Skip to Main Content
This paper details an advanced set of analyses designed to drive specific process variable setpoint adjustments or maintenance actions required for cost effective process control using the Dynamic Neural Controller™ (DNC) wafer-to-wafer advisories for semiconductor manufacturing advanced process control. The new analytic displays and metrics are illustrated using data obtained on a LAM 4520XL at STMicroelectronics as part of a SEMATECH SPIT beta test evaluation. The DNC represents a comprehensive modeling environment that uses as its input extensive process chamber information and history of the time since maintenance actions occurred. The DNC uses a neural network to predict multiple quality output metrics and a closed-loop risk-based optimization to maximize process quality performance while minimizing overall cost of tool operation and machine downtime. The software responds in an advisory mode on a wafer-to-wafer basis as to the optimal actions to be taken. In this paper, we present three specific instances of patterns arising during wafer processing over time that signal the process or equipment engineer to the need for corrective action: either a process setpoint adjustment or specific maintenance actions. Based on the controller's recommended corrective action set with the overall risk reduction predicted by such actions, a metric of corrective action "urgency" can be created. The tracking of this metric over time yields different pattern types that signify a quantified need for a specific type of corrective action. Three basic urgency patterns are found: 1. a pattern in a given maintenance action over time showing increasing urgency or "risk reduction" capability for the action; 2. a pattern in a process variable specific to a given recipe indicating a chronic request over time to only adjust the variable setpoint either above or below the current target; 3. a pattern in a process variable existing over all recipes processed through the chamber indicating chronic request to adjust the variable setpoint in either or both directions over time. This pattern is a pointer to the need for a maintenance action that is either corroborated by the urgency graph for that maintenance action, or if no such action has been previously take- n, a guide to the source of the equipment malfunction.