Charge trap flash
Charge trap flash is a non-volatile semiconductor memory technology that stores data by trapping electrical charges, typically electrons or holes, in discrete sites within a dielectric layer such as silicon nitride (SiN), modulating the threshold voltage of a transistor to represent binary states.[1] Unlike traditional floating-gate flash memory, which relies on a conductive polycrystalline silicon floating gate surrounded by insulating oxide layers to store charge, charge trap flash uses a non-conductive trapping layer that isolates charges more effectively and reduces inter-cell interference.[2] This mechanism enables programming and erasure through voltage-induced charge injection or tunneling into the trap sites, with data retention achieved by the deep energy wells of the traps that prevent charge leakage over time.[2]
The concept of charge trap memory was first proposed in 1967 by John Szedon and Ting Chu at Westinghouse Research Laboratories, who recognized charge trapping in metal-oxide-semiconductor (MOS) transistors as a potential basis for non-volatile storage to replace magnetic core memory with integrated circuits.[3] This evolved into structures like metal-oxide-nitride-oxide-silicon (MONOS) in the late 1960s and silicon-oxide-nitride-oxide-silicon (SONOS) by Fujitsu around 1977, but widespread adoption in commercial flash memory began in 2002 with Spansion's MirrorBit NOR flash devices.[3] By the mid-2000s, the shift accelerated with Toshiba's 2007 proposal for 3D NAND architectures, where charge trap cells proved essential for vertical stacking due to their superior scalability beyond the planar limits of floating-gate technology.[3]
Charge trap flash offers several key advantages over floating-gate designs, including higher endurance (up to thousands of program/erase cycles), reduced susceptibility to charge leakage and oxide damage, lower power consumption during operations, and faster programming speeds enabled by thinner tunnel oxides.[2] These benefits have made it the dominant technology in modern 3D NAND flash, powering high-density solid-state drives (SSDs) in enterprise and consumer applications, with leading implementations including Samsung's 280-layer V-NAND (2024), Micron's 276-layer NAND (2025), and SK Hynix's 321-layer NAND (2025).[4][5][6] By 2019, the total number of charge trap bits produced annually had surpassed those of floating-gate bits, with charge trap technology becoming dominant in modern NAND flash production.[3]
Fundamentals
Definition and principles
Charge trap flash (CTF) is a non-volatile semiconductor memory technology that stores data by capturing electrons in discrete traps within an insulating charge storage layer, rather than in a continuous conductive electrode. This approach typically employs a silicon nitride (Si₃N₄) layer as the charge trap material, integrated into a metal-oxide-nitride-oxide-silicon (MONOS) or oxide-nitride-oxide-silicon (SONOS) device stack. The trapped charges modulate the threshold voltage of the underlying transistor, enabling binary or multi-level data representation without power consumption for retention.[7][8]
The core structure of a CTF cell includes a thin tunnel oxide (typically 2-5 nm SiO₂) directly on the silicon channel for charge injection, followed by the charge trap layer (e.g., 5-10 nm Si₃N₄), a thicker blocking oxide (5-10 nm SiO₂ or high-k dielectric) to prevent charge escape toward the gate, and a polysilicon or metal control gate. Charge localization in isolated traps within the nitride layer reduces lateral charge spreading, allowing cell dimensions to shrink effectively and supporting higher areal densities in advanced nodes.[7][9]
CTF offers key advantages for modern memory applications, including superior scalability below 20 nm feature sizes by mitigating capacitive coupling and interference between adjacent cells, endurance exceeding 10⁵ program/erase cycles due to distributed trap sites that limit localized stress, and inherent suitability for vertical 3D stacking in NAND architectures. These benefits stem from the discrete nature of charge storage, which avoids the electron percolation issues in continuous conductors at sub-20 nm scales.[8][10][11]
At its foundation, CTF relies on quantum mechanical tunneling for electron injection and extraction through the tunnel oxide under applied bias, with traps in the nitride layer—arising from atomic-scale defects like dangling bonds—providing energy wells of approximately 1-2 eV depth. Typical electron trap densities in optimized Si₃N₄ layers reach around 10¹² cm⁻², ensuring reliable charge capture while maintaining low leakage for data retention over a decade.[12][13]
Comparison to floating-gate flash
In traditional floating-gate flash memory, charge is stored on a conductive polysilicon island that delocalizes electrons across the entire gate, resulting in high coupling ratios between adjacent cells and significant interference during programming and reading operations.[14] This delocalized storage leads to challenges such as charge redistribution and increased susceptibility to defects, which exacerbate crosstalk in densely packed arrays.[15]
Charge trap flash (CTF) addresses these scaling limitations by using discrete electron traps within an insulating nitride layer, which confines charge locally to individual sites and prevents lateral leakage between neighboring cells.[14] This isolation enables CTF to scale effectively to feature sizes below 10 nm, where floating-gate etching processes become unreliable due to imprecise patterning and high aspect ratios that cause incomplete removal of the polysilicon layer.[15] In contrast, floating-gate technology struggles with these fabrication issues beyond the 10 nm node, limiting planar NAND density without transitioning to 3D architectures.[16]
