Hubbry Logo
Multi-level cellMulti-level cellMain
Open search
Multi-level cell
Community hub
Multi-level cell
logo
8 pages, 0 posts
0 subscribers
Be the first to start a discussion here.
Be the first to start a discussion here.
Multi-level cell
Multi-level cell
Multi-level cell
View on Wikipedia
from Wikipedia
SLC, MLC, TLC, QLC, PLC shown with all possible bit combinations per cell type
The differences of the memory cells in comparison

In electronics, a multi-level cell (MLC) is a memory cell capable of storing more than a single bit of information, compared to a single-level cell (SLC), which can store only one bit per memory cell. A memory cell typically consists of a single floating-gate MOSFET (metal–oxide–semiconductor field-effect transistor), thus multi-level cells reduce the number of MOSFETs required to store the same amount of data as single-level cells.

Triple-level cells (TLC) and quad-level cells (QLC) are versions of MLC memory, which can store three and four bits per cell respectively. The name "multi-level cell" is sometimes used specifically to refer to the "two-level cell". Overall, the memories are named as follows:

  1. Single-level cell or SLC (1 bit per cell)
  2. Multi-level cell or MLC (2 bits per cell), alternatively double-level cell or DLC
  3. Triple-level cell or TLC (3 bits per cell) or 3-Bit MLC
  4. Quad-level cell or QLC (4 bits per cell)
  5. Penta-level cell or PLC (5 bits per cell) – currently in development[1]

Notice that this nomenclature can be misleading, since an "n-level cell" in fact uses 2n levels of charge to store n bits (see below).

Typically, as the "level" count increases, performance (speed and reliability) and consumer cost decrease; however, this correlation can vary between manufacturers.

Examples of MLC memories are MLC NAND flash, MLC PCM (phase-change memory), etc. For example, in SLC NAND flash technology, each cell can exist in one of the two states, storing one bit of information per cell. Most MLC NAND flash memory has four possible states per cell, so it can store two bits of information per cell. This reduces the amount of margin separating the states and results in the possibility of more errors. Multi-level cells that are designed for low error rates are sometimes called enterprise MLC (eMLC).

New technologies, such as multi-level cells and 3D Flash, and increased production volumes will continue to bring prices down.[2]

Single-level cell

[edit]

Flash memory stores data in individual memory cells, which are made of floating-gate MOSFET transistors. Traditionally, each cell had two possible states (each with one voltage level), with each state representing either a one or a zero, so one bit of data was stored in each cell in so-called single-level cells, or SLC flash memory. SLC memory has the advantage of higher write speeds, lower power consumption and higher cell endurance. However, because SLC memory stores less data per cell than MLC memory, it costs more per megabyte of storage to manufacture. Due to higher transfer speeds and expected longer life, SLC flash technology is used in high-performance memory cards. In February 2016, a study was published that showed little difference in practice between the reliability of SLC and MLC.[3]

A single-level cell (SLC) flash memory may have a lifetime of about 50,000 to 100,000 program/erase cycles.[4]

A single-level cell represents a 1 when almost empty and a 0 when almost full. There is a region of uncertainty (a read margin) between the two possible states at which the data stored in the cell cannot be precisely read.[5]

Multi-level cell

[edit]

The primary benefit of MLC flash memory is its lower cost per unit of storage due to the higher data density, and memory-reading software can compensate for a larger bit error rate.[6] The higher error rate necessitates an error-correcting code (ECC) that can correct multiple bit errors; for example, the SandForce SF-2500 flash controller can correct up to 55 bits per 512-byte sector with an unrecoverable read error rate of less than one sector per 1017 bits read.[7] The most commonly used algorithm is Bose–Chaudhuri–Hocquenghem (BCH code).[8] Other drawbacks of MLC NAND are lower write speeds, lower number of program/erase cycles and higher power consumption compared to SLC flash memory.

Read speeds can also be lower for MLC NAND than SLC due to the need to read the same data at a second threshold voltage to help resolve errors. TLC and QLC devices may need to read the same data up to 4 and 8 times respectively to obtain values that are correctable by ECC.[9]

MLC flash may have a lifetime of about 1,000 to 10,000 program/erase cycles. This typically necessitates the use of a flash file system, which is designed around the limitations of flash memory, such as using wear leveling to extend the useful lifetime of the flash device.

The Intel 8087 used two-bits-per-cell technology for its microcode ROM,[10] and in 1980 was one of the first devices on the market to use multi-level ROM cells.[11][12] Intel later demonstrated 2-bit multi-level cell (MLC) NOR flash in 1997.[13] NEC demonstrated quad-level cells in 1996, with a 64 Mbit flash memory chip storing 2 bits per cell. In 1997, NEC demonstrated a dynamic random-access memory (DRAM) chip with quad-level cells, holding a capacity of 4 Gbit. STMicroelectronics also demonstrated quad-level cells in 2000, with a 64 Mbit NOR flash memory chip.[14]

MLC is used to refer to cells that store 2 bits per cell, using 4 charge values or levels. A 2-bit MLC has a single charge level assigned to every possible combination of ones and zeros, as follows: When close to 25% full, the cell represents a binary value of 11; when close to 50%, the cell represents a 01; when close to 75%, the cell represents a 00; and when close to 100%, the cell represents a 10. Once again, there is a region of uncertainty (read margin) between values, at which the data stored in the cell cannot be precisely read.[15][5]

As of 2013, some solid-state drives use part of an MLC NAND die as if it were single-bit SLC NAND, giving higher write speeds.[16][17][18]

As of 2018, nearly all commercial MLCs are planar-based (i.e. cells are built on silicon surface) and so subject to scaling limitations. To address this potential problem, the industry is already looking at technologies that can guarantee storage density increases beyond today’s limitations. One of the most promising is 3D Flash, where cells are stacked vertically, thereby avoiding the limitations of planar scaling.[19]

In the past, a few memory devices went the other direction and used two cells per bit to give even lower bit error rates.[20]

Enterprise MLC (eMLC) is a more expensive variant of MLC that is optimized for commercial use. It claims to last longer and be more reliable than normal MLCs while providing cost savings over traditional SLC drives. Although many SSD manufacturers have produced MLC drives intended for enterprise use, only Micron sells raw NAND Flash chips under this designation.[21]

