Flash storage: A cheat sheet

Flash storage: A cheat sheet

With over a dozen form factors, knowing what the best flash memory card and SSD technology is available can be daunting

TechRepublic’s flash storage cheat sheet is an overview of the key information you need to know about the topic. This article will be updated when new formats and post-flash technologies are introduced.

What is flash storage?

Flash memory is an electronic, solid-state storage medium developed by Fujio Masuoka while working at Toshiba, circa 1980. Masuoka first publicly demonstrated the invention in 1987, with Intel producing the first commercial flash chip in 1988.

Masuoka’s invention covers two related types of non-volatile memory: NOR and NAND. NOR flash takes longer to write or erase, but provides byte-level random access, making it a suitable replacement for read-only memory (ROM) chips. NAND flash provides faster write and erase times, as well as more dense storage capabilities. NAND flash is written and read at a block level, making it unsuitable for embedded use cases that require byte-level access.

Generally, NOR flash is restricted to mission-critical applications and embedded use cases such as the firmware of a computer or an embedded electronic device. Flash memory cards and solid-state drives (SSDs) utilize NAND flash for mass storage.

What are the drawbacks of flash storage?

Flash storage has a finite lifespan in terms of the number of times a block can be erased and rewritten. As NAND flash memory becomes more dense through the use of multi-level cell technology, this lifespan has decreased sharply.

SLC NAND, for example, offers relatively low capacities, though it can withstand approximately 100,000 write/erases per block. MLC NAND (two-bit) offers 1,000 to 3,000 cycles in high-capacity applications and 5,000 to 10,000 cycles in medium-capacity applications, while TLC NAND (three-bit) offers approximately 1,000 cycles.

3D NAND offers higher write/erase cycles, with 3D MLC NAND rated between 6,000 to 40,000 cycles, 3D TLC NAND rated between 1,000 to 3,000 cycles, and 3D QLC NAND (four-bit) rated from 100 to 1,000 cycles.

Manufacturers measure the lifespan of SSDs in terms of “total bytes written,” or TBW. A 2 TB Intel 660p SSD, which uses 3D QLC NAND, is rated for 400 TB. In contrast, a 2 TB Intel 760p SSD, which uses 3D TLC NAND, is rated for 1152 TB.

How is flash storage different from a solid-state drive (SSD)?

SSDs utilize NAND flash technology for mass storage, though this is not the only component used in an SSD. Typically, SSDs are comprised of a disk controller, a DRAM cache, and NAND flash. The disk controller is used to manage the efficient use of NAND, such as preventing uneven wear levels of flash cells, extending the lifespan of a drive.

What portable flash storage form factors exist?

Different form factors of flash storage cards exist to accommodate the variety of devices that the cards are used with. While most consumer-level devices now center around Secure Digital (SD), prosumer and professional devices require higher speeds than SD is readily able to provide.

Secure Digital (SD)

Secure Digital (SD) is the industry standard flash memory card; it was introduced in 1999, and it is used in practically every category of consumer electronic devices since its inception.

The original format, Secure Digital Standard Capacity (SDSC), was nominally limited to 2 GB. In 2006, the Secure Digital High Capacity (SDHC) format was introduced to allow for cards up to 32 GB. The Secure Digital eXtended Capacity (SDXC) format was introduced in 2009 to allow for cards up to 2 TB. In June 2018, the Secure Digital Ultra Capacity (SDUC) format was introduced to allow for cards greater than 2 TB; the specification permits cards up to 128 TB.

Presently, the largest commercially available SD card is 1 TB.

Devices that support SD cards are essentially backward compatible, with some limited forward compatibility. Devices that claim only support for SDHC are often able to use larger SDXC cards by reformatting the card (using a computer) to use the FAT32 file system. For a device to be sold with a SDXC or a SDUC logo, it must support the patent-encumbered exFAT file system.

