Dynamic comparators are essential components in analog-to-digital converters (ADCs), but are prone to offset voltage errors due to process-induced mismatches. This paper presents a hybrid calibration technique that combines traditional capacitor-based offset calibration with a neural network (NN) to effectively minimize offset. The neural network-based offset calibration technique significantly improves efficiency by automatically predicting optimal capacitor array configurations for dynamic comparators, eliminating time-consuming manual tuning procedures, and achieving offset compensation. The time-based offset polarity classification approach is made, where the zero-crossing timing is extracted using a derivative circuit followed by a time-to-digital converter (TDC). The TDC output is then fed into the neural network, which predicts the optimal capacitor array configuration.
The complete design,occupying 447.2 µm² and consuming 783 µW at 1 GHz, highlights the effectiveness of combining machines.
