On the implementation of RTD based CNNs

Resonant tunnelling diodes (RTD's) have found various applications in high-speed digital and analog circuits due to their specific advantages associated with the unique folded-back negative differential resistance (NDR) I-V characteristics. As a result of the nonlinearity of RTD's, cellular neural networks (CNNs) designed with RTD's can achieve higher integration density and higher processing speed in comparison to standard CMOS based implementations. This paper describes two implements of RTD's based CNNs: one with RTD's only where RTD's are represented by current sources describing physics-based models, and the other with RTD's and FET's configured in well-known monostable-bistable logic elements (MOBILEs) circuitry. The paper also proposes a new and simple cell structure of MOBILE based CNN for connected component detection. Several image processing operations have been successfully simulated for these two types of CNNs. Simulation results show that RTD based CNNs have excellent performance in terms of complexity, speed and compactness.