Resumen de Reliability-aware circuit design to mitigate impact of device defects and variability in emerging memristor-based applications
In the last decades, semiconductor industry has fostered a fast downscale in technology, propelling the large scale integration of CMOS-based systems. The benefits in miniaturization are numerous, highlighting faster switching frequency, lower voltage supply and higher device density. However, this aggressive scaling trend it has not been without challenges, such as leakage currents, yield reduction or the increase in the overall system power dissipation. New materials, changes in the device structures and new architectures are key to keep the miniaturization trend. It is foreseen that 2D integration will eventually come to an insurmountable physical and economic limit, in which new strategic directions are required, such as the development of new device structures, 3D architectures or heterogeneous systems that takes advantage of the best of different technologies, both the ones already consolidated as well as emergent ones that provide performance and efficiency improvements in applications.
In this context, memristor arises as one of several candidates in the race to find suitable emergent devices. Memristor, a blend of the words memory and resistor, is a passive device postulated by Leon Chua in 1971. In contrast with the other fundamental passive elements, memristors have the distinctive feature of modifying their resistance according to the charge that passes through these devices, and remaining unaltered when charge no longer flows. Although when it appeared no physical device implementation was acknowledged, HP Labs claimed in 2008 the manufacture of the first real memristor. This milestone triggered an unexpectedly high research activity about memristors, both in searching new materials and structures as well as in potential applications. Nowadays, memristors are not only appreciated in memory systems by their nonvolatile storage properties, but in many other fields, such as digital computing, signal processing circuits, or non-conventional applications like neuromorphic computing or chaotic circuits. In spite of their promising features, memristors show a primarily downside: they show significant device variation and limited lifetime due degradation compared with other alternatives.
This Thesis explores the challenges that memristor variation and malfunction imposes in potential applications. The main goal is to propose circuits and strategies that either avoid reliability problems or take advantage of them. Throughout a collection of scenarios in which reliability issues are present, their impact is studied by means of simulations. This thesis is contextualized and their objectives are exposed in Chapter 1. In Chapter 2 the memristor is introduced, at both conceptual and experimental levels, and different compact levels are presented to be later used in simulations. Chapter 3 deepens in the phenomena that causes the lack of reliability in memristors, and models that include these defects in simulations are provided. The rest of the Thesis covers different applications. Therefore, Chapter 4 exhibits nonvolatile memory systems, and specifically an online test method for faulty cells. Digital computing is presented in Chapter 5, where a solution for the yield reduction in logic operations due to memristors variability is proposed. Lastly, Chapter 6 reviews applications in the analog domain, and it focuses in the exploitation of results observed in faulty memristor-based interconnect mediums for chaotic systems synchronization purposes. Finally, the Thesis concludes in Chapter 7 along with perspectives about future work.
