Deep learning is transforming real-time biomedical applications, but deploying these powerful models on edge devices faces significant hurdles. These devices have strict limits on computational resources, memory, and energy. We’ve developed a lightweight, hardware-optimized 1D Convolutional Neural Network (CNN) architecture specifically for classifying electrocardiogram (ECG) signals. By strategically using knowledge distillation and quantization, we’ve dramatically compressed the model. It now occupies just 0.08 MB with roughly 66,000 parameters, yet it maintains an exceptional 99.6% classification accuracy. Our CNN accelerator design also demonstrates impressive efficiency. It achieves 594.49 GOP/s/MeLUT in logic efficiency and 32.94 GOP/s/W in power efficiency. These results show that our architecture makes efficient, low-power biomedical inference on edge devices entirely feasible, especially benefiting deployments where resources are scarce.