AVS 71 Session TF+CPS+MS+EM-ThM: Thin Films for Microelectronics I

Solid-state drives (SSDs) continue to serve as the foundation of long-term active data storage in modern data centers. Over the past decade, conventional vertical NAND (vNAND) technology has achieved a remarkable 50× increase in storage density, enabled by advances in physical scaling (x–y and z dimensions) and logic scaling (from multi-level cell, MLC, to quad-level cell, QLC). The explosive growth of artificial intelligence (AI)—with models like GPT-4 surpassing a trillion parameters—has further accelerated the demand for high-capacity, high-performance storage systems to support petabyte-scale datasets.

Today’s state-of-the-art vertical NAND devices offer densities nearing 30 Gb/mm² with over 300 stacked layers. However, extending this scaling trajectory to 1000 layers and beyond—targeting storage densities exceeding 100 Gb/mm²—poses significant challenges. Chief among these are reliability concerns intrinsic to charge-trap flash technologies, such as lateral charge migration and the poor endurance of higher logic level operations.

To overcome these limitations, ferroelectric field-effect transistors (FeFETs) have emerged as a promising alternative, enabling further z-direction scaling with improved reliability. This presentation will highlight recent advances in atomic layer deposition (ALD)-based ferroelectric gate stack engineering, and how these innovations can support the development of next-generation NAND architectures capable of 1000-layer integration and ultra-high-density storage.

Ferroelectric NAND (FeNAND) using Hf_0.5Zr_0.5O₂ (HZO) offers increased memory density, speed, and decreased operation voltage of NAND compared to charge trap flash technology.^[1] However, to compete with charge trap flash, the memory window of FeNAND must be increased above 6 V for 3 bit/cell operation or above 8 V to enable 4 bit/cell operation.^[1] Since the memory window is directly related the ferroelectric coercive field (E_c), finding pathways to increase the coercive field of HZO is critical to enable FeNAND. Prior studies show that inserting a dielectric interlayer can increase the coercive field, but the mechanism driving the increase in E_c beyond just adding a capacitor in series is still unclear.^[2]

The goal of this work is to test the hypothesis that the increased defect states in the dielectric-HZO interface cause in-built fields that then increase the coercive field.^[3] First, we fabricate 19 nm HZO both with and without Al₂O₃interlayers or adjacent layers. Varying the layer thicknesses and positions via atomic layer deposition allows for the determination of how the device structure impacts the ferroelectric switching. Polarization-electric field hysteresis loops and positive-up-negative-down (PUND) show ferroelectric switching for each of the films, with a remnant polarization (2P_r) up to 27.4 µC/cm². The coercive field increases from 1.01 MV/cm in devices without an additional dielectric layer (19 nm HZO) to 3.11 MV/cm in a 3 nm Al₂O₃ interlayer inserted between two 8 nm layers of HZO (8 nm HZO-3 nm Al₂O₃-8 nm HZO). First-order reversal curve (FORC) analysis reveals an increase in internal bias field in devices with dielectric layers, potentially due to defects at the Al₂O₃ - HZO interface. X-ray photoelectron spectroscopy valence band measurements confirm an increase in mid-gap defect states at this interface compared to bulk of the film. Additionally, temperature-dependent modulus spectroscopy is used to evaluate the activation energy and defect concentration in samples with and without a dielectric layer. These findings provide key insights into mechanisms to modulate coercive field in HZO, enabling the design of FeNAND devices with larger memory windows.

References

[1] G. Kim et al., J. Mater. Chem. C 2022

[2] L. Fernandes et al., IEEE Trans. Electron Devices 2025

[3] D. Das et al., Int. Electron Devices Meet. IEDM, 2023

One major limitation of 2D transition metal dichalcogenide (TMD) based FETs is the high contact resistance between metallic electrodes and semiconducting channels, particularly for p-type contacts. In this presentation I will address how PEALD of p-type TMDs can be used to improve this contact resistance. First, I will go over controlled doping strategies to form p-type 2D TMD contact materials using PEALD, with an emphasis on Nb Doped WS₂. Our recent results show contact resistancevalues as low as 0.30 ± 0.26 kΩ·μm between Pd and PEALD NbxW1-xS2, demonstrating that low resistance contacts between metal and p-type TMDs are possible. Then, I will discuss reducing unintentional p-doping introduced during PEALD of TMDs. PEALD TMDs typically contain some level of hydrogen impurities that leads to unintentional p-doping. We have shown that these impurities can be reduced by introducing an Ar plasma C step in the standard PEALD TMD process. Finally, the use of remote plasmas in PEALD for contact deposition can lead to the creation of undesired impurities and defects in the 2D TMD channel, possibly impacting electronic behavior. I will present some first insights into the defects that are created during PEALD on 2DTMDs and how we can reduce the number of plasma-induced impurities and defects.

