COSMO: COmpressed Sensing for Models and Logging Optimization in MCU Performance Screening

In safety-critical applications, microcontrollers must meet stringent quality and performance standards, including the maximum operating frequency $F_{\max}$. Machine learning models have proven effective in estimating $F_{\max}$ by utilizing data from on-chip ring oscillators. Previous research has shown that increasing the number of ring oscillators on board can enable the deployment of simple linear regression models to predict $F_{\max}$. However, the scarcity of labeled data that characterize this context poses a challenge in managing high-dimensional feature spaces; moreover, a very high number of ring oscillators is not desirable due to technological reasons. By modeling $F_{\max}$ as a linear combination of the ring oscillators’ values, this paper employs Compressed Sensing theory to build the model and perform feature selection, enhancing model efficiency and interpretability. We explore regularized linear methods with convex/non-convex penalties in microcontroller performance screening, focusing on selecting informative ring oscillators. This permits reducing models’ footprint while retaining high prediction accuracy. Our experiments on two real-world microcontroller products compare Compressed Sensing with two alternative feature selection approaches: filter and wrapped methods. In our experiments, regularized linear models effectively identify relevant ring oscillators, achieving compression rates of up to 32:1, with no substantial loss in prediction metrics.

I. Introduction

In safety-critical sectors such as automotive and aerospace, ensuring the reliability of microcontrollers (MCUs) is crucial. This involves identifying devices that fulfill specific criteria. Within this context, the maximum operating frequency ($F_{\max}$) holds particular significance due to its impact on the overall performance of the MCU. $F_{\max}$ influences the speed at which the MCU can process data, which is crucial in applications where timely processing is essential. If the $F_{\max}$ is not reached, this can lead to problems such as timing violations and system instability, which can endanger the reliability and safety of the entire system. Traditional approaches to determine $F_{\max}$ require extensive testing at varying clock frequencies (Speed Binning, [1], [2]). This method is time-intensive, relies on costly test setups, and merely yields binary pass/fail outcomes. In response to these challenges, machine learning (ML) regression models have been proposed to predict $F_{\max}$ of MCUs based on alternative easy-to-acquire on-chip measurements, offering significant time savings compared to traditional methods [1], [2], [3]. In particular, previous research showed that frequency values from ring oscillators (ROs) can be linked with the device's speed $F_{\max}$ and used as features for ML models [3], [4].