Multi-condition alternate test of analog, mixed-signal, and RF systems

Abstract

This work proposes a generic path to improve Alternate Test strategies. It demonstrates that multi-condition test increases the amount of information present in the test data and consequently decreases the prediction error of the trained models. The ambition of this paper is to be a methodological contribution to the field of AMS-RF test, and formal guidelines are provided that justify the interest of the approach. For the sake of validation, the proposed methodology has been applied to several alternate test strategies for analog, mixed signal, and RF circuits. Promising results are found for the following case studies: an analog filter, a I:ll. AID converter, and an RF LNA.

Brownian distance correlation-directed search: A fast feature selection technique for alternate test

“…Such set of features can be efficiently computed on-chip in the digital domain [27], and can be related to performance parameters of the DUT such as its frequency response and linearity figures. In addition, following the multi-condition test strategy in [6] to enhance the feature space, we repeat the test at a 75% power supply, and at a ten times higher clock frequency, leading to a total of 18 features. The goal of the machine-learning based test is to predict the Total Harmonic Distortion of the filter.…”

“…Initially, we thus have a 6dimension input space that we want to map onto the unidimensional output space formed by the SNDR. In addition, again using a multi-condition strategy [6], we repeat the six tests at 80% of the power supply in order to stimulate the limited output range of the amplifiers and the settling tests at 120% of the nominal frequency in order to observe more important settling errors (we do not repeat the leakage tests in this case because we do not expect any change). In this way, the dimension of the input space grows to 15.…”

“…For circuits of medium complexity, this readily represents a significant search space. Repeating the measurements under different operating conditions (like the supply voltage or the operating frequency) to stress the circuit and extract additional information can further increase the dimensionality of the candidate space [6].…”