Historical Foundation and Practical Guideline for Ferroelectric Switching Kinetic Studies

Abstract: Electrical measurements of ferroelectric switching kinetics are widely used to probe the dynamics of polarization reversal, yet the influence of the measurement circuit is often underappreciated. In this paper, we show that the interplay between ferroelectric capacitors and circuit elements produces distorted, time-dependent voltage waveforms across the device, particularly in the sub-ns regime. We examine how these circuit contributions affect polarization transients extracted from PUND measurements. The resulting distortions scale with supply voltage, capacitor dimensions, and lumped circuit elements, but are not accounted for in conventional experimental analyses or analytical model fitting. We then critically assess existing nucleation and growth models and show that neglecting the time-varying voltage profile can lead to unphysical interpretations of switching kinetics, most notably in the extracted growth dimensionality represented by the Avrami exponent. Finally, we outline guidelines for future studies, emphasizing the need for direct voltage monitoring and circuit-aware de-embedding, as well as modeling frameworks that incorporate voltage-dependent nucleation and growth rates based on intrinsic material parameters.

• The paper demonstrates that voltage distortions from circuit parasitics can introduce up to 9% error in extracting intrinsic switching parameters in HZO capacitors.
• The methodology critiques prevailing PUND protocols and classical nucleation-growth models, showing that idealized voltage assumptions lead to misinterpretations of kinetic data.
• The study advocates for real-time voltage monitoring and dynamic-field driven models to improve measurement accuracy and guide advanced FE device engineering.

Critical Analysis of "Historical Foundation and Practical Guideline for Ferroelectric Switching Kinetic Studies" (2604.05328)

Introduction

The study systematically revisits the experimental and analytical framework for ferroelectric (FE) switching kinetics, with a focus on the interaction between intrinsic material properties and extrinsic circuit effects in electrical switching measurements. The paper extensively critiques prevailing methodologies, especially the assumptions embedded in PUND (positive-up-negative-down) measurement protocols and classical nucleation and growth (N&G) modeling. Its analysis reveals that overlooked circuit-induced voltage distortions produce significant artifacts in switching transient data, leading to misinterpretation of intrinsic kinetic parameters.

Experimental Artifacts in Ultrafast Ferroelectric Switching

Ferroelectric switching studies commonly utilize PUND pulse measurements to extract the time-resolved polarization response. The canonical analysis assumes that the voltage applied to the FE capacitor is an ideal square pulse, and that circuit parasitics are negligible. This work rigorously deconstructs these assumptions.

Irreducible Circuit Effects

The study shows that actual voltage waveforms across the FE capacitor are substantially distorted from intended programmatic pulses as a result of device capacitance, series resistance, and measurement circuit parasitics. In particular, the sub-ns switching regime is strongly affected by finite rise-time and RC effects, yielding nonlinear, time-dependent voltage drops correlated with displacement (switching) currents. The error in extracted polarization can reach up to 9% in moderate-scale Hf$_{0.5}$Zr$_{0.5}$O$_2$ (HZO) capacitors, with the largest artifacts observed for high supply voltages and larger capacitor areas.

Parasitic Linear Components and Material Defects

Beyond the idealized FE capacitor structure, real devices exhibit parasitic capacitance (from isolation dielectrics), series resistance/inductance (from electrodes and probe layout), and in many cases, dead layers and leakage paths. Series capacitance from dead layers (non-polar interface phases) systematically underestimates apparent remanent polarization and skews switching transients. Leakage currents introduce further errors, especially in ultra-thin or defective films, as they are convoluted with displacement currents in the time-domain data.

Limitations of Classical Nucleation and Growth Models

Switching transients—subject to the described voltage distortions—are typically fit to analytical N&G models (KAI, NLS, SNNG) to extract mechanistic parameters such as the Avrami exponent, nucleation rate, and characteristic time.

Assumptions and Misuse of Kinetic Models

The paper highlights that all prevalent kinetic models (KAI, NLS, SNNG) are derived under the assumption of strictly constant applied voltage during switching. A time-invariant electric field is the basis for connecting extracted parameters (e.g., switching time) to the activation field via Merz law. In practice, the real-time voltage profile is non-constant—rendering fits to these models physically ambiguous. The resultant extracted dimensionality parameters (Avrami exponent $n$) can become unphysically high or low, particularly in larger devices or with high-voltage/slow-ramp pulses. This undermines the validity of assigning specific N&G mechanisms based solely on transient fits, as is widely practiced.

Apparent Avrami Exponent Anomalies

The authors dissect the so-called "Avrami exponent mystery": the widespread observation of effective $n$ well outside the range implied by physical growth dimensions (i.e., $1 \leq n \leq 4$), and in some cases, negative or highly dynamic values. They show that when voltage ramp and switching timescales are comparable, or when field is dynamically evolving, these anomalous exponents are natural artifacts of the analysis protocol, not evidence of exotic switching physics. Compensation via empirical broadening parameters in NLS or SNNG models does not restore physical validity without explicit voltage path dependence.

Toward Rigorous, Circuit-Aware, and Voltage-Dependent Modeling

A core recommendation of the paper is the integration of explicit real-time voltage monitoring and the de-embedding of all circuit elements in transient measurements, followed by kinetic modeling that incorporates voltage-dependent nucleation and growth rates. Static mapping of Merz-type scaling to characteristic times is inappropriate; future frameworks must treat rate processes as functions of instantaneous electric field and potentially include feedback from circuit response.

Dynamic-Field Driven Nucleation and Growth (DFNG) Modeling

The authors reference the recent development of DFNG models [49], which treat nucleation rates and domain growth velocity as explicit functional dependencies on the real, time-varying voltage. Such models yield a single, intrinsic set of material parameters, successfully capturing transients under diverse drive conditions and providing quantitative predictions relevant for advanced device operation (e.g., in-memory computing, neuromorphic architectures). The DFNG paradigm enables realistic circuit co-design and forward modeling for arbitrary control waveforms, in contrast to the post hoc empirical fitting of legacy models.

Implications for Device Engineering and AI Hardware

Practically, these insights necessitate a paradigm shift in FE device engineering for AI and logic-in-memory applications. Not only must device scaling and interface engineering address leakage/dead layers, but circuit/system-level design must provide for waveform fidelity and accurate extraction of polarization state. Analytical models for system simulation must be validated in a circuit-aware context; otherwise, hardware-accelerated AI primitives built on FE switching risk unpredictable operation, limited endurance, and computational error.

Conclusion

This work provides a comprehensive critique of the conventional experimental and analytical approaches to FE switching kinetics. It establishes that reliable extraction of intrinsic parameters is impossible without correcting for measurement artifacts and incorporating real-time voltage profiles. The classical practice of fitting switching transients to static-field models is shown to yield physically ambiguous or misleading results. The paper advocates for dynamic, voltage-dependent kinetic models and rigorous measurement protocols. These advances are critical for the continued development of FE devices for high-speed, low-power memory and AI computation, and suggest new avenues for predictive, materials-to-circuit modeling architectures.