Definition

flash memory

Contributor(s): Garry Kranz, Carol Sliwa, Steve Collins and Julian de Silva

Flash memory, also known as flash storage, is a type of nonvolatile memory that erases data in units called blocks. A block stored on a flash memory chip must be erased before data can be written or programmed to the microchip. Flash memory retains data for an extended period of time, regardless of whether a flash-equipped device is powered on or off.

Flash memory is used in enterprise server, storage and networking technology, as well as in a wide range of consumer devices, including USB flash drives, mobile phones, digital cameras, tablet computers, PC cards in notebook computers and embedded controllers. For instance, NAND flash-based solid-state drives are often used to accelerate the performance of I/O-intensive applications. NOR flash memory is often used to hold control code, such as the basic input/output system (BIOS), in a PC.

Flash memory is seeing growing use for in-memory computing to help speed performance and increase the scalability of systems that manage and analyze increasingly large sets of data.

Origins of flash storage technologies

Dr. Fujio Masuoka is credited with the invention of flash memory when he worked for Toshiba in the 1980s. Masuoka's colleague, Shoji Ariizumi, reportedly coined the term flash because the process of erasing all the data from a semiconductor chip reminded him of the flash of a camera.

Flash memory evolved from erasable programmable read-only memory (EPROM) and electrically erasable programmable read-only memory (EEPROM). Flash is technically a variant of EEPROM, but the industry reserves the term EEPROM for byte-level erasable memory and applies the term flash memory to larger block-level erasable memory.

Devices using flash memory erase data at the block level and rewrite data at the byte level -- NOR flash -- or multiple-byte page level -- NAND flash. Flash memory is widely used for storage and data transfer in consumer devices, enterprise systems and industrial applications.

Flash memory module
Flash memory consists of a transistor and a floating gate that stores the electric current.

How does flash memory work?

Flash memory architecture includes a memory array stacked with a large number of flash cells. A basic flash memory cell consists of a storage transistor with a control gate and a floating gate, which is insulated from the rest of the transistor by a thin dielectric material or oxide layer. The floating gate stores the electrical charge and controls the flow of the electrical current. 

Electrons are added to or removed from the floating gate to change the storage transistor's threshold voltage. Changing the voltage affects whether a cell is programmed as a zero or a one.

Floating gate flash cell

A process called Fowler-Nordheim tunneling removes electrons from the floating gate. Either Fowler-Nordheim tunneling or a phenomenon known as channel hot-electron injection traps the electrons in the floating gate.

With Fowler-Nordheim tunneling, data is erased via a strong negative charge on the control gate. This forces electrons into the channel, where a strong positive charge exists.

The reverse happens when using Fowler-Nordheim tunneling to trap electrons in the floating gate. Electrons manage to forge through the thin oxide layer to the floating gate in the presence of a high electric field, with a strong negative charge on the cell's source and the drain and a strong positive charge on the control gate.

Fowler-Nordheim tunneling diagram

Channel hot-electron injection, also known as hot-carrier injection, enables electrons to break through the gate oxide and change the threshold voltage of the floating gate. This breakthrough occurs when electrons acquire a sufficient amount of energy from the high current in the channel and the attracting charge on the control gate.

Channel hot-electron injection diagram

Electrons are trapped in the floating gate whether or not a device containing the flash memory cell is receiving power as a result of electrical isolation created by the oxide layer. This characteristic enables flash memory to provide persistent storage.

EPROM and EEPROM cells operate similarly to flash memory in how data is written, or programmed, but differ from flash memory in how data is erased. An EPROM is erased by removing the chip from the system and exposing the array to ultraviolet light. An EEPROM erases data electronically at the byte level, while flash memory erases data electronically at the block level.

NOR vs. NAND flash memory

There are two types of flash memory: NOR and NAND.

NOR and NAND flash memory differ in architecture and design characteristics. NOR flash uses no shared components and can connect individual memory cells in parallel, enabling random access to data. A NAND flash cell is more compact and has fewer bit lines, stringing together floating gate transistors to increase storage density.

NAND is better suited to serial rather than random data access. NAND flash process geometries were developed in response to planar NAND reaching its practical scaling limit.