In order to accommodate the shrinking sizes of consumer electronics, smaller variants of SD cards were developed. The miniSD standard was introduced in 2003, though enjoyed only limited support before being discontinued after about five years.

microSD, formerly called TransFlash, was introduced in 2005. microSD cards are commonly found in Android smartphones and tablets, game systems like the Nintendo Switch, and single-board computers (SBCs) like the Raspberry Pi. microSD cards can be used in legacy devices that support only miniSD or full-size SD cards using passive adapters.

CompactFlash (CF)

CompactFlash was introduced in 1994 and is designed as a 50-pin subset of the 68-pin PCMCIA standard. Despite the age of the standard, it is still used on professional DSLR cameras from Canon and Nikon.

The largest capacity currently available is 512 GB.

In terms of signaling, CompactFlash is similar to ATA hard disks. This was leveraged by the IBM Microdrive, which implemented a miniature hard disk on the CompactFlash interface, with capacities up to 8 GB. Microdrive and similar competitors have long since been discontinued.

CFexpress

CFexpress, announced in September 2016, is the successor to CompactFlash. It is used exclusively in professional-grade cameras. CFexpress is based on the PCI Express 3.0 interface, and leverages NVM Express to provide low latency and processing overhead. 1 TB CFexpress cards were previewed at CES 2019 and are expected to be available in retail channels this year.

Version 1.0 of the specification provides a PCIe 3.0 x2 connection for CFexpress cards, for a maximum speed of 1.96 GB/s.

Though the specification permits different form factors, CFexpress cards inherit the XQD form factor, which was developed by Sony, Nikon, and SanDisk, though the latter declined to actually produce any XQD cards. Because of IP licensing encumbrances, XQD cards not produced by Sony were limited, in turn limiting widespread adoption. Sony has pledged to manufacture CFexpress cards, creating an industry consensus around CFexpress.

Universal Flash Storage (UFS) card

The UFS card standard was published in March 2016, and was intended to replace microSD cards. Though the standard was updated in January 2018, no UFS cards are commercially available, and as of March 2019, no devices are marketed as supporting interoperability with the format.

USB flash drives

USB flash drives encompass a wide variety of similar products. On the lower end, these drives typically include a simple controller for wear-leveling and USB connectivity; higher-end USB flash drives include DRAM for write caching. Typically, these drives are available in the “gumstick” form factor, with the USB port attached to the drive, leading to popularly being called USB sticks.

Early USB flash drives typically included a write-protect switch. This feature has been relegated to the specialty market, though manufacturers such as Kanguru produce USB 3.0 flash drives with write-protect switches.

Drives with larger capacities do exist, such as the Samsung T5 SSD, which requires the use of an external cable to connect to a computer. The Samsung T5 SSD is technically a USB-connected portable SSD that combines an mSATA SSD with a SATA to USB3 bridge in a rugged enclosure.

Legacy form factors

SmartMedia cards were a Toshiba format launched in 1995 and used for digital cameras and to a lesser extent in PDAs and MP3 players. The largest SmartMedia card was 128 MB and was produced in (older) 5V and (newer) 3.3V variants. The digital camera industry transitioned away from SmartMedia in 2003.

xD-Picture Cards were used in Fujifilm and Olympus digital cameras from 2002 to 2009, in capacities from 16 MB to 2 GB. xD and SmartMedia cards are functionally raw NAND flash chips without any disk controller.

Memory Stick was a flash memory format developed by Sony in 1998. Though it was licensed to other companies, Memory Stick was limited primarily to Sony electronics from 1998 to 2012. Sony made the majority of Memory Stick products, though the format was also produced by SanDisk and Lexar. Seven Memory Stick form factors were introduced, with capacities from 128 MB to 32 GB. The PlayStation Vita used a proprietary Memory Stick form factor, produced only by Sony. The PlayStation Vita was discontinued in March 2019.

What solid-state drive (SSD) form factors exist?

Compared to a traditional platter hard drive—which requires the use of a drive head to be moved by an actuator across a platter to read and write data—NAND flash storage has no moving parts, making these solid-state drives.

Solid-state drives used in laptops, desktops, and servers were originally engineered as drop-in replacements for traditional hard drives; as a result, early SSDs are somewhat shoehorned to work in existing standards.

