Analog VLSI Circuits for the Perception of Visual Motion

Preface

It was 1986 when John Tanner and Carver Mead published an article describing one of the first analog VLSI visual motion sensors. The chip proposed a novel way of solving a computational problem by a collective parallel effort amongst identical units in a homogeneous network. Each unit contributed to the solution according to its own interests and the final outcome of the system was a collective, overall optimal, solution. When I read the article for the first time ten years later, this concept did not lose any of its appeal. I was immediately intrigued by the novel approach and was fascinated enough to spend the next few years trying to understand and improve this way of computation - despite being told that the original circuit never really worked, and in general, this form of computation was not suited for aVLSI implementations.

Luckily, those people were wrong. Working on this concept of collective computation did not only lead to extensions of the original circuit that actually work robustly under real-world conditions, it also provided me with the intuition and motivation to address fundamental questions in understanding biological neural computation. Constraint satisfaction provides a clear way of solving a computational problem with a complex dynamical network. It provides a motivation for the behavior of such systems by defining the optimal solution and dynamics for a given task. This is of fundamental importance for the understanding of complex systems such as ...