Year: 2025
Type: article
Abstract: The study explores the topic of neuromorphic computing which is a hardware-based paradigm to realizing energyefficient artificial intelligence (AI). In contrast to conventional von Neumann architectures that processes are performed sequentially and memory bottlenecks are employed, neuromorphic systems implement an event-driven model of spiking neurons and massively parallel architecture through biological and neural dynamics. The descriptive-analytical design provided in this case is a synthesis of existing chip implementations (Intel Loihi, IBM TrueNorth, SpiNNaker) and models to examine performance, scalability and energy consumption. Results show that spiking neural networks (SNNs) implemented on a neuromorphic substrate can use as much as 100x energy than using a GPU-based deep learning and still achieve similar accuracy in classification and pattern-recognition problems. Moreover, neuromorphic chip provides scalability in edge AI implementation (IoT, robotics and sensory processing) where power and real time are essential. The paper highlights the fact that neuromorphic models are a viable direction of AI in the future, as they integrate performance, flexibility, and energy awareness.
Author: Dr. Diwakar Ramanuj Tripathi
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7083435136
FWCI:
Cites:
Cited by:
Related to: