Ferroelectric hafnium oxide films are one of the most promising functional materials for a new generation of nonvolatile memory for data storage because of a number of excellent performance features, including nanosecond speed, high endurance, low power consumption, full scalability, and perfect compatibility with Si technology. However, the commercialization of this memory is hindered by its limited retention time, which still does not meet the 10-year electronics-industry standard. To date, much effort has been focused on the improvement of retention via material and interface engineering. Meanwhile, the evaluation and comparison of engineering results have been hampered, since there is still no unified retention test. Indeed, a specific feature of the retention performance is that it cannot be measured directly, because 10 years is required to measure the actual polarization loss. In this paper, we present a physical model that predicts the retention loss in ${\mathrm{Hf}\mathrm{O}}_{2}$-based memory during long-term information storage. The time-dependent polarization loss originates from the phenomenon of polarization imprint, which consists in the temporal development of a preferential polarization state in a poled ferroelectric. For high reliability of the retention calculation, the model takes into account the specific features of imprint evolution in the ferroelectric material ${\mathrm{Hf}\mathrm{O}}_{2}$. Among these is the origin of the imprint in this material, which consists in the interplay of two physical mechanisms, specifically, charge injection into interface traps and migration of mobile charged defects across the functional layer. The second property of ${\mathrm{Hf}\mathrm{O}}_{2}$ taken into account is the different rates of imprint evolution in two types of ${\mathrm{Hf}\mathrm{O}}_{2}$ film that exist within each memory cell. These two types of film originate from two populations of domains that appear during crystallization of the ${\mathrm{Hf}\mathrm{O}}_{2}$ and persist during the whole lifetime of the memory cell. The high accuracy of the retention model makes it possible to implement an express retention test, i.e., to predict the retention time in record short time (1 h) for any particular sample, whereas usually a retention test takes at least several days. For ${\mathrm{Hf}}_{0.5}{\mathrm{Zr}}_{0.5}{\mathrm{O}}_{2}$-based capacitors, the retention model shows excellent agreement with experimental results. The proposed model can serve not only for the implementation of this express retention test, but also as a tool for intelligent material engineering in the field of nonvolatile ferroelectric memory.
Read full abstract