United States

ULTRARAM™ is an emerging charge-based memory that is non-volatile whilst also having ultra-low-energy logic-state switching.1,2 These seemingly contradictory properties are achieved by a triple-barrier resonant tunnelling (TBRT) structure, which utilises the extremely large InAs/AlSb conduction band offset (2.1 eV). Much like flash memory, ULTRARAM™ uses a floating gate to store charge, with non-destructive read of the logic state by a proximate channel. However, due to the TBRT structure only 2.5 V is required to program/erase, which is an order of magnitude lower than flash. Switching times and endurance are predicted to be similar to DRAM. ULTRARAM™ has many potential applications including conventional computing (large-scale and personal), artificial intelligence and the internet-of-things. However, for these to be realised, low-cost mass production must be achieved, necessitating growth on 300-mm Si substrates. Growth of III-V’s on Si presents challenges related to differences in lattice constant, thermal expansion coefficients and the polar/non-polar interface. These degrade epilayer quality due to the formation of threading defects and anti-phase domains.3

Here we report on recent progress on the growth of the III-V ULTRARAM™ on Si substrates by molecular beam epitaxy (MBE). A high-quality GaSb surface is required prior to deposition of the In(Ga)As and AlSb memory epilayers. Several growth techniques, including migration-enhanced epitaxy, interface-misfit epitaxy and two-step temperature growth were used to form two different buffer structures on 4˚ offcut substrates: (A) GaSb/GaAs/Ge/Si, where the Ge layer was deposited by chemical vapour deposition, and (B) GaSb/AlSb/Si, where all layers were grown by MBE. A comparison of these buffer layers, as well as a GaSb/GaAs control sample, is shown in the table below. The GaSb/AlSb/Si buffer was found to have the highest quality surface, exceeding that of the control sample grown on GaAs, and was used for growth of the full memory structure (TEM images below). This material has been processed into the first ULTRARAM™ memories on Si, with testing currently underway.

Materials Challenges for Memory

III-V non-volatile ULTRARAM™ memory on Si

non-volatile

charge-based

ultraram


![](https://assets.underline.io/uploads/markdown_image/1/image/2e10c515564e9ec3287ee4c664011c20.png)

The energy cost of storing and moving data is enormous, and is growing at such a rapid pace that within a few years the power demand may overwhelm some developed nations. Non-volatile memory (NVM) and neuromorphic computing are enabling critical advances in this area, but there are many scientific and technological challenges. This conference will cover the materials challenges of different memory technologies, from the fundamental processes and mechanisms, to industry perspectives on materials specifications and manufacturing.

poster

Recently, we have successfully employed the single-band tight-binding model to predict the noncollinear spin torque effect in insulator and spin-filter based magnetic tunnel junctions [1]. However, for real complex magnetic heterojunctions, the injected spin-polarized electrons from ferromagnetic electrode can be strongly influenced by the complicated interfacial spin-polarized charge transfer, which is normally ignored in the tight-binding model can causes discrepancy between computational results and experimental findings. 

Our newly developed “JunPy” [2] package has successfully combined the self-consistent Hamiltonian by using the first-principles calculation with the non-equilibrium Green’s function (NEGF) method to obtain the noncollinear spin torque effect in nm-scale magnetic heterojunctions. The divide-and-conquer (DC) method was first applied to reveal the oscillatory decay of layer-resolved spin torques away from the MgO/Fe interface, and suggests a very thin Fe layer thickness below 2 nm to preserve the efficient current-driven magnetization switch [3]. We further developed the Landau-Lifshitz-Gilbert (LLG) equation implemented by our calculated spin transfer torque (STT) effect to simulate the magnetization switching in Fe/MgO/Fe MTJ in the presence of perpendicular magnetic anisotrpy (PMA), external magnetic field, and current-induced spin transfer torque (STT). The resistance as a function of magnetic field (R-H loop) and of bias (R-V loop) are thus obtained and well comparable with experimental results. On the other hand, our “JunPy” package is applied to predict the giant field-like spin torque (FLST) effect in the amine-ended single-molecule magnetic junction [4], which can efficiently assist the molecule scale magnetic proximity effect via interfacial spin filter effect and may open a new avenue for multifunctional manipulation in next-generation organic FLST-MRAMs with lower power consumption. We believe that this newly developed “JunPy” calculation process not only can efficiently resolve current self-consistent difficulties in first-principles calculation for magnetoelectric and magnetotransport properties, but also may inspire future experimental explorations in novel magnetic heterojunctions for spintronics applications. This work is supported by the Ministry of Science and Technology of Taiwan (MOST 107-2633-M-008-004 and 108-2628-M-008-004-MY3).

