Analogue neuromorphic systems
This thesis addresses a new area of science and technology, that of neuromorphic systems, namely the problems and prospects of analogue neuromorphic systems. The subject is subdivided into three chapters. Chapter 1 is an introduction. It formulates the oncoming problem of the creation of highly computationally costly systems of nonlinear information processing (such as artificial neural networks and artificial intelligence systems). It shows that an analogue technology could make a vital contribution to the creation such systems. The basic principles of creation of analogue neuromorphic systems are formulated. The importance will be emphasised of the principle of orthogonality for future highly efficient complex information processing systems. Chapter 2 reviews the basics of neural and neuromorphic systems and informs on the present situation in this field of research, including both experimental and theoretical knowledge gained up-to-date. The chapter provides the necessary background for correct interpretation of the results reported in Chapter 3 and for a realistic decision on the direction for future work. Chapter 3 describes my own experimental and computational results within the framework of the subject, obtained at De Montfort University. These include: the building of (i) Analogue Polynomial Approximator/lnterpolatoriExtrapolator, (ii) Synthesiser of orthogonal functions, (iii) analogue real-time video filter (performing the homomorphic filtration), (iv) Adaptive polynomial compensator of geometrical distortions of CRT- monitors, (v) analogue parallel-learning neural network (backpropagation algorithm). Thus, this thesis makes a dual contribution to the chosen field: it summarises the present knowledge on the possibility of utilising analogue technology in up-to-date and future computational systems, and it reports new results within the framework of the subject. The main conclusion is that due to its promising power characteristics, small sizes and high tolerance to degradation, the analogue neuromorphic systems will playa more and more important role in future computational systems (in particular in systems of artificial intelligence).