ACTIVE NEUTRAL NETWORK CONTROL OF WAFER ATTRIBUTES IN A PLASMA ETCH PROCESS

We have investigated the use of a neural network to compute plasma etch end point times based on in situ monitoring of the optical-emission trace. The network was trained using data from a complementary metal-oxide semiconductor production facility. In some cases, even though an end-point system is used, the overetch that follows has to be adjusted continuously to reflect machine conditions as it ages after a clean. This can be done manually by measuring wafers after the etch and adjusting the overetch time for the next run. At its present level of training, the neural network shows equivalent performance without the need for such intervention. As the network's learning database grows, we can expect its performance to improve.