[1] Y. –H. Tang et al., Phys. Rev. Lett. 103, 057206 (2009); Sci. Rep. 5, 11341 (2015).
[2] https://labstt.phy.ncu.edu.tw/junpy 
[3] B. -H. Huang et al., AIP Advacnes 11, 015036 (2021). This paper was selected as an Editor’s Pick. 
[4] Y. –H. Tang and B. –H. Huang, J. Phys. Chem. C 122, 20500 (2018).


Magnetotransport Properties in Magnetic Heterojunctions: Combined First-Principles Calculation with NEGF Method and Spin Dynamics Simulation

Topological insulators (TI) possess topologically protected surface conducting states and enormous spin-orbit coupling (SOC) [1,2]. This makes TIs ideal candidates for Spin-Orbit Torque (SOT) based Non-volatile Memory. A TI adjacent to a ferromagnetic (FM) thin film generates SOT on the magnetization in FM [2-5] because of Spin Hall Effect (SHE). SOT in TI/FM (ferromagnet) films have been reported previously with TI films epitaxially grown with MBE [2-5], which is not ideal for industrial complementary metal oxide (CMOS) processes. Although, there have been reports of SOT from sputtered TIs [6-8], effects of crystalline disorder in SOT characteristics have not been extensively studied. We performed magnetic and Spin-Torque Ferromagnetic Resonance (ST-FMR) measurements to investigate the effect of sputtering growth temperature and hence crystalline ordering on SOT characteristics of Bi2Te3 coupled with NiFe (Permalloy) and CoFeB as FM layers. The Bi2Te3/FM heterostructures revealed a steady increase in out of plane (OOP) magnetic anisotropy with the increase in crystalline order of Bi2Te3. Further, STFMR measurements revealed a nonlinear trend in damping-like spin orbit torque with increase in sputtering growth temperature. We believe that at room temperature growth, smaller grain size in Bi2Te3 causes quantum-confinement governed enhancement in SOT [6]. However, a giant enhancement in SOT characteristics is observed for highly crystalline c-axis oriented Bi2Te3 heterostructure where Bi2Te3 was grown at 250 ºC. Our results provide a clear path towards integration of crystalline TI based spintronic devices in industrial CMOS compatible processes.

Crystalline Order Dependent Spin-Orbit Torque in Sputter Deposited Topological Insulator /Ferromagnet Heterostructures

With the rapidly growing interest in cryogenic electronics, which is driven by the recent progress in spintronics and quantum / superconducting computing, there is an urgent need to develop logic and memory devices that can operate efficiently at low temperatures. To underpin these new developments there is a need for magnetic semiconductors which are compatible with cryogenic temperatures. We will discuss our recent progress in fabricating exactly these materials.

The rare-earth nitride series offer many intrinsically ferromagnetic semiconductors. They have already been incorporated into prototype field-effect transistors, spin-filters and tunnelling structures. They can be doped by controlling the inclusion of nitrogen vacancies, which donate electrons into the conduction band. They have a finite, and tuneable, conductivity even as the temperature approaches 0 K. Furthermore, the electronic and magnetic properties are not derived from the same electrons, as they are in the dilute magnetic semiconductors, i.e. the saturation magnetization and electron concentration can be tuned independently.

We will show electrical transport and magnetic measurements on GdN and DyN displaying their facility to be doped, along with density functional theory based calculations showing that the dopant states do indeed lie in the conduction band at 0~K. These materials provide cryogenically compatible ferromagnetic semiconductors for application in magnetic memory or spintronics devices.

