A blind identification approach to digital calibration of analog-to-digital converters for built-in-self-test

A major bottleneck in analog and mixed-signal built-in-self-test (AMBIST) is the difficulty in generating high precision input stimuli that are required in most existing AMBIST schemes. This paper presents a new approach to AMBIST of ADCs using low precision input stimuli. With mild qualitative assumptions on input signals and the model of the ADC, the blind identification algorithms identify both the non-precision part in input signals as well as correction codes for the ADC from the ADC output codes. Initial simulation results show that a 12-bit ADC can be calibrated to achieve INL at the +0.5/-0.5 LSB level from an uncalibrated 25 LSB INL level with input stimuli having only 7-8 bit accuracy