Performance-wise, CTF typically requires lower program and erase voltages, around 15-20 V compared to over 20 V for floating-gate devices, due to the thinner effective oxide thickness and reduced stress on the tunnel layer.[17] Read speeds in CTF can reach up to 100 ns in optimized designs with high sensing currents, offering faster access than traditional floating-gate NAND, though early implementations may exhibit higher initial leakage currents if trap engineering is suboptimal.[18] Reliability in CTF is enhanced by minimizing charge loss over time, as discrete traps resist the delocalized leakage paths inherent in floating gates, leading to improved data retention under thermal stress.[2]
History
Early charge trap EEPROM
The concept of charge trap memory was first proposed in 1967 by John Szedon and Ting Chu at Westinghouse Research Laboratories, who recognized charge trapping in metal-oxide-semiconductor (MOS) transistors as a potential basis for non-volatile storage.[3] The foundational development of charge trap technology in EEPROM devices originated in the 1960s through early metal-nitride-oxide-silicon (MNOS/MONOS) concepts pioneered by researchers at Westinghouse and other institutions, which utilized silicon nitride as a trapping layer to enable non-volatile charge storage without relying on conductive floating gates.[3] These structures addressed scaling limitations of earlier floating-gate approaches by distributing charge in discrete traps within the nitride insulator, allowing for more reliable retention in low-density, electrically erasable memories.[19]
Significant early progress occurred in the 1970s with silicon-oxide-nitride-oxide-silicon (SONOS) structures developed by Fujitsu around 1977, followed by prototypes in the 1980s that demonstrated exceptional charge retention exceeding 10 years at 85°C, validating the technology's potential for commercial non-volatile applications despite initial concerns over leakage paths in the nitride layer.[3] These prototypes highlighted the advantages of charge trapping for embedded systems requiring byte-level rewritability, with endurance capabilities supporting over 10^5 program/erase cycles under controlled conditions.[20][21]
Key technical challenges in these early devices, including efficient charge injection and extraction, were overcome by employing modified Fowler-Nordheim tunneling for write and erase operations, where electrons or holes tunnel through a thin oxide barrier into traps within the nitride layer characterized by depths of approximately 1-2 eV. This mechanism ensured deep capture sites that minimized detrapping, with electron traps around 1.5 eV and hole traps near 1 eV contributing to stable threshold voltage shifts.[22][23]
The byte-alterable architecture of these EEPROMs subjected the charge traps to intensive stability testing via repeated individual bit alterations, confirming their robustness against charge loss and degradation, which laid the groundwork for adapting the technology to higher-density, block-erasable flash memories.[24]
Initial flash experiments
In the late 1990s, researchers began exploring nitride-based charge trapping mechanisms for flash memory architectures to address scaling limitations of floating-gate technologies, with initial experiments focusing on silicon-oxide-nitride-oxide-silicon (SONOS) structures and localized trapping cells fabricated using 0.25 μm processes. These efforts demonstrated the potential for 1-2 bits per cell storage by exploiting discrete charge traps in the nitride layer, enabling multi-level operation without the charge-sharing issues inherent in floating gates.[25][20]
Key prototypes from 2000 to 2002 showcased promising performance, including endurance exceeding 10^4 cycles and program times under 10 μs achieved through channel hot electron injection for writing and Fowler-Nordheim tunneling or hot hole injection for erasing. For instance, early SONOS cells with a 2 nm tunnel oxide, 4.5 nm nitride trap layer, and 5.5 nm blocking oxide exhibited stable operation after deuterium annealing to passivate interface traps, while localized trapping designs like NROM confirmed block-erasable functionality in array configurations. These results highlighted charge trap flash's suitability for high-density applications, with retention projected beyond 10 years at elevated temperatures following extensive cycling.[25][20]
Early implementations faced challenges with trap uniformity in the nitride layer, resulting in threshold voltage shifts (ΔV_th) of approximately 2-3 V due to non-uniform charge distribution and de-trapping during read/program operations. Such variability arose from defects at the oxide-nitride interfaces, exacerbating over-erase or read disturb effects in initial arrays. Solutions emerged through the use of nitrided oxides for the tunnel and blocking layers, which improved trap density control and reduced interface state generation, thereby enhancing V_th stability and window margins to over 3 V for reliable multi-bit sensing.[26]
Proof-of-concept demonstrations included the first block-erasable charge trap flash arrays, which verified superior scalability over floating-gate designs at sub-100 nm nodes by localizing charge storage to mitigate inter-cell interference and capacitive coupling. These experiments laid the groundwork for denser, more reliable flash architectures, confirming that nitride trapping could support aggressive scaling while maintaining block-level erase uniformity essential for NAND and NOR variants.[25]
Commercialization via MirrorBit
