VLSI CMOS Memristor Building Blocks for Future Neuromorphic Processors

ABSTRACT:Both civilian and military personnel live in a world awash in information while our military commands reconnaissance and weapons platforms of all shapes and sizes over a global and increasingly congested theater. We need new technology to help us sort, prioritize, make sense, and act on the growing streams of information.In Phase 1 we have proven feasible a core CMOS+Memristor circuit capable of supporting larger neuromorphic processors with applications in perceptual encoding, classification and planning.In Phase II we are developing the processors by providing a rigorous mapping between low-level CMOS+Memristor circuit simulations and high-level functional models that describe them. We will then use the functional models to build a software-based Neural Engine API which we will benchmark against existing systems.BENEFIT:Commercial applications for our Neural Engine API include replacements for existing pattern classification systems to include real-time adaptation for use in biomedical applications particularly high-throughput automation, image processing for surveillance, cyber intrusion detection, document classification, fault detection and quality control in manufacturing, and others.Commercial applications for our CMOS+Memristor neural processors include integrated sensory motor processing for UAV's and autonomous vehicles, large-scale weapons allocation, large-scale data fusion and others.