Experimental and Industrial Evaluation of Variability Resilient Schemes
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
As semiconductor devices shrink further, the effects of process variation become more and more pronounced. They negatively impact the speed and power consumption of Integrated Circuits (ICs). Traditionally, a worst-case based design methodology is used to compensate for their potential impact. For modern technology this method means an increased penalty in area and power consumption. Therefore, Variability Resilient Schemes (VRSs) are introduced. They are used to improve the speed of ICs with slow processing and decrease the power consumption of ICs with fast processing. Thanks to these techniques a Better-Than-Worst-Case (BTWC) design methodology can be used. It improves performance, area, and power consumption. The work in this thesis is performed within the Variability Resilient Architectures (VRA) project at NXP Semiconductors. In this project the potential of body biasing, clock stretching, and error avoidance flip-flops as VRSs is investigated. Circuits with VRSs run closer to their optimal performance. This means more extensive testing is necessary compared to worst-case based designs. Also, the usage of error avoidance flip-flops bring new testability challenges. In order for designs with VRSs to be interesting yield and testability need to be high enough. This thesis evaluates the testability challenges of the VRSs used in the VRA project. A cost model is proposed to compare the cost of worst-case based designs with designs with VRSs. The testability of the error avoidance flip-flop is evaluated and a solution is proposed for stuck-at and path delay testing of the error output. Finally, the potential of body biasing to compensate process variation is investigated, by comparing the performance of a worst-case based design and a BTWC design, which has an area reduction of 25%. The performance of both designs is compared by measuring path delays in both simulations and measurements. It is shown that with around 0.2 V forward body biasing the performance of the BTWC design is comparable to the worst-case design.