reach 46 billion yuan, and the follow-up is expected to drive the demand for back-end inspection equipment of 45 billion yuan.
In such a huge market, what we have to face is the lack of independent technology.
For a long time, my country`s semiconductor testing field has almost always been monopolized by foreign equipment vendors such as Teradyne, Xcerra, and Advantest. However, this situation is being dominated by a startup company from China. break in.
"Matrixtime Technology" (www.matrixtime.com) is a local Chinese machine vision manufacturer that aims to provide industrial inspection solutions for high-end manufacturing scenarios such as semiconductors with artificial intelligence technology.
As a hard technology AI company that is deeply engaged in deep learning and computer vision, Jushi Technology is committed to using AI technology to solve difficult scene problems in the manufacturing industry. The fully deep learning-driven machine vision inspection system developed by it has been successfully applied to defect detection and quality control in the semiconductor backend field, and has been effectively verified in customer production. At present, the technology of Jushi Technology can detect the internal structure of a 14-nanometer chip, which is equivalent to looking at a nail on the earth from space.
In terms of core technology, Jushi Technology adopts the research and development of core technologies such as deep learning, computer vision, and motion planning, applies AI technology and algorithms to the field of industrial intelligent manufacturing, and develops scene-oriented industrial AI products. The solution of Jushi Technology is a highly productized and agile project development and delivery mechanism. At this stage, Jushi Technology has two major products: AI vision detection MatrixSemi® and industrial robot "eyes" MatrixRobot®. Starting from the customization business, it penetrates into vertical industries and gradually forms relatively standardized products, and based on this, it is customized for various industrial scenarios. End-to-end technical solutions.
Take the current application in the field of semiconductor back-end inspection as an example. At present, there are dozens of product defects that often occur in the semiconductor manufacturing industry such as insufficient corrosion, over corrosion, sand hole, silver deficiency, and plating offset. Jushi Technology has launched MatrixSemi equipment level solution. This solution is an AI detection platform for semiconductor defects launched by Jushi Technology specifically for the "full deep learning drive" semiconductor defects, including special lighting imaging, AI detection algorithms, underlying acceleration, ADC systems and other functions. Among them, special lighting imaging can provide micron and sub-micron special lighting imaging; semiconductor-specific deep learning models, including dozens of deep learning models, 3D/2D vision models and algorithms, specific vision algorithms and quality inspection rules analysis engine. At the same time, the solution covers the latest full-stack AI acceleration technology, supports real-time edge computing of semiconductor high-resolution vision, and reduces CT time. Provide automatic machine learning defect analysis technology ADC to realize the closed loop of semiconductor quality management. Based on this core technology, Jushi Technology independently developed a semiconductor AOI (Automatic Optical Inspection) system-Juxin 2000, which is the first set of semiconductor AI vision inspection system equipment at home and abroad. It covers more than 30 deep learning model algorithms, and the inspection is accurate. The degree of subversive improvement has been achieved, which is 10 times higher than the current foreign system. When the missed detection rate is 0, the actual false detection rate of the detection system can be maintained between 1% and 5%. Specific classification, visual positioning, and quantitative inspection of defects, and different inspection standards are set according to different defects and different customers. In addition, for different product testing, Polycore 2000 can also automatically adapt and change the testing standards. At present, it is mainly used for defect detection in advanced semiconductor manufacturing and complex 3C precision manufacturing, including front-end/back-end, wafer-related processes, lead frame stamping and etching, LED chip inspection, SiP package inspection, complex 3C manufacturing and liquid crystal panel inspection, etc. Scenes.
"The use of deep learning and complex machine vision to improve the level of quality inspection and control is of great significance to high-end semiconductor manufacturing," said Zheng Jun, founder and CEO of Jushi Technology. "The more complex the scene, the more obvious our AI inspection advantage. In the field of industrial machine vision, we have surpassed traditional monopolies such as Cognex, Keyence, Germany and other traditional monopolies in complex scenarios, and won a large number of orders from well-known customers, which proves that Polycore 2000 has global competitiveness. ."
In addition, compared with the leading technology level, the advantage of Jushi Technology is that its products have extremely high adaptability and convenience.
As semiconductor design and manufacturing become more and more specialized, almost every fab needs a dozen different large-scale equipment to complete hundreds of steps in the wafer production process, which makes any additional process flow It has become very difficult, especially for general-purpose equipment in the testing process, which often requires complex transformation of the production line and long-term training for employees.
The Juxin 2000 developed by Jushi Technology can realize one-step "dumb" operation through the AI system-workers or engineers can quickly get started and use directly without training. On the other hand, the unique embedded design of Jushi Technology enables the finished and defective products to be clearly separated at the discharge end once the Juxin 2000 is connected to the normal production line, without the need to send the products to the inspection room separately. And the laboratory for testing. This alone can reduce inspection costs by more than 10 times for enterprises and integrators.
This ability to commercialize advanced technologies through engineering capabilities and product thinking is also the core competitive barrier of Jushi Technology. At present, Polycore 2000 has launched the production line of Ningbo Kangqiang Electronics, China's largest semiconductor lead frame manufacturer, in September 2019. The operation effect is good and it has been widely praised by the enterprise.
From the perspective of the team, Jushi Technology's member configuration can also be called "hard core". The company does not even have direct sales personnel. More than 90% of the team are product R&D and technical personnel, including 22 PhDs, and there are no shortage of technical experts from specific production lines such as Siemens and Foxconn. Founder and CEO Zheng Jun once worked at Bell Labs and has contact experience with Foxconn, SAIC and other B-end production lines.
At present, many startups in the market have applied machine vision to industrial inspection, but most startups will choose to cut into machine vision inspection from the easier 3C and auto parts industries. Only Jushi Technology has chosen the most difficult AI application. , With high-precision semiconductor testing as a breakthrough.
Everything seems to be just as the founder Zheng Jun said, "Use the most interesting technology to do the most meaningful things, and the most difficult things are often the most valuable."
At present, Jushi Technology has deployed more than 40 patents in deep learning and machine learning, computer 2D/3D vision, computer graphics, lighting imaging and intelligent mechanical control.
Through technology advancement and independent research and development, Jushi Technology has successfully opened an innovative hole in the field of chip testing that foreign giants are waiting for, and has made useful explorations for domestic semiconductor high-end manufacturing. Zheng Jun said that this year the company will continue to complete the product research and development of algorithms, while carrying out market replication and business development. In 2020, Jushi Technology will focus on the three tracks of semiconductor, photovoltaic and automotive precision manufacturing.
(Source: Provided by Jushi Technology)
