• Radhika Khorana
• Adrija Mitra
• S Kumar Swamy
• Sahana Madhusudan
• Sharath P C

Edge Artificial Intelligence (Edge AI) is changing smart device ecosystems by allowing on-device processing that lowers latency by 50–90% compared to cloud-based systems and reduces bandwidth consumption by 70%. This review presents new Edge AI device development uses and hardware designs, including novel ultra-low-power VLSI processors with power budgets of less than 1W and custom AI accelerators achieving 2–5 TOPS/W (Tera Operations Per Second per Watt) energy efficiency. We discuss hardware optimization techniques such as 8-bit/4-bit model quantization (achieving 75-90% reduction in model size), lightweight deep neural networks, and federated learning for use on resource-constrained devices having memory footprints of approximately 100KB-5MB. Real-world implementations in smart cities, healthcare, and industrial automation have shown increases in bandwidth efficiency of between 60 and 80%, 100% local data processing for privacy, and energy consumption reduced by approximately 40 to 65% when compared with cloud-centric architectures. For real-time applications, Edge AI deployments have reached inference speeds of 10 - 100ms, enabling video analytic workloads at 30 FPS on embedded devices consuming under 2W. There remain major challenges in establishing strong security frameworks (current edge devices are three to five times more vulnerable), achieving cross-platform interoperability across more than fifty different hardware architectures, and scalability to accommodate the expected 30 to 50 billion IoT connected devices anticipated by 2030. Emerging solutions make use of standardized AI frameworks that support 95%+ model portability, trusted execution environments, and adaptive edge-cloud orchestration with dynamic workload distribution (achieving 30–40% performance optimization). To achieve scalable, reliable, and secure Edge AI solutions that can process 1–10 trillion edge inferences per day across next-generation smart device networks, this review summarizes recent advancements, quantifies performance benchmarks, and identifies research directions that are essential.

Edge AI, Smart Device, On-device Inference, Low-power AI Processors, Federated Learning, Real-time Analytics

"Edge AI For Smart Devices: A Comprehensive Review of Hardware Architectures, Software Optimizations, And Real-Time Applications", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.12, Issue 11, page no.c704-c712, November-2025, Available :http://www.jetir.org/papers/JETIR2511286.pdf

"Edge AI For Smart Devices: A Comprehensive Review of Hardware Architectures, Software Optimizations, And Real-Time Applications", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.12, Issue 11, page no. ppc704-c712, November-2025, Available at : http://www.jetir.org/papers/JETIR2511286.pdf