Triple-level cell

[edit]
Image of a 2TB-3D-NAND-SSD
A triple-level cell storage

A triple-level cell (TLC) is a type of NAND flash memory that stores 3 bits of information per cell. Toshiba introduced memory with triple-level cells in 2009.[22]

With current technology a maximum lifetime of up to 3,000 program/erase cycles is achievable.[23]

Samsung announced a type of NAND flash that stores 3 bits of information per cell, with 8 total voltage states (values or levels), coining the term "triple-level cell" ("TLC"). Samsung Electronics began mass-producing it in 2010,[24] and it was first seen in Samsung's 840 Series SSDs.[25] Samsung refers to this technology as 3-bit MLC. The negative aspects of MLC are amplified with TLC, but TLC benefits from still higher storage density and lower cost.[26]

In 2013, Samsung introduced V-NAND (Vertical NAND, also known as 3D NAND) with triple-level cells, which had a memory capacity of 128 Gbit.[27] They expanded their TLC V-NAND technology to 256 Gbit memory in 2015,[24] and 512 Gbit in 2017.[28]

Enterprise TLC (eTLC) is a more expensive variant of TLC that is optimized for commercial use.

Quad-level cell

[edit]
A grey SSD with the text Samsung Solid State Drive"
The Samsung 870 QVO: A QLC SSD with 8 TB storage

Memory that stores 4 bits per cell is commonly referred to as quad-level cell (QLC), following the convention set by TLC. Prior to its invention, the term "QLC" was synonymous with MLC in referring to cells that can have 4 voltage states, i.e. ones that store 2 bits per cell – what is now unambiguously referred to as DLC.[citation needed]

Due to the exponentially increasing number of required voltage stages for higher level flash the lifetime of QLC is further reduced to a maximum of 1,000 program/erase cycles.[23]

In 2009, Toshiba and SanDisk introduced NAND flash memory chips with quad-level cells, storing 4 bits per cell and holding a capacity of 64 Gbit.[22][29]

SanDisk X4 flash memory cards, introduced in 2009, was one of the first products based on NAND memory that stores 4 bits per cell, commonly referred to as quad-level-cell (QLC), using 16 discrete charge levels (states) in each individual transistor. The QLC chips used in these memory cards were manufactured by Toshiba, SanDisk and SK Hynix.[30][31]

In 2017, Toshiba introduced V-NAND memory chips with quad-level cells, which have a storage capacity of up to 768 Gbit.[32] In 2018, ADATA, Intel, Micron and Samsung have launched some SSD products using QLC NAND memory.[33][34][35][36]

In 2020, Samsung released a QLC SSD with storage space up to 8 TB for customers. It is the SATA SSD with the largest storage capacity for consumers as of 2020.[37][38]

Enterprise QLC (eQLC) is a more expensive variant of QLC that is optimized for commercial use.

See also

[edit]

References

[edit]
[edit]
Revisions and contributorsEdit on WikipediaRead on Wikipedia

Multi-level cell

View on Grokipedia
from Grokipedia
A multi-level cell (MLC) is a type of non-volatile cell that stores multiple bits of data—typically two bits—per cell by representing distinct binary states through varying levels of charge on a floating gate or charge trap, allowing for four possible voltage thresholds (e.g., corresponding to 00, 01, 10, or 11). This contrasts with single-level cells (SLC), which store only one bit per cell using two states, enabling MLC to achieve higher storage density and lower cost per bit while requiring more precise programming and sensing mechanisms to distinguish the finer voltage differences. The concept of multi-level cells emerged in the early 1990s as a means to reduce the silicon area and manufacturing costs of , with initial beginning at in 1992 to address the growing demand for denser non-volatile storage in applications like embedded systems and portable devices. By 1994, announced the technology and demonstrated a 32 Mbit test chip at the International Solid-State Circuits Conference (ISSCC) in 1995, showcasing reliable operation without the need for error correction through innovations in charge placement and on-chip reference sensing. Commercial production followed in 1997 with the launch of 's 64 Mbit StrataFlash memory, fabricated on a 0.4 μm process, marking the first widespread adoption of MLC in NOR flash architectures and setting the stage for its integration into NAND flash. While MLC offers a balance of capacity and performance—storing twice the data of SLC at roughly half the cost per bit—it trades off some and speed, with cells typically supporting fewer program/erase cycles (around 10,000 compared to SLC's 100,000) and slower read/write operations due to the need for more accurate voltage discrimination. Over time, the multi-level approach has evolved beyond traditional 2-bit MLC to include triple-level cells (TLC) with 3 bits per cell and quad-level cells (QLC) with 4 bits, further increasing density for consumer SSDs and enterprise storage, though these extensions amplify reliability challenges like charge retention and error rates. Today, MLC and its extensions dominate the market, powering everything from smartphones to data centers, with ongoing advancements in 3D NAND stacking to mitigate scaling limitations.

Overview

Definition and Basic Concept

A multi-level cell (MLC) in refers to a storage unit capable of representing multiple bits of by utilizing distinct levels of charge within a single floating-gate transistor. Floating-gate transistors form the core of cells, where is stored non-volatily by trapping electrons in the floating gate—a conductive layer isolated by oxide layers—which modifies the transistor's (Vth). The number of trapped electrons determines the Vth shift, enabling the cell to retain information without power. In NAND flash architecture, these floating-gate cells are connected in series within strings to create efficient, high-density arrays organized into pages and blocks, serving as the foundational structure for scalable . Single-level cells (SLC) limit storage to one bit per cell by distinguishing just two Vth states (erased and programmed), whereas MLC and higher variants increase by partitioning the Vth range into multiple discrete states—such as four states for two bits in MLC or eight states for three bits. This multi-state encoding allows each cell to hold more information without expanding the physical footprint. The reliability of these states relies on threshold voltage distributions, visualized as non-overlapping Gaussian curves along the Vth axis, where each curve corresponds to a specific state and is separated by guard bands to prevent read errors. In a conceptual , these distributions appear as bell-shaped curves clustered at different Vth levels (e.g., low for erased states, progressively higher for programmed states), with voltages positioned in the inter-curve gaps for state detection. This foundational concept has extended to higher-level cells like triple-level (TLC) and quad-level (QLC) for further density improvements in modern NAND devices.

