YURUNOX / Memory selection guide

SLC vs MLC vs TLC vs QLC NAND Flash: Which Fits Your Workload?

For mixed workloads, start by qualifying a TLC device. Evaluate QLC when capacity and read intensity dominate; use native SLC or a supported pSLC configuration when repeated writes, deterministic latency or harsh conditions justify the density trade-off. Do not approve any option until its exact part, capacity, firmware, post-cache behavior, endurance, retention and power-loss scope match the real workload.

For hardware engineers, embedded-system teams and component buyers
Primary manufacturer sources reviewed September 5, 2026 · Product examples, not YURUNOX test results

Illustrative NAND selection map A two-axis map places SLC and pSLC toward higher write intensity and latency sensitivity, QLC toward higher capacity priority, and MLC and TLC between them. Exact device evidence is still required. Start with the workload, not the acronym Capacity priority and read dominance Write intensity and latency sensitivity SLCor qualified pSLC MLC TLC QLC Illustrative direction only — exact device evidence decides
Cell type narrows the shortlist. Capacity, controller, firmware, temperature and the measured write envelope decide whether a specific device can be qualified.

Which NAND Type Fits Your Workload?

Start with the application’s hardest write, retention and recovery condition. TLC is often a practical mixed-workload shortlist; QLC needs an explicit write envelope; native SLC or supported pSLC can justify their capacity trade-off in compact, write-intensive designs; and 2-bit MLC may remain correct for a platform already qualified around an exact part.

A storage device that rarely writes can still fail an application requirement if a large update takes too long. Conversely, a higher-density device may be a good fit when its specified endurance, write latency and retention comfortably cover the job.

Scroll sideways to compare all columns →

A starting-point matrix, not an approval rule. Qualification remains part- and workload-specific.
Workload conditionStarting pointEvidence requiredStop or hold when
Small dataset, frequent writes, tight write latency or harsh conditionsEvaluate native SLC or a manufacturer-supported pSLC modeExact ordering code, usable capacity, endurance and retention conditions, controller/ECC responsibilitiesThe quote says only “SLC grade,” or capacity loss and mode support are undocumented
Mixed operating-system, application and data writesQualify TLC firstCapacity-specific post-cache speed, TBW/DWPD basis, temperature, firmware and power-loss behaviorOnly peak speed is supplied or the tested capacity/firmware does not match the quote
Large, mostly read-only dataset with controlled replacement windowsEvaluate QLCLongest write burst, intended fill level, cache limits, refill time, retention and service-life write budgetThe update cannot finish after cache exhaustion or retention evidence omits wear and temperature
Existing platform qualified around 2-bit MLCProtect the exact MLC design or qualify a controlled alternateGeometry/interface match, firmware impact, PCN/lifecycle information and requalification plan“Same family” or “same package” is offered instead of exact compatibility evidence

Here, MLC means 2-bit-per-cell NAND. Some manufacturers also use "multi-level cell" as a broad term that includes TLC and QLC.

01 / Understand the mechanism

How Do SLC, MLC, TLC and QLC Store Data Differently?

NAND stores information through electrical charge that changes a cell's threshold voltage. A cell storing n bits must represent 2n states. As the number of states rises, reading and programming those states becomes more demanding. The controller, error correction and NAND generation therefore matter alongside the bit count. KIOXIA explains the underlying multi-level-cell mechanism.

One cell / Four ways to encode data
SLC2 states
MLC4 states
TLC8 states
QLC16 states
State-count illustration only: the blocks are not a measured voltage scale or actual threshold distributions.

At the same cell count, QLC stores 4/3 as many nominal bits as TLC—about 33% more, not four times as much. Finished-drive capacity and price also depend on spare space, die configuration, controller design and other costs.

How Is 3D NAND Different From Bits per Cell?

3D describes the stacking of cells; SLC, MLC, TLC and QLC describe bits per cell. A 3D device may use TLC or QLC. Comparing old planar MLC with newer 3D TLC using bit count alone overlooks changes in cell geometry and technology. Micron's automotive 3D TLC discussion provides a historical example of why generation matters; it is not a guarantee for every TLC part.

02 / Match the workload

Where Does Each NAND Type Fit Best?

When Is SLC Worth the Capacity Trade-Off?