2.5″ SATA drives

Early consumer-facing solid-state drives most often took the form of 2.5″ disk drives, the same form factor used in small form factor (SFF) PCs and notebook computers, connected over SATA. Because of the limitations of SATA and AHCI, the maximum speed a SATA-linked drive can achieve is approximately 550 MB/sec.

Early 2.5″ drives commonly filled the entire 2.5″ space internally and used the standard 9.5mm height of traditional drives. Newer SATA drives, like the Crucial MX500, use only part of the 2.5″ space internally and have a 7mm enclosure, with a spacer included for use in devices (typically, notebooks) that require the extra 2.5mm to be filled.

SSDs in a 3.5″ form factor are exceedingly rare. The OCZ Colossus LT in 2010 utilized the extra space to reach 1 TB, at a premium price of $4,000. Newer 3.5″ SSDs like the Nimbus Data 100 TB ExaDrive are built exclusively for enterprise applications.

mSATA

Notebooks and SFF PCs were often built with support for mSATA cards, which provide the same signaling as 2.5″ SATA drives in a 30 x 50.95 mm package. Computer manufacturers largely abandoned the format in 2015 in favor of M.2 drives.

Samsung continued to support the format into early 2018 with the 860 EVO, the last Samsung SSD to be released for the form factor, in capacities up to 1 TB.

M.2

M.2, formerly Next-Generation Form Factor (NGFF), is a versatile form factor used for high-performance SSDs, as well as other peripherals including Wi-Fi and Bluetooth networking cards, WWAN (4G LTE, 5G modems), and other devices. M.2 SSDs can use legacy SATA connections or PCI Express with AHCI or NVMe.

M.2 SSDs are 22 mm wide and are available in lengths of 30, 42, 60, 80, and 110 mm—42 and 80 mm are the most common. Sizes are typically denoted in the format M.2-WWLL such as M.2-2280.

M.2 is becoming the industry default for SSDs and are used extensively in notebook and SFF PCs. On desktops, enthusiast PC motherboards typically include one or more M.2 PCIe x4 slots.

Presently, the largest capacity M.2-2280 SSD is 2 TB.

NGSFF, formerly NF1, is a hot-swappable version of M.2 devised by Samsung and intended for data centers that require the ability to swap SSDs on running systems.

U.2

U.2, formerly SFF-8639, is technically an interface—essentially, it provides the PCIe x4 lane of M.2 for the same physical form factor as SATA drives. In theory, it would also be possible to use the interface for 3.5″ drives.

PCI Express drives

SSDs that plug directly into PCIe slots on motherboards exist, though the format has fallen into disuse. By late 2010, mainstream consumer SSDs were bottlenecked by the comparatively limited throughput granted by SATA, prompting early PCI Express drives to be produced. These drives used AHCI rather than NVMe, providing higher potential read/write speeds but not appreciably better performance in terms of random I/O.

Enterprise-targeted PCI Express drives exist, such as the Intel SSD DC P3608, though this drive has since been discontinued. Intel continues to offer the mid-2015 SSD 750 series as PCI Express cards, though these are slower than newer M.2 options.

For a single drive, PCI Express to M.2 adapter boards are readily available. In terms of electrical signaling, M.2 provides a full PCIe x4 connection, making the conversion essentially passive—no performance loss is incurred through the use of an adapter. Some manufacturers offer these adapters with their SSDs often at a nominal additional cost. PCIe to M.2 adapters are manufacturer agnostic—an aftermarket adapter can be used for any PCIe M.2 drive.

Disk-on-Module (DOM)

DOM SSDs are intended to replace Parallel ATA (PATA) disks used in legacy equipment. Production of PATA HDDs ended more than a decade ago, and continued reliance on these disks invites the potential for data loss. DOM SSDs plug directly into a PATA port, allowing airflow obstructing IDE cables to be removed.

How can I tell if a flash storage card or SSD is fast enough for my use case?

Different standards and speed classes exist for different flash memory cards, though the number of differing rating systems for different use cases can become confusing.