History of Development

Flash memory technology originated in 1984 with Fujio Masuoka at developing the first NOR-type using single-level cell (SLC) designs storing one bit per cell, enabling non-volatile storage with electrical erasure. Commercial NAND flash, also pioneered by Masuoka and in 1987, prioritized reliability and speed for early applications like embedded systems and portable devices. Multi-level cell (MLC) technology emerged in the 1990s to increase storage density. began research in 1992 and announced MLC in 1994, demonstrating a 32 Mbit test chip; commercial production followed in 1997 with the 64 Mbit StrataFlash NOR memory. The shift toward MLC in NAND flash occurred in the early 2000s, driven by the need for higher storage density in . SanDisk and jointly introduced the first MLC NAND flash in 2001, with a 1 Gbit device that stored two bits per cell, enabling the 1 GB card as the initial commercial product. This innovation reduced costs per bit and facilitated wider adoption in devices like digital cameras and players. By the early 2010s, triple-level cell (TLC) technology emerged, with launching the world's first consumer SSD using 3-bit TLC NAND in its 840 series in 2012, significantly boosting capacity for mainstream solid-state drives. A pivotal advancement occurred in 2013 with the commercialization of 3D NAND by , stacking memory cells vertically to overcome planar scaling limits and enhance the viability of higher-level cells by improving density without excessive voltage precision demands. This transition paved the way for quad-level cell (QLC) NAND, which and Micron commercialized in 2018 with the industry's first 1 Tb QLC die, offering four bits per cell and 33% greater bit density than TLC for read-intensive applications like . Recent years have seen the emergence of penta-level cell (PLC) prototypes, pushing to five bits per cell amid ongoing 3D stacking innovations. In 2023, demonstrated a 1.67 Tb PLC device using 192-layer floating-gate 3D NAND, achieving a bit of 23.3 Gb/mm² and featuring advanced read algorithms for margin improvement. Industry research, including by Corp. (YMTC), has advanced PLC feasibility, while showcased next-generation NAND strategies at the OCP Global Summit 2025, targeting petabyte-scale SSDs via enhanced layer counts and cell densities with QLC technology to support exabyte-level storage systems.

Cell Types

Single-Level Cell (SLC)

Single-Level Cell (SLC) NAND represents the simplest form of non-volatile storage in flash technology, serving as a foundational reference for more advanced multi-bit variants. Each SLC stores exactly one bit of per cell by distinguishing between two voltage threshold states: the erased state (typically uncharged, representing logic '1') and the programmed state (charged via injection, representing logic '0'). This binary mechanism enables reliable through the floating-gate structure, where the presence or absence of trapped s modulates the cell's conductivity during reads. SLC's design contributes to exceptional , with cells typically supporting 50,000 to 100,000 program/erase (P/E) cycles before becomes limiting, far exceeding that of higher-density alternatives. This robustness stems from the wider margin between the two voltage states, reducing susceptibility to charge leakage and disturb effects over repeated operations. Historically, SLC NAND found primary applications in early enterprise storage solutions and embedded systems, such as industrial controllers and mission-critical servers, where high reliability and consistent performance were essential over cost efficiency. These uses leveraged SLC's ability to maintain in environments demanding frequent writes without frequent replacements. Key advantages of SLC include rapid read and write operations, with typical read latencies under 50 μs and program times around 200-300 μs, alongside inherently low raw bit error rates (on the order of 10^{-9} to 10^{-11}) that necessitate only basic error-correcting codes (ECC). In contrast to multi-level cells like MLC or TLC, SLC trades lower storage density for superior speed and longevity, making it ideal for performance-oriented baselines.

Multi-Level Cell (MLC)

Multi-level cell (MLC) NAND represents the foundational multi-bit storage technology in , storing two bits of data per cell by distinguishing among four distinct voltage states, typically encoded as 00, 01, 10, and 11. This approach doubles the storage density compared to single-level cell (SLC) NAND, which uses only two states, while building on SLC's established binary reliability for charge trapping mechanisms. Achieving these four states requires precise control of injection and retention within the floating gate or charge trap layer, as even minor variations in can shift the cell's state during reads or over time. Introduced commercially in the early 2000s, MLC NAND enabled significant cost reductions in high-density storage applications, with SanDisk and announcing the first 1 Gbit MLC NAND device in 2001, which powered early products like 1 GB cards. This innovation facilitated the proliferation of affordable USB drives and nascent solid-state drives (SSDs) for consumer markets, where cost-per-bit became a primary driver over enterprise-grade durability. By the mid-2000s, MLC had become the standard for mainstream storage, balancing density gains with acceptable performance for everyday use cases like personal computing and portable media. A key trade-off in MLC design is its , rated at approximately 3,000 to 10,000 program/erase (P/E) cycles per cell, which supports reliable operation in consumer SSDs but falls short of SLC's higher cycle counts for intensive workloads. This limitation arises from the repeated stress on the tunnel oxide during multi-state programming, accelerating wear compared to binary operations. MLC faces specific reliability challenges due to narrower voltage margins between its four states, which are roughly half the width of SLC's two-state margins, resulting in elevated raw bit error rates (BER) on the order of 10^{-3}. These tighter distributions demand more robust sensing circuitry and calibration during read operations to distinguish states accurately, particularly as cells age or under variations. Despite these hurdles, MLC's density advantages made it pivotal for scaling flash storage in the era.

Triple-Level Cell (TLC)

Triple-level cell (TLC) NAND flash memory stores three bits of data per cell by distinguishing among eight distinct voltage states, enabling three times the storage density of single-level cell (SLC) NAND, which uses only two states for one bit. This multi-level approach builds briefly on multi-level cell (MLC) technology by adding an additional bit, allowing for greater capacity at lower costs while maintaining compatibility with existing NAND architectures. TLC has become the dominant choice for consumer storage in the 2020s, striking an optimal balance between cost per and performance for mainstream applications like laptops and gaming systems. The commercial adoption of TLC began with 's introduction of the 840 series SSD in 2012, marking the first widespread use of TLC in consumer drives and paving the way for higher-capacity options. further advanced TLC integration in with its V-NAND technology in the 850 EVO series, which enabled the production of 1TB and larger consumer SSDs by leveraging vertical stacking to increase cell density without sacrificing too much reliability. These milestones shifted TLC from niche to mainstream, powering the growth of affordable, high-capacity solid-state drives that now dominate the market. TLC NAND typically offers an endurance of 1,000 to 3,000 program/erase (P/E) cycles per cell, sufficient for consumer workloads but lower than SLC or MLC due to the finer voltage distinctions that accelerate cell wear. Write operations in TLC are slower than in lower-level cells because they rely on sequential programming, where each of the eight states is set incrementally through multiple voltage s, often taking 200-500 μs per state transition in the incremental step programming process. This results in overall page program times of around 900-1,350 μs, contributing to reduced write speeds in sustained workloads compared to SLC but still adequate for typical user scenarios.

