[Session 1: Alternative MI] Measurement of Fine Patterns by Utilizing Machine Learning in Semiconductor Industry
As the trend of technology shrinkage and HAR in semiconductor process has become excessive, the level of difficulty of process is increasing. As a result, the need for metrology and inspection for development and management is becoming more important. However, difficulties in metrology and inspection are also increasing for the same reason. Equipment development is being attempted to overcome the limitations, but this attempt is met with physical limitations. Recently, new concepts of equipment such as CD-SAXS are being developed to overcome these limitations, but the pace of development is very slow. The need for instrumentation for HAR structures such as 3D NAND is increasing, but the current in-line measurement technology is difficult to respond to. Most of the measurements are based on the destructive analysis, which puts a heavy burden on the application of mass production.
This presentation will discuss measurement methods based on machine learning that are being developed recently to solve these measurement problems. The application of machine learning to OCD spectrum to solve the problem of profile measurement to HAR structure is actively studied. This technique can also be useful for the measurement of overlay or roughness after etch process. Another example of applying machine learning to this field is an analysis of the e-beam image. In particular, this can be used for rapid and accurate classification for inspection. However, proper reflection of the Fab situation and the raw signal of equipment is essential for application of machine learning to the MI field.