Neuronal Processing, Reconfigurable Neural Networks and Stochastic Computing

-This paper proposes and studies the premise of three-dimensional (3D) reconfigurable vector neural networks (^3DVNNs). We research a neurocomputing paradigm to accomplish efficient computing. Our overall objective is to advance engineered (human-devised) processing and computing by developing and applying a theory of massive vector processing in a three-dimensional space. Our neurocomputing paradigm and theoretical advancements contribute to natural computing by enhancing the knowledge on processing in living systems. The proposed developments in the fundamental areas of theoretical computer engineering/science and neuroscience are inspired by natural processing and emerging molecular engineering. Our specific objectives are to: (1) Develop enabling design methods thereby advancing the theory of computing and neuroscience; (2) Establish sound and practical CAD-supported tools to design engineered molecular processing platforms (^MPPs); (3) Foster preeminent technology-centric design algorithms. This will allow one to synthesize computing hardware (circuits, processing platforms, etc.) guarantying efficient computing and processing. Our goals are to advance models and principles of computation and to devise-develop-and-demonstrate a sound neurocomputing paradigm supported by a set of highly effective methods, algorithms and tools.