Quad-Level Cell (QLC)

Quad-level cell (QLC) NAND flash memory stores four bits of data per cell by distinguishing among 16 discrete voltage states, ranging from the lowest (representing 1111 in Gray coding) to the highest (representing 0000). This approach quadruples the storage density compared to single-level cell (SLC) NAND, which uses only two states for one bit per cell, allowing manufacturers to pack more capacity into smaller dies without increasing cell size. As a result, QLC has become pivotal for ultra-high-density applications, particularly archival storage where maximizing terabytes per drive is essential. QLC builds on the progression from triple-level cell (TLC) NAND by adding a fourth bit, pushing further while prioritizing read-intensive scenarios. Its is limited to approximately 100–1,000 program/erase (P/E) cycles per cell, far lower than SLC or TLC, which suits it for workloads like cold data storage, media archives, and backups where writes are minimal and reads dominate. However, this low endurance makes QLC unsuitable for write-heavy tasks in advanced SSDs, as frequent writes accelerate wear and reduce lifespan due to narrower voltage margins and charge leakage. This enables cost-effective, high-capacity solutions but requires careful workload matching to avoid premature wear. Commercialization of QLC began in 2018 with Intel and Micron delivering the first 1 Tb QLC die based on 64-layer 3D NAND technology, followed by Micron's shipment of the 5210 ION SSD series for enterprise use. By the early 2020s, consumer-grade QLC drives achieved capacities over 8 TB, exemplified by Samsung's 870 QVO SATA SSD, which targeted affordable mass storage. In 2025, the technology advanced to over 200-layer 3D stacking, with SK Hynix initiating mass production of 321-layer QLC NAND, enhancing density for petabyte-scale archival systems in data centers. A key drawback of QLC is its elevated raw (BER), reaching around 10−210^{-2} due to the narrow voltage margins among 16 states, which demands sophisticated error-correcting codes (ECC) like low-density parity-check (LDPC) to achieve reliable operation and meet uncorrectable BER targets below 10−1610^{-16}. These advanced ECC mechanisms increase controller complexity and overhead but are essential for maintaining in high-density environments. Additionally, QLC production faces yield issues due to manufacturing complexities in achieving precise voltage distributions across multiple states, often requiring built-in self-repair techniques to improve reliability. In stacked configurations, such as high-layer 3D NAND, reliability concerns arise from increased charge leakage and retention loss, potentially limiting QLC's use in high-performance SSD applications without mitigation strategies like caching or hybrid modes.

Penta-Level Cell (PLC)

The penta-level cell (PLC) represents an emerging advancement in NAND flash memory technology, capable of storing 5 bits of data per cell by utilizing 32 distinct voltage states to represent different charge levels. This configuration enables a theoretical storage density five times greater than that of single-level cells (SLC), which store only 1 bit per cell, thereby addressing the density limitations of prior technologies like quad-level cells (QLC). Development of PLC has progressed through prototypes and lab demonstrations, with Solidigm (a SK hynix subsidiary) unveiling the industry's first working PLC-based SSD prototype in 2022, showcasing its potential for higher-capacity storage solutions. Ongoing research as of 2025 includes efforts by YMTC toward transitioning to PLC architectures as part of broader 3D NAND density improvements, with lab demonstrations focusing on integration for high-capacity applications such as AI data centers. Despite these advances, PLC faces substantial technical hurdles due to the extremely narrow voltage margins—approximately 3% between adjacent states, compared to 6% in QLC—resulting in projected of fewer than 500 program/erase (P/E) cycles and elevated raw bit error rates (BER) exceeding 10^{-1}. These limitations stem from the increased susceptibility to and charge leakage in higher-level cells, necessitating advanced mitigation strategies. Feasibility of practical PLC deployment in 2025 relies heavily on AI-driven error correction techniques to manage the high BER and ensure reliable operation in dense storage environments.

Technical Principles

Storage Mechanism

Multi-level cells (MLCs) in NAND store multiple bits of data by modulating the (Vth) of the memory cell , achieved through controlled charge storage in either floating-gate or charge-trap structures. In floating-gate flash, electrons are injected into or removed from an isolated polysilicon gate via Fowler-Nordheim (FN) tunneling, a quantum mechanical effect that occurs under high (typically 10 MV/cm) when a voltage pulse (around 15-20 V) is applied between the control gate and substrate. This tunneling adjusts the Vth, shifting it from an erased state (low Vth, representing logic 1) to programmed states (higher Vth, representing logic 0 patterns). Charge-trap flash, increasingly adopted in modern designs, replaces the floating gate with a nitride layer (e.g., ) that traps electrons more discretely, reducing issues like charge lateral migration and enabling better scaling. For multi-bit storage, MLCs define discrete Vth distributions corresponding to each possible state (e.g., four states for 2-bit MLC), where the separation between adjacent distributions (ΔVth) must exceed approximately 3σ—three times the standard deviation (σ) of the in the distribution—to minimize read errors from overlap. These distributions are modeled as Gaussian with , ensuring reliable state discrimination via read reference voltages placed in the margins. The evolution to 3D NAND addresses 2D scaling limits below 15 nm by vertically stacking charge-trap layers (over 300 as of ) in a pillar-like , increasing density without further shrinking cell size laterally. This vertical integration uses materials like silicon-nitride for charge trapping within stacked word lines, mitigating inter-cell interference and enabling higher areal densities (>25 Gb/mm²). Key noise sources degrading Vth stability in MLCs include charge leakage (retention loss), where trapped electrons gradually escape over time due to thermal emission or stress-induced leakage, modeled as an in Vth; read disturb, which incrementally raises Vth in unselected cells during read operations from pass voltage stress; and program disturb, elevating Vth in untargeted cells via coupling fields during programming. These effects widen distributions and erode margins, particularly in higher-level cells.

