Design And Implementation Of Storage And Analysis For WAT Data

There is a process in the chip production line called Wafer Accept Test(WAT),which mainly tests the electrical parameters of the transistor.A company proposed Dense Array technology,a high-density test chip design and rapid test technology,which can put about 30 million devices under test in a wafer,and can measure one million test values in one minute,compared with existing test solutions have nearly a100-fold increase in device density and test speed.Therefore,for the WAT test of Dense Array technology,and the test data volume of a product reaches TB level.The efficiency of test data storage,query and statistical analysis needs to be solved urgently.This paper proposes a wafer test data storage and analysis system based on Cassandra+My SQL+Spark framework.The main work of this paper is as follows:1)The characteristics,query and analysis requirements of the test data are analyzed,and the Cassandra distributed cluster storage scheme is designed in detail and the database is modeled,so as to realize the fast storage and reading of the test data.For structured data such as accounts and configurations with a small amount of data,MySQL is used for database modeling,which makes the system operation more smooth.2)In this paper,through the realization of functions such as threshold voltage calculation,abnormal point judgment,and data statistical calculation,applications such as device failure analysis and wafer yield analysis have been completed.For the system statistical calculation function,this paper uses the Spark engine to read the test data in the Cassandra database,and creates an elastic data set for parallel data processing,which solves the problem of statistical efficiency of large-scale test data.3)The entire system was built using automated deployment scripts and Docker container services,and the system's functions,performance,scalability and fault tolerance were fully tested.The analysis and application are carried out through the actual test data of a foundry,and the results show that the system can complete the failure analysis of hundreds of millions of devices.