Solving problems of low dimensional devices at the system level

Abstract

We describe an approach to an electronic system based on assumptions made regarding low dimensional devices. Systems using those devices will have three major obstacles: The variation of the parameters is increasing, the reliability of the individual switching elements decreases and the interconnection between those elements will prove to be more difficult than ever. To overcome those obstacles we consider an approach using locally interconnected very simple elements in a neuromorphic way. Because analog implementations of weighting and addition can be made very simple, we examine a theoretical analog implementation of linear threshold gates as the simple elements needed for a neuromorphic approach. We derive the dependence of the tolerance against statistical variations on the interconnect density and show, that one can increase the tolerance against parameter variations with an increase of the interconnect density. Finally we conclude the discussion with an example of a Full Adder based on monostable bistable transition logic elements (MOBILE) build with resonant tunneling diodes (RTD). We show that it is possible to trade off the number of interconnections and computing elements against the robustness against statistical variations.