Design and simulations of spintronic devices for conventional applications and unconventional computing

Spintronic technologies have emerged as promising candidates for ultralow power and CMOS-compatible technology. In addition, spintronics can have a role in facing the von Neumann architecture bottlenecks. The main potential applications of spintronics are microwave detection/generation, biomedicine, energy harvesting, memories – with solutions where the information can be stored in topological magnetic textures (i.e. skyrmions) –, neuromorphic computing, and unconventional computing applications such as probabilistic computing. This thesis deals with two main topics: 1) the design and modeling of magnetic tunnel junctions (MTJs), one of the key spintronic devices, which are characterized by compact size, high-speed operation (up to tens of GHz), and versatility; 2) the exploration of the static and dynamic properties of topological textures (i.e. magnetic skyrmions) for their potential use in many applications and for the development of skyrmionics as a promising technology for hybrid skyrmionics-CMOS technology. Chapter 1 introduces the micromagnetic formalism, fundamental for the understanding of the magnetization behavior in ferromagnetic materials at the nanoscale. This chapter describes a micromagnetic solver developed in C/C++ language able to efficiently perform data-driven tests of MTJ designs. In addition, a user-friendly interface has been developed and benchmarked. The main outcomes of this chapter include solutions for the development of MTJ-based electromagnetic energy harvesting, verified with experimental activities implemented in the laboratory of Prof. Yang at National University of Singapore, and spintronic accelerometers. Chapter 2 presents static and dynamic properties of magnetic skyrmions stabilized in different magnetic materials and heterostructures. In particular, we identified a protocol for the characterization of the type of skyrmion, pure Neel vs Hybrid skyrmion. This activity has been implemented in collaboration with Prof. Wanjung Jiang’s group at Tsinghua University of China. The last part of the chapter deals with skyrmion manipulation driven by temperature gradients. Chapter 3 introduces the main concepts of neuromorphic computing and deep learning, with a particular focus on how spintronics can be used to develop the field of neuromorphic spintronics. In particular, we show how to use intrinsic nonlinearity of time nonlocality to implement different types of neurons and operations, including the extraction of dark knowledge. Chapter 4 provides a brief introduction on the idea behind the Ising model and its use in facing combinatorial optimization problems. The primary focus is the implementation of probabilistic Ising machines based on the idea of probabilistic computing with p-bits (PC). We present a PC solver developed in C/C++ with CUDATM acceleration that incorporates several energy minimization algorithms, such as standard annealing and parallel tempering. Many potential solutions of probabilistic bits with MTJs are also explored and an experimental implementation of PC achieved in collaboration with Prof. Pedram Khalili Amiri at Northwestern University (USA) is presented.