UoA Competitive Research Funding (Ref. 2024-24): Combination of Approximate Computing and Approximate Stacking Memory for Low-power Neuromorphic Computing (2024-2025)

Brief description

In this research, we investigate designing a low-power neuromorphic computing solution for IoT and Edge devices by combining Approximate Computing and Approximate Stacking Memory. As IoT and edge devices have become ubiquitous, it is necessary to have the ability to calculate the neural networks on the devices using dedicated chips/modules. However, current neural network architectures are power intensive, especially for deep neural networks, which prevents them from being able to be widely adopted. As a result, there is a need for low-power solutions for Neuromorphic Computing. In this research proposal, we combine Approximate Stacking Memory with Approximate Computing to improve energy efficiency further while maintaining overall accuracy.

Output

1. Ryoji Kobayashi^* and Khanh N. Dang, '’An Efficient Hardware Implementation of Spiking Neural Network Using Approximate Izhikevich Neuron’‘. 2024 9th IEEE International Conference on Integrated Circuits, Design, and Verification, June 6-8, 2024.