Computational energy consumption has been increasing exponentially, making energy-efficient microelectronics and computing an urgent need. Three-dimensional integrated circuits (3D ICs) and neuromorphic computing promise to revolutionize information technology by drastically reducing the energy consumption of computers, and two-dimensional (2D) semiconductors like molybdenum disulfide (MoS₂) can enable such technologies. However, scalable and controllable manufacturing processes are still needed to realize the technology’s full potential. Here, we demonstrate the uniform and controlled deposition of few-layered MoS₂ using atomic layer deposition (ALD) for the purposes of memtransistor fabrication. By leveraging the equilibrium shift from material deposition to material etching, a self-limiting deposition of MoS₂ is achieved where material growth is stopped after the initial few layers. The resulting few layer MoS₂ was characterized using Raman spectroscopy and X-ray photoelectron spectroscopy, and was used to fabricate and test memtransistors. This deposition strategy is straightforward, robust and more scalable compared to other methods such as powder CVD and exfoliation.

Recent advances in integrated nonreciprocal components—such as isolators and circulators—have enabled transformative wireless communication and sensing technologies, including full-duplex radio, in-band self-interference cancellation, and protected high-power transmission systems. While commercial ferrite-based isolators offer low insertion loss and high power handling, their reliance on kOe-level bias fields, high-temperature ferrite growth, and bulky permanent magnets severely limits their compatibility with CMOS processes and low-power applications.

To address these limitations, magnetoacoustic isolators have emerged as a promising class of passive, CMOS-compatible, and power-efficient nonreciprocal devices. These isolators consist of magnetic heterostructures integrated within the propagation path of surface acoustic waves (SAWs) on piezoelectric substrates. Magnetoelastic and magnetorotational coupling mechanisms enable strong spin wave–acoustic wave interactions, generating hybrid magnetoacoustic waves with dramatically asymmetric damping rates in opposite directions. This asymmetry yields unidirectional transmission, fundamental to nonreciprocal operation.

Despite progress, early devices suffered from weak non-reciprocity, primarily due to a mere helicity mismatch effect and an inherent symmetric spin wave dispersive relation in single-layer magnetic films. Recent efforts have focused on engineering magnetic stacks with nonreciprocal spin wave dispersion. Key examples include: (i) interfacial Dzyaloshinskii–Moriya interaction (iDMI) stacks like CoFeB/Pt, (ii) interlayer dipolar-coupled (IDC) stacks such as FeGaB/SiO₂/FeGaB, and (iii) RKKY synthetic antiferromagnets like CoFeB/Ru/CoFeB. These architectures achieve nonreciprocity strengths up to 250 dB/mm. Recent demonstrations using shear-horizontal waves in LiTaO₃ substrates coupled to ferromagnetic and anti-magnetostrictive bilayers have yielded nonreciprocity levels of 60–82 dB/mm with simpler fabrication.

Nevertheless, a persistent challenge remains in reducing insertion loss while maintaining wide bandwidth and high isolation. We will introduce our recent efforts in a fundamental mode SAW-driven magnetoacoustic isolator with giant non-reciprocity and a wideband nonreciprocal magnetoacoustic isolator based on non-collinear dipolar-coupled ferromagnetic stacks. The talk will provide a comprehensive overview of the mechanisms, material platforms, and experimental breakthroughs driving the field of magnetoacoustic isolators. We will highlight the path toward integrated, low-loss, and high-performance nonreciprocal components for future quantum, RF, and IoT systems.

Magnetoresistive sensors are widely used to detect magnetic fields by measuring changes in electrical resistance. One such effect, extraordinary magnetoresistance (EMR), arises from the geometry of semiconductor-metal hybrid structures that combine high-mobility semiconductors with highly conductive metals. EMR strongly depends on both the semiconductor’s mobility and the quality of the metal-semiconductor contact (ohmic contact with low contact resistance). The device geometry further influences boundary conditions and current paths under magnetic fields, enabling flexible design and performance tuning. While most previous EMR studies have focused on III-V semiconductors and 2D materials, there has been limited exploration of oxide-based systems.

Here, we demonstrate EMR in high-quality La-doped SrTiO₃ thin films grown on SrTiO₃ (001) substrates using hybrid molecular beam epitaxy (MBE). We grow films with carrier concentrations ranging from ~2×10¹⁷ cm^-3 to ~1×10²⁰ cm^-3, achieving Hall mobilities from ~300 cm²/(V·s) up to over 50,000 cm²/(V·s) at 1.8 K. Using an asymmetric device geometry that breaks mirror symmetry between voltage probes, we observe corresponding asymmetry in magnetoresistance (MR) measurements. With embedded metals that are ohmic contacts to SrTiO₃, we achieve an MR ((R(B) – R(0))/R(0), where R(B) is the measured resistance at magnetic field B) approaching 9000% at 9 T and 1.8 K, which is over 3900% higher than the intrinsic MR of SrTiO₃ – a world record for an oxide-based EMR device! Finite element simulations of current flow and MR in these SrTiO₃-based hybrid structures align well with experimental data, validating the design principles. These results establish the potential of complex oxide systems for low-temperature EMR sensors and open opportunities for integrating oxide heterostructures in future magnetoelectronic devices. In this presentation, we will discuss the hybrid MBE growth and microfabrication of high-mobility SrTiO₃ thin films, along with device optimization strategies and detailed magnetotransport measurements across various temperature and magnetic field ranges.