|
AI technology has developed rapidly in recent years, garnering a great deal of attention. So, what exactly is AI? |
AI is short for Artificial Intelligence. In other words, it is technology that enables manmade machines to exhibit human-like intelligence. Typically, this involves the use of ordinary computer programs to demonstrate nearly human intelligence. The main purpose of AI is to assist people with their work by improving efficiency, saving time and reducing human error. The history of AI can be traced back eighty years to the very invention of computers in the 1940s. With the development of learning technologies in the 21st century, however, the development of AI technology has accelerated, and its fields of application are expanding rapidly.
|
The Application of AI to Materials Analysis—The Best Way to Tackle Huge Volumes of Data |
The history of Materials Analysis is even longer than that of AI. For example, the first TEM (Transmission Electron Microscope) machine was born more than 90 years ago in 1931. Later, the development of the semiconductor, optoelectric, electronics and nanotechnology industries led to more and more opportunities and needs for its use. It soon became an indispensible technology. Its main purpose is to analyze physical and chemical properties such as material composition, structure and morphology. Aside from the TEM, other tools used for materials analysis include the SEM (Scanning Electron Microscope), the FIB (Focus Ion Beam Microscope), the SIMS (Secondary Ion Mass Spectrometer), the XPS (X-ray Photoelectron Spectroscope), the AES (Auger Electron Spectroscope) and more.
Materials analysis results are typically presented in a variety of forms, including analysis data, graphs, engineer drawings, photos and graphic files. With the rapid development of the technology industry, the amount of analysis results to be processed is also increasing. This enormous amount of data and graphics is making manual processing difficult. In fact, it can be said to have exceeded the amount that can be handled by humans. Trying to handle all this data manually is time-consuming and labor-intensive in addition to being at risk of input errors. Accuracy also decreases as the amount of data increases. Therefore, AI computer assistance is needed to ameliorate the situation and better enable us to respond to a large number of analysis results and customer needs. With this in mind, MA-tek set out to develop methods for processing large amounts of analytical data many years ago, focusing mainly on combining AI and materials analysis technology.
Take, for example, the automatic measurement software that MA-tek has developed, which applies software algorithms to TEM images to define graphic boundaries then carry out a series of procedures, such as measurement, labeling and output, thus reducing the need for human interpretation and man-hours. Common Edge Detection Algorithms include: (1) Laplacian Edge Detection, (2) Sobel Edge Detection, and (3) Canny Edge Detection, etc. Among these algorithms, Canny is noteworthy as it is a composite algorithm that combines four other algorithms—the Gaussian Filter for reducing noise, gradient detection, non-maximum suppression and judgment boundaries—to perform edge detection. Its advantages include a low error rate, accurate positioning and high resolution. It is a well-known and mature edge detection algorithm that is widely used in industrial image processing and image recognition applications.
Figure 1. Canny Edge Detection Algorithm Schematic Diagram [2] |
|
Applications of AI Materials Analysis |
MA-tek began with materials analysis and now has a whole range of analytical technologies suitable for MA (Materials Analysis), FA (Failure Analysis), RA (Reliability Analysis), SA (Surface Analysis), and CA (Chemical Analysis). MA-tek’s AI materials analysis technology too has continued to develop according to our customers’ needs and can be applied to the analysis of the advanced semiconductor GAA (gate all around) structure, FinFET structure, semiconductor process monitoring, machine validation analysis, grain size measurement, and nanoparticle size measurement. In addition to statistical analysis, large-scale data analysis and graphics, we can also provide automated services such as dedicated analysis software.
1. Multilayer Film Measurements
Figure 2 shows the automatic measurement of multilayer films. Multilayer films are common structures used in semiconductors, optoelectric components and III-V compound semiconductors. Human errors are common when measurements are performed manually, especially when the amount of data increases. This leads to a corresponding decrease in accuracy and consistency. By using AI automatic measurement, however, manual errors can be reduced, and the thickness of different structures can be better measured according to each customer’s needs.
 Figure 2. Results of the Automatic Measurement of Multilayer Film Structures (a) TEM Image; (b) Boundary After Image Processing; (c) Measurement Parameters |
|
2. Via Profile Measurement Figure 3 shows the automatic measurement of via profile etching results in an IC manufacturing case. The software is used to pinpoint the via’s bottom position, measure the width of the CD (Critical Dimension) at 3nm intervals, and repeatedly measure other vias and photo data. With the help of AI software, a large number of measurements can be made in a short time. The measurement results can then be exported as CSV text files, enabling clients to perform subsequent big data analysis using Excel so that they can find the key factors in a process in relation to each experiment design. This allows the identification of optimum setting values, making the adjustment of etching process parameter settings faster and more effective.
3. Grain Size Analysis When analyzing the grain sizes of metal structures, metal samples are usually ground down. Grain sizes and distributions are then investigated using a metallographic optical microscope (OM). If the grain sizes are too small to be observed using the OM, a FIB SIM image can be used instead, or further observation and analysis can be performed using the TEM. Due to the irregular shapes of the grains, manual measurement requires grains to be selected manually, which can easily lead to a data bias. Software-assisted analysis, on the other hand, enables the measurement of a large number of grains and the outputting of the statistical results of grain size distributions. This allows researchers to easily analyze and compare the performance of different grain sizes under different process conditions. |
Figure 3. Schematic Diagram of Via profile Automatic Measurements and Statistical Analysis Results |
|
Figure 4. FIB Observation of Grain Distribution |
Figure 5. Grain Size Analysis and Statistics |
As AI technology develops, materials analysis technology too is pushed to new heights. The combination of these two technologies allows us to effectively solve various materials analysis problems so that we can continue to develop and innovate.
