Nano-magnetic devices for computation

The continuous scaling down of the metal-oxide-semiconductor field-effect transistor (MOSFET) has improved the performance of electronic appliances. Unfortunately, it has come to a stage where further scaling of the MOSFET is no longer possible due to the physical and the fabrication limitations. This has motivated researchers towards designing and fabricating novel devices that can replace MOSFET technology. Carbon Nanotube Field-Effect Transistors, Single Electron Tunneling Junctions, Nano-Magnetic Devices, and Spin Field-Effect Transistors are some prospective candidates that could replace MOSFET devices. In this dissertation, we have studied the computational performance of Nano−Magnetic Devices due to their attractive features such as room temperature operation, high density, robustness towards thermal noise, radiation hardened nature and low static power dissipation. In this work, we have established that data can be propagated in a causal fashion from a driver cell to the driven cells. We have fabricated a ferromagnetic wire architecture and used a magnetic force microscopy (MFM) tip to provide localized magnetic inputs. This experiment validated two important phenomena; (1) a clocking field is essential to propagate data and (2) upon removal of the clocking field data can be propagated according to the input data. Next, we have fabricated and captured MFM images of a nano-magnetic logic architecture that has computed the majority of seven binary variables. The architecture was designed by interconnecting three three-input majority logic gates with ferromagnetic and antiferromagnetic wire architectures. This seven input majority logic architecture can potentially implement eight different logic functions that could be configured in real-time. All eight functions could be configured by three control parameters in real-time (by writing logic one or zero to them). Even though we observed error-free operations in nano-magnetic logic architectures, it became clear that we needed better control (write/read/clock) over individual single layer nano-magnetic devices for successful long-term operation. To address the write/clock/read problems, we designed and fabricated amultilayer nano-magnetic device. We fabricated and performed a set of experiments with patterned multilayer stacks of Co/Cu/Ni80Fe20 with a bottom layer having a perpendicular magnetization to realize neighbor interactions between adjacent top layers of devices. Based on the MFM images, we conclude that dipolar coupling between the top layers of the neighboring devices can be exploited to construct three-input majority logic gates, antiferromagnetic and ferromagnetic wire architectures. Finally, we have experimentally demonstrated a magnetic system that could be used to solve quadratic optimization problems that arise in computer vision applications. We have harnessed the energy minimization nature of a magnetic system to directly solve a quadratic optimization process. We have fabricated a magnetic system corresponding to a real world image and have identified salient features with true positive rate more than 85%. These experimental results feature the potentiality of this unconventional computing method to develop a magnetic processor which solves such complex problems in few clock cycles.