Relevance vector and feature machine for statistical analog circuit characterization and built-in self-test optimization

Abstract

Aiding design and test optimization of analog circuits requires accurate models that can reliably capture complex dependencies of circuit performances on essential circuit and device parameters, and test signatures. We present a novel Bayesian learning technique, namely relevance vector and feature machine (RVFM), for characterizing analog circuits with sparse statistical regression models. RVFM not only produces accurate models learned from a moderate amount of simulation or measurement data, but also computes a probabilistically inferred weighting factor quantifying the criticality of each parameter as part of the overall learning framework, hence offering a powerful enabler for variability modeling, failure diagnosis, and test development. Compared to other popular learning-based techniques, the proposed RVFM produces more accurate models, requires less amount of training data, and extracts more reliable parametric ranking. The effectiveness of RVFM is demonstrated in terms of the statistical variability modeling of a low-dropout regulator (LDO) and the built-in self-test (BIST) development of a charge-pump phase-locked loop (PLL).

Power Management for Multicore Processors via Heterogeneous Voltage Regulation and Machine Learning Enabled Adaptation

“…Such ML module must come with sufficient accuracy, low area/power overhead, and should incur low processing latency to enable HVR adaptation at fine temporal granularity. In this work, we adopted a recently developed sparse Bayesian-based ML algorithm, namely, sparse relevance kernel machine (SRKM) [34], [35] as the ML algorithm. As a kernel machine, SRKM predicts the target value y of a new input vector x using N training samples…”

“…Similar to the relevance vector machine (RVM), the SRKM model is treated as probabilistic, whereby the model parameters w are considered as random variables, which are optimally inferred in the training process. It has been demonstrated the advantages of SRKM for a variety of applications in [34] and [35]. Unlike the widely adopted support vector machine (SVM) and RVM, SRKM can achieve sparsity in both the (training) sample and (parameter) feature space.…”

“…The resulted sparse model is very appealing for achieving low processing latency and hardware overhead in our application. More details about SRKM can be found in [34] and [35].…”

Turbine bearing fault diagnosis based on embedded system

“…In addition to analyzing vibration signals, audio signals also contain much information [6] . Existing fault detection methods based on audio detection are mainly based on detecting the physical level of the captured audio [7][8][9][10] , manually inputting speech, and observing whether the output is consistent with the input [11] . Among them, the physical level detection method tests the output level by designing specialized hardware circuits and then compares the test results with the results when passing through a fault-free audio data acquisition circuit.…”