Spansion, formed in 2003 as a joint venture between AMD and Fujitsu, commercialized charge trap flash technology through its MirrorBit NOR flash products, building on initial developments from AMD in 2002.[27][3] This marked the first major entry of charge trapping into high-volume production, shifting from floating-gate designs to nitride-based storage for improved scalability. MirrorBit enabled two bits per cell by utilizing discrete charge trapping sites within a silicon nitride layer, allowing independent programming of adjacent bits in a single transistor channel.[28]
The technology debuted at the 130 nm process node and advanced to 90 nm by 2005, supporting densities up to 1 Gb in single-chip devices optimized for code execution and data storage.[29][30] This architecture provided manufacturing advantages, including higher yields and faster time-to-production compared to floating-gate equivalents, which facilitated significant cost reductions per bit through denser integration without complex lithography adjustments.[28][31]
MirrorBit gained rapid market traction in mobile devices, such as wireless handsets for firmware and applications, and embedded systems like PDAs and automotive electronics, sustaining adoption into the 2010s.[32][33] By 2005, MirrorBit contributed over 20% of Spansion's memory sales, reflecting record growth amid a global flash market exceeding 5 billion units annually, and underscoring its role in driving NOR flash volume.[34][35]
The success of MirrorBit established charge trap flash as a viable alternative for high-density non-volatile memory, influencing subsequent generations of NOR and paving the way for broader CTF applications in embedded and consumer markets.[36][30]
Operating Principles
Charge trapping mechanism
In charge trap flash (CTF) memory, the charge trapping mechanism relies on discrete defect sites within the charge trap layer, typically silicon nitride (Si₃N₄), where electrons are captured and stored to represent data states. These traps are primarily silicon dangling bonds in the nitride material, which act as deep electron traps with energy levels approximately 1-3 eV below the conduction band, enabling stable charge retention by preventing thermal excitation at operating temperatures. Shallow traps, with energies closer to the band edges (around 0.9-1.2 eV), primarily capture holes and contribute less to long-term electron storage due to easier thermal detrapping.[37][38][39]
The storage principle involves electrons tunneling from the channel or substrate into these isolated trap sites during programming, forming localized potential wells that confine the charge without significant lateral spreading along the layer. This localization arises from the discrete nature of the traps—such as Si dangling bonds in Si-rich nitride—creating Coulombic barriers that inhibit charge migration, unlike the conductive polysilicon floating gate where charges can redistribute delocalized across the entire node. The trapped electrons modify the local electric field in the tunnel oxide, altering the transistor's threshold voltage to encode binary or multi-level states.[37][40]
Detection of trapped charge occurs through a measurable shift in the transistor's threshold voltage (ΔV_th), which quantifies the stored charge density. The shift is given by the equation:
\Delta V_{th} = \frac{q \cdot N_{trapped}}{C_{ox}}
where q is the elementary charge, N_{trapped} is the areal density of trapped electrons (in cm⁻²), and C_{ox} is the capacitance per unit area of the control oxide stack (in F/cm²). To derive this, consider the trapped charge inducing an equivalent gate voltage offset: the total trapped charge Q_{trapped} = q \cdot N_{trapped} acts like a fixed charge sheet, producing an electric field that shifts the flat-band voltage by \Delta V_{FB} = -Q_{trapped} / C_{ox}; for n-channel devices, this translates directly to \Delta V_{th} under the approximation of uniform trap distribution near the tunnel oxide interface, neglecting higher-order effects like charge centroid depth. This linear relationship allows reliable readout, with typical shifts of 2-5 V for multi-bit cells depending on trap density (∼10¹²-10¹³ cm⁻²).[37][40]
Stability of the trapped charge is governed by factors that minimize detrapping, including trap passivation via hydrogen incorporation during deposition, which saturates dangling bonds and reduces trap density by up to 50%. Retention characteristics are modeled using the Arrhenius equation for thermal emission from traps:
t_r = \tau_0 \exp\left(\frac{E_a}{kT}\right)
where t_r is the retention time, \tau_0 is the attempt-to-escape time (∼10⁻¹³ s), E_a is the trap activation energy (1-2 eV for deep electron traps), k is Boltzmann's constant, and T is temperature. Deeper traps exhibit longer retention (e.g., >10 years at 85°C for E_a \approx 1.8 eV), but Si-rich nitrides may show faster detrapping due to increased lateral hopping between occupied sites.[37][40][38]
Programming the charge trap layer
Programming the charge trap layer in charge trap flash (CTF) devices primarily involves injecting electrons into the nitride or similar trap layer to store data, shifting the threshold voltage of the memory cell positively. A common method, particularly in NOR-type CTF, is channel hot electron (CHE) injection, where a drain bias of 5-7 V accelerates electrons in the channel to energies of approximately 3.5 eV, enabling them to overcome the silicon dioxide barrier and be captured by traps in the charge trap layer near the drain end.[41][42] This process relies on hot carrier generation through impact ionization, with the gate voltage typically set around 8-10 V to enhance the vertical electric field for injection efficiency.
