Overview of Semiconductor Manufacturing Technology and Application of AI Technology

Machine Learning Artificial Intelligence Natural Language Processing Probabilistic Generative Models Algorithm ICT Technology Computer Architecture IT Infrastructure Digital Transformation Deep Learning Mathematics Computer Hardware DX Case Study  Navigation of this blog
About Semiconductor Manufacturing Technology

In the previous issue of “Computational Elements and Semiconductor Chips that Compose Computers,” it was mentioned that the integrated circuit of a CPU, which can be considered the heart of a computer, is made up of a large number of transistors as shown below.

In “Semiconductor Design, AI, and Chips for AI,” we discuss the design of semiconductor chips and semiconductor chips for AI. In this article, we will mainly discuss their manufacturing technologies.

Semiconductor manufacturing can be broadly divided into two categories: front-end processes and back-end processes. Front-end processes include wafer fabrication, cleaning, film deposition, lithography, etching, and impurity diffusion, while back-end processes include dicing, mounting, bonding, molding, marking, bumping, and packaging.

These processes are described below.

Si substrates (wafers) are made by cutting and polishing rods made of high-purity crystals called silicon ingots (bottom left in the figure above). Silicon ingots are made from silica rock (common rock) from which impurities are removed to produce 98% pure metallic silicon, from which 99.9999999999% (eleven nines) pure polycrystalline silicon (a collection of microcrystals with random crystal orientation) is made. This polycrystalline silicon is formed by melting this polycrystal in a crucible at a high temperature (1420°C), dipping the seed crystal into the molten liquid, and slowly pulling it up (known as the Choklarsky (CZ) method).

Since the ingots are extremely hard, they are separated into individual wafers using a special diamond blade or similar tool. This technology requires a high level of know-how, and DISCO, a company founded in Kure City, Hiroshima Prefecture, where “Angel Ball” took place, holds a 70% share of the global market. The wafers are then polished before being shipped as finished products (wafer sizes range from 200mm to 450mm and thickness from 0.2mm to 1.0mm).

  • Cleaning Process: Semiconductor processes require a very clean environment. Therefore, the process of washing wafers before and after processing to clean them of dirt is repeated many times. This contamination includes dirt (about 10 particles of 0.1 µm in 1 cubic foot in the cleanest class), metal contamination (Na molecules in sweat from workers, trace amounts of heavy metal atoms in chemicals used in the factory, etc.), organic contamination (carbon from workers’ bodies, trace amounts of carbon in chemicals used in the wafer process, etc.), and oil and grease (carbon from the workers’ bodies, trace amounts of carbon in chemicals used in the wafer process, etc.). ), organic contamination (e.g., carbon from the bodies of workers and trace amounts of carbon in chemicals used in the wafer process), and grease (oil from workers and manufacturing equipment).

Wet cleaning is the most widespread type of cleaning equipment, and consists of a series of tanks containing chemicals and pure water, into which wafers are immersed sequentially to dissolve, neutralize, and rinse off contaminants, and then dried. This equipment requires special know-how as well as wafer fabrication, in which SCREEN and Tokyo Electron have a 42% and 25% share, respectively, of the Japanese market.

  • Film Deposition Process : When an LSI is fabricated on an Si wafer, it is necessary to create layers (films) of silicon oxide and aluminum, which are the materials for electrical separation (insulating film) and wiring (metal wiring film) on the transistor element structure that constitutes the LSI. The film deposition methods can be broadly classified into the following three categories.

(1) Thermal oxidation method (a high-temperature furnace containing gases such as oxygen (O2) and water vapor (H20) is used from the surface to the inside of the semiconductor to form silicon dioxide (SiO2) by causing a reaction between the silicon of the substrate and oxygen)

(2) Sputtering method (A sputtering method in which a vacuum is created in a reactor and inert gas (Ar, etc.) flows through it, ionizing the inert gas and striking the material plate, which is the raw material for forming thin films, with high energy. The film deposition method in which atoms ejected by being hit at this time are attached to the wafer surface)

(3)CVD (Chemical Vapor Deposition: A method of depositing a desired film on the wafer surface in a reactor through a chemical reaction between the wafer and a weight loss gas to be deposited. Depending on the method used to promote the chemical catalytic reaction, there are thermal CVD using thermal energy, plasma CVD using plasma energy, optical CVD using light, etc.)

  • Lithography process : Lithography is a photocatalytic process for processing silicon wafers and deposited thin films. In the main flow of the process, silicon wafers and deposited thin films are processed through photosensitive agent application, exposure, development, etching, and other photoengraving-applied processing technologies to form fine patterns for semiconductor devices.

The superiority or inferiority of this lithography process determines the superiority or inferiority of semiconductor manufacturing technology. In terms of power consumption, the smaller the transistor size, the less charge is transferred during switching, which reduces dielectric loss (dielectric loss). In addition, the wiring can also be shortened, which reduces resistance loss, and the overall loss (dielectric loss and resistance loss) can be greatly reduced.

