An Evaluation of Deadlock-Handling Strategies in Semiconductor Cluster Tools

We study the deadlock handling performance of a single-blade cluster tool and present a comparison of average flowtimes under deadlock detection and resolution versus prevention. Two detection and resolution policies are tested: either resolve permanent deadlocks or resolve both permanent and transient deadlocks. A permanent deadlock requires external intervention to resolve the deadlock, whereas a transient deadlock has a positive probability that the deadlock will resolve itself over time. Prevention averts deadlock by providing sufficient in-process buffer spaces. Our experiments indicate interplay of process and robot transfer times dictate the choice of deadlock strategy. Under low robot transfer times relative to process times, providing sufficient in-process buffer to prevent deadlock or resolving both permanent and transient deadlocks can be equally effective. We conclude with some practical guidelines for operating and designing cluster tools under deadlock conditions.