A machine-learning based regression model of analog and mixed-signal (AMS) circuit presents an alternative design methodology against the rapidly increased design complexity. The more advanced technology structures, such as FinFET or SOI, are proposed, the more powerful computation engine is required to fulfill the different design specification ensuring an operational robustness. In this work, we applied a supervised learning artificial neural network (ANN) to characterize the regression model of AMS, thus it enables fast exploration of the complex design space including the performance change due to the PVT variations. Moreover, this approach saves significant computation cost compared to SPICE simulations. To prove the concept, successive approximation register analog-to-digital converter (SAR ADC) with various specifications in 14nm predicted technology model (PTM) is designed to illustrate the effectiveness of our approach.