Oxide-based ferroelectric materials are expected to have a key role in the future of semiconductor devices. Tim Böscke, Ulrich Böttger and Uwe Schroeder recount how ferroelectricity in hafnium oxide thin films was first discovered.

We systematically studied device-to-device variability and retention in pseudo metal-ferroelectric-metal-insulator-semiconductor (MFMIS) based ferroelectric memory. Threshold voltage ($\boldsymbol{V}_{\text {th }}$) variability is found to increase with shrinking MFM capacitor area, which is consistent with the random polarization switching in polycrystalline $\text{Hf}_{1-\mathrm{x}} \text{Zr}_{\mathrm{x}} \mathrm{O}_{2}$. In contrast, the variability can be suppressed by scaling down the total transistor channel area, revealing a universal dependence of the variability on the MFM-to-transistor area ratio. Retention characteristics were measured with various ferroelectric and gate-oxide thicknesses. Fast memory window degradation was observed for thinner HZO or thinner gate oxides, indicating that charge loss through these dielectric layers, rather than intrinsic ferroelectric depolarization, dominates retention characteristics. Furthermore, devices programmed with a reduced memory window exhibit markedly improved retention, even with thinner dielectrics. These results establish comprehensive understanding for variability and retention, providing insight into the design methodologies for ferroelectric memories and computing-inmemory applications.

$\text{Zr}_{\mathrm{x}} \text{Hf}_{1-\mathrm{x}} \mathrm{O}_2$ with an optimized Hf-to-Zr ratio has demonstrated a dielectric constant exceeding 55 within the DRAM operating voltage range. This enhancement can reduce the power consumption of DRAM and is expected to meet the growing demands driven by the rapid expansion of AI services. This work investigates the Hf-to-Zr composition in $6 \text{nm} \text{Zr}_{\mathrm{x}} \text{Hf}_{1-\mathrm{x}} \mathrm{O}_{2}$ metal-insulator-metal capacitors to achieve a high dielectric constant. The physical mechanisms responsible for dielectric enhancement are analyzed, and comprehensive reliability characterizations are conducted, including voltage polarity-dependent dielectric constant loss, retention behavior, and temperature stability. These results define the usable operating conditions and practical limitations of dielectric constant enhancement observed in $\mathbf{Z r}_{\mathbf{x}} \mathbf{H f}_{\mathbf{1}}, { }_{x} \mathrm{O}_{2}$ for DRAM applications.

Nanoscale domain nucleation and domain-wall propagation govern switching in hafnia-based ferroelectrics. However, it is not known how electrode interfaces and thermal processing govern these processes at the grain scale. Here, bias-dependent piezoresponse force microscopy and pulsed switching measurements reveal how bottom electrodes and annealing control polarization reversal in 10-nm-thick Hf0.5Zr0.5O2 capacitors. TiN promotes rapid nucleation and cross-grain propagation, NbN leads to gradual intragrain switching, and MoO2 strongly suppresses ferroelectric switching. The fastest reversal is obtained for TiN/HZO/TiN annealed at 600 °C, consistent with a fine-grained microstructure and grain boundaries that support cooperative propagation. Quantitative analysis of pulse-switching kinetics using a modified Kolmogorov-Avrami-Ishibashi framework yields comparable time constants for nucleation and growth, indicating that both processes proceed concurrently. These results establish a grain-scale mechanistic link between interfaces, microstructure, and switching kinetics in Hf0.5Zr0.5O2 capacitors.

The current market launch of HfO2 -based ferroelectric devices relies on the control of the inherent oxygen vacancies (OVs) and their impact on the ferroelectric performance. Due to the necessary stabilization of the ferroelectric phase by doping, several dopants are investigated for their applicability to control the vacancy concentration. Hf signatures in X-ray photoemission spectra are often used as an indication of OVs for both qualitative and quantitative analysis. The analysis of Y doped HfO2 (Y:HfO2) as investigated by hard x-ray photoelectron spectroscopy (HAXPES) reveals the inapplicability of the Hf signature for a quantitative determination of OVs in the case of heterovalent doping and is restricted to pure HfO2 or isoelectronic substitution of Hf by, for example, Zr.