SLC is worth considering for a compact data logger or embedded system that frequently changes a relatively small dataset. Its lower state count generally favors programming simplicity and cell-level endurance. However, the application still needs suitable error correction, bad-block handling and retention specifications. "SLC" alone does not certify a system for safety-critical use.

When Should a Design Stay With 2-Bit MLC?

Two-bit MLC can remain the right choice when a platform's firmware and qualification are tied to a particular device. For a new design, compare actual alternatives before treating MLC as the automatic industrial option. Availability, product-change notices and requalification cost can matter more than a theoretical cell-level advantage.

When Is TLC the Practical Starting Point?

TLC is a useful shortlist for systems combining operating-system activity, applications and regular data writes. Compare the exact capacity: a larger model may have different parallelism, cache behavior and endurance. An industrial temperature grade or documented power-loss feature must be specified separately; neither follows automatically from "TLC."

When Does QLC Make Sense?

QLC is worth evaluating when capacity is important and most access is reading. Enterprise products such as Solidigm's D5-P5336 demonstrate that QLC is not limited to entry-level client drives. That does not make an enterprise QLC device interchangeable with a consumer QLC SSD.

Ask how much data arrives in one burst, how full the drive becomes and how quickly it must be ready again. A dataset read all month but replaced overnight is still a sustained-write application during that replacement.

03 / Decode the SLC claim

Are Native SLC, pSLC and SLC Cache Equivalent?

Native SLC is designed for one-bit-per-cell storage. Pseudo-SLC, or pSLC, operates supported higher-bit NAND in a one-bit mode. A temporary SLC write cache uses that mode as a staging area, then moves data into the main TLC or QLC storage area. Kingston describes this caching approach.

Host writesA burst arrives from the application
SLC-mode cacheFast, finite staging capacity
Main TLC / QLC areaData is stored at higher density
This flow describes temporary caching, not a permanently configured pSLC storage area.

A cache can make a short benchmark look excellent. After the available cache is exhausted, writes depend on the device's uncached behavior and background work. Dynamic cache capacity may also shrink as the drive fills. Ask for sustained performance at the intended occupancy, not just an empty-drive peak.

Persistent pSLC trades capacity for one-bit operation: the same TLC cells hold roughly one-third of their nominal bits, and QLC cells one-quarter, before other overhead. Actual usable capacity and endurance depend on the supported product configuration. Do not assume every SSD lets the buyer enable this mode.

04 / Documented product comparison

What Happens When an SSD’s SLC Cache Is Full?

Rear identification label of a Samsung 870 QVO 1TB SATA SSD
The 1TB capacity matters: another capacity in the same family can have different post-cache specifications.Photo: Dinkun Chen / Wikimedia Commons · CC BY-SA 4.0. Uncropped; resized for display.

Samsung's 870 EVO and 870 QVO datasheets offer a concrete reason to read the footnotes. Both list up to 530 MB/s sequential writes, but their 1TB post-TurboWrite figures differ sharply.

This is a comparison of published client-SATA specifications, not a YURUNOX benchmark and not a prediction for every TLC or QLC product.

Scroll sideways to compare both drives →

Samsung 870 EVO Rev. 1.1 and 870 QVO Rev. 1.1; 1TB models only.
Published specification870 EVO 1TB870 QVO 1TB
NAND wording3-bit MLC V-NAND = TLC4-bit MLC V-NAND = QLC
Sequential write, up to530 MB/s530 MB/s
Write figure after TurboWrite530 MB/s80 MB/s
Endurance600 TBW360 TBW
Limited warranty period5 years3 years

Sources: 870 EVO datasheet, technical specifications and footnotes; 870 QVO datasheet, page 5. Performance depends on test conditions, firmware and host configuration. Warranty coverage is subject to the manufacturer's terms and TBW limit.

The purchasing lesson: identical peak speeds do not establish identical behavior during an image deployment, video ingest or backup. Also, Samsung’s “3-bit MLC” wording means TLC here—not the 2-bit MLC category discussed elsewhere in this guide. Use this example to learn how to read footnotes, not to generalize every TLC or QLC product.

05 / Translate workload into a requirement

How Much Endurance Does Your Application Need?

A generic endurance number for "all TLC" or "all QLC" is not a reliable purchasing specification. Use the exact device's rating and test conditions. Cell wear, host writes and years of service are related, but they are not interchangeable units.

How Do P/E Cycles, TBW and DWPD Relate?

