Ever since the advent of Atomic Force Microscope (AFM) the community of its users is expanding which led to a broad spectrum of applications. Most of these applications require AFM instruments with fast, accurate, and reliable performance. The sensors and actuators that are utilized in AFM have linear and nonlinear dynamics that hinder the performance criteria. The controller which interconnects these elements can compensate for the dynamics of comprising elements and surpass the performance limitations. This relies on a controller design process that accounts for the undesired dynamics present in the AFM. A data-driven controller design approach is adopted to address the limitations in various control loops of AFM. Presence of nonlinearities, unknown dynamics, and limited measurements in AFM instruments make data-driven methods a suitable choice for this purpose. The control loops in AFM are categorized into three groups: the vertical, the lateral, and the control loop in dynamic modes. In this approach, the controller design process breaks down into solving an optimization problem formed by the dynamics of sensors and actuators and the performance criteria, the solutions of which define the controller. The data-driven methods are divided into on-line and off-line groups, the former using data while AFM is operating and the latter acquiring data for identification while the system is not operating. To address the accuracy limitations in the lateral scan direction which stems from nonlinear dynamics of piezo-actuators, a data-driven feedforward controller is designed. The pair of forward and backward images are used to identify hysteresis in the lateral actuators. The on-line identification data is implemented in a feedforward controller. The proposed sensor-less solution proves to be easy-to-design and suitable for high-speed AFM instruments. The snap-off ringings of cantilever occur in the dynamic off-resonance tapping mode of AFM when the damping of the cantilever and the environment is low. This can cause damages to tip and sample and limits the speed as it sets a limit on the excitation frequencies. A feedback controller is proposed to counteract these undesired oscillations upon the tip-sample separation. The parameters of the controller are tuned based on the force-curves acquired during AFM operation. The implementation of the controller show its effectiveness in cantilevers with various actuation mechanisms. The nonlinear tip-sample interaction limits the tracking performance in vertical direction due to parachuting effect. A gain scheduling controller is proposed to substitute conventional proportional-integral controller. It enables tracking sample topographies without the need for reducing scan speed or setting high force set-points on the system. The lightly damped resonances of the vertical actuator impose constraints on the maximum attainable bandwidth setting a limit on the imaging speed. A data-driven controller designed with off-line identification cancels these dynamics surpassing former limits. To surpass the range-bandwidth limitations, a dual-actuation controller architecture is proposed to run two actuators with both large-range and high-bandwidth characteristics. To further increase the bandwidth limits and reduce the effects of mechanical couplings, high-order controllers are applied on the setup. The combination of dual-actuation with data-driven designed controllers results in improvements in imaging speeds.