Scaling ferroelectric Hf0.5Zr0.5O2 (HZO) films below 10 nm is critical for low-voltage non-volatile memory but remains challenging due to phase instability and interface-related depolarization fields. Here, we demonstrate that the electrode-ferroelectric interface is the key factor for stabilizing the ferroelectric orthorhombic phase in sub-10 nm HZO films. By comparing films down to 5 nm thickness with TiN and W electrodes, we reveal that W electrodes induce significantly lower in-plane tensile strain due to the formation of an amorphous, conductive WOx interfacial layer. This strain relaxation suppresses the non-polar tetragonal phase favored in ultrathin films, whereas standard TiN electrodes generate high tensile strain that stabilizes the undesirable t-phase. Moreover, the conductive nature of the WOx layer suppresses the depolarization fields typically caused by dielectric TiOxNy interfaces. Consequently, 5 nm HZO films with W electrodes exhibit higher remanent polarization, lower coercive fields, and negligible wake-up effects compared to those with TiN electrodes. Furthermore, we show that the strain-induced performance loss in films with TiN electrodes can be reduced by modifying the Hf:Zr stoichiometry, effectively compensating for the interface strain. These findings establish a critical design rule for interface and strain engineering, providing a pathway to reliable sub-10 nm hafnium-based ferroelectric devices.

ZrxHf1-xO2-based materials have attracted considerable interest due to their excellent scalability and compatibility with complementary metal-oxide-semiconductor technology. Following the discovery of ferroelectricity in these materials, it has opened new avenues not only redefining ferroelectric memory but also as a promising high-k dielectric for advanced non-volatile and volatile memory applications by utilizing specific crystalline phases to tune the electrical parameters. Various fundamental factors influencing the formation of ferroelectricity in ZrxHf1-xO2 have been identified and distinguished from those in the classical perovskite ferroelectrics. Notably, a sharp increase in the dielectric constant near 0 V observed in Zr-rich ZrxHf1-xO2 films achieved by tuning the fabrication parameters has been attributed to the presence of a morphotropic phase boundary. This study investigates 6 nm Zr-rich ZrxHf1-xO2 thin film metal-insulator-metal capacitors using a combination of experimental methods and machine learning-based molecular dynamics simulations. The study provides insight into the physical mechanisms that enhance the dielectric constant near 0 V and attributes it to the orthorhombic phase rather than a morphotropic phase boundary. The work discusses the limitations in the practical application of the high dielectric constant observed near 0 V. Additionally, it highlights similarities and differences between ZrxHf1-xO2 and the well-known morphotropic phase boundary in PbZrxTi1-xO3.

HfxZr1-xO2 thin films have excellent complementary metal-oxide semiconductor compatibility and scalability compared to other ferroelectric materials. This makes them a promising candidate for non-volatile memory applications. However, the polymorphism of the materials presents a challenge in stabilizing the ferroelectric properties. Since the wake-up free non-volatile memory applications require the presence of ferroelectric properties in the pristine state of the films without additional electric field cycling, it is necessary to understand how to promote the ferroelectric orthorhombic phase formation. In this work, the interaction between in-plane tensile strain and phase formation of atomic layer deposition grown HfxZr1-xO2 thin films with different thicknesses and different compositions is demonstrated. By combining the biaxial in-plane tensile strain with the electric switching field and remanent polarization, it is observed that the best ferroelectric properties correlated with an in-plane tensile strain range of 0.4-0.6%. Moreover, the observed correlation between strain and phase formation indicates that strain exerts an influence on phase formation in the pristine state, and that phase formation, in turn, affects strain during electrical field cycling. This work is expected to be helpful to improve the ferroelectric properties in HfxZr1-xO2 films, which can be processed for different memory devices with specialized requirements.

Wurtzite‐structured ferroelectrics, such as aluminum scandium nitride (Al 1‐x Sc x N), are among the most promising candidates for implementing innovative nonvolatile memory concepts into commercial technologies. However, the opposite state (OS) retention limits the long‐term retention performances. Since verifying the retention requirement directly up to 10 years, as typically targeted by commercial technologies, is timely unfeasible, developing a model to predict the 10‐year OS retention performances of wurtzite‐structured ferroelectrics is of the utmost importance for validating their reliable long‐term operation. This work demonstrates the imprint as the primary factor in determining the Al 1‐x Sc x N OS retention performances. A model to predict the 10‐year OS retention performances of Al 1‐x Sc x N is developed by directly correlating the coercive field (E c ) increase with the slowing down of the switching dynamics through the relationship between the characteristic switching time and applied electric field magnitude to E c ratio (E/E c ). The model is verified with OS retention measurements performed on Al 0.85 Sc 0.15 N capacitors after baking for up to 2 weeks at 150 °C. The E/E c that guarantees reliable 10‐year OS retention performances is extrapolated for pulse widths down to the nanosecond range. Finally, electric field switching cycling is proven as a viable strategy for recovering from the imprint that degrades OS retention performances.