The dominant programming approach in NAND architectures employs Fowler-Nordheim (FN) tunneling, which applies a high electric field exceeding 8 MV/cm across the tunnel oxide to enable quantum mechanical tunneling of electrons directly into the charge trap layer, providing more uniform charge distribution compared to CHE.[43][44] This method is particularly suited for NAND architectures, where gate voltages of 15-20 V are used to generate the required field, though it demands higher voltages and can lead to greater oxide stress. The FN injection current density J is described by the equation
J = A E^2 \exp\left(-\frac{B}{E}\right),
where E is the electric field, and A and B are material-dependent constants derived from the barrier height and effective electron mass (typically A \approx 1.54 \times 10^{-6} \, \mathrm{A \cdot eV/V^2} and B \approx 6.83 \times 10^7 \, \mathrm{V/cm} for silicon dioxide barriers).[45][46]
For multi-level cell (MLC) operation in CTF, programming uses incremental voltage pulsing schemes to precisely control the trapped charge density, enabling storage of 2-4 bits per cell by establishing distinct threshold voltage levels.[47][48] Pulses with durations of 10-100 μs and stepwise increasing gate voltages (e.g., starting from 15 V and incrementing by 0.5-1 V per pulse) allow fine-tuned electron injection until the desired state is reached, often verified via incremental step pulse programming (ISPP). Overall programming efficiency is influenced by the capacitive coupling ratio between the control gate and the channel, typically ranging from 0.6 to 0.8, which determines the fraction of applied voltage effectively dropped across the tunnel oxide for injection.[49]
Erasing the charge trap layer
In charge trap flash (CTF), erasing removes electrons from the nitride trap layer to reset the cell's threshold voltage (V_th), shifting it negatively relative to the programmed state and reversing the ΔV_th change from electron trapping. The predominant method is Fowler-Nordheim (FN) tunneling of holes from the substrate or channel into the trap layer, achieved by applying a negative bias to the control gate (typically -15 to -22 V) while grounding the substrate, enabling hole injection through the tunnel oxide to neutralize trapped electrons.[50][51][52]
An alternative approach is hot hole injection, often via band-to-band tunneling (BTBT) at the drain junction under positive drain bias and zero gate voltage, which generates energetic holes for faster erasure compared to FN tunneling—typically around 1 ms versus 10 ms for FN in block operations—while reducing the required voltage.[53]
Erase operations occur at the block level in CTF arrays, with all cells in a block processed in parallel by biasing the shared p-well or channel region uniformly, allowing simultaneous hole injection across thousands of cells. Uniformity is maintained through iterative erase-verify loops, where read operations check V_th distributions after each pulse, continuing until all cells pass the erase verify threshold.[54][55]
A major challenge is over-erase, where some cells exhibit excessive negative V_th shifts due to uneven hole injection; this is mitigated by applying a positive substrate bias post-erase to selectively program over-erased cells back to neutral V_th. Additionally, endurance degrades after about 10^5 program/erase cycles, as repeated hole tunneling during erase generates interface and bulk traps in the tunnel oxide, narrowing the memory window and causing V_th instability.[51][56]
Manufacturing
Material differences
Charge trap flash (CTF) devices differ fundamentally from floating-gate flash in their core materials, shifting from a conductive storage mechanism to discrete trapping sites for improved scalability and reduced interference. The primary distinction lies in the storage layer, where CTF employs silicon nitride (Si₃N₄) as the charge trap layer, featuring a trap density of 10^{18}–10^{19} cm⁻³ that enables localized electron or hole capture without the continuous conductivity of the polysilicon (poly-Si) floating gate used in traditional devices.[57][58]
The dielectric stack in CTF also incorporates thinner tunnel oxides, typically 4–6 nm of SiO₂ or HfO₂, which support efficient quantum mechanical tunneling for charge injection while minimizing coupling ratio losses seen in thicker oxides of floating-gate structures. Complementing this, high-k blocking layers such as Al₂O₃ (10–15 nm thick) are used to enhance capacitive coupling to the control gate and suppress leakage, offering superior voltage scaling over the conventional SiO₂ blocking dielectrics in poly-Si-based flash.[59][60]
Electrode materials further diverge, with CTF favoring metal gates like TiN or W over poly-Si to achieve uniform work function and lower resistivity, critical for high-density 3D architectures where poly-Si would introduce variability and higher resistance.[61]
Si₃N₄'s wide bandgap of approximately 5 eV facilitates deeper trap states for stable charge retention, and its deposition processes align with thermal budgets under 800°C, enabling seamless integration into advanced logic-compatible manufacturing flows unlike the higher-temperature requirements for doped poly-Si gates.[62][63]
Process variations from floating-gate
The fabrication of charge trap flash (CTF) devices introduces several process variations compared to floating-gate (FG) flash, tailored to support scaling below 20 nm while maintaining reliability in the oxide-nitride-oxide (ONO) dielectric stack. A primary difference lies in the deposition of the charge trap layer, where low-pressure chemical vapor deposition (LPCVD) is employed for the silicon nitride film at temperatures of 700–800 °C, yielding a high-quality layer with controlled trap density and conformality suitable for thin geometries.[64][65] This LPCVD process circumvents the etch selectivity challenges inherent to FG devices, where polycrystalline silicon deposition and patterning require stringent control to avoid damaging the underlying tunnel oxide during subsequent etches.[64][66]