NOR flash is fast on data reads, but it is typically slower than NAND on erases and writes. NOR flash programs data at the byte level. NAND flash programs data in pages, which are larger than bytes, but smaller than blocks. For instance, a page might be 4 kilobytes (KB), while a block might be 128 KB to 256 KB or megabytes in size. NAND flash consumes less power than NOR flash for write-intensive applications.

NOR flash is more expensive to produce than NAND flash and tends to be used primarily in consumer and embedded devices for boot purposes and read-only applications for code storage. NAND flash is more suitable for data storage in consumer devices and enterprise server and storage systems due to its lower cost per bit to store data, greater density, and higher programming and erase (P/E) speeds.

Devices, such as camera phones, may use both NOR and NAND flash, in addition to other memory technologies, to facilitate code execution and data storage.

Flash memory form factors

Flash-based media is based on a silicon substrate. Also known as solid-state devices, they are widely used in both consumer electronics and enterprise data storage systems.

There are three SSD form factors that have been identified by the Solid State Storage Initiative:

  • SSDs that fit into the same slots used by traditional electromechanical hard disk drives (HDDs). SSDs have architecture similar to that of an integrated circuit.
  • Solid-state cards that reside on a printed circuit board and use a standard card form factor, such as Peripheral Component Interconnect Express (PCIe).
  • Solid-state modules that fit in a dual inline memory module (DIMM) or small outline dual inline memory module using a standard HDD interface, such as the Serial Advanced Technology Attachment (SATA).

An additional subcategory is a hybrid hard drive that combines a conventional HDD with a NAND flash module. A hybrid hard drive is generally viewed as a way to bridge the divide between rotating media and flash memory.

All-flash and hybrid flash memory

The advent of flash memory fueled the rise of all-flash arrays. These systems contain only SSDs. They offer advantages in performance, and sometimes reduced operational costs, compared to all disk-based storage arrays. The chief difference, aside from the media, is in the underlying physical architecture used to write data to a storage device.

HDD-based arrays have an actuator arm that enables data to be written to a specific block on a specific sector on the disk. All-flash storage systems do not require moving parts to write data. The writes are made directly to the flash memory, and custom software handles data management.

A hybrid flash array blends disk and SSDs. Hybrid arrays use SSDs as a cache to speed access to frequently requested hot data, which subsequently is rewritten to back-end disk. Many enterprises commonly archive data from disk as it ages by replicating it to an external magnetic tape library.

Flash plus tape, also known as flape, describes a type of tiered storage in which primary data in flash is simultaneously written to a linear tape system.

In addition to flash memory arrays, the ability to insert SSDs in x86-based servers has increased the technology's popularity. This arrangement is known as server-side flash memory and it enables companies to sidestep the vendor lock-in associated with purchasing expensive and integrated flash storage arrays.

The drawback of placing flash in a server is that customers need to build the hardware system internally, including the purchase and installation of a storage management software stack from a third-party vendor.

Pros and cons of flash memory

Flash is the least expensive form of semiconductor memory. Unlike dynamic random access memory (DRAM) and static RAM (SRAM), flash memory is nonvolatile, offers lower power consumption and can be erased in large blocks. Also, on the plus side, NOR flash offers fast random reads, while NAND flash is fast with serial reads and writes.

An SSD with NAND flash memory chips delivers significantly higher performance than traditional magnetic media, such as HDDs and tape. Flash drives also consume less power and produce less heat than HDDs. Enterprise storage systems equipped with flash drives are capable of low latency, which is measured in microseconds or milliseconds.

The main disadvantages of flash memory are the wear-out mechanism and cell-to-cell interference as the dies get smaller. Bits can fail with excessively high numbers of program/erase cycles, which eventually break down the oxide layer that traps electrons. The deterioration can distort the manufacturer-set threshold value at which a charge is determined to be a zero or a one. Electrons may escape and get stuck in the oxide insulation layer, leading to errors and bit rot.

Anecdotal evidence suggests NAND flash drives are not wearing out to the degree once feared. Flash drive manufacturers have improved endurance and reliability through error correction code algorithms, wear leveling and other technologies.

