Photolithography is one process used for integrated circuit manufacturing where geometric patterns are transferred onto the semiconductor wafer. The exposed surfaces are then chemically etched resulting in the circuit. One key subsystem is the wafer scanner which is responsible for accurately positioning both the wafer stage containing the silicon wafer and a reticle stage housing the image mask. In order to meet the increasing demands for denser integrated circuits, tighter tolerances are placed on the wafer scanning subsystem. One way to address these needs is to utilize the fact that photolithography is a batch process. In this dissertation, we use Iterative Learning Control (ILC) to exploit the repetitive nature of the wafer scanner and meet the high performance tracking control requirements. ILC is a data-driven method for tuning a feedforward control signal. It uses information from previous experiments (iterations) to update the current control signal by using a model based update (learning) law. In this dissertation, we first describe the single degree of freedom wafer stage prototype which is used as a test stand for developing new control algorithms. A new software design is proposed using a finite state machine architecture which allows users to automate operation and limit the amount of downtime between experiments. We also develop a new hardware interface between the controller and one of the transducers which provides scalability for future hardware upgrades. Recent ILC research efforts have been toward developing increasingly sophisticated ILC algorithms aimed at improving the rate of ILC convergence over standard methods. We present here four practical issues with ILC with the intent on reducing the complexity of the ILC algorithm. We first address the issue of choosing a system model for the ILC algorithm. We present a model based learning filter design which uses a finite impulse response learning filter. We show that the wafer stage system model based on the closed loop continuous time dynamics can be used to construct a simple learning filter. With minor loss of high frequency model fidelity we show that these new learning filters perform as well as standard ILC algorithms. Second, we study the effect of ILC in a feedback control loop. Since it is common to first implement a feedback controller, the ILC command signal can either be placed in parallel or series to the feedback controller. We develop here the differences between the ILC architectures and show that the primary difference between them is how well the learning filter matches the system inverse. Third, we study monotonically convergent P-Type ILC algorithms. Motivated by the fact that our experiments result in good performing ILC algorithms, we then look at the analytical methods for designing monotonically convergent ILC algorithms. We observe that the sufficient conditions for monotonic stability are considerably conservative and is largely biased toward high frequency content. Lastly, we present a method for integrating ILC and system identification techniques. This method is motivated by the fact that traditional ILC algorithms use system model information which is obtained before the ILC algorithm begins learning. In this method, we perform a least squares estimation of the linear system dynamics each iteration and use this system model for the model based ILC algorithm. We implement this on the wafer stage prototype and show that the new method outperforms the traditional, static ILC algorithm.