Towards Green AI: Energy-Efficient and Secure Circuit Design for Next-Generation Neural Hardware

Abstract

Artificial Intelligence (AI) systems are increasingly deployed in energy- and security-constrained environments, from mobile devices to autonomous vehicles. The explosive growth in neural network workloads has led to a pressing demand for sustainable hardware that minimizes power consumption while ensuring trustworthiness. This paper explores the emerging paradigm of "Green AI," emphasizing energy-efficient and secure circuit design as the foundation for next-generation neural hardware. By combining novel circuit techniques, low-power device architectures, and integrated hardware-level security mechanisms, it is possible to achieve scalable AI computation with reduced environmental impact and robust resilience against attacks. Experimental evaluations demonstrate that reconfigurable energy-efficient circuits, when combined with built-in cryptographic primitives and lightweight security layers, can reduce energy consumption by up to 42% compared to traditional CMOS-based accelerators, while also mitigating hardware-based adversarial vulnerabilities. The results highlight the synergy between energy efficiency and security as dual objectives in advancing sustainable neural hardware design.