Leveraging pre-silicon data to diagnose out-of-specification failures in mixed-signal circuits

Abstract

Diagnosing out-of-specification failures in mixed-signal circuits has become increasingly challenging due to: (1) failures caused by interactions between input-signal conditions and design uncertainties, and (2) the need to identify critical input and uncertainty conditions that cause these regions. We propose a simulation-driven approach that first uses ensemble learning to extract if − then rules that naturally solve both problems. By ranking, pruning and clustering these rules, we then construct non-linear failure regions which can be directly employed for pre-silicon debug, as demonstrated on a phase-locked loop circuit. Furthermore, these regions can be used to guide test pattern generation and/or assist with post-silicon debug.

Using Presilicon Knowledge to Excite Nonlinear Failure Modes in Large Mixed-Signal Circuits

“…The simplest such knowledge representation scheme is a set of hypercubes lying in a highdimensional parameter space such as the 2-D example in Figure 1. Such "failure regions" of the form R i ¼ \ j a ij G x j G b ij can be constructed from simulation data using standard machine learning algorithms [1]. While any model that returns PðF jxÞ may be used for the purpose of failure excitation, an advantage to the "failureregion"-based approach is that it is easy to map chosen tests to preidentified failure modes.…”

“…In this section, we describe how the failure regions we discuss previously may be used to rank parameters to guide design-for-test decisions. Each failure region R i in [1] already contains some amount of information regarding a given parameter x j . If the parameter x j is critical for region R i , ½a ij ; b ij does not default to ½À1; 1, where a ij x j b ij defines bounds for x j in R i .…”

“…More importantly, multiple parameters may interact nonlinearly to cause out-of-specification failures. Detecting such out-ofspecification failure mechanisms in the presilicon domain [1] helps, but presilicon assumptions about design uncertainty may not be entirely accurate.Moreover, downstream effects of out-ofspecification failures on system-level performance are not always known, making postsilicon testing of such failure mechanisms critical.It is impossible to test for functional failures under every possible input and parametric variation, making design for test (DfT) indispensable. Identifying what parameters to control, and what values to excite, is especially difficult in highdimensional, highly nonlinear systems since linearity and worst case corner assumptions no longer apply.…”

“…Traditional sensitivity-based approaches find it difficult to scale as a result. We instead start with prior work on identifying presilicon failure mechanisms [1] and demonstrate a novel way to excite these failure mechanisms. More specifically, the contributions made in this work are twofold.1) We develop an information-theoretic parameter ranking scheme to assist with design-for-test decisions.…”

“…More importantly, multiple parameters may interact nonlinearly to cause out-of-specification failures. Detecting such out-ofspecification failure mechanisms in the presilicon domain [1] helps, but presilicon assumptions about design uncertainty may not be entirely accurate.…”