Kim Hyung-Hwan, Director of SK Hynix's Process Base Technology, said, “We’re striving to improve productivity by applying machine learning techniques to our process lines. This is the task we’ve been focusing on recently” at the Semiconductor and Display Technology Roadmap Seminar held at El Tower in Yangjae-dong, Seoul, on the 26th.
Director Kim said, "We aim to combine the data from the equipment in the production line (source parameters) and the data from the wafer output (response parameters) to create meaningful information.” He explained that once the process is completed, the productivity of the process will be increased.
Director Kim emphasized that it is becoming very difficult to improve production efficiency due to the miniaturization of the process.
According to Director Kim, the current memory process is an extremely refined process that controls the 4000th of a human hair. DRAMs must design and produce very high aspect ratio (A/R) capacitors. Also, as the number of laminate layers increases, 3D NAND flashes must be drilled deeper.
Because atomic layer-level processes must be controlled precisely, a decrease in productivity is inevitable. According to Director Kim, the current memory production lines must control more than 8000 process parameters. The number of pieces of equipment is over 2800, and the number of process staff members is more than 600. It takes two or months from inserting the empty wafer to the finished process. This figure is increasing continuously.
Director Kim emphasized that this is why machine learning technology is being applied in the production line. He said, “It is difficult to overcome various technical challenges with only the efforts of device makers such as SK Hynix. Therefore, collaboration is more important than anything else."
Director Kim said, "The memory industry has experienced tremendous success until last year. However, if we fail to catch up with the constantly changing circumstances, we may collapse. Let’s face the reality and prepare for the future.”