Read and Write Operations

In multi-level cell (MLC) NAND , the write operation, also known as programming, relies on Incremental Step Programming (ISPP) to precisely control the (Vth) of each cell to one of multiple target levels corresponding to the stored data states. ISPP involves applying a series of short, successive high-voltage to the control gate of the selected cell, with each incrementally increasing in amplitude—typically by a step of about 0.2 V—while starting from an initial program voltage often in the range of 15-20 V, depending on the technology node. After each , a verify operation reads the cell's Vth to determine if it has reached the desired level for the programmed state; if not, the next is applied, ensuring tight Vth distributions essential for distinguishing multiple levels without overlap. This iterative process, grounded in charge trapping mechanisms, allows for fine-grained control over injection via Fowler-Nordheim tunneling, accommodating variations in cell characteristics. The erase operation in MLC NAND flash resets an entire block of cells simultaneously to a low Vth state, representing the erased data value (typically all 1s). This block-level erasure uses Fowler-Nordheim tunneling by applying high negative voltage pulses (around -15 to -22 V) to the substrate or p-well while grounding the control gates, which extracts electrons from the charge storage layer across the tunnel oxide. Verify steps follow each erase pulse to confirm that all cells in the block have fallen below a predefined erase verify voltage, ensuring uniform reset without individual cell addressing, which is a fundamental limitation of NAND . Reading an MLC involves determining the cell's Vth state by comparing it against multiple reference voltages that separate the Vth distributions of adjacent data states. For basic MLC (2 bits per cell), reads at three reference voltages are required to distinguish the four states, but for higher-level cells like triple-level cell (TLC, 3 bits), multi-pass reads—typically 3 to 4—are employed, where the cell is sensed multiple times at different reference levels to resolve the exact state. Algorithms such as binary search or successive approximation register (SAR) are used to efficiently identify the Vth window by iteratively adjusting the reference voltage and observing the cell's conduction behavior, minimizing the number of sensing operations while accurately decoding the multi-bit value. To mitigate errors from small shifts in Vth that could cause misreads between adjacent states, MLC NAND employs Gray coding in the data-to-state mapping, ensuring that neighboring Vth levels differ by only one bit. This scheme confines potential decoding errors to a single bit flip rather than multiple, enhancing reliability in the presence of or wear without altering the core read procedure.

Performance Characteristics

Density and Capacity

Multi-level cells (MLCs) enhance storage in NAND flash memory by encoding multiple bits of data within each individual cell, as opposed to single-level cells (SLCs) that store only one bit per cell. This approach directly scales capacity: for example, quad-level cells (QLCs), which store 4 bits per cell, provide up to four times the storage of SLCs within the same die area, allowing manufacturers to pack more data without expanding the physical footprint. Similarly, triple-level cells (TLCs) with 3 bits per cell offer a threefold increase over SLCs, balancing and capacity in mainstream applications. The integration of 3D NAND technology compounds these gains by vertically stacking memory layers, transitioning from planar 2D layouts to high-aspect-ratio structures. As of 2025, commercial 3D NAND stacks commonly feature 200-350 layers, with ongoing advancements toward even higher counts, such as plans for 400-layer NAND targeted beyond 2029-2031, enabling exponential capacity growth when paired with multi-level cells. This vertical scaling, combined with the logical density from MLC variants, has driven chip capacities into the terabyte range per die, facilitating overall system-level storage expansions. In practice, these innovations have resulted in 2025-era solid-state drives (SSDs) achieving 16 TB or higher capacities through QLC-based 3D NAND, as seen in offerings from manufacturers like with 2 Tb QLC dies enabling dense form factors. Such milestones underscore how multi-level cell density directly translates to larger, more efficient storage solutions for data-intensive environments. However, these capacity advancements often against other operational attributes like .

Speed and Latency

Multi-level cells (MLCs) in NAND flash memory exhibit read latencies that increase with the number of voltage states per cell due to the need for more precise sensing to distinguish finer voltage thresholds. In single-level cells (SLCs), which store one bit per cell using two voltage states, typical read latency is around 25 μs, as reported for early Micron SLC NAND devices. For triple-level cells (TLCs) and quad-level cells (QLCs), which store three or four bits per cell with eight or sixteen states respectively, read latency rises to 50-100 μs, primarily because multi-pass sensing is required to accurately resolve the narrower voltage margins between states. Write speeds in MLCs are notably slower than reads and degrade further with higher cell levels, as programming involves incremental adjustments to voltage levels to avoid overshoot. For TLCs, write times per cell typically range from 500 μs to 1 ms, driven by multiple iterations of incremental step-pulse programming (ISPP), where voltage pulses are applied in steps to fine-tune the for each of the eight states. This iterative process, essential for precision in multi-level storage, contrasts with the simpler single-pulse programming in SLCs, which completes in under 300 μs. Several factors influence these latencies in MLCs, including the time required for sensing amplifiers to detect subtle voltage differences, which scales with state complexity and can add tens of microseconds per pass. To mitigate sustained write slowdowns, many NAND controllers employ caching modes that emulate SLC operation in a portion of the TLC or QLC array, enabling burst write speeds up to several GB/s until the cache fills, after which performance drops to native multi-level rates. These operational steps, such as multi-pass sensing and ISPP, inherently contribute to the delays observed in MLC reads and writes. As of 2025, advancements in 3D NAND architectures have reduced effective latencies through enhanced parallelism, such as interleaving operations across multiple channels and dies to overlap sensing and data transfer. For instance, next-generation 3D flash from and SanDisk achieves a 33% improvement in NAND interface speed via higher parallelism, lowering overall system latency for multi-level cells in enterprise SSDs.