Etching procedures also diverge significantly, with the ONO stack in CTF patterned via anisotropic plasma etching to achieve vertical profiles in high-aspect-ratio structures, enabling precise definition of memory cells without excessive lateral undercutting.[64] For gates narrower than 20 nm, sidewall protection is incorporated through the deposition and anisotropic etching of spacer materials, such as additional nitride layers, to shield the stack edges during source/drain formation and prevent shorting or charge leakage paths.[67][68]
To accommodate aggressive scaling and 3D architectures, CTF fabrication incorporates self-aligned contacts, which align source/drain regions directly to the gate without additional lithography steps, reducing overlay errors in vertical NAND strings.[64] Complementing this, damascene processes are utilized for wordline and channel formation, involving selective etching of trenches followed by metal or polysilicon fill, which enhances compatibility with stacked 3D structures by minimizing topography variations.[64][69] Recent innovations as of 2025 include airgap integration and charge trap layer separation to enable z-pitch scaling below 50 nm in next-generation 3D NAND, improving density and reducing parasitic capacitance.[61] These adaptations have facilitated seamless integration with CMOS logic processes starting from the 45 nm node, leveraging shared front-end tools for gate stack formation.[66]
Yield optimization in CTF production emphasizes low defect densities, typically below 0.1/cm² for the ONO stack and channel regions, achieved through refined plasma chemistries and in-line monitoring to suppress particulates and interface traps that could degrade retention.[70][71]
Connection to nanocrystal approaches
Silicon nanocrystal memory represents a discrete trapping approach closely related to charge trap flash (CTF), where charge storage occurs in isolated silicon dots embedded within an oxide matrix rather than a continuous dielectric layer. These nanocrystals, typically 5-10 nm in diameter, function as localized charge traps that mimic the behavior of a floating gate but with metallic-like conduction due to their quantum-confined structure, enabling efficient electron storage and reduced lateral charge migration compared to uniform nitride layers in CTF.[72][73]
The connection between nanocrystal approaches and CTF stems from their shared goal of overcoming scaling limitations in floating-gate memories, with nanocrystals offering a hybrid solution that addresses charge leakage in continuous traps. In early prototypes during the 2000s, such as those explored by Intel, silicon nanocrystals were integrated into non-volatile memory structures to minimize leakage paths through the tunneling oxide, achieving better isolation than continuous nitride CTF by confining charges to discrete sites. This discrete nature reduces the impact of defects in the oxide, as charge loss requires tunneling from isolated dots rather than percolating through a uniform layer.[74][72]
Advantages of nanocrystal methods include superior charge isolation, leading to extrapolated retention times exceeding 10 years under typical operating conditions, attributed to the Coulomb blockade effect that suppresses charge emission at low temperatures. The Coulomb blockade energy for a spherical nanocrystal is given by
E_c = \frac{e^2}{8\pi \epsilon r},
where e is the electron charge, \epsilon is the permittivity of the surrounding medium, and r is the dot radius; for r \approx 5 nm, this yields E_c on the order of 0.1-0.5 eV, enhancing stability. However, disadvantages arise from fabrication complexity, involving precise control of nanocrystal size, density, and uniformity via methods like chemical vapor deposition or self-assembly, which hinders scalability for mass production.[75][76][72]
Over time, the field evolved from nanocrystal-based prototypes to pure CTF for commercial adoption, as the continuous nitride layer in CTF simplifies manufacturing processes like chemical vapor deposition, avoiding the variability in nanocrystal formation while retaining many benefits of discrete trapping at larger scales. This transition prioritized production yield and cost-effectiveness, with nanocrystals influencing hybrid designs but largely supplanted in mainstream NAND and NOR flash by the 2010s.[72][73]
Key Technologies
MirrorBit cell operation
The MirrorBit cell, developed by Spansion as a pioneering two-bit-per-cell charge trap flash design, features a structure with two adjacent charge traps in a silicon nitride layer along the channel of a single transistor. This oxide-nitride-oxide (ONO) dielectric stack allows independent storage of two bits by localizing charge in discrete nitride regions near the source and drain, minimizing interference between the bits compared to floating-gate alternatives.[77]
Programming in the MirrorBit cell relies on selective channel hot electron (CHE) injection to target either the left or right nitride trap, achieved through drain/source asymmetry by reversing the source and drain connections during the operation. Electrons are accelerated under a high gate voltage (typically around 9-10 V) and injected into the desired trap site, altering the local threshold voltage without affecting the opposite bit. This asymmetric biasing enables precise control over individual bit programming in the cell.[77]
The cell encodes four distinct states—00, 01, 10, and 11—corresponding to the charge occupancy in the two traps, where each programmed bit induces a threshold voltage shift (ΔV_th) of approximately 2 V relative to the erased state. These states are defined by the combination of charged or neutral traps, providing a reliable multi-bit storage mechanism within the NOR flash architecture.[77]
The operation cycle of the MirrorBit cell supports page-mode read and write functionalities, optimized for compatibility with NOR array configurations, where multiple cells can be accessed in parallel for efficient data handling. This includes sequential programming of bits within a page while maintaining array-level integrity.[78]
Multi-bit storage and reading
Charge trap flash (CTF) supports multi-level cell (MLC) operation, enabling the storage of 2 to 4 bits per cell by establishing 4 to 16 discrete charge levels within the nitride trap layer, each corresponding to a distinct threshold voltage that represents different data states.[79] This approach increases storage density compared to single-level cells while leveraging the localized charge trapping in the insulating nitride to minimize interference between levels.[80]