In this study, a detailed investigation of the redox behavior of cobalt(II/III) oxide-hydroxide is presented. With the enhanced sensitivity and unique pulse nature of square-wave voltammetry (SWV), two distinct electrode processes could be observed, proposing a two-step oxidation of Co(OH)(2) according to: Co(OH)(2)-> CoO(OH)-> CoO2. In the initial scan of cyclic voltammetry, Co(OH)(2) is irreversibly oxidized to CoO(OH) with the quasireversible Co3O4/CoO(OH) redox system prevailing in subsequent scans. The system was further kinetically characterized. Theoretical studies and the unique peak splitting in SWV revealed that the electrode reaction is associated with an anodic charge transfer coefficient of alpha=0.59 and an apparent standard rate constant of k(s,app)=(2.9 +/- 0.1) & sdot; 10(-5) cm s(-1). To cope with the complexity of the electrode process pertinent to the redox couple Co3O4/CoO(OH), the recently introduced and advanced multi-frequency electrochemical Faradaic spectroscopy (MEFS) was applied. With rapid measurement times of only 4 s, compared to multiple, hour-long experiments for square-wave and cyclic voltammetry, an apparent standard rate constant of k(s,app)=(2.2 +/- 0.2) & sdot; 10(-5) cm s(-1) was obtained with MEFS, which is quite close to the established methods, highlighting the advantages of this novel square-wave derived technique.

Conventional FeRAM readout methods are destructive, requiring polarization switching of the FE capacitor (FeCAP) and write-back, which reduces endurance, increases latency, and energy consumption. Prior works on non-destructive readout (NDRO) relied on capacitance memory window (MW), which is slow and requires asymmetric FeCAP structure, compromising retention and increasing circuit complexity. Here, we present a novel NDRO method utilizing ultrafast transient response which applies to both symmetric and asymmetric structures. We experimentally demonstrate sub-ns read operations without altering the polarization state, achieving >10(13) read cycles (limited by test time). This structureagnostic method improves retention (tested at 125 degrees C), endurance, and simplifies implementation, thus paving the way for fast, energy-efficient FeRAM-based solutions.

Abstract Zr x Hf 1‐x O 2 ‐based materials have attracted considerable interest due to their excellent scalability and compatibility with complementary metal‐oxide‐semiconductor technology. Following the discovery of ferroelectricity in these materials, it has opened new avenues not only redefining ferroelectric memory but also as a promising high‐k dielectric for advanced non‐volatile and volatile memory applications by utilizing specific crystalline phases to tune the electrical parameters. Various fundamental factors influencing the formation of ferroelectricity in Zr x Hf 1‐x O 2 have been identified and distinguished from those in the classical perovskite ferroelectrics. Notably, a sharp increase in the dielectric constant near 0 V observed in Zr‐rich Zr x Hf 1‐x O 2 films achieved by tuning the fabrication parameters has been attributed to the presence of a morphotropic phase boundary. This study investigates 6 nm Zr‐rich Zr x Hf 1‐x O 2 thin film metal‐insulator‐metal capacitors using a combination of experimental methods and machine learning‐based molecular dynamics simulations. The study provides insight into the physical mechanisms that enhance the dielectric constant near 0 V and attributes it to the orthorhombic phase rather than a morphotropic phase boundary. The work discusses the limitations in the practical application of the high dielectric constant observed near 0 V. Additionally, it highlights similarities and differences between Zr x Hf 1‐x O 2 and the well‐known morphotropic phase boundary in PbZr x Ti 1‐x O 3 .

ZrO2 is a promising high-k dielectric for SiC power devices due to its favorable bandgap alignment with SiC. However, it exhibits low breakdown fields and excessively high leakage currents for thicker layers. This paper presents an approach to suppress this leakage current by integrating thin interlayers of Al2O3, Y2O3, or La2O3 into the ZrO2 film. These interlayers significantly reduce the charge carrier transport through ZrO2 films and, thereby, the leakage current of the stack. Among the investigated interlayers, Al2O3 shows the most pronounced effect, reducing the leakage of ZrO2-based thick films by 2 orders of magnitude and achieving a breakdown field of 7.4 MV/cm. This is comparable to the value measured for pure Al2O3 (7.7 MV/cm). These improvements can be attributed to the amorphous nature of the laminated oxide as the crystallization temperature could be increased from 350 °C for pure ZrO2 up to 750 °C for the nanolaminate. Notably, the dielectric constant of this optimized stack is 13, which is twice as high as that of pure Al2O3. No additional charge trapping due to the interlayers was detected by Capacitance-Voltage hysteresis measurements. Furthermore, by additional optimization of the stack's deposition conditions, the charge trapping was reduced by 50% compared to pure ZrO2 films.

