Recent Progresses in Hafnia-Based Materials and Devices

With the rapid development of big data, autonomous driving, and artificial intelligence technologies, competition among countries in the AI field has become increasingly fierce. Advanced AI models represented by OpenAI and DeepSeek rely on massive amounts of unstructured data (such as speech and images) for training. however, conventional computing architectures encounter limitations such as the "memory wall" and "power wall" during large-scale data computation. Therefore, breaking through the limitations of traditional architectures and exploring high-performance computing paradigms has become an urgent necessity. The efficient information processing mechanism of the human brain provides inspiration for this effort. Inspired by this, neuromorphic computing has emerged. Its core concept is based on an in-memory computing architecture, integrating data storage and computational functions within the same computational unit to reduce energy loss caused by data migration and enhance computational parallelism. It exhibits broad application potential in areas such as artificial intelligence and large-scale data analysis. Nevertheless, breakthroughs in materials and devices remain crucial for realizing high-performance neuromorphic computing systems.

Hafnia-based materials have emerged as highly promising candidates for neuromorphic computing devices due to their exceptional compatibility with complementary metal oxide semiconductor (CMOS) technology, high dielectric constants, and stable ferroelectric properties at ultrathin scales. In fact, as early as 2007, non-ferroelectric HfO₂ was widely applied in CMOS as a high-dielectric-constant (K) gate insulator. Therefore, the discovery of ferroelectricity in hafnia-based materials is critical for advancing CMOS compatibility with ferroelectric materials. This has also altered the conventional understanding of fluorite crystal-structured materials and opened up possibilities for realizing ferroelectricity at ultrathin scales. Subsequently, the crystal structures and origins of ferroelectricity in hafnia-based oxides were systematically studied through theoretical calculations and experimental approaches. The crystal structures of hafnia-based materials are primarily classified into two categories: polar and nonpolar. Current research indicates that their ferroelectricity primarily arises from the non-centrosymmetric orthorhombic ferroelectric phase. Furthermore, recent studies have also reported polar rhombohedral phases (R3m and R3). In fact, our recent studies have demonstrated rich phase-transition pathways between polar orthorhombic phases and nonpolar monoclinic and tetragonal phases, highlighting the structural complexity of this material. In this article, we systematically introduce various structural phase-transition techniques for hafnia-based materials, such as temperature control, doping, oxygen vacancy engineering, and stress and interface engineering methods. It should be noted that a single factor (e.g., doping alone or solely oxygen vacancy modulation) might not significantly alter the phase composition of HfO₂ films; however, when doping, oxygen defect modulation, mechanical stress, and interface engineering act together, they can significantly modify the free-energy landscape, kinetically "freezing" originally non-ferroelectric equilibrium phases into metastable ferroelectric phases. Device fabrication is also crucial for realizing high-performance neuromorphic computing. Device structures such as ferroelectric RAM (FeRAMs), ferroelectric field-effect transistors (FeFETs), and ferroelectric tunneling junctions (FTJs) have demonstrated remarkable potential in neuromorphic computing applications. FeRAM devices utilize reversible polarization switching to achieve binary and multilevel memory operations, demonstrating significant potential in terms of operation speed, low power consumption, and integration density. FeFET devices, benefiting from their ultrathin ferroelectric gate layers, exhibit advantages including low power consumption, rapid switching speeds, and miniaturization capability. Similarly, FTJ devices, characterized by ultra-thin ferroelectric barriers, enable tunable tunneling resistances and non-destructive readout. These properties provide substantial benefits for neuromorphic computing applications, facilitating multilevel data storage and high-density device integration.

Although significant achievements have been made in hafnia-based ferroelectric materials and their device applications, several fundamental and practical challenges persist.

The preparation of thin-film devices based on HfO₂ faces significant challenges related to electric field cycling stability. Early cycling phases experience the wake-up effect, attributed to field-induced phase changes, charge detrapping and redistribution, domain wall depinning, and reduction of depolarization fields from breakdown in nonpolar layers near electrodes. Later cycling stages suffer fatigue, indicated by a reduction in remanent polarization (P_r), often ending in device breakdown.

Specifically, FeFET devices encounter reliability challenges related to their metal-ferroelectric-insulator-semiconductor (MFIS) gate stack; the interfacial layer (IL), due to high permittivity and coercive fields, undergoes increased stress during polarization switching, causing premature fatigue and short retention times driven by depolarization fields. Optimizing the IL is thus crucial for improving FeFET reliability. FeRAM, despite demonstrating promising retention stability, still faces limitations in commercialization due to polarization fatigue, imprint effects, wake-up phenomena, and substantial device-to-device variability, necessitating further optimization of electrode design, composition, interface treatments, and crystallization anneals. Meanwhile, FTJs rely on ferroelectric polarization-modulated tunneling current, with main challenges including low tunnel current due to relatively thick tunneling barriers, as well as endurance issues. Although FTJ technology holds significant potential for 3D stacking and in-memory computing, it remains at an early development stage, requiring further research to assess future applicability.

The discovery of ferroelectricity in hafnia oxide has opened numerous promising pathways in nanoelectronics, enabling significant advancements in non-volatile memories, sensors, actuators, and neuromorphic computing. Future research efforts should focus on overcoming current challenges such as material fatigue and reliability issues inherent in FeFET, FeRAM, and FTJ devices. Continuous advances in material science, device engineering, and integration technology will thus play a pivotal role in establishing hafnia-based ferroelectric materials as cornerstone components of next-generation computing platforms.