During readout, the memory cell operates in source follower mode, with a gate voltage (V_g) of 1-3 V applied to bias the transistor above its threshold, allowing the drain-source current (I_ds) to flow proportionally to the stored charge level.[81] Sense amplifiers then compare this I_ds against multiple reference currents to accurately discriminate the threshold voltage states and decode the multi-bit data.[82]
Reliability in MLC CTF is enhanced through integrated error correction codes (ECC), which correct bit errors arising from charge detrapping or noise, ensuring robust operation despite narrower voltage margins between states.[83] Read operations typically achieve latencies around 50 ns, supporting high-speed access suitable for embedded applications.[84]
Advancements
Adoption in NAND flash
The adoption of charge trap flash (CTF) in NAND memory marked a pivotal transition from traditional floating-gate architectures, primarily driven by the physical scaling limitations encountered in planar (2D) NAND designs. As NAND feature sizes approached the 15 nm node around 2010, floating-gate cells suffered from severe capacitive coupling between adjacent gates, leading to increased interference, charge leakage, and reliability degradation that hindered further density improvements. Samsung pioneered CTF integration in 2D NAND prototypes during the early 2010s, announcing a 20 nm-class 64 Gb 3-bit-per-cell device in 2010 that leveraged CTF to mitigate these issues by storing charge in discrete traps within a silicon nitride layer, reducing lateral electron migration and enabling tighter cell spacing.[85] By 2013, CTF achieved full commercial adoption in 1x nm (approximately 15-19 nm) nodes for 2D NAND production, allowing Samsung to fabricate larger 128 Gb dies with improved yield and reduced manufacturing complexity compared to floating-gate alternatives.[86]
This shift accelerated with the introduction of 3D architectures, where CTF became essential for vertical stacking. Samsung initiated mass production of its first-generation V-NAND in 2013, featuring 24 stacked layers and CTF-based cells that delivered 128 Gb MLC density while offering 2-10 times better reliability and endurance over equivalent 10 nm-class floating-gate planar NAND.[87] The technology's non-conductive charge storage minimized inter-cell interference in the vertical dimension, facilitating rapid layer scaling—from 24 layers in 2013 to over 238 layers by 2023 in Samsung's ninth-generation V-NAND—without the etching and deposition challenges of floating gates in 3D.[88] Other manufacturers, including SK Hynix and Kioxia, followed suit, adopting CTF for their 3D NAND roadmaps to achieve similar scaling benefits.
By 2025, CTF has become the dominant technology in the NAND flash industry, accounting for nearly all production as planar floating-gate devices phased out due to obsolescence and the near-universal shift to 3D stacking. This widespread adoption supports advanced multi-bit storage modes, such as triple-level cell (TLC) at 3 bits per cell and quad-level cell (QLC) at 4 bits per cell, which balance high density with practical performance in consumer and enterprise applications. In these configurations, CTF-enabled 3D NAND typically achieves endurance ratings of 10^3 to 10^4 program/erase cycles, enhanced by optimized nitride trap materials that improve charge retention and reduce wear compared to earlier generations.[89]
3D and vertical structures
Charge trap flash (CTF) has been integral to the transition from planar to three-dimensional (3D) NAND architectures, enabling vertical scaling through structures that stack memory cells along the z-axis. Kioxia's Bit Cost Scalable (BiCS) technology exemplifies this approach, utilizing a stacked deck of alternating oxide and wordline layers where high-aspect-ratio channel holes are etched through the entire stack to form vertical polysilicon channels surrounded by the CTF layer.[90] This configuration allows memory cells to be formed sequentially along the channel hole, with plate-shaped wordlines wrapping around multiple channels for efficient control. Similarly, Samsung's V-NAND employs vertical channels etched into the stacked layers, with the CTF nitride layer deposited conformally around the channel to provide discrete charge trapping sites, facilitating reliable operation in tall, cylindrical cell strings.[91]
The adoption of CTF in these vertical structures addresses key challenges in scaling, particularly the non-uniformity of charge distribution in polycrystalline silicon channels. Unlike floating-gate cells, which suffer from grain boundary variations in vertical polysilicon, CTF's insulating nitride layer enables uniform electron trapping across the channel height, improving threshold voltage consistency and reducing program/erase variability.[92] Layer scaling has advanced rapidly, with SK Hynix announcing mass production of 321-layer 3D NAND in 2024, leveraging CTF to maintain performance in denser stacks while achieving capacities up to 2 Tb per die in quad-level cell (QLC) configuration as of 2025.[93][94]
These 3D CTF-based designs deliver significant benefits in density and reliability. Bit densities exceeding 10 Tb per square inch have been realized through aggressive layer stacking and channel shrinking, as seen in recent generations where areal densities surpass 20 Gb/mm², enabling multi-terabyte SSDs in compact form factors.[95] Interference between adjacent cells is mitigated via slit-etched wordlines, which divide the stack into isolated blocks, minimizing lateral charge leakage and wordline-to-wordline coupling during operations.[96]
As of 2025, efforts to further enhance vertical scaling focus on z-pitch reduction, with IMEC demonstrating airgap integration between wordlines to lower parasitic capacitance and interference, targeting pitches below 30 nm for stacks exceeding 400 layers while preserving CTF reliability.[61] This approach, combined with charge trap layer cuts to prevent lateral migration, supports continued density gains toward 1,000-layer architectures by the end of the decade.
