Neuromorphic circuits, Design and Dynamics

Sammanfattning

The rapid advancements in artificial intelligence have prompted a fundamental rethinking of traditional hardware design and computing architectures. As AI realized by software grows, there is an increasing need for hardware that offers high performance, flexibility, and low power consumption. Neuromorphic engineering addresses this challenge by drawing inspiration from the human brain, the most efficient and adaptable computing system known to humankind.This thesis explores the design of analog silicon neurons and synapses capable of mimicking the dynamics of their biological counterparts. By the deploying semiconductors combined with other electrical components, the behaviour of nonlinear thresholding neurons can be generalized and replicated, forming what is so called low-power neuron circuits. In parallel, a feedforward neural network (FFNN) was trained to optimize resistor weights for synaptic connectivity, enabling flexible and scalable inter-neuronal communication.The contribution of this work is aimed towards the development of realistic neuromorphic systems, laying the foundation for future AI hardware that is both efficient and brain-like in function.