Circuits-Informed Machine Learning Technique for Blind Open-Loop Digital Calibration of SAR ADC

TL;DR

This paper introduces a circuits-informed machine learning method for blind digital calibration of SAR ADCs, using a low-speed reference ADC and neural networks to significantly improve linearity without prior error knowledge.

Contribution

It proposes a novel ML-based calibration technique that does not require prior error knowledge, leveraging circuit insights and a low-speed reference ADC for improved calibration accuracy.

Findings

SFDR improved by over 38dB

Calibration consumes 25.8fJ per conversion

Effective for 28nm, 12-bit, 84MHz SAR ADCs

Abstract

This work presents a supervised machine-learning (ML) approach for blind digital calibration of SAR ADCs without requiring prior knowledge of errors. A low-speed reference ADC is used to train a shallow neural network (NN) to estimate errors in a high-speed ADC by comparing the outputs of the ADCs when their sampling instants align and subtracting these errors in the back-end. The proposed NN-calibration improves SFDR of a 28nm, 12-bit, 84MHz ADC by >38dB while consuming 25.8fJ/conversion-step.