With the increasing demand for precision and accuracy in the high-tech industry, limiting thermal dissipation has become a critical challenge. In high-precision positioning systems, heat dissipation causes thermal expansion of components, which degrades positioning accuracy and ultimately compromises overall performance. These shortcomings led to the development of a Tunable Magnet Actuator(TMA). TMA can sustain the required force without a continuous current supply by utilizing the tunable magnets, which can be magnetized along different states. To minimize thermal dissipation, accurate control of the magnet’s magnetization state is essential. Existing magnetization state tuning methods, however, are subpar, falling short in efficiency, accuracy, or both simultaneously. Moreover, many of the existing state-of-the-art control algorithms utilize the look-up table, resulting in a feedforward control. To generate these look-up tables, a comprehensive data collection is required, as well as the calibration of the obtained results, which is undesirable. To mitigate this, a feedback control strategy similar to the traditional Pulse Width Modulation algorithm has been developed. While accurate, the PWM control remains highly inefficient. The aforementioned necessitates the development of a new feedback control algorithm that would be as accurate but more efficient. The neural networks used for the prediction of the hysteresis loops have been largely explored in the past years. However, the implementation of the neural network in tandem with a real-time control remains novel. This thesis proposes a new neural network-based PWM control algorithm to address the challenges in accurately controlling the remanent state of the tunable magnet. Initially, a system characterization is done to train the neural network model. Afterwards, a new control algorithm that utilizes the developed neural network models is created and implemented.