Endurance and Reliability

Endurance in multi-level cells (MLCs) refers to the number of program/erase (P/E) cycles a cell can withstand before significant degradation impairs its functionality. As the number of voltage states per cell increases to store more bits, the precision required for charge placement and retention diminishes, leading to reduced . Single-level cells (SLCs), storing 1 bit per cell, typically support 50,000 to 150,000 P/E cycles. In contrast, MLCs (2 bits per cell) endure around 3,000 to 10,000 cycles, triple-level cells (TLCs, 3 bits) 1,000 to 3,000 cycles, and quad-level cells (QLCs, 4 bits) 100 to 1,000 cycles. This lower endurance in QLC NAND particularly limits its suitability for write-heavy tasks in advanced SSDs, where frequent program/erase operations accelerate wear and reduce overall lifespan. Degradation in MLCs primarily arises from oxide wear during repeated P/E operations, where high cause trap generation in the tunnel , accelerating charge leakage. Additionally, charge detrapping from the storage layer leads to (Vth) shifts, narrowing the voltage margins between states and increasing error susceptibility. For instance, retention loss can result in Vth shifts of several hundred millivolts after one year at 85°C, particularly in higher-level cells like TLCs and QLCs, due to accelerated de-trapping at elevated temperatures. In QLC NAND, these degradation mechanisms are exacerbated in stacked 3D configurations, where increasing layer counts lead to greater charge leakage and retention loss, raising reliability concerns. Reliability metrics for MLCs highlight the escalating challenges with cell density. The raw (RBER) rises from approximately 10^{-4} in SLCs to 10^{-3} in MLCs and up to 10^{-2} in QLCs after moderate P/E cycling, reflecting tighter voltage distributions prone to interference and retention errors. This elevated RBER in QLC necessitates advanced error correction codes, such as low-density parity-check (LDPC) schemes with higher computational overhead, to maintain data integrity. Production of QLC NAND also faces yield issues, as evidenced by delays in mass production due to design flaws and performance inconsistencies in high-layer stacked architectures. In enterprise solid-state drives (SSDs) using MLC NAND, (MTBF) is typically rated at 2 to 2.5 million hours, providing a statistical measure of drive-level reliability under continuous operation. Without appropriate mitigation techniques, such as enhanced caching or hybrid modes, these challenges limit QLC's application in high-performance SSDs, particularly for workloads requiring consistent speed and durability. As of 2025, material engineering in 3D NAND architectures has improved endurance and reliability through innovations like high-k dielectrics (e.g., Al₂O₃ and HfO₂) in blocking layers, which reduce oxide stress and enhance charge confinement, extending P/E cycles by 20–50% in TLC and QLC implementations. Antiferroelectric materials, such as HfZrO₂, further mitigate Vth instability by boosting capacitance and minimizing detrapping, supporting denser stacking without proportional reliability losses.

Applications and Challenges

Commercial Uses

Multi-level cells, particularly triple-level cell (TLC) and quad-level cell (QLC) variants, dominate consumer storage applications due to their ability to provide high capacities at reduced costs compared to single-level cell (SLC) options. In 2025, TLC NAND is the standard in solid-state drives (SSDs) for laptops, enabling configurations like 2TB drives in models such as the HP OmniBook X and series, which support demanding tasks including and gaming without compromising everyday performance. The rising demand for AI features in smartphones has driven NAND prices up by 5-10% as of Q3 2025, increasing adoption of TLC for efficient, high-capacity storage. QLC NAND extends this trend to portable devices, powering high-capacity USB flash drives from manufacturers like SanDisk, where terabyte-scale storage facilitates data backups and media transfer for personal use. These deployments leverage the density advantages of multi-level cells to meet growing consumer needs for affordable, large-scale storage. In enterprise settings, SLC and multi-level cell (MLC) hybrids are employed in server SSDs to deliver the endurance required for mixed workloads, with techniques like pseudo-SLC emulation on MLC NAND boosting write cycles for databases and . QLC finds its niche in cold storage within data centers, where its high density suits infrequently accessed data and helps minimize costs for long-term retention. This application highlights QLC's role in scaling petabyte-level repositories economically. In AI servers, demand for NAND is driven by high-capacity QLC SSDs utilized for backend storage, supporting large-scale data processing and inference workloads in generative AI applications. This trend remains ongoing, with projections indicating QLC SSDs will comprise 30% of the enterprise SSD market by 2026 due to continued AI infrastructure expansion. Embedded systems also benefit from multi-level cells' balance of performance and reliability. TLC NAND is commonly integrated into smartphones, providing durable storage for operating systems and apps in devices from and Apple, where space constraints demand efficient density. In the automotive sector, TLC NAND powers and data logging systems in vehicles like Tesla models, handling real-time data, autonomous driving features, and media while enduring vibration and temperature extremes. As of Q3 2025, market estimates indicate TLC commands over 50% of the NAND flash sector, driven by its versatility in and , while QLC captures around 20% through expansions in high-capacity storage; MLC's share is declining as manufacturers like phase out production in favor of higher-density alternatives.

Mitigation Techniques

To mitigate the higher bit error rates (BER) inherent in triple-level cell (TLC) and quadruple-level cell (QLC) NAND flash memories, low-density parity-check (LDPC) codes are widely employed as advanced error correction mechanisms. These codes provide robust correction capabilities, approaching the theoretical Shannon limit, and are particularly effective for the elevated error rates in multi-level cells compared to single-level cells (SLC). For QLC NAND, which can exhibit raw bit error rates (RBER) up to 10^{-2}, hardware-accelerated LDPC engines with extended codeword lengths and configurable decoding iterations are utilized to handle complex error patterns efficiently, ensuring data integrity in high-density configurations. For instance, LDPC implementations can correct hundreds to thousands of bits per , such as over 100 bits per 1KB sector, enabling reliable operation in high-density storage environments. Wear leveling algorithms address the limited program/erase (P/E) cycle endurance of multi-level cells by evenly distributing write operations across all available blocks, preventing premature wear on frequently used areas and thereby extending overall device lifespan. Dynamic and static techniques identify and relocate "hot" (frequently written) and "cold" (infrequently written) to balance usage, with dynamic methods focusing on active data movement during writes. An example is dynamic SLC caching, where a portion of TLC or QLC storage is temporarily operated as higher-endurance SLC to handle intensive writes, combined with wear leveling to reallocate blocks and optimize efficiency. To further enhance endurance in QLC for broader applications, including high-performance SSDs, techniques such as Flexible Data Placement (FDP) and aging-aware data placement are employed, reducing write amplification and optimizing NAND usage under sustained write activity. Retention management techniques counteract charge leakage over time in multi-level cells, which exacerbates errors in higher-density configurations, by implementing periodic refresh cycles to rewrite data and restore voltage levels. These refreshes are scheduled based on factors like P/E cycle count and time since last write, reducing retention-induced errors without excessive overhead. Temperature compensation is integrated to adjust retention parameters dynamically, as elevated temperatures accelerate charge loss; for example, data retention time can decrease by orders of magnitude from 25°C to 85°C, prompting adaptive monitoring and correction in SSD controllers. For reliability concerns in stacked QLC configurations, advanced error monitoring and predictive maintenance mechanisms detect wear-out proactively, addressing issues like charge leakage and retention loss in high-layer stacks. In 2025 advancements, AI-optimized error correction codes, such as machine learning-based algorithms, enhance LDPC efficiency by predicting error patterns and adapting decoding strategies, reducing computational overhead and improving latency in NAND flash systems. Additionally, dynamic over-provisioning in SSD controllers allocates variable spare capacity based on workload, enhancing for multi-level cells by providing more flexible space for garbage collection and . To improve yield in QLC production and enable use in high-performance applications, advanced repair techniques and bad block management are applied during manufacturing, combined with controller-level mitigations like scalable logical-to-physical mapping to maintain performance in dense configurations. These techniques collectively address reliability challenges in and retention, as detailed in related performance characteristics.