Bus interface standards for SD cards

The bus interface standard determines how an SD card connects to a host device, giving a peak potential performance, but it is not indicative of how individual cards perform. Prior to the introduction of Ultra High Speed (UHS) classes, SD cards were limited to either 12.5 MB/s or 25 MB/s, though no standardized marking exists to indicate the difference.

  • UHS-I offers 50 MB/s in half- or full-duplex mode, or 104 MB/s in half-duplex mode.
  • UHS-II offers 156 MB/s in full-duplex mode, or 312 MB/s in half-duplex mode.
  • UHS-III offers 312 MB/s or 624 MB/s in full-duplex mode. It has no half-duplex mode.

SD Express (also marked SD-Express I) offers 985 MB/s as a PCIe 3.0 x1 lane, with NVMe support.

Per specifications, UHS and SD Express cards must be backward-compatible, though the maximum speed possible is determined by the rating of the host device.

Speed class ratings for SD cards

Overlapping and conflicting standards for SD cards for different use cases often causes confusion when selecting an SD card. Here’s what the different speed class ratings mean in real-world terms.

Min. sequential write speedSpeed ClassUHS Speed ClassVideo Speed Class
2 MB/sClass 2 (C2)
4 MB/sClass 4 (C4)
6 MB/sClass 6 (C6)Class 6 (V6)
10 MB/sClass 10 (C10)Class 1 (U1)Class 10 (V10)
30 MB/sClass 3 (U3)Class 30 (V30)
60 MB/sClass 60 (V60)
90 MB/sClass 90 (V90)

Nominally, shooting video in 4K requires at least a V6-rated card, while 8K requires a V30-rated card. Requirements vary between devices.

Application Performance Class primarily measures IOPS, with the intent of providing a standard for running apps from smartphones.

ClassMin. random readMin. random writeMin. sequential write speed
A11500 IOPS500 IOPS10 MB/s
A24000 IOPS2000 IOPS10 MB/s

Speed ratings for SSDs

With the use of DRAM caches, it is possible for SSDs, particularly SATA SSDs, to saturate the available bandwidth of the bus they are connected to. These speed ratings are useful only for sequential read/write, as other factors explained in the section below influence how SSDs handle bulk operations.

Connection typeMax. speedWorking state
SATA 2.x (3 Gb/s)300 MB/sLegacy
SATA 3.x (6 Gb/s)600 MB/sCurrent
PCIe 3.0 x1985 MB/sCurrent
PCIe 3.0 x21970 MB/sCurrent
PCIe 3.0 x43940 MB/sCurrent

How do I select the best flash storage for my device?

In part, this is subjective. If you want to carry a FLAC-encoded music collection on your smartphone, a larger card with slower write speeds is adequate for your use case. For running applications stored on a flash card, or for use in a game system like the Nintendo Switch, an A2-rated card would theoretically provide higher performance.

Purchasing a genuine, name-brand product is important. For SD cards, off-brand or knockoff products sold by disreputable sellers incorrectly report their size. In practice, this means a card purported to be 128 GB may only have 16 GB storage. A device may write beyond that 16 GB capacity, but newly-written data would be lost or overwrite existing data.

For SSDs, consider the number of writes needed for a task. QLC SSDs offer a higher storage capacity but can only endure a limited number of writes. PCIe SSDs are always higher performance than SATA SSDs. For desktops, always use an M.2 or M.2 to PCIe bridge for your SSD, if you have PCIe connections available. For laptops, devices with M.2 capabilities typically ship with a M.2 drive integrated, limiting upgradability options.

It is possible to purchase SSDs at a storage capacity equivalent to that of traditional platter-based hard disk drives, though the cost per gigabyte is significantly higher—presently, a 15.36 TB Seagate Nytro SSD is over $6,000, though a 14 TB Seagate IronWolf Pro is $570. This effect is magnified when building enterprise storage arrays. In many use cases, all-flash arrays are inefficient from a cost perspective, though tiered storage solutions may add unnecessary complexity.

 

 



Source link