Design Implementation of an Efficient Low Power Hybrid LIF Neuron Architecture with Spiking Neural Network for Neuromorphic Applications

Abstract

The convergence of emerging technologies in digital circuit design and neuromorphic newlinesystems has resulted in transformative advancements in power-efficient and highperformance newlinecomputational architectures. This thesis investigates a series of novel hybrid newlinedesigns that combine the distinct capabilities of complementary technologies such newlineas CMOS, memristors, and Carbon Nano Tube Field Effect Transistors (CNTFETs). newlineFirst, a Synergistic Arithmetic Logic Unit (SALU) employing the modified Gate Diffusion newlineInput (m GDI) technique is proposed. This proposed architecture demonstrates newlinesignificant reductions in power consumption, component count, and delay, leveraging newlinethe high density and scalability of memristors. The SALU obtains outstanding newlineperformance metrics when simulated on 45 nm technology, including a power-delay newlineproduct (PDP) of 2.0789Eand#8722;14 joules, demonstrating its appropriateness for energyconstrained newlineapplications. Furthermore, the design incorporated just 33 PMOS transistors newlineand 66 memristors, which reduces the number of components in comparison newlinewith the bassline architecture. As a proof of concept of the proposed combination, a newlinehybrid multiplexer architecture employing memristors and CNTFETs is designed to newlineoptimize subthreshold operation, to enhance power dissipation in line with the reduced newlinedelay. Further to reduce the overall power consumption of the architecture, newlineintroducing a CNTFET-based power-gating technique, the 8:1 multiplexer achieves newlinean unprecedented power dissipation of 1.263 and#956; W, showcasing the efficacy of hybrid newlinedesigns in surpassing traditional CMOS approaches. The thesis also explores the incorporation newlineof cognitive properties in Spiking Neural Networks (SNNs) through a newlinehybrid Leaky Integrate-and-Fire (LIF) neuron model. The neuron composed of CNTFETs newlineand memristors, achieves an operational frequency of 1.54 MHz with an energy newlineconsumption of 0.273 fJ per spike. With these characteristics the proposed neuron newlinemodel mimics the spiking behavior of biological neurons with the minimum leakage newlinecurrent.