This study introduces a novel artificial intelligence (AI)-integrated circuit co-design aimed at enhancing the efficacy of closed-loop neuromodulation devices, which are crucial for treating neurological disorders like Parkinson’s disease, epilepsy, and depression. Traditional devices often suffer from overstimulation and lack of adaptive precision, whereas current closed-loop systems are hindered by high false alarm rates, leading to unnecessary stimulations and increased energy consumption. The proposed solution employs two optimized neural network models—a binary neural network for minimal computation with high sensitivity and a convolutional neural network designed to minimize false alarms to as low as 0.1/h. These models are implemented on a dedicated low-power processor fabricated in 55 nm technology, resulting in an application-specific integrated circuit (ASIC) that occupies a mere 5mm² of silicon area and consumes an average power of 142 μW. This approach significantly reduces false alarm rates and power consumption, showcasing a substantial improvement over existing technologies.