UCLA

The past decade has seen a sharp rise in the development and manufacture of different hardware frameworks to meet the ever-rising computational demands of the machine learning software community. Conventional computing architectures require massive server farms and consume large quantities of energy to perform these tasks. Consequently, this necessitates that end users must connect wirelessly to powerful servers capable of performing complex machine learning tasks. The aforementioned shortcomings have sparked a pursuit for the development of energy efficient hardware capable of successfully performing complex computational tasks offline.Self-organized nanowire arrays of memristive materials, known as atomic switch networks, are the collection of billions of individual memristive elements randomly intertwined as an interconnected network of electrically active junctions. The resulting morphology of the network has a number of attractive neuromorphic properties and emergent phenomena which yield an intrinsic capacity to perform complex computational tasks on a physical substrate. Under an external stimulus these networks exhibit a dynamic, non-equilibrium modulation of conductance across the entire network. The resultant non-linear dynamics are capable of being utilized as both logic and memory components operating in parallel through a technique called in-materio computing. This form of computing enables a physical substrate to be utilized as a dynamic reservoir capable of transforming a simple external stimulus into higher dimensional non-linear outputs. The output layer is then mapped onto a desired computational task through a technique called reservoir computing (RC). Silver selenide (Ag2Se) and silver iodide-based (AgI) nanowire networks were characterized and implicated as efficient memristive materials for RC applications. Both materials were successfully employed within an RC framework for waveform regression, spoken digit recognition and handwritten digit classification tasks. Conventional techniques for nanoscale manufacturing have also begun to hit their limit of resolution, garnering interest in developing new techniques capable of manufacturing materials with molecular and/or atomic precision. Atomically precise manufacturing (APM) aims to implement scanning probe microscopy techniques in conjunction with tailor made molecular tools as a powerful system capable of realizing atomic scale 3D printing. The preliminary state of APM is explored using molecular tools for the abstraction/donation of individual atoms.