A New Method for Classifying High Speed Chip Using Machine Learning

We developed a model that predicts chips that will pass the high speed test using machine learning. The fast speed of DRAM (Dynamic Random Access Memory) is receiving global attention, and the current maximum speed is Samsung's LPDDR5X 8.5Gbps (high speed). As more and more places require high speed chips such as mobile devices, it has become important to produce many chips equivalent to high speed Therefore, the need to develop wafer control and chip classification technologies for fast chip production has emerged, and traditionally, the speed of chips has been classified using tPD (Propagation Delay time) values. However, only 40 to 50% of chips classified as fast based on tPD actually pass the speed test. In order to solve these losses and increase the actual test pass rate, a machine learning model has been developed to predict the chip to pass the test. Also, this model showed 88% performance, exceeding the performance of traditional methods, and derived applicable wafer control conditions for high speed chip production. Also the most important item in determining chip speed turned out to be the value of item. This methodology is applicable to all products, and the optimal model currently built in the experiment is applicable even if it there is a revision on the product.