References

  1. https://semiconductor.samsung.com/support/tools-resources/dictionary/semiconductors-101-part-2-different-types-of-nand-flash/
  2. https://www.intel.com/content/dam/www/public/us/en/documents/research/1997-vol01-iss-4-intel-technology-journal.pdf
  3. https://ieeexplore.ieee.org/document/848644/
  4. https://www.kingston.com/en/blog/pc-performance/difference-between-slc-mlc-tlc-3d-nand
  5. https://semiconductor.samsung.com/news-events/tech-blog/a-brief-history-of-data-placement-technologies/
  6. https://www.researchgate.net/publication/2985190_Flash_Memory_Cells-An_Overview
  7. https://www.usenix.org/system/files/conference/fast12/grupp.pdf
  8. https://www.usenix.org/system/files/conference/fast14/fast14-paper_jeong.pdf
  9. https://irds.ieee.org/images/files/pdf/2023/2023IRDS_MDS.pdf
  10. https://spectrum.ieee.org/chip-hall-of-fame-toshiba-nand-flash-memory
  11. https://www.global.toshiba/ww/news/corporate/2001/11/pr1202.html
  12. https://semiconductor.samsung.com/consumer-storage/world-no1-flash-memory-episode2/
  13. https://semiengineering.com/3d-nand-race-faces-huge-tech-and-cost-challenges/
  14. https://www.hpcwire.com/2018/05/22/intel-micro-up-the-ante-for-flash-memory/
  15. https://ieeexplore.ieee.org/document/10149089/
  16. https://news.skhynix.com/sk-hynix-presents-next-generation-nand-storage-product-strategy-at-ocp-2025/
  17. https://userpages.umbc.edu/~squire/images/ssd1.pdf
  18. http://www.ccs.neu.edu/home/simona/fast10-flash.pdf
  19. https://americas.kioxia.com/en-us/business/memory/slc-nand.html
  20. https://www.micron.com/products/storage/nand-flash/slc-nand
  21. https://www.kioxia.com/en-jp/rd/technology/multi-level-cell.html
  22. https://www.intelligentmemory.com/fileadmin/Editors/pdf/White_Paper_NAND_Flash_Tech_01.2023.pdf
  23. https://newsletter.semianalysis.com/p/the-history-and-timeline-of-flash
  24. https://americas.kioxia.com/content/dam/kioxia/en-us/about/asset/KIOXIA_35th_Anniv_Milestones_Infographic.pdf
  25. https://www.ssstc.com/knowledge-detail/nand-flash-types-comparison/
  26. https://www.oscoo.com/news/nand-flash-slc-mlc-tlc-and-qlc/
  27. https://www.thessdreview.com/ssd-guides/learning-to-run-with-flash-2-0/understanding-slc-mlc-tlc-and-qlc-nand-flash-learning-to-run-with-flash-2-0/
  28. https://www.purestorage.com/knowledge/what-is-3d-nand.html
  29. https://oretonstorage.com/blog/understanding-nand-types-ssd
  30. https://www.thessdreview.com/our-reviews/samsung-840-series-240gb-ssd-review-the-worlds-first-tlc-ssd-takes-the-stage/
  31. https://www.storagereview.com/news/samsung-begins-mass-production-of-industrys-first-tlc-v-nand
  32. https://www.silicon-power.com/knowledge-detail/ssd-buying-guide/
  33. https://buy.advantech.com/whatsnew/mktg/Download/Industrial_Storage_Guidebook.pdf
  34. https://www.simms.co.uk/Uploads/Resources/50/f4366381-b992-425a-bec4-cca409b51a6c.pdf
  35. https://www.ibm.com/think/insights/qlc-vs-tlc
  36. https://www.enterprisestorageforum.com/hardware/understanding-multilayer-ssds/
  37. https://www.techtarget.com/searchstorage/tip/4-QLC-workloads-and-why-theyre-a-good-fit-for-QLC-NAND-flash
  38. https://hackaday.com/2024/07/08/the-flash-memory-lifespan-question-why-qlc-may-be-nand-flashs-swan-song/
  39. https://investors.micron.com/news-releases/news-release-details/micron-and-intel-extend-their-leadership-3d-nand-flash-memory
  40. https://redmondmag.com/articles/2018/05/24/micron-first-quad-level-cell-nand-ssd.aspx
  41. https://www.techpowerup.com/340276/sk-hynix-begins-mass-production-of-321-layer-qlc-nand-flash
  42. https://www.researchgate.net/publication/351743634_BER_Evaluation_System_Considering_Device_Characteristics_of_TLC_and_QLC_NAND_Flash_Memories_in_Hybrid_SSDs_with_Real_Storage_Workloads
  43. https://www.solidigm.com/products/technology/qlc-nand-ready-for-mainstream-use-in-data-center.html
  44. https://www.allaboutcircuits.com/industry-articles/solving-the-qlc-nand-flash-ssd-scaling-challenge/
  45. https://ieeexplore.ieee.org/document/11218053
  46. https://news.solidigm.com/en-WW/217006-solidigm-demonstrates-world-s-first-penta-level-cell-ssd-at-flash-memory-summit/
  47. https://www.globenewswire.com/news-release/2025/06/17/3100387/0/en/Memory-and-Storage-Technology-Research-Report-2025-High-Bandwidth-Memory-Set-to-Dominate-DRAM-Market-by-2030-Amid-AI-Boom-and-3D-NAND-Advancements-Long-term-Forecast-to-2036.html
