Flash storage is any type of drive, repository or system that uses flash memory to keep data for an extended period of time. Flash memory is ubiquitous in small computing devices and common today in large business storage systems. The size and complexity of flash-based storage varies in devices ranging from portable USB flash drives, smartphones, cameras and embedded systems to enterprise-class all-flash arrays (AFAs). Flash is packaged in a variety of formats for different storage purposes.
Flash storage's memory is a form of electrically erasable programmable read-only memory, although flash differs from conventional EEPROM in the way that it erases data. Flash erases whole blocks of data at a time, rather than on a bit-by-bit level, as conventional EEPROM does. A block stored on a flash memory chip must be erased before new data can be written to the microchip. Unlike dynamic RAM (DRAM), standard EEPROM and flash are nonvolatile memory (NVM). This means they do not require power to preserve stored data with integrity, so a system can be turned off -- or lose power -- without losing data.
Flash is solid-state storage (SSS) and stores data using a charge on a capacitor to represent a bit. It is most often packaged in surface-mounted chips attached to a printed circuit board. There are no moving mechanical parts involved, which reduces power consumption. A typical serial Advanced Technology Attachment (SATA) flash drive consumes 50% or less of the power required by mechanical SATA hard disk drives (HDDs) and may be capable of sequential read speeds of more than 500 MB per second in consumer drives -- faster than even the fastest enterprise-class mechanical HDDs. That is only a part of the picture because access times are where flash shines. Flash drives have no mechanical limitation for file access, which enables access times in microseconds, rather than the millisecond seek times required by mechanical HDDs -- which is several orders of magnitude less in latency.
Most flash storage systems are composed of memory chips and a flash controller. The memory chips store data, while the controller manages access to the storage space on the memory unit. The flash controller is often multichannel, working with a RAM cache. The cache buffers the data going to and from a number of chips, which enhances speed.
Flash memory is in wide use in consumer devices. Smartphones and MP3 players have abandoned the mechanical HDD; flash provides advantages in compactness and power consumption. In notebook computers, flash storage offers the additional boon of being more resistant to the high gravitational acceleration bumps and drops these devices often receive in their mobile lives. This rugged nature enables the drives to maintain function through these events, which protects data. Flash is more prevalent in notebooks than desktop computers.
Flash is also the standard form of storage in digital cameras, tablets and digital camcorders. Photolithographic shrinks and the development of denser types of flash have enabled an increase in capacity, making flash suitable for miniaturized applications.
Flash storage adoption continues to grow in enterprise storage systems. Initial deployments focused on the acceleration of input/output (I/O)-intensive applications, such as databases and virtual desktop infrastructures (VDIs). Use cases have since expanded to general enterprise workloads and mission-critical applications as the cost of flash has dropped and businesses have attempted to take advantage of its performance and low-latency benefits.
The history of flash storage
Dr. Fujio Masuoka is credited with inventing NOR and NAND flash, the two main types of flash memory, while he worked for Toshiba in the 1980s. In comparison to the slow process used by EEPROM, the new format's ability to be programmed and erased in large blocks reminded a colleague of Masuoka's of a camera flash. NOR and NAND are named for the way the floating gates of the memory cells that hold data are interconnected in configurations that somewhat resemble a NOR or a NAND logic gate.
Intel's interest was piqued as NOR flash served as a higher-functioning replacement for the EEPROM chips the company was shipping at that time. The company released the first NOR flash chips in 1988. Toshiba followed with the first NAND flash chips in 1989.
By the mid-2000s, it looked like NAND flash would hit a hard scaling limit. That's because the photolithographic processes used to shrink transistors would no longer suffice to continue the price declines and performance improvements the industry and its customers had become accustomed to. In 2006, Toshiba developed a new process called Bit Cost Scaling (BiCS) to overcome these issues.
Instead of continuing to try to shrink transistors, BiCS enabled manufacturers to greatly increase the number of transistors on a chip by building them vertically rather than horizontally, as is done with standard planar NAND technology. 3D NAND is backward-compatible with planar NAND, so any devices that support the latter can read and write data to the former and vice versa.
Major manufacturers of NAND flash memory chips include Intel, Micron Technology, Samsung, SK Hynix, Toshiba and Western Digital's SanDisk division. Major manufacturers of NOR flash memory include Cypress Semiconductor, Macronix, Microchip Technology, Micron Technology and Winbond.
