Leakage currents in hafnia-based ferroelectric capacitors: modeling and validation

The recent advancements in machine learning, along with its increasingly widespread applications, have soon highlighted the limitations of the conventional Von-Neumann architecture, particularly excessive power consumption and high delay. An emerging computing paradigm inspired by the brain referred to as Neuromorphic computing promises to address these challenges. Some benefits of this new paradigm stem from the usage of innovative memory elements, such as ferroelectric capacitors (FeCaps). Hafnia-based ferroelectric memories are among the most promising emergent memory technologies due to their high endurance, high switching speed and low power consumption. Of particular relevance for the characterization of hafnia-based FeCaps is the study and modeling of the leakage currents flowing through the capacitor stack. If leakage is not modeled properly, when the FeCap model is used in the context of circuit design, the functionality of the circuit could be severely impaired. To address this need, this work aims at modeling the leakage currents flowing through the FeCap stack. The leakage model has been implemented inside a FeCap compact model for analog circuit simulations. A compact model provides a mathematical description of the physics of a specific electronic device. The core of a compact model describes the fundamental behavior of the device, and can be enriched by including additional phenomena such as leakage currents. Compact models are fundamental for integrated circuits development, because their simplicity and high accuracy allow for the adoption of computationally intensive Monte Carlo methods, which are essential for designing scalable and reliable integrated circuits. This work started by updating the initial, resistance-based implementation of leakage to include the relevant conduction mechanisms in nanometric thin films. I evaluated the impact of three different leakage models by comparing the outputs of computer simulations with available experimental data. This comparison demonstrated that a physical leakage model incorporating three different conduction mechanisms - Poole-Frenkel, Fowler-Nordheim tunneling, and Schottky emission - provided the most accurate explanation of the leakage occurring in the FeCap stack. Subsequently, I studied the physical model in detail, highlighting how a restricted number of parameters profoundly influence the behavior of each leakage current. Quasi-static DC measurements were then conducted on existing devices with different ferroelectric thickness to validate the model. For each thickness, multiple measures were taken on different FeCaps, in order to obtain a statistically meaningful dataset for calibrating the leakage model. Considering that Poole-Frenkel was the dominant leakage mechanism, I extracted physically plausible ranges for the leakage parameters by calibrating the leakage model with experimental data. The results confirmed that Poole-Frenkel was clearly dominant, although the sample with the thinnest ferroelectric layer showed a current trend which deviated from the linear current-thickness trend seen in the other samples with thicker layers. This study provided a reliable foundation for further refinement of the model FeCaps in future research.