UC San Diego

The continued device scaling trend and the aggressive integrated circuit design style have shifted the major device failure mechanism from stuck-at fault types to marginal failures induced by timing uncertainty and signal noise. The production test methodologies currently employed by industry, however, are still based on the traditional structural test schemes that focus on the detection of permanent defects, failing to account for emerging failure mechanisms in nanometer scale designs. The inability of current test methodologies in adapting to the failure mechanism shift imposes critical challenges to the IC providers, mainly observed as significant product quality degradation and yield loss. To make things worse, the marginal failures result in highly ambiguous failure syndromes, invalidating traditional assumptions employed in silicon debugging. The degraded test quality and yield, combined with inaccurate failure diagnosis, lead to a lengthened design-fabrication-debugging cycle needed for ramping up the yield and quality for final production, significantly slowing down the time-to-market and boosting the overall product cost. Maintaining high quality yet low cost production test for nanometer scale integrated circuits necessitates a comprehensive examination of marginal failure scenarios while minimizing yield loss. Reducing the time-to-market cycle relies on an accurate identification of marginal failure locations and causalities to pinpoint the design and fabrication weaknesses that have gross quality impact. The challenges, though, are the resolution to the paradox between overscreening and underscreening that are simultaneously taking place in today's industrial testing practice, and the extraction of sensible diagnostic signals from highly ambiguous fault behaviors of marginal failures. The presented thesis work overcomes these challenges through the proposition of an innovative marginal failure aware test and diagnosis scheme, capable of thoroughly targeting the functional mode failure scenarios with a low cost structural test platform and the accurate identification of failure-induced feature change in large volume test data. A comprehensive production ramp-up flow, constructed based on the proposed test and diagnosis schemes, is furthermore presented to guide the silicon debugging, test optimization, and yield/quality learning activities, so as to minimize the time-to-market. From a technical point of view, this thesis work analyzes the power ground noise in functional and testing modes and its impact on circuit timing robustness, with a focus on the differentiation of the functional mode timing failures from the pure testing mode ones, thus enabling a clear decomposition of the noise treatment strategies for different operation scenarios. A set of tightly-coupled approaches, including 1) noise resilience in testing related circuitry for overscreening minimization, 2) approximation of worst-case functional mode noise in structural testing for marginal timing failure detection, and 3) diagnosis of noise- induced timing failure diagnosis in scan paths and scan clock trees for design optimization, are presented to attain the overall goal of high yield, low test escape rate, and fast silicon re-spin. These techniques are developed with the consideration of enabling a seamless adaptation of industrial flows by delivering maximal compatibility to mainstream design-for-testability architectures and testing platforms employed in nanometer scale designs. The successful incorporation of these techniques will significantly expedite the silicon production ramp-up process with highly reduced risk and cost