SLAM, or Simultaneous Localization and Mapping, is the combined problem of constructing a map of an agent’s environment while localizing, or tracking that same agent’s pose in tandem. It is among the most challenging and fundamental tasks in computer vision, with applications ranging from augmented reality to robotic navigation. With the increasing capability and ubiquity of mobile computers such as cell phones, portable 3D SLAM systems are becoming feasible for widespread use. The Microsoft Hololens, Google Project Tango, and other 3D aware devices are modern day examples of the potential of SLAM and the challenges it has yet to face. The ICP, or Iterative Closest Point Algorithm, is a popular solution for retrieving the relative transformation between two scans of the same object. It has gained a resurgence in popularity due to the rise of affordable depth sensors such as the Kinect in robotics and augmented reality research. ICP, while providing a high certainty of correctness given similar point clouds, is challenging to implement in real time due to its computational complexity. In this thesis, a basic 3D SLAM algorithm is implemented and evaluated, and two proposed FPGA architectures to accelerate the Nearest Neighbor component of ICP for use in a mobile ARM-based System-on-Chip (SoC) are presented. These architectures are predicted to achieve speedups of up to 7.89x and 17.22x over a naive embedded software implementation.