In this paper, we present an ASIC based on 22FDX/FDSOI technology for the detection of atrial fibrillation in human electrocardiograms using neural networks. It consists of a RISC-V core and a machine learning IP core for computationally intensive inference. The ASIC was designed for energy efficiency. A feature of the ML-IP is its modular, generic and scalable design. This allows to perform a optimization of hardware design and architecture of the neural network. A multi-objective optimization of the system is performed with respect to computational efficiency at a given classification accuracy which is carried out using a probabilistic surrogate model.