A nonvolatile solid-state drive (SSD) uses solid-state flash memory to persistently store data. Its main components include NAND flash memory chips and a flash controller. The SSD controller is designed and optimized to provide for high read and write performance for both random and sequential data requests. Manufacturers achieve varying densities and capacities in SSDs by stacking chips in a grid.
Every flash memory cell used in an SSD has a storage transistor, called a floating gate transistor (FGT), with a floating gate and a control gate separated by a thin oxide layer to control the flow of electrical currents. It is the FGT that prevents volatility, as it enables SSDs to retain stored data even when they are not connected to a power source.
Every FGT in an SSD contains a single bit of data. If a bit is designated as a 1, it is a charged cell. If a bit is designated a 0, that means the cell lacks an electrical charge. Adding (charging) or removing (discharging) electrons from the floating gate is how flash memory works.
Adding, or trapping, electrons in the floating gate is done through one of two processes: Fowler-Nordheim tunneling or channel hot electron injection.
EMC, now Dell EMC, is attributed with being the first vendor to integrate SSDs in an enterprise storage product, with its 2008 Symmetrix disk arrays. Apple's 2005 iPod was the first notable use of an SSD in a consumer device.
Flash storage vs. traditional HDDs
NAND flash storage offers advantages over traditional HDDs. HDDs carry a lower cost per stored data bit, but flash drives can provide significantly higher performance, lower latency and reduced power consumption. Their compact size also makes flash suitable for small consumer devices.
In enterprise systems, flash can enable a business to consolidate storage and lower the total cost of ownership (TCO). Fewer SSDs are needed to process transactions and to deliver a comparable level of performance to systems using slower HDDs. Enterprises, in turn, can realize savings on rack space, system management, maintenance, and power and cooling costs. Data reduction technologies, such as inline deduplication and compression in all-flash storage systems, also enable businesses to reduce their data footprints.
As interest in flash storage has grown, industry watchers have noted a frequently overlooked caveat with flash. While speed and random read access is far superior in flash than in traditional hard drives, longevity may be reduced in heavy use with high write workloads. This reduction in endurance is due to flash's relatively limited tolerance for write-erase cycles. Manufacturers use features such as wear leveling and DRAM/nonvolatile RAM caching to provide flash storage with better performance, while reducing flash SSD write wear to improve reliability.
Unlike HDDs, which are constrained by their movable parts, SSDs come in a variety of form factors. Available at multiple heights with support for SATA, Serial-Attached SCSI (SAS) and nonvolatile memory express (NVMe) protocols, the 2.5-inch SSD is the most common type of SSD. It falls into the traditional HDD form factor type of SSD -- as it fits into the same SAS and SATA slots in a server -- as identified by the Storage Networking Industry Association's Solid State Storage Initiative.
Other major types of SSDs include solid-state cards and solid-state modules. The former comes in the form of a standard add-in card -- such as those that, for example, use a Peripheral Component Interconnect Express (PCIe) serial port card on a printed circuit board -- and bypasses host bus adapters (HBAs) to speed up storage performance. A U.2 SSD is an example of such a device. The latter type of SSD, also known as nonvolatile dual inline memory module (NVDIMM) cards, uses a DIMM or small outline dual inline memory module (SO-DIMM) format.
Flash storage formats
NOR flash offers memory addressing on a byte scale, enabling true, random access and good read speeds. It was this addressability that interested Intel in NOR, because the technology matched the requirements for basic input/output system (BIOS) and Extensible Firmware Interface (EFI) applications. NOR flash is more expensive per gigabyte than NAND because of its larger, individual cell size. NOR has slower write and erase times than NAND as well.
Both NAND and NOR use quantum tunneling of electrons to move electrons through the dielectric insulating material of the cell wall, which degrades the material over time. NOR flash is erasable, which makes it a great replacement for EEPROM- or ROM-based firmware BIOS and EFI chips where addressability and read speed is a boon, while the rewrite durability is less of a concern.