This trend of increasing semiconductor performance with miniaturization is known as Moore’s Law (the number of chips that can be made from a single wafer doubles in 18-month cycles). However, when considering the 2nm miniaturization target of the current state-of-the-art process, for example, the size of a single atom is approximately 0.1nm, making it difficult to physically control a structure packed with 20 atoms, and the hurdle to realization is extremely high.

The exposure technology that transfers and burns the mask pattern onto the wafer occupies an important position in the lithography technology. In the early days of lithography, the wafer and mask were one-to-one and the pattern was formed on the entire surface of the wafer in a single exposure, but as miniaturization made it difficult to create a one-to-one mask, the approach was changed to one where the mask pattern is reduced and exposed using optics, and these reduced exposures are then used to move the exposure range while exposing specific areas on the wafer. These reduced exposures were replaced by a method that exposes a specific area on the wafer and then moves the exposure area to expose the entire surface (such an exposure device is called a stepper).

Light is a wave, and shorter wavelengths are needed to form fine patterns. Currently, light sources such as ArF excimer lasers (193nm wavelength) and Extreme Ultraviolet (EUV) light sources are used.

In addition, the space between the wafer and the lens is filled with liquid to increase the numerical aperture NA of the projection lens to improve resolution.

  • Etching process : The etching process is a process in which excess areas are removed from the pattern formed in the lithography process by using chemical reactions (corrosive action) of chemicals and ions.

Wet etching technology, which uses chemicals, has been widely used because it is relatively inexpensive and has the advantage of high productivity, such as the ability to process dozens of pieces at a time, but it has the characteristic of isotropic corrosion (corrosion proceeds in all directions in the same amount), which makes it difficult to remove the excess area even in the horizontal direction under the mask. However, wet etching is isotropic (corrosion proceeds in all directions in the same amount), so etching also proceeds in the horizontal direction under the mask, and the etching thickness direction becomes narrower from the periphery, making it impossible to process fine patterns.

In contrast, a method that does not use chemicals, but rather bombards the unmasked areas with ions, etc., is now used to remove the unmasked areas. This is called Dry etching. Dry etching enables anisotropic control (pattern formation proceeds in only one direction), and is almost the mainstream method for LSIs today.

As shown in the figure below, this process is achieved by launching impurities into a mask-patterned wafer by means of diffusion or ion implantation, removing the mask, and then diffusing the impurities deeply and uniformly by thermal treatment or other means.

  • Back-end process: After all processing is completed, wafers go through wafer inspection, dicing process to cut them into chips, mounting process to fix the cut chips into packages, bonding process to connect electrodes and chips of packages, molding process to seal packages, and finally, marking process to finish the process as shown in the figure below. Finally, marking is performed as the final step.

As described above, semiconductor chips are formed through a very complex process. Finally, we will discuss how AI technology can be applied to this semiconductor manufacturing technology.

Application of AI technology to semiconductor manufacturing technology

Examples of applications of AI technology in semiconductor manufacturing technology include

  • Quality control: AI technology can be used to control product quality. For example, image data generated in the semiconductor manufacturing process can be analyzed to detect product defects. Data mining using AI technology can also be used to identify the causes of product defects and lead to quality improvement.

For specific approaches to these issues, see “Image Information Processing Technology,” “Anomaly Detection and Change Detection Technology,” “General Machine Learning and Data Analysis,” etc.

  • Manufacturing process optimization: AI technology can be used to optimize manufacturing processes. For example, data from the manufacturing process can be analyzed to find optimal manufacturing conditions to reduce the defect rate of products. Also, predictive maintenance using AI technology can predict machine failures and optimize maintenance schedules.

For specific approaches to these issues, see “Anomaly Detection and Change Detection Technology,” “General Machine Learning and Data Analysis,” and “Workflow and Service Technologies.

  • Automation of manufacturing processes: AI technology can be used to automate manufacturing processes. For example, robot control using AI technology can automate manufacturing processes. Also, a self-learning control system using AI technology can automatically optimize the manufacturing process.

For specific approaches to these issues, see “Theory of Artificial Intelligence Technology and Basic Algorithms,” etc.

  • Integrated Data Management: AI technology can be used to manage the vast amount of data generated in the manufacturing process in an integrated manner. For example, data from manufacturing processes can be analyzed and used for product quality control, optimization of manufacturing processes, and automation of manufacturing processes. In addition, big data analysis using AI technology can be used to analyze product development and market trends.

For specific approaches to these issues, please refer to “Sensor Data & IOT Technology,” “Time-Series Data Analysis,” “Machine Learning of Data Streams (Time-Series Data) and System Architecture,” etc.

As described above, in semiconductor manufacturing technology, the application of AI technology will enable more advanced and efficient manufacturing through quality control, optimization of manufacturing processes, automation of manufacturing processes, and integrated data management.

コメント

Exit mobile version
タイトルとURLをコピーしました