Charged oxygen vacancies are thought to be responsible for fatigue effects in HfO2- and ZrO2-based ferroelectrics, while also supporting the formation of the ferroelectric phase. We investigate the possible influence of intrinsic electric fields generated by predominantly doubly positively charged oxygen vacancies that accumulate near the electrode and violate charge neutrality. Our calculations are based on a meticulous kinetic Monte Carlo simulation, which simultaneously treats the movement of electrons, the diffusion of oxygen defects, and their interactions that change the charge state in a self-consistent manner. It is shown that with a realistic vacancy concentration, electric fields of several MVcm−1 can form within 2nm of the electrode, supporting the formation of the ferroelectric phase. This effect becomes significant for thin films below 10nm. The strength of the effect depends on the position of the defect energy level relative to the valence band offset. This dependency could be used to control the effect.

Ferroelectric materials have great potential for applications in information technology. With the discovery of ferroelectricity in semiconductor process-compatible materials, the activities in both research and industry to drive ferroelectric functionalities toward commercialization have exploded. This paper will discuss the recent developments both with respect to applications in standard semiconductor memories as well as emerging in-memory computing and neuromorphic computing applications.

Recently ferroelectric properties have been found in hafnia-based nanosized films. Such films are of the utmost interest for development of a universal memory, which combines the advantages of random access memory and flash memory. The paper studies optical properties of hafnia-zirconium oxide films H-fxZryO2 and lanthanum-alloyed hafnia-zirconium oxide films La:HfxZryO2. Fluctuations of thickness in HfxZryO2 do not exceed 3.5

How and why the reliability of ferroelectric HfO2- and HZO (Hf0.5Zr0.5O2)-based memory devices strongly depends on the choice of electrode materials is currently under intense discussion. Interface conditions such as band alignment, defect formation, and doping are recognized as decisive and interrelated parameters, but a unified picture of the physical mechanisms is still missing. Here, two opposite scenarios of band alignment are found in TiN/HZO/TiN and IrO2/HZO/IrO2 using hard X-ray photoelectron spectroscopy, revealing on the one hand the conditions for a stable device performance, and the origin of their degradation on the other. As a key difference, TiN electrodes scavenge oxygen from the HZO, while IrO2 electrodes supply it. Considering the electronic doping limit of HfO2, a key condition for the stability of ferroelectric devices can be identified: The alignment of the charge neutrality levelwith respect to the metallic Fermi level, which is pinned by the doping limit. Stable device performance can only be achieved for oxygen-deficient HfO2-based interfaces, where the Fermi level of the metal electrode is close to the conduction band of the ferroelectric insulator. This empirical model explains the fatigue behavior of HfO2-based capacitors using either oxygen-scavenging TiN or oxygen-supplying IrO2 electrodes.

To address the increasing computational demands of artificial intelligence (AI) and big data, compute-in-memory (CIM) integrates memory and processing units into the same physical location, reducing the time and energy overhead of the system. Despite advancements in non-volatile memory (NVM) for matrix multiplication, other critical data-intensive operations, like parallel search, have been overlooked. Current parallel search architectures, namely content-addressable memory (CAM), often use binary, which restricts density and functionality. We present an analog CAM (ACAM) cell, built on two complementary ferroelectric field-effect transistors (FeFETs), that performs parallel search in the analog domain with over 40 distinct match windows. We then deploy it to calculate similarity between vectors, a building block in the following two machine learning problems. ACAM outperforms ternary CAM (TCAM) when applied to similarity search for few-shot learning on the Omniglot dataset, yielding projected simulation results with improved inference accuracy by 5%, 3x denser memory architecture, and more than 100x faster speed compared to central processing unit (CPU) and graphics processing unit (GPU) per similarity search on scaled CMOS nodes. We also demonstrate 1-step inference on a kernel regression model by combining non-linear kernel computation and matrix multiplication in ACAM, with simulation estimates indicating 1,000x faster inference than CPU and GPU.