As CMOS technologies have shrunk to tens of nanometers, aging problems have emerged as a major challenge. There has been tremendous progress in developing new methods for modeling and diagnosing reliability at the level of individual transistors, but much less work on propagating these models to higher levels of abstraction to analyze and optimize the reliability of larger circuits. This talk will provide an introduction to various circuit aging mechanisms and will then discuss research that develops computer-aided design techniques for estimating and enhancing the reliability of large digital circuits, examining solutions that could practically be applied to analyze or improve the lifetime of a design while maintaining consistency to accurate device-level models and the associated physics.CIRCUIT-LEVEL ANALYSIS OF AGING Aging effects are described by the classical bathtub curve: after a steep initial failure rate, the failures level off for a while before rising again, resulting in a bathtub-like shape for the number of failures as a function of time. Aging is strongly sensitivity to process parameter variations, as well as to onchip temperature and supply voltage level changes. Here, we overview analysis methods for prominent aging mechanisms, based on both conventional and newer physical models. Bias temperature instability (BTI) is a device aging phenomenon that causes threshold voltage shifts over long periods of time in the presence of voltage stress at the gate, eventually causing the circuit to fail to meet its specifications. A PMOS transistor in an inverter experiences negative BTI stress when its gate node is at logic 0, and the resulting increase in the threshold voltage is partially reversed when the voltage stress is removed (i.e., a logic 1 is applied). A similar phenomenon of positive BTI affects the threshold voltage of NMOS devices when they are stressed, and relaxes the degradation on the removal of stress. The magnitude of degradation depends on the ratio of the stressed time to unstressed time, i.e., the signal probability (SP) of the gate input. There are two widely-used theories for BTI. The reaction-diffusion (R-D) model [1,2] explains the degradation due to the accumulation of positive charges as Si-H bonds at the interface are broken and hydrogen diffuses away; the removal of this stress allows partial restoration of these bonds. The charge-trapping (CT) model [3] provides an alternative explanation, where defects in gate dielectrics can capture charged carriers, causing the threshold voltage to degrade. Neither theory fully explains experimental observations, and it has been posited that R-D and CT mechanisms coexist. The CT model predicts high variability in small devices, making variability a major factor for memories. For logic circuits, however, large transistors and averaging effects on critical paths significantly mitigate variability [4]. Hot carrier injection (HCI) in MOSFETs is caused by the acceleration of carriers (electrons/holes) under lateral electric fields in th...
