Niet-lineaire diëlektrische respons in ferro-elektrica en de toepassing in geheugencellen

With the discovery of a ferroelectric phase in HfO2 about a decade ago, a new line of research has emerged for the application of ferroelectrics in nonvolatile memory cells as well as in the context of machine learning such as in-memory computing. Considering the very low power consumption of the ferroelectric switching mechanism, this line of research fits perfectly in the current trend towards less power-hungry and therefore more sustainable electronic systems. When switching a ferroelectric, a displacement charge is generated on its electrodes that can be captured by conventional sense amplifiers and transformed into an output voltage. The drawback of such a sensing mechanism is that it requires the data to be rewritten after every (destructive) read operation. This is a major disadvantage in a memory structure as the write-endurance is now essentially the same as the read-endurance while typically much higher read cycles are needed as compared to write cycles. The purpose of this thesis is to investigate the possibility of reading the stored data non-destructively (i.e., sensing the polarization state) without actually switching the ferroelectric capacitor. This is possible due to the highly non-linear behavior of the dielectric response in a ferroelectric, even before reaching the coercive field, which in-turn gives rise to a non-volatile capacitive memory window. At this moment, the nature of the non-linear dielectric response in this elastic regime, and, therefore, the origin of the capacitive memory window and limitations of the non-destructive read are not completely understood. The intention is to study their dependence on various parameters such as temperature, built-in electrical fields, pulse frequency as well as the material parameters. This should lead to an in-depth understanding of the physics at play and provide the guidelines to design ferroelectric capacitors (such as choice of ferroelectrics, electrode configurations, insertion of other materials in the stack, etc.) for optimum read performance in memory and machine learning applications.