Development of self-referencing thermally assisted magnetic random access memory (MRAM) cells.

Summary The objective of this thesis was the modeling and experimental demonstration of the read and write functionalities of a new thermally assisted magnetoresistive random access memory structure, the self-referenced MRAM. The magnetic stacking of the self-referenced MRAM is obtained from that of the thermally assisted MRAM by removing the reference antiferromagnetic layer, thus replacing the trapped reference layer by a free reading layer. By indirectly switching the magnetization of the readout layer via an external field, the magnetization direction of the trapped storage layer, and thus the stored logic level, can be measured in-situ. Thanks to the possibility of individually programming the two magnetic layers, the self-referencing MRAM can be considered as a magnetic logic unit, combining memory functionality with comparative logic in one device, which opens up new fields of application. The functionality of the read and write modes of the self-referencing MRAM has been demonstrated experimentally on a first set of samples. However, the required fields proved to be incompatible with an application in a functional industrial product. In order to optimize the required fields for writing and reading, a macrospin model, inspired by the Stoner-Wohlfarth model of magnetization reversal, was developed. By introducing the phenomena of magnetostatic coupling, RKKY and exchange between ferromagnetic and antiferromagnetic materials, a general form of the energy applicable to any MRAM magnetic stack has been obtained. A low field strength writing mode, based on the magnetostatic coupling between the read and storage layers, has been predicted by the model and experimentally demonstrated on a new batch of samples. An excellent agreement was obtained between the model and the experimental measurements. In order to study the reproducibility of writing, the influence of thermal activation was introduced by calculating the energy barriers related to the magnetic transitions performed during writing, and then compared to the experimental measurements of the writing probability of a new batch of samples. Once again, excellent agreement was obtained between the model and the experiment. Using the developed and validated model, a roadmap defining the magnetic stacks allowing to keep low operating fields for memory points up to 45 nm has been established. Due to fundamental technological limitations in field-switched MRAMs, it appeared essential to increase the individual storage capacity of each memory point to achieve higher storage density. A new angular storage method exploiting the mobility of the readout layer magnetization has been explored. Using the previously developed model, suitable samples were produced and experimentally demonstrated a storage capacity of up to 4 bits per individual memory point. However, the required operating ranges were found to be far above what is compatible with an industrial application. Using the model, a new writing method was proposed and allowed to establish a second roadmap towards the 45 nm technology node. Double barrier mirror structures were then studied, with an experimental demonstration of the feasibility of their fabrication, as well as their functionalities. In particular, a low-field writing mode, similar to the one observed in single-barrier self-referencing MRAMs, has been obtained. Finally, the adaptation of angular storage to these mirror structures has been modeled, leading to the proposal of a method allowing to store up to 8 bits per memory point.