An N-bit multicell-encoded cellular neural network for multidata processing

Significant disadvantages of Cellular Neural Networks (CNNs) include the implementation of very large CNN arrays and the time required for loading/unloading the CNN data. This research addresses both of these problems by developing a new CNN architecture for processing the data of multiple CNN cells simultaneously. Several significant contributions of this research include a mathematical formalism for this new CNN, a CMOS implementation of a specific test architecture, and the design of the equivalent test architecture using predominately Quantum-Well Device (QWD) circuits.;For a digital CNN with n bits per cell, this new architecture is an n-Bit Multi-Cell Encoded Cellular Neural (nB-MCE CNN). The specific test architecture selected for in-depth investigation is a 1B-4CE CNN which processes the data of four 1B CNN cells simultaneously. A mathematical description of the 1B-4CE CNN templates and a mathematical expression of the 1B-4CE CNN dynamics in terms of 1B-4CE CNN variables and newly defined 1B-4CE CNN templates is presented. System-level simulations for both for both Connected-Component Detection (CCD) cases and Edge Detection cases validate this new CNN architecture.;A CMOS implementation of a 1B-4CE CNN system is presented, including the design and operation of the circuits and functional blocks of this system. Experimental results of a one-dimensional 1B-4CE-CNN CMOS chip are presented for CCD cases and Edge Detection cases using both one-chip and two-chip systems. These test results indicate the correct encoding, processing and decoding of single-data CNN cells from multiple 1B-4CE cell components and between 1B-4CE CNN cells.;A 1B-4CE CNN is also designed and simulated using predominately QWDs. Mathematical models for several QWDs, including the multiple-gated Resonant Tunneling Diode (mgRTD), are developed. A general mgRTD comparator is analyzed in detail using the mgRTD model presented. This comparator is then customized to meet the specific requirements of many of the required 1B-4CE CNN functional blocks. System-level simulations of this QWD-based 1B-4CE-CNN for both CCD test cases and Edge Detection test cases match the expected 4B encoded and 1B decoded outputs at operating speeds at 4GHz. These simulation results demonstrate the compatibility of this new, advanced CNN with a futuristic device technology.