Downloads: 17 | Views: 1562 | Weekly Hits: ⮙1 | Monthly Hits: ⮙1

Review Papers | Computer Science & Engineering | India | Volume 12 Issue 4, April 2023 | Popularity: 6 / 10

Rate This Article!      

An Analysis of Neuromorphic Computing in Modern Technology

Karan Chawla

Abstract: According to IT experts, neuromorphic computing may play a key role in bringing about the fourth AI revolution. Throughout time, as the hardware industry has expanded, we have seen neuromorphic chips take control. It will improve other chips' hardware platforms so that each can handle the unique AI workloads for which it was intended. Several experts think neuromorphic computing has the potential to alter the strength, effectiveness, and capacities of algorithms in artificial intelligence while also revealing new information about cognition. Energy economy, execution speed, resilience against local failures, and learnability are key advantages of neuromorphic computing over conventional methods. Neuromorphic computing may be a game-changer for space applications where mission success depends on rapid and autonomous processing of a wide range of incoming information from various sources. It was shown that the SIF model had a 91.5% accuracy rate and had reduced the number of steps by adopting an early exit strategy in order to explore design space. The RIF model was only able to achieve less than 85% accuracy, no matter the number of steps taken. For design space exploration, there are design and control time knobs that, while lowering inference latency, provide accuracy that is comparable to or slightly below that of full precision models. In edge artificial intelligence, Spiking neural networks (SNNs), motivated by biological neurons, have been investigated as a possible neuromorphic computing solution for the incorporation of AI algorithms in edge devices due to their low energy consumption, in order to meet this difficulty. Due to the LFNL approach's use of spike activation, which requires a limited number of training time steps (T) to optimize, classification accuracy is slightly lower than that of traditional federated learning-based ANNs. This review article examines the use of neuromorphic computing in three domains: unattended ground sensors, space, and wireless edge artificial intelligence.

Keywords: Ground Sensors, Space, Wireless Edge Artificial Intelligence, Neural Networks, Neuromorphic Computing

Edition: Volume 12 Issue 4, April 2023

Pages: 1359 - 1364

DOI: https://www.doi.org/10.21275/SR23424160205

Make Sure to Disable the Pop-Up Blocker of Web Browser

Click to Cite this Article

Similar Articles