Alternative Architectures for NonBoolean Information Processing Systems

Abstract

In this chapter we explore algorithms and develop an architecture based on emerging nanodevice technologies for cognitive computing tasks such as recognition, classification, and vision. Though the performance of computing systems has been continuously improving for decades, for cognitive tasks that humans are adept at, our machines are usually ineffective despite the rapid increases in computing speed and memory size. Moreover, given the fact that CMOS technology is approaching the limits of scaling, computing systems based on traditional Boolean logic may not be the best choice for these special computing tasks due to their massive parallelism and complex nonlinearities. On the other hand, recent research progress in nanotechnology provides us opportunities to access alternative devices with special nonlinear response characteristics that fit cognitive tasks better than general purpose computation. Building circuits, architectures, and algorithms around these emerging devices may lead us to a more efficient solution to cognitive computing tasks or even artificial intelligence. This chapter presents several nonBoolean architecture models and algorithms based on nano-oscillators to address image pattern recognition problems.The idea of special hardware architectures for cognitive tasks is not new. Equivalent circuits for a neuron model go back at least to 1907, where the model was just simply composed of a resistor and a capacitor [1]. Since then, various models have been implemented based on different technologies. In the late 1980s, Carver Mead proposed the term neuromorphic system in reference to all the electronic implementations of neural systems [2]. Meanwhile, with the development of digital computing systems, theoretical and computational models for neural