CAFE & Test

Scientific Background

The integration capabilities offered by current nanoscale CMOS technologies enable the fabrication of highly complex, fully integrated mixed-signal systems on a single chip. However, manufacturing processes are prone to imperfections and defects that can degrade—sometimes catastrophically—the intended functionality of the fabricated circuits. Extensive production testing is therefore necessary to separate these defective or unreliable parts from the functionally correct devices. Unfortunately, the co-integration of blocks of very different types (analog, mixed-signal, digital, RF, MEMS, etc.) as well as limited access to the internal nodes of an integrated system make testing these devices a very difficult and costly task.

Standard test methods for analog, mixed-signal, and RF (AMS-RF) circuits are based on the direct measurement of complex circuit specifications, relying on the use of expensive dedicated test equipment. Testing AMS-RF blocks integrated into complex systems has become a difficult, costly, and time-consuming task that has been identified as one of the main bottlenecks in the production of current and future integrated systems.

Project positioning

Machine learning-based testing, also known as indirect testing or alternative testing, is a promising strategy for overcoming these challenges. Indirect testing reduces the complexity and cost of production testing by replacing conventional functional tests on the production line with a set of inexpensive indirect observations, often called signatures. Test results are then inferred by post-processing these signatures through the construction of nonlinear multidimensional regression models. The underlying idea is that signatures are easier to measure than specifications and can be extracted using inexpensive equipment or, even more advantageously, using simple on-chip test instruments that can be integrated into the device under test. In other words, indirect testing naturally enables an effective and practical built-in self-test (BIST) for AMS-RF circuits. The combination of machine learning-based testing and integrated test instruments can help overcome many of the challenges associated with current AMS-RF testing.

Furthermore, it should be noted that AMS-RF BIST techniques, if feasible, not only reduce the complexity and cost of production-line testing, but also enable valuable features such as field testing, in-line testing, diagnostic capabilities, adaptive self-calibration, and self-healing, which can have a significant socioeconomic impact across a wide range of applications. Indeed, these features are key attributes for circuits deployed in applications that require high reliability (e.g., harsh environments, systems with limited or no access), or that are safety-critical (e.g., automotive, avionics, space, healthcare).

Objectives

As part of the Advanced Data Mining research focus area of the Labex PERSYVAL, the proposed exploratory project brings together the expertise of microelectronics designers, test engineers, and mathematicians specializing in data mining with the aim of exploring, identifying, and developing systematic methodologies for reliable and accurate integrated indirect test strategies for complex AMS-RF systems. To this end, the project is based on collaboration among three research institutions: the TIMA laboratory in Grenoble, France (Joint Research Unit 5159), IMSE-CNM in Seville, Spain (Joint Center of the University of Seville and the CSIC), and LIRIS in Lyon, France (UMR 5205).

Activities and News

Seminar on Bayesian Networks and Causal Inference

• By Prof. Alexander Aussem, LIRIS, Lyon (April 3, 2016).
• Abstract: The central objective of many studies in the social and medical sciences is to elucidate cause-and-effect relationships between variables or events. Although the appropriate methodology for extracting such causal relationships from data remains an open (and fiercely debated) question, graphical models provide a simple and practical means of communicating causal claims. The causal diagram represents the researcher’s understanding of the causal influences between measurable—but partially observed—quantities in the domain. In this presentation, I review the basics of causal inference in a graphical model and demonstrate, using simple examples, how non-trivial causal phenomena, paradoxes, and controversies in causal analysis can be understood, illustrated, and analyzed using the do-calculus developed by Judea Pearl. I also show how selection bias and missing data mechanisms can be represented in the graph and discuss the conditions under which consistent causal (or probabilistic) inferences can be made from such corrupted datasets.

Publications

G. Leger and M. J. Barragan, "Questioning the reliability of Monte Carlo simulation for machine learning test validation," 2016 21st IEEE European Test Symposium (ETS), Amsterdam, 2016, pp. 1–6.

G. Leger and M. J. Barragan, "Brownian distance correlation-directed search: A fast feature selection technique for alternative testing," *Integration, the VLSI Journal*, vol. 55, September 2016, pp. 401–414

M. J. Barragan, G. Leger, A. Gines, E. Peralias, A. Rueda, "On the limits of machine learning-based testing: a calibrated mixed-signal system case study," Design Automation and Test in Europe 2017 (DATE'17), accepted for publication.