Scroll sideways to read the metric definitions →

MetricWhat it describesWhat it does not establish
P/E cyclesSpecified program/erase cycling at the NAND levelA complete device's host-write budget without controller and workload context
TBWA device's rated written-data amount under stated conditionsAn exact date or byte count at which failure must occur
DWPDFull-drive-equivalent writes per day over a stated periodEndurance independent of capacity, years or workload definition

For decimal units, DWPD = TBW ÷ (capacity in TB × 365 × years). Always carry the period with the number. A five-year planning target does not extend a three-year warranty. Kingston's endurance guide explains TBW and DWPD.

The Samsung examples both normalize to about 0.33 DWPD over their respective periods: 600 ÷ (1 × 365 × 5) and 360 ÷ (1 × 365 × 3). That similarity does not erase their different time limits or post-cache write specifications.

Planning tool / Host writes

Estimate your write budget

Enter average host writes, including application, operating-system and logging activity. This calculation estimates a requirement; it does not qualify a drive.

Decimal GB; average over the operating schedule.
365 days per year for this estimate.
1.5 means 50% headroom; choose for your uncertainty.
Use the capacity basis used in the device rating.

Lifetime host-write budget, with allowance547.5 TB

Required DWPD, with allowance0.150

Base estimate: 365 TB over 5 years. Without the planning allowance: 0.100 DWPD on a 2 TB device.

Formula: GB/day × 365 × years ÷ 1,000 × allowance. No NAND-level write amplification is added. Confirm workload compatibility, retention, temperature and the actual warranty period separately.

In the illustrated plan, 200 GB/day becomes 365 TB over five years, or 547.5 TB with 50% headroom. A hypothetical 600 TBW device clears that numerical screen, but this alone says nothing about latency, retention or suitability for the actual workload.

When Should Write Amplification Be Included?

Write amplification is physical NAND writes divided by host writes. For example, 200 GB of host writes at a write-amplification factor of 2.5 produces 500 GB of NAND writes. This can help explain internal wear, but do not multiply host writes by that factor again when comparing them with a host-write TBW rating. Instead, check whether the rating's workload assumptions match yours.

Public engineering account / Micron

Why Must Capacity-Specific Evidence Be Reviewed?

In a first-person product-management account, Micron describes an endurance issue affecting one capacity of an SSD under development. A waiver was initially raised; the team instead worked through the issue. The account illustrates why qualification belongs to the specific capacity and workload, not just a product-family name.

Micron's published account is manufacturer-reported experience, not an independent test. It does not disclose enough before-and-after measurements to reproduce the result; no such measurements are assumed here.

06 / Look beyond write endurance

How Do Retention, Temperature and Power Loss Change the Choice?

Endurance asks how much writing a device supports. Retention asks how long stored data remains recoverable under specified conditions. Temperature, accumulated wear and time without refresh all matter. A lightly written device is not automatically qualified for years in an unpowered warehouse.

For equipment shipped with preloaded firmware, specify the longest unpowered storage interval and expected storage temperature. For a field recorder, include its worn state at end of service. KIOXIA's automotive NAND retention paper explains why wear and thermal history must be considered together.

Powering a device briefly is not proof that every block was refreshed. Confirm the controller's documented refresh behavior and the conditions needed to complete it. Keep independent backups for data that cannot be recreated.

Power-loss protection is another axis. Ask whether it protects acknowledged user writes, volatile buffers, mapping metadata, or only part of that path. NAND type does not answer this. Test the complete system's flush, commit and recovery behavior under the intended shutdown conditions.

07 / Identify what you are buying

Should You Buy Raw NAND, eMMC, UFS or an SSD?

Samsung eMMC package shown from the marking side and the BGA solder-ball side
eMMC packages combine NAND with a controller. The interface and management model are separate from bits per cell.Photo: Toniperis / Wikimedia Commons · CC BY-SA 4.0. Unmodified.

With raw NAND, the host generally carries the management burden: error correction, bad blocks and wear leveling. On-die ECC can move one task into the chip without making it fully managed.

With eMMC, UFS or an SSD, a controller manages the NAND internally. The host still needs compatible commands, correct power handling and a qualified software stack. Micron distinguishes these NAND architectures.

A raw-NAND alternate needs checks for voltage, interface timing, page and block geometry, addressing, ECC requirements and bad-block conventions. For a managed device, examine capacity, firmware, supported features, endurance, sustained behavior and power-loss scope. Matching package dimensions or a "TLC" description is not enough for either substitution.