Emerging materials and innovations
Recent advancements in charge trap flash (CTF) memory have focused on integrating two-dimensional (2D) materials to enhance performance in low-power and high-speed applications. Molybdenum disulfide (MoS₂) channels have emerged as a promising alternative to polycrystalline silicon in 3D NAND structures, enabling high-density storage with reduced power consumption due to MoS₂'s low bandgap and high carrier mobility. A 2025 study demonstrated MoS₂-based CTF devices achieving multilevel cell operation with improved gate-coupling ratios, supporting energy-efficient in-memory computing for AI workloads.[97] As of November 2025, IMEC's research on airgap structures further advances z-pitch scaling, reducing interlayer capacitance by up to 30% in vertical CTF architectures for stacks beyond 300 layers.[61]
Further innovations leverage 2D materials for ultrafast programming, addressing latency bottlenecks in next-generation non-volatile memory. Graphene-based Dirac channels, combined with 2D-enhanced hot-carrier injection, enable sub-nanosecond program speeds as low as 400 picoseconds while maintaining non-volatile retention and endurance exceeding 5.5 × 10⁶ cycles. This approach supports both electron and hole trapping, offering bidirectional operation suitable for advanced neuromorphic systems.[98]
Process innovations include airgap integration to minimize parasitic capacitance in vertical CTF architectures, facilitating tighter z-pitch scaling in 3D NAND beyond 300 layers. IMEC's 2025 research highlighted airgap structures on the channel side, reducing interlayer capacitance by up to 30% and improving word-line stacking density without compromising reliability. Complementing this, nanoscale engineering of the charge trap layer (CTL), such as optimizing silicon nitride composition and trap energy distributions, allows precise control of trap density to mitigate lateral charge migration in vertical channels. These techniques, informed by machine learning models, enhance retention and suppress initial voltage shifts in multi-level cells.[61][99]
Charge-trapping models have evolved to predict and optimize electron dynamics during vertical NAND migration, incorporating barrier-engineered layers to reduce program/erase cycling-induced diffusion. Simulations from 2020–2025 studies reveal that tailored trap profiles minimize inter-cell interference, enabling reliable operation in stacked architectures with over 200 layers.[100]
The surge in AI-driven data storage demands is projected to propel the CTF-enabled NAND flash market beyond $100 billion by 2030, fueled by needs for scalable, high-density memory in data centers and edge computing.[101]
Applications
Embedded charge trap memories
Embedded charge trap flash (CTF) is widely utilized in embedded systems as eFlash within microcontrollers (MCUs) and system-on-chips (SoCs), particularly for automotive and Internet of Things (IoT) applications where low-power, non-volatile storage is essential for code and data retention. These implementations provide densities ranging from 1 to 64 Mb, seamlessly integrated with logic circuitry to enable compact, on-chip memory solutions for real-time control in vehicles and connected devices.[102][53]
A primary advantage of embedded CTF over discrete NOR flash is its significantly smaller footprint, achieved through monolithic integration that eliminates the need for separate memory packages, thus reducing overall system size and board space in power-constrained environments. Additionally, CTF exhibits strong compatibility with advanced CMOS processes at 28 nm and beyond, including high-k metal gate technologies, allowing co-fabrication with logic transistors without requiring specialized floating-gate steps.[102]
Key implementations leverage split-gate CTF architectures, such as split-gate metal-oxide-nitride-oxide-silicon (SG-MONOS), which support byte-level alterability for efficient in-system reprogramming of firmware and configuration data. These cells deliver endurance exceeding 10 million program/erase cycles at junction temperatures up to 170°C, making them suitable for reliable code storage in demanding automotive environments.[103][53]
As of 2025, trends in embedded CTF emphasize deeper integration into AI edge devices, where multi-bit storage schemes—typically 2-3 bits per cell—enhance density and efficiency for on-device machine learning inference, supported by in-flash processing techniques to accelerate low-latency computations.[104][102]
Bandgap-engineered devices
Bandgap-engineered charge-trapping devices represent an advanced variant of charge trap flash (CTF) memory, where the dielectric stack is modified to precisely control charge trapping and detrapping through tailored energy barriers and trap sites. The core technique involves stacked dielectrics, such as alternating layers of silicon dioxide (SiO₂) and silicon nitride (Si₃N₄) multilayers, which create engineered potential wells for electrons with trap depths typically ranging from 2 to 4 eV below the conduction band in the nitride layer.[58][105] This multilayer structure, exemplified by the bandgap-engineered silicon-oxide-nitride-oxide-silicon (BE-SONOS) configuration with triple oxide-nitride-oxide (ONO) layers, enhances charge confinement by leveraging differences in material bandgaps to form discrete trapping regions, improving overall device scalability and reliability compared to conventional single-layer nitride traps.
