A Neuron-MOS-Based VLSI Implementation of Pulse-Coupled Neural Networks for Image Feature Generation

<para xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> An analog circuit for implementing pulse-coupled neural networks (PCNNs) in very-large-scale integration (VLSI) hardware has been developed using the Neuron-MOS <formula formulatype="inline"><tex Notation="TeX">(ν MOS)</tex></formula> technology. PCNNs are biologically inspired models having powerful ability for image feature generation. With the <formula formulatype="inline"> <tex Notation="TeX">ν MOS</tex></formula> technology, weighted sum of multiple input signals, which is an essential of PCNNs, is implemented simply by the capacitive coupling effect in a <formula formulatype="inline"> <tex Notation="TeX">ν MOS</tex></formula> block. By employing the switched floating gates in the <formula formulatype="inline"><tex Notation="TeX">ν MOS</tex></formula> blocks as temporary analog memories, the storage of image data is simply realized. Moreover, the function of decay generation, which is crucial for emulating PCNNs neuronal dynamics, is also merged into a <formula formulatype="inline"><tex Notation="TeX">ν MOS</tex> </formula> block by utilizing the input-terminal capacitors in it. With such techniques, the circuit achieves a purely voltage-mode implementation of PCNNs in a compact structure. Inheriting the merits of PCNNs, the circuit has good discriminability against different patterns as well as robustness against rotation and translation of identical patterns, which is analogous to human image perception. The performance of the circuit has been verified by the measurements of a proof-of-concept chip fabricated in a 0.35-<formula formulatype="inline"> <tex Notation="TeX">μm</tex></formula> double-polysilicon CMOS technology. </para>