08 / Turn a shortlist into a qualification plan

How Should You Qualify a Candidate Under the Real Workload?

Illustrative scenario / Not a customer test

Can a Read-Heavy System Still Need High Write Throughput?

Imagine a content appliance that mostly serves reads but must ingest a 600 GB update in 30 minutes. The required average is about 333 MB/s before allowing for verification and other overhead: 600,000 MB ÷ 1,800 seconds.

Its low monthly write total does not settle the choice. The candidate must sustain the update at its intended fill level, including any cache exhaustion. This is an acceptance requirement to test—not a prediction that a specific QLC or TLC drive will pass.

  1. Measure the workload. Record daily host writes, longest bursts, random versus sequential traffic, block sizes, queue depth and acceptable tail latency.
  2. Freeze the configuration. Record the exact part, capacity, firmware, host platform and temperature. Precondition the device and test at realistic occupancy.
  3. Run beyond the cache. Use a dedicated test device and enough data to expose sustained behavior. Write tests consume endurance and can overwrite data; keep production storage out of the test.
  4. Check recovery and aging conditions. Validate shutdown recovery, health reporting and the required retention conditions through the manufacturer's qualification guidance.
  5. Control future substitutions. Tie acceptance to the approved configuration. Define which NAND, controller or firmware changes trigger review and requalification.

Industrial suitability includes consistency, environmental limits and change control, not just a cell label. Swissbit's industrial-versus-consumer guidance discusses these broader distinctions.

09 / Convert the requirement into evidence

What Must Buyers Verify Before Ordering NAND or an SSD?

Supply the exact manufacturer part number if the design is already qualified. If an alternate is allowed, describe the application well enough to prevent a sales label such as “industrial TLC” or “high-endurance NAND” from substituting for engineering evidence.

Scroll sideways to review every RFQ field →

Minimum RFQ inputs and evidence. Add project-specific safety, regulatory and qualification requirements.
Decision fieldBuyer providesSupplier evidence requested
IdentityExact MPN or controlled alternate rules, capacity and quantityFull ordering code, manufacturer, capacity, NAND/controller identification where disclosed, firmware revision and traceability
ArchitectureRaw NAND, eMMC, UFS, SATA SSD, NVMe SSD or another managed formInterface revision, geometry/timing and ECC requirements for raw NAND, or supported commands/features for managed storage
WorkloadGB/day, longest burst, read/write ratio, block size, queue depth, fill level and latency limitPerformance conditions, cache size or policy when published, and steady-state or post-cache results for the quoted capacity
EndurancePlanned service life and headroom policyTBW, DWPD or P/E basis with period, workload assumptions, warranty boundary and applicable test conditions
EnvironmentOperating/storage temperatures, duty cycle and longest unpowered intervalTemperature grade, retention conditions, health or refresh behavior and any end-of-life limitations
Power and recoveryShutdown sequence, hold-up time and which data cannot be lostDocumented power-loss protection scope, flush/commit behavior and recovery guidance
LifecycleProduction years, change-notification needs and alternate approval processProduct status, PCN/PDN path, change-control information and the exact conditions that require requalification

Hold the order when the quoted capacity, firmware or device architecture differs from the evidence. Reject the candidate when a mandatory interface, temperature, retention, endurance or recovery condition fails. Move to sample qualification only when the evidence package and test plan match the exact configuration.

Before approving an alternate, align the technical evidence with incoming checks and YURUNOX quality-assurance requirements. Lower unit cost is not a saving if it creates an undocumented substitution or an unplanned qualification cycle.

YURUNOX / Electronic-component sourcing

Bring the exact part number and the workload

Share the required device class, MPN or alternate rules, capacity, quantity, need date, temperature range, write profile, retention interval and power-loss requirement. YURUNOX can then source against a defined evidence request instead of selecting by the NAND acronym alone.

Discuss your NAND sourcing requirements

Which Primary Sources Support This Comparison?

The Samsung comparison is a reading of published specifications. The write-budget and update-window examples are illustrative calculations, not measured device results. No manufacturer source below proves that a different product or capacity will behave the same way.

Sources accessed September 5, 2026. Manufacturer specifications may change; confirm the latest documentation for the exact ordering code, capacity and firmware before qualification or purchase. Photo credits and licenses appear beside each external image.

Cart (0 items)