In terms of physics, the bandgap offset between the nitride and oxide layers plays a crucial role in charge retention and programming efficiency. The bandgap of Si₃N₄ is approximately 5 eV, while that of SiO₂ is about 9 eV, resulting in a ΔE_g = E_{g,nitride} - E_{g,oxide} ≈ -4 eV overall, but the effective conduction band offset at the SiO₂/Si₃N₄ interface is around 1 eV, with the nitride's conduction band lying lower than the oxide's. This offset enhances electron confinement within the nitride traps during storage, as illustrated in the band diagram below, where the valence band offset (≈2.5 eV) and conduction band offset (≈1.5 eV) create a potential barrier that suppresses leakage while allowing Fowler-Nordheim tunneling for program/erase operations under high fields (>8 MV/cm).
Conduction Band Energy
Si Substrate Tunnel Oxide (SiO₂) Trap Layer (Si₃N₄) Blocking Oxide (SiO₂) Gate
┌────────────┐ ┌────────────┐ ┌────────────┐ ┌────────────┐ ┌────────────┐
│ │ │ │ │ │ │ │ │ │
│ E_c,Si │ │ E_c,O₂ │ │ E_c,N₄ │ │ E_c,O₂ │ │ E_c,gate │
│ │ │ (higher) │ │ (lower by │ │ (higher) │ │ │
└────┬───────┘ └────┼───────┘ └────┼───────┘ └────┼───────┘ └────┼───────┘
│ │ │ │ │
│ │ │ Trap depth │ │
│ │ │ 2-4 eV │ │
Valence │ │ │ │ │
Band │ E_v,Si │ E_v,O₂ │ E_v,N₄ │ E_v,O₂ │ E_v,gate
│ │ (offset ~2.5 eV)│ (offset ~1 eV CB) │ │
┌────┴───────┐ ┌────┴───────┐ ┌────┴───────┐ ┌────┴───────┐ ┌────┴───────┐
│ │ │ │ │ │ │ │ │ │
└────────────┘ └────────────┘ └────────────┘ └────────────┘ └────────────┘
Conduction Band Energy
Si Substrate Tunnel Oxide (SiO₂) Trap Layer (Si₃N₄) Blocking Oxide (SiO₂) Gate
┌────────────┐ ┌────────────┐ ┌────────────┐ ┌────────────┐ ┌────────────┐
│ │ │ │ │ │ │ │ │ │
│ E_c,Si │ │ E_c,O₂ │ │ E_c,N₄ │ │ E_c,O₂ │ │ E_c,gate │
│ │ │ (higher) │ │ (lower by │ │ (higher) │ │ │
└────┬───────┘ └────┼───────┘ └────┼───────┘ └────┼───────┘ └────┼───────┘
│ │ │ │ │
│ │ │ Trap depth │ │
│ │ │ 2-4 eV │ │
Valence │ │ │ │ │
Band │ E_v,Si │ E_v,O₂ │ E_v,N₄ │ E_v,O₂ │ E_v,gate
│ │ (offset ~2.5 eV)│ (offset ~1 eV CB) │ │
┌────┴───────┐ ┌────┴───────┐ ┌────┴───────┐ ┌────┴───────┐ ┌────┴───────┐
│ │ │ │ │ │ │ │ │ │
└────────────┘ └────────────┘ └────────────┘ └────────────┘ └────────────┘
This diagram depicts the approximate band alignment, where electrons injected into the Si₃N₄ traps are confined by the surrounding oxide barriers, minimizing thermal emission and lateral charge migration.[106][107] The design enables ultra-low power operation in bandgap-engineered charge-trapping memory (BE-CTM) devices, with programming voltages below 10 V, making them suitable for embedded applications requiring minimal energy consumption.[108]
Research on these devices advanced significantly in the 2010s through 2025, with contributions from TSMC focusing on integration into CMOS processes for enhanced endurance and scalability. For instance, TSMC explored bandgap-engineered structures in embedded non-volatile memory prototypes, achieving improved retention through optimized multilayer stacks that reduce charge loss via improved barrier heights.[108][109] Research has also investigated high-k variants of charge-trap devices for radiation tolerance, where the discrete trap nature minimizes single-event upsets in harsh environments. These developments have led to applications in radiation-hardened embedded flash for aerospace systems, where devices demonstrate total ionizing dose tolerance >300 krad(Si) and negligible bit errors under gamma irradiation, outperforming traditional floating-gate alternatives in some metrics.[53][110]