NAND offers greater write speeds than NOR flash, along with a lower cost per gigabyte. The lower cost is a result of the NAND memory cell's string design, saving die space and reducing the overall size of a chip per gigabyte. NAND can come in single-level cell (SLC) and multi-level cell (MLC) forms, which include enterprise MLC (eMLC) and triple-level cell (TLC). SLC stores a single bit of information per cell. SLC generally offers greater speeds -- especially when it comes to writes -- greater longevity and fewer bit errors. MLC provides storage capacity for more data, as its cell is capable of more levels of charge (or states), which enable it to store multiple bits of data per cell. MLC can double capacity over SLC; TLC provides a third bit. The extra levels of charge, along with smarter flash controllers and firmware, can enable bit error correction as well.
Flash storage interfaces
Flash storage for computer memory comes in a variety of interfaces, including USB, SAS, SATA, M.2 and PCIe. USB 3.1 Gen 2, known as SuperSpeed USB 10 Gbps, became available in 2013 and sees general use in flash drives, enclosures and mobile devices.
SATA is a common format in desktops and notebook computers, and the 6 Gb version can eliminate bandwidth bottlenecks. Volume shipments of 12 Gbps SAS began in 2015. SAS-based SSDs are in wide use in enterprise storage systems.
PCIe-connected flash storage provides sufficient bandwidth to enable future expansion and represents the extreme end of speed-demanding offerings.
NVMe technology in use with PCIe-based SSDs further reduces latency, increases input/output operations per second (IOPS) and lowers power consumption through the streamlining of the I/O stack.
Flash in the data center
Data centers with I/O-intensive applications, such as high-transaction rate databases and credit card processing systems, are increasingly turning to flash storage as an efficient and cost-effective way to increase throughput without having to add more servers.
Major storage system manufacturers offer all-flash systems and hybrid arrays, which are equipped with SSDs and HDDs. Numerous all-flash storage specialists have also emerged to challenge the incumbents. Servers equipped with flash storage are also increasingly common and can further reduce latency.
Data center managers looking for ways to address the energy drain represented by HDDs are examining flash storage as a way to achieve green computing or green data center benchmarks. Flash SSDs provide high bandwidth at much lower power consumption than HDDs, making them a good choice for this application.
The future of the flash storage market
Recent advances in flash storage include the growing number of storage array vendors that have added support for NVMe to their products to greatly speed up the transfer of data between flash storage and servers. NVMe exploits the high-performance PCIe bus to enable applications and flash storage to directly communicate.
Emerging and growing in popularity alongside NVMe and PCIe-connected SSDs are NVDIMMs. These devices, which integrate NAND flash with DRAM and a power supply, plug directly into a standard DIMM slot on a memory bus. The flash storage on NVDIMM cards is used to back up and restore data in DRAM, the power supply to maintain nonvolatility.
There's also 3D XPoint memory developed by Intel and Micron. 3D XPoint isn't supposed to replace DRAM or NAND flash but rather serve as a middle memory layer to enhance the cost performance of computers. So far, Intel has released the only 3D XPoint product, its Optane SSDs using the PCIe NVMe interface. While 3D XPoint promises 1,000 times better performance than NAND flash, Intel Optane is only six to eight times speedier. A DIMM interface is required to maximize 3D XPoint performance, but no such product has been released yet -- although Intel announced in August 2018 it had shipped some of these DIMMs to Google.
Other types of memory -- such as phase-change memory (PCM), which goes back to 1970 -- have emerged as possible NAND flash replacements. Some examples of these new memory types include:
- magnetoresistive RAM (MRAM)
- ferroelectric memory (FRAM)
- conductive-bridging memory (CBRAM)
- resistive RAM (ReRAM)
- oxide-based resistive RAM (OxRRAM)
- Nano-RAM (NRAM)
These technologies aim to deliver high performance, nonvolatility and low power. They also promise to move the flash storage industry beyond DRAM and NAND flash scaling limits.
Flash for hobbyists
Many enthusiasts have adopted flash storage. These users often have their operating system (OS), a few games and data-intensive applications, such as audio, video and image editing software, on a flash drive or even a flash RAID array.
Main storage of files that don't require speed could be stored on a cost-effective and potentially massive conventional hard drive. Other enthusiasts may use all-flash, depending on whether they are excited about flash itself or its attributes, including silence, speed and the elimination of bottlenecks.