  48. https://www.computer.org/csdl/magazine/co/2024/09/10660602/1ZPP7VjWRQk
  49. https://www.computerweekly.com/feature/PLC-flash-The-next-generation-or-a-mirage
  50. https://ieeexplore.ieee.org/document/10873308/
  51. https://www.mdpi.com/2079-9292/10/24/3156
  52. https://users.ece.cmu.edu/~omutlu/pub/flash-memory-voltage-characterization_date13.pdf
  53. https://news.skhynix.com/one-team-spirit-the-miracle-of-321-layers/
  54. https://americas.kioxia.com/en-us/business/news/2025/20250220-1.html
  55. https://library.techinsights.com/search/sectioned-blog-viewer/57985145-c9b1-4c27-b200-e97702d4df6b
  56. https://people.engr.tamu.edu/ajiang/Publications/2014/p2262-li.pdf
  57. https://ieeexplore.ieee.org/document/535460/
  58. https://www.sciencedirect.com/topics/engineering/fowler-nordheim-tunneling
  59. https://arxiv.org/pdf/2202.05661
  60. http://reports-archive.adm.cs.cmu.edu/anon/2018/CMU-CS-18-101.pdf
  61. https://www.mdpi.com/2079-9292/11/10/1585
  62. https://www.sciencedirect.com/science/article/abs/pii/S0026269216307169
  63. https://www.micron.com/products/storage/nand-flash/qlc-nand
  64. https://www.techtarget.com/searchstorage/tip/The-truth-about-SLC-vs-MLC
  65. https://www.imec-int.com/en/articles/unlocking-z-pitch-scaling-next-generation-3d-nand-flash
  66. https://www.trendforce.com/news/2025/11/04/news-sk-hynix-unveils-2029-2031-roadmap-featuring-hbm5-gddr7-next-and-400-layer-nand/
  67. https://www.tomshardware.com/pc-components/ssds/16tb-m2-ssds-will-soon-grace-the-market
  68. https://arxiv.org/pdf/1208.6391
  69. https://www.tomshardware.com/pc-components/ssds/custom-pcie-5-0-ssd-with-3d-xl-flash-debuts-special-optane-like-flash-memory-delivers-up-to-3-5-million-random-iops
  70. https://ghose.cs.illinois.edu/papers/18sigmetrics_3dflash.pdf
  71. https://people.inf.ethz.ch/omutlu/pub/Reducing-SSD-Read-Latency-by-Optimizing-Read-Retry_asplos21.pdf
  72. https://www.atpinc.com/blog/SSD-cache-what-is-it
  73. https://apac.kioxia.com/en-apac/business/news/2025/20250220-1.html
  74. https://www.arasan.com/blog/nand-flash-interleaving/
  75. https://research.ece.cmu.edu/safari/pubs/1706.08642.pdf
  76. https://www.usenix.org/legacy/events/fast11/tech/full_papers/Pan.pdf
  77. https://pdfs.semanticscholar.org/1739/8e85c39fcc657757abc961545425669443c2.pdf
  78. https://www.enterprisestorageforum.com/hardware/ssd-lifespan-how-long-will-your-ssd-work/
  79. https://www.techradar.com/pro/understanding-the-differences-between-enterprise-and-client-ssds
  80. https://ieeexplore.ieee.org/document/10979979/
  81. https://www.trendforce.com/news/2025/09/16/news-samsung-reportedly-delays-v9-qlc-nand-to-1h26-as-sk-hynix-hits-321-layers-in-august/
  82. https://www.cell.com/device/fulltext/S2666-9986(24)00639-2
  83. https://www.hp.com/us-en/shop/vwa/laptops/stor=2-TB
  84. https://www.dell.com/en-us/shop/dell-laptops/scr/laptops/appref=2-tb-hard-drive-size
  85. https://smartphones.gadgethacks.com/news/ai-boom-drives-smartphone-prices-up-20-in-2025/
  86. https://shop.sandisk.com/topics/ssd/what-is-qlc
  87. https://www.trendforce.com/presscenter/news/20251014-12744.html
  88. https://www.trendforce.com/presscenter/news/20251203-12813.html
  89. https://www.trendforce.com/news/2025/05/27/news-samsung-reportedly-to-end-mlc-nand-shipments-for-consumers-by-june-shift-to-automotive/
  90. https://www.eetimes.com/flash-wearout-drives-tesla-recall/
  91. https://www.industryresearch.biz/market-reports/nand-market-101457
  92. https://www.trendforce.com/presscenter/news/20240722-12228.html
  93. https://www.astutegroup.com/news/industrial/samsung-phases-out-mlc-nand-driving-industry-shift-to-tlc-and-qlc/
  94. https://www.szyunze.com/ldpc-codes-power-the-core-reliability-of-modern-ssds/
  95. https://americas.kioxia.com/content/dam/kioxia/en-us/business/memory/mlc-nand/asset/KIOXIA_Managed_Flash_BOS_P3_Understanding_Wear_Leveling_Tech_Brief.pdf
  96. https://www.cervoz.com/technology/dynamic-SLC-cache/detail
  97. https://users.ece.cmu.edu/~omutlu/pub/flash-correct-and-refresh_iccd12.pdf
  98. https://americas.kioxia.com/content/dam/kioxia/en-us/business/memory/mlc-nand/asset/KIOXIA_NAND_Flash_Data_Retention_Technical_Brief.pdf
  99. https://www.atpinc.com/blog/how-to-over-provisioning-SSD-benefits-endurance-wai-tbw
Add your contribution
Related Hubs
User Avatar
No comments yet.