A production-loss based maintenance plan is proposed to minimize the cost due to unscheduled ATE (Automatic Test Equipment) downtime in the back-end process of semiconductor manufacturing. This paper suggests two methods, active redundancy and cold standby redundancy, to expedite the ATE system repair time for returning the system back to production. This strategy is different from other reliability improvement methods such as corrective actions and preventive maintenance. By reducing the repair time, the system upper time actually increases and hence the production loss is minimized. The optimization is formulated to minimize the production loss considering the system depreciations, lost sales and idle labor when they are subject to maintenance budget and volume constraint. To solve this optimization problem, Genetic Algorithm is used to find the near-optimal solution. The illustrative example demonstrates that using redundant modules is a very effective way to minimize the semiconductor production loss due to unscheduled system downtime.