Ferromagnetic semiconductors for low temperature devices: The rare earth nitrides

The Pt/Hf-based hydroquinone (HQ)/Al2O3/TiN/Si bilayer hybrid memory was prepared by molecular layer deposition (MLD)/atomic layer deposition (ALD). The synthesis of Polymerized hybrid Hf-HQ film is demonstrated by XPS and IR. Free of electroforming process, a typical bipolar resistive switching characteristic has been confirmed with low SET/RESET voltages of -0.85 V/+1.87 V. 8 distinguishable resistance states were obtained, allowing the multi-binary data storage. Long retention time (>104 s) and good switching endurance (2500 times) with ON/OFF ratio around 70 were achieved. The I-V curve of hybrid memory units follows space-charge-limited-current theory, indicating the RS mechanism of conducting filaments. The polymer-like structure of hybrid Hf-HQ can improve the intrinsic flexibility of devices. On the basis, we’ll explore the application of polymerized hybrid Hf-based hydroquinone/Al2O3 bilayer structure in flexible RRAM devices.


Polymerized hybrid Hf-based hydroquinone /Al2O3 bilayer structure by molecular/atomic layer deposition for non-volatile resistive random access memory

Materials that demonstrate high spin-orbit torque are central to the functionality of next-generation spintronic memory and logic technologies. A promising class of materials for this purpose are topological semimetals. While first experimental results have shown the high spin hall angle of Weyl semimetals such as WTe2, these materials are not readily CMOS-compatible and cannot be easily integrated into existing industry-standard cleanroom processes on a 300 mm wafer level. Cobalt silicides are transition metal silicides that are already well-established in the semiconductor industry. Further applications have opened up with the recent identification of cobalt monosilicide (CoSi) as a chiral topological semimetal.

Here, we present material properties of CoSi thin films fabricated on 300 mm wafers using industry-standard cleanroom tools. Metalorganic chemical vapor deposition (MOCVD) was used to deposit Co layers of varied thickness onto wafers with an amorphous Si (a-Si) layer. The stacks were then annealed in the same tool in a vacuum atmosphere. We were able to control the thickness of the fabricated silicide films by adjusting the thickness of the a-Si and deposited Co layers. Relative to the a-Si thickness, the Co thickness was adjusted to obtain silicide films with a 1:1 Co:Si stoichiometry on a nanometer scale. The sensitivity of the composition of the formed silicide to the initial Co to a-Si ratio was investigated using GIXRD. The resistivity of the silicide films was also investigated.

Therefore, we speculate that the CoSi film may possess topological properties for efficient charge-to-spin conversion for use in low-power spintronic applications. This methodology paves the way for fabricating large-scale silicide films on a CMOS platform for spintronic applications.

In vacuo fabrication of thin-film cobalt monosilicide (CoSi) on 300 mm wafer level

ULTRARAM™ is an emerging III-V universal memory that exploits quantum resonant tunnelling to achieve non-volatility and fast, ultralow-low-energy switching per unit area. In this poster, we present characteristics of memories grown on Si substrates which match or exceed recent results for devices grown on GaAs substrates, revealing a path to ULTRARAM™’s commercial viability.

III-V non-volatile ULTRARAM™ 2x2 memory arrays

Temperature-dependent current-voltage (I-V) characteristics of crossbar type Pt/NiO/Pt devices with different bottom electrode thickness were measured at
cryogenic temperatures. Unipolar resistive switching occurs at a lower temperature as the thickness of the bottom electrode is reduced. The results provide important insight into the role of the bottom electrode thickness on the creation of thermally stable conductive filaments.

Thermal cycling driven changes of the resistive switching in Pt/NiOx/Pt crossbar arrays

Ferroelectric materials are used in building non-volatile memories due to the existence of spontaneous electric polarization of which direction could be reversed by applying an electric field. This functionality represents a classic application of these materials as a binary system, but new challenges are brought by the evolution of nowadays computation needs (as increased speed or increased amount of stored data). In this way, the new paradigms of computing appeared and they address all classes of materials used in these technologies. 
In this study, we experimentally prove that multilayered ferroelectric structures present a different switching behavior than simple ferroelectric thin films. By using alternating layers of ferroelectric and isolator materials architectures we can have multiple step-like polarization switching. This behavior could be used to separately access more than 2 polarization states, which presents good time stability. In this way, we could obtain 2n different states, where n is the number of isolator layers. Experimentally, we obtained structures with 4 and 8 polarization states. Moreover, these different states could be differentiated by different capacitive values. The process of measuring the capacitance is based on applying a low voltage signal and does not determine a reversing/changing of polarization, being in this way a non-destructive reading method. The way the system could pass by different polarization states enables the algorithm to be performed for simple binary computation functions such as OR/NOR and AND/NAND. This proves that we can have data processing and storing the result on the same device, which represents a parallel computing system based on a ferroelectric memcapacitor. The current circuit architecture based on classic logic gates can be simplified and optimized using a multilayered ferroelectric capacitor, leading also to significantly reduced power consumption and reduced cost of manufacturing.


