Series 2. The significance of sensing and RHEED Image Data
IVWorks combines epitaxy technology with artificial intelligence (AI) to provide differentiated, high-quality Epi wafer foundry services. Three articles are presented in this series in which we divide semiconductor manufacturing and AI series into data, model, and platform, respectively. Let us explore the appearance of a semiconductor from the perspective of a researcher directly involved in DOMM AI Epitaxy System research.
IVWorks combines data-based decision-making to implement high-quality semiconductor materials by controlling the physical and chemical changes that occur in compound semiconductor manufacturing. As described in detail in the last column, a reflection high-energy electron diffraction (RHEED) pattern consists of changes in the lattice constant between compound semiconductors of angstrom units. Because high-quality data are directly linked to an effective AI model application, we must better understand RHEED image data.
In compound semiconductor manufacturing, the RHEED pattern contains the surface state information of the growth of films on substrate. The pattern shape varies depending on the surface state and can be classified as “ring,” “spot,” “streaks,” “modulated streaks,” “satellite streaks,” and others. Because compound semiconductor manufacturing is primarily , heteroepitaxy all types of patterns can occur based on specific situations (i.e., growth and chamber conditions).
Mirror that Reflects the Growth Stage : “Pattern”
The ring pattern indicates an amorphous crystalline. Patterns of an amorphous crystalline are those that should be avoided as much as possible in compound semiconductor manufacturing. They are highly likely to occur during initial epitaxial growth because of large mismatch(lattice, thermal expansion, etc.) between substrates and compound semiconductors. If no ring pattern appears during initial growth, it is highly unlikely that a ring pattern will occur afterwards. The DOMM AI Epitaxy System is trained to recognize the process in which ring patterns occur and can significantly increase productivity by sending warnings (Level 3) to engineers prior to the occurrence of ring patterns, or it can modify recipes (Level 5) it previously produced.
Unlike ring patterns, spot patterns can appear anywhere (depending on growth and chamber conditions) over the course of the manufacturing process. Ring patterns must be avoided at all cost. However, depending on the situation, spot patterns may play a positive role. For example, if a nucleation layer is formed uniformly during initial epitaxial growth, defects and strains can be reduced. In addition, quantum dots can be formed on defect site during compound semiconductor manufacturing to prevent defects propagation from occurred at initial l growth stage. They can also increase the crystallinity of the grown layer.
Unintended spot patterns such as ring patterns should be avoided. Because unintended spot patterns vary depending on the chamber situation (e.g., the types of sensing data involved, source and substrate temperatures, chamber pressure, residual gas, etc.), collection and analysis of sensing data are also crucial. The DOMM system collects, manages, and trains all sensing data from the DOMM edge device in a time series for application purposes.
A streak pattern indicates that the surface of a compound semiconductor is growing very flat. If the streak pattern is continuously maintained in the section where the layer (i.e., composition) changes, it indicates that the manufacturing of the compound semiconductor is proceeding without problems.
Even if the streak pattern is maintained, DOMM has not completed its task. This is because observation of gap changes due to differences in the lattice constant and thermal expansion coefficient are critical. As noted in the last column, a gap in the RHEED image data implies changes to the lattice constant and thermal expansion coefficient. Therefore, DOMM analyzes the composition and strain by examining the spacing in the section where the layer changes. It does this without terminating the rotation of the substrate.
“Sensing Data” : Signals Sent by Eqipment
Because the appearance of a RHEED pattern implies changes to various aspects of current situations (e.g., temperature, pressure, residual gas) within a specified time, the prediction model can be developed and managed without sensing data by assessing analytical results related to manufactured compound semiconductors. Here, “sensing data” refers to data generated from various pieces of equipment when compound semiconductor materials are grown.
Because of this, sensing data may seem unnecessary. However, this is far from true. To analyze the results of a prediction model through DOMM or to develop and modify an advanced model, the sensing data must be matched one-to-one with the RHEED image data in a time-series. The types of sensing data managed include temperature (growth temperature, raw material temperature), the open/closed state of the source shutter, the amount of gas flow, pressure, and the amount of residual gas. Sensing data are collected from the DOMM edge device using Modbus (TCP/IP) communication protocols and then matched with RHEED image data. By matching and managing both the sensing and RHEED data in real-time, it is possible to analyze and strengthen the prediction model, send warnings about equipment anomalies before the RHEED image data change, and send notifications about scheduled times for equipment maintenance.
Real-time data collection and management are essential to achieve a high level of recipe modification and development through AI. Currently, IVWorks uses a service that properly recognizes the patterns of RHEED images with high probability during compound semiconductor manufacturing and visualizes them in real-time. IVWorks can be expected to take the AI manufacturing technology to the next level by combining sensing data with the vast amount of recognized data to determine pattern changes in advance.
Byung-Guon Park / Artificial Intelligence team
What are the alternative solutions of the semiconductor industry in the new era of “mandatory environment”?
Exploration of a Platform for the Utilization of Artificial Intelligence in Semiconductor Manufacturing