h1

TY  - THES
AU  - Bengel, Christopher Reinhard
TI  - Variability-aware compact modeling of valence-change-mechanism based devices for computation-in-memory
PB  - Rheinisch-Westfälische Technische Hochschule Aachen
VL  - Dissertation
CY  - Aachen
M1  - RWTH-2023-08986
SP  - 1 Online-Ressource : Illustrationen, Diagramme
PY  - 2023
N1  - Veröffentlicht auf dem Publikationsserver der RWTH Aachen University
N1  - Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023
AB  - Today’s modern society relies on semiconductors for a wide range of fields, especially in the domain of information technology. Advancements there are heavily reliant on performance and efficiency improvements of the underlying hardware, especially memory technologies and processing units. For these two types of technology, most of the performance improvements have been achieved through the miniaturization of their core element, the Metal Oxide Semiconductor Field Effect Transistor (MOSFET). Further miniaturization is however becoming increasingly difficult and expensive, as fundamental physical laws are greatly increasing the challenges for technologists and circuit designers. Another path to increase the performance of computing systems is optimizing their architecture, by further utilizing parallel architectures or by performing computation-in-memory. One proposal for the technological and architectural problems is the usage of resistive switching devices, such as Valence Change Mechanism (VCM) based devices to enable computation in memory. In these devices the information of the cell is stored in the resistance of the cell and can be varied over orders of magnitude in a binary or analog fashion, making them promising candidates for computing applications. Independent of the specifics of the various applications, all applications require a deep understanding of the device behavior and their variability. Furthermore, the impact of these physical properties and the variability on the performance of the applications is important. This thesis investigates the physical modeling, including variability, of filamentary and bipolar switching VCM cells, as well as their applications for Computation-in-Memory. The compact modeling of the device-to-device, cycle-to-cycle and read-to-read variability is verified through extensive experimental investigations considering the cell statistics. The compact model developed in this thesis is shown to be able to describe various filamentary switching VCM systems, such as HfOx, ZrOx and TaOx. Initially, the qualitative behavior could be matched, e.g. for the SET and RESET kinetics as well as for IV measurements. in a second step, the difference between device-to-device and cycle-to-cycle variability could be demonstrated. Finally, the focus was on describing reliability, which is affected by read disturb and read noise. The space of Computation-in-memory applications can be split into logic applications, machine learning accelerators and neuromorphic computing applications. All of these fields will be studied in this thesis. In general, the approach is to start by defining the most important device characteristics for each application and by then showing, that this characteristic or combination of characteristics can be described using the compact model. The applications are then built from the ground up, while respecting lessons learned from the lower detail levels.
LB  - PUB:(DE-HGF)11
DO  - DOI:10.18154/RWTH-2023-08986
UR  - https://publications.rwth-aachen.de/record/969153
ER  -