Over the past few decades, machine learning has become integral to our daily lives. Deep learning has revolutionized industry and scientific research, enabling us to solve complex problems that were previously intractable. Similarly, as computer components have become smaller and more efficient, humanity has gained unprecedented access to affordable hardware. However, the looming fundamental limits on transistor sizes have sparked widescale investigation into alternative means of computation that can circumvent the restrictions imposed by conventional computer architecture. One such method, called reservoir computing, maps sequential data onto a higher dimensional space by using deep neural networks or nonlinear dynamical systems found in nature.

Networks of nanowires are currently under consideration for a wide range of electronic and optoelectronic applications, and have recently been pursued as potential devices for reservoir computing. Nanowire devices are usually made by sequential deposition, which inevitably leads to the stacking of wires on top of one another. This thesis builds a fully three dimensional simulation of a nanowire network and demonstrates the effect of stacking on the topology of the resulting networks. Perfectly 2D networks are compared with quasi-3D networks, and both are compared to the corresponding Watts Strogatz networks, which are standard benchmark systems. By investigating quantities such as clustering, path length, modularity, and small-world propensity it is shown that the connectivity of the quasi-3D networks is significantly different to that of the 2D networks. This thesis also explores the effects of stacking on the performance in two reservoir computing tasks: memory capacity and nonlinear transformation. After developing a dynamical model that describes the connections between individual nanowires, a comparison of reservoir computing performance is made between 2D and quasi-3D networks. Most previous simulations use the signals from every wire in the network.

In this thesis an electrode configuration is used that is a more physically realistic representation of nanowire networks. The result is that the two different network types have a strikingly similar performance in reservoir computing tasks, which is surprising given their radically different topologies. However, there also exist key differences: for large numbers of wires the upper limit on the performance of the 3D networks is significantly higher than in the 2D networks. In addition, the 3D networks appear to be more resilient to changes in the input parameters, generalizing better to noisy training data. Since previous literature suggests that topology plays an important role in computing performance, these results may have important implications for future applications of nanowire networks.