Ferroelectric based multilayered thin films structures for multi-bit memory and memcomputing

One of the most promising contenders for ultralow-energy electronic devices is resistive switching memory (RRAM). It delivers sustainably-scalable ‘neuromorphic’ computing, potentially capable of reducing energy consumption in IT by > 50% [1]. Currently, a poor understanding of the nanoscale kinetics of the switching mechanisms is preventing its commercial utilization. RRAMs are mostly studied by electron microscopy, which is destructive, invasive, and under drastically different conditions, so is not sufficient for finding true understandings [2]. 

We develop a novel non-destructive fully-optical technique to observe in-situ and in real-time switching in memristive materials under ambient conditions, otherwise invisible to conventional electron microscopy techniques [3-4]. The formation of conductive paths in a memristive material is captured within a tightly-confined plasmonic hotspot, which greatly enhances optical signals coming from a hotspot only [4]. This allows the optical measurements to be intimately linked to the electrical and material properties of the switching channel, hence accessing memristive dynamics directly. So far, we have characterized memristive switching in RRAMS based on SiO2 (electrochemical metallization mechanism (ECM)) [3] and SrTiO3 (valence change mechanism (VCM)) [4] .

For SiO2-based ECM cells, we consistently observed a change of the optical spectra only upon switching, which we correlated with SET and RESET cycles inside the memristive material in a plasmonic hotspot. This resulted in the first in-situ optical observations of the nanoscale kinetics of RRAMs under ambient conditions. Our findings indicated that the filament nucleation position can differ from cycle to cycle in the SiO2 based RRAMs [3].

The investiagtions of the SrTiO3 based RRAMs have demonstrated the method’s ability to track oxygen vacancies motion and identify the breaking mechanisms in valence change mechanism (VCM) cells. Our numerical framework couples each spectral resonance peak to the drift of oxygen vacancies and to the formation of an O2 reservoir at the STO/TiO2 interface. The reservoir might be the cause of the device breakdown upon cycling [4].

Our future plans involve investigating switching in h-BN and HfO2 based memristors.


References
[1] Miller, D. A. B., (2009) IEEE Proceedings, 97, 7, 1166.
[2] Tappertzhofen, S.; Di Martino, G.; Hofmann, S., (2019) Physica Status Solidi (a), 217, 6, 190058.
[3] Di Martino, G. et al, (2016) Small, 12, 10, 1334.
[4] Di Martino, G. et al, (2020) Nature Electronics, 3, 687.

Real-Time In-Situ Optical Tracking of Memrisitive Switching

HfOx-based resistive random access memories (RRAMs) have attracted much attention for future non-volatile memory (NVM) to realize a neuromorphic system. They demonstrated successful CMOS compatibility, good stability, and low power consumption. At room temperature and low temperature (77 K) or cryogenic temperature (4 K), HfOx-based RRAM exhibited their potential for low power memory devices. However, there are several problems to be solved to apply HfOx-based RRAM for low temperatures. For operating the devices at low temperature, a vacuum environment is needed to protect from contamination. But due to oxygen scarcity in the vacuum environment, reduction of oxygen vacancies in HfOx film can be disturbed, then redox-based HfOx RRAM characteristics are degraded. Thus, the buffer layer which plays a role as an oxygen reservoir is important in a vacuum environment.
In this work, the device performance dependence on the buffer layer thickness in a vacuum environment was studied. The titanium thin film was used as a buffer layer because of its oxygen affinity and its thickness was varied from 30 nm to 150 nm to examine the effect of oxygen gettering ability in a vacuum environment. As a result, devices show a decrement in switching voltages in a vacuum environment and thicker buffer layer, because the high concentration of oxygen vacancies makes leakier HfOx film which needs less energy for switching.


Premium content

Next from Materials Challenges for Memory

Magnetotransport Properties in Magnetic Heterojunctions: Combined First-Principles Calculation with NEGF Method and Spin Dynamics Simulation

Similar lecture

III-V non-volatile ULTRARAM™ 2x2 memory arrays