Artificial Neurons – Transistor Based Implementation

In-memory computing architectures have emerged as a possible answer to the associated memory and processing difficulties posed by advancements in AI. The significance of artificial neurons implemented in transistor-based neural network applications is emphasized in this article. These implementations provide benefits such as parallel processing, speed, efficiency, energy conservation, adaptability, scalability, and reduced latency by utilizing VLSI, making them ideal for applications requiring rapid decision-making and fine-tuned control in comparison to biological neurons. The applications of artificial neurons in real-time scenarios and its coordination in hardware capabilities with the dynamic requirements of modern artificial intelligence systems are shown. The FET- based Spiking Neuron and a circuit that uses less power of a leaky integrate-and-fire neuron are discussed, addressing energy efficiency and efficient spiking behavior. A self-mending CMOS neuromorphic circuit contributing to robustness and fault tolerance in artificial neural systems and a CMOS implementation of LIF and Spike Timing Dependent Plasticity (STDP) mechanisms that reflect biological learning processes are shown. The exploration of the revised Axon-Hillock (AH) Artificial Neuron and its utility extends the understanding of neuron models, while the characteristics of a threshold switching memristor offer insights into potential hardware enhancements for neural network architectures.