Technology scaling into the sub-100nm domain implies that the effects of process, voltage, and temperature variations have a resounding effect on the performance of digital circuits. An increase in complexity has resulted in challenges in design and manufacturing of these circuits, as well as guaranteeing their accurate and reliable performance. Two key challenges in present-day circuit design are to ensure the long term reliability of circuits and to accurately estimate the arrival times and margins of the various paths in the circuit during timing analysis, under the presence of variations, at all operating conditions. Bias Temperature Instability (BTI), a long-term transistor degradation mechanism has escalated into a growing threat for circuit reliability; hence its exact modeling and estimation of its effects on circuit performance degradation have become imperative. Techniques to mitigate BTI and ensure that the circuits are robust over their lifetime are also becoming vital. Similarly, the impact of process variations has prompted new areas of research, to determine the timing and power estimates of the circuit, as accurately as possible. Since the effects of variations can no longer be ignored in high-performance microprocessor design, sensitivity of timing slacks to parameter variations has become a highly desirable feature to allow designers to quantify the robustness of the circuit at any design point. Further, the effect of varying on-chip temperatures, particularly in low voltage operation, has led to inverted temperature dependence (ITD) in which the circuit delay may actually decrease with increase in temperature. This thesis addresses some of these major issues in present day design.
We propose a model to estimate the long term effects of Negative Bias Temperature Instability (NBTI). We initially present a simple model based on an infinitely thick gate-oxide in transistors, to compute the asymptotic threshold voltage of a PMOS transistor, after several cycles of stress and recovery. We then augment the model to handle the effects of finite-oxide thickness, and justify the findings of several other experimental observations, particularly during the recovery phase of NBTI action. Our model is robust and can be efficiently used in a circuit analysis setup to determine the long-term asymptotic temporal degradation of a PMOS device, over several years of operation.
In the next chapter, we use this model to quantify the effect of NBTI, as well as Positive Bias Temperature Instability (PBTI), a dual degradation mechanism in PMOS devices, on the temporal degradation of a digital logic circuit. We provide a technique for gaging the impact of signal probability on the delay degradation numbers, and investigate the problem of determining the maximal temporal degradation of a circuit under all operating conditions. In this regard, we conclude that the amount of overestimation using a simple pessimistic worst-case model is not significantly large. We also propose a method to handle the effect of correlations, in order to obtain a more accurate estimation of the impact of aging on circuit delay degradation.
The latter part of this chapter proposes several circuit optimization techniques that can be used to ensure reliable operation of circuits, despite temporal degradation caused by BTI. We first present a procedure of combating the effects of NBTI during technology mapping in synthesis, thereby guardbanding our circuits with a minimal area and power overhead. An adaptive compensation scheme to overcome the effects of BTI through the use of adaptive body bias (ABB) over the lifetime of the circuit is explored. A combination of adaptive compensation and BTI-aware technology mapping is then used to optimally design circuits that meet the highest target frequency over their entire lifetime, and with a minimal overhead in area, average active power, and peak leakage power consumption. Lastly, we present a simple cell-flipping technique that can mitigate the impact of NBTI on the static noise margin (SNM) of SRAM cells.
In the final chapter of this thesis, we first present a framework for block based timing sensitivity analysis, where the parameters are specified as ranges - rather than statistical distributions which are hard to know in practice. The approach is validated on circuit blocks extracted from a commercial 45nm microprocessor design. While the above approach considers process and voltage variations, we also show that temperature variations, particularly in the low voltage design space can cause an anomalous behavior, i.e., the circuit delays can decrease with temperature. Thus, it is now necessary to estimate the maximum delay of the circuit, which can occur at any operating temperature, and not necessarily at a worst case corner setting. Accordingly we propose a means to efficiently compute the maximum delay of the circuit, under all operating conditions, and compare its performance with an existing pessimistic worst-case approach.