In addition, SSDs do not wear out without warning. They typically alert users in the same way a sensor might indicate an underinflated tire.

NAND flash memory storage types

NAND flash semiconductor manufacturers have developed different types of memory suitable for a wide range of data storage uses cases. The following chart explains the various NAND flash types.

Types of NAND flash memory storage

Description

Advantages

Disadvantages

Primary use

Single-level cell (SLC)

Stores one bit per cell and two levels of charge.

Higher performance, endurance and reliability than other types of NAND flash.

Higher cost than other types of NAND flash.

Enterprise storage, mission-critical applications.

Multi-level cell (MLC)

Can store multiple bits per cell and multiple levels of charge. The term MLC equates to two bits per cell.

Cheaper than SLC and enterprise MLC (eMLC), high density.

Lower endurance than SLC and eMLC, slower than SLC.

Consumer devices, enterprise storage.

Enterprise MLC (eMLC)

Typically stores two bits per cell and multiple levels of charge; uses special algorithms to extend write endurance.

Less expensive than SLC flash, greater endurance than MLC flash.

More expensive than MLC, slower than SLC.

Enterprise applications with high write workloads.

Triple-level cell (TLC)

Stores three bits per cell and multiple levels of charge. Also referred to as MLC-3, X3 or 3-bit MLC.

Lower cost and higher density than MLC and SLC.

Lower performance and endurance than MLC and SLC.

Mass storage consumer applications, such as USB drives, flash memory cards, smartphones, and client SSDs, and data center SSDs for read-intensive workloads.

Vertical/3D NAND

Stacks memory cells on top of each other in three dimensions vs. traditional planar NAND technology.

Higher density, higher write performance and lower cost per bit vs. planar NAND.

Higher manufacturing cost than planar NAND; difficulty in manufacturing using production planar NAND processes; potentially lower data retention.

Consumer and enterprise storage.

*Quad-level cell (QLC)

Uses a 64-layer architecture that is considered the next iteration of 3D NAND. Not widely available as of November 2017.

Stores four bits of data per NAND cell, potentially boosting SSD densities.

More data bits per cell can affect endurance; increased costs of engineering.

Mostly write once, read many (WORM) use cases.

Note: NAND flash wear-out is less of a problem in SLC flash than it is in less expensive types of flash, such as MLC and TLC, for which the manufacturers may set multiple threshold values for a charge. The commonly cited industry wear-out figures are 100,000 program/erase -- write/erase -- cycles for SLC NAND flash; 30,000 for eMLC; 10,000 or fewer for MLC; and 3,000 or fewer for TLC. Actual endurance figures may be higher.

* Samsung and Western Digital have disclosed or previewed preliminary designs for their respective QLC architectures.

Vendor breakdown of enterprise NAND flash memory products

Major manufacturers of NAND flash memory chips include Intel Corp., Micron Technology Inc., Samsung Group, SanDisk Corp. -- now owned by Western Digital Corp. -- SK Hynix Inc. and Toshiba Memory Corp.

Sampling of NAND flash memory vendors

A NAND flash shortage is causing disruption in the market. The shortfall is causing SSD prices to rise and lead times to lengthen. The demand outstrips supply largely due to soaring demand from smartphone makers.

Other turmoil is exerting an impact on the market. As of November 2017, leading flash supplier Toshiba agreed to sell its chip making unit to a group of corporate and institutional investors led by Bain Capital. Toshiba sold the flash business as part of its effort to cover financial losses and to avoid being delisted on the Tokyo Stock Exchange.

However, Toshiba partner Western Digital (WD) has vowed to fight the $18 billion deal. WD inherited a stake in Toshiba's chip business following its acquisition of SanDisk Corp., which had a joint venture with Toshiba.

NOR flash memory types

The two main types of NOR flash memory are parallel and serial, also known as serial peripheral interface. NOR flash was originally available only with a parallel interface. Parallel NOR offers high performance, security and additional features; its primary uses include industrial, automotive, networking, and telecom systems and equipment.

NOR cells are connected in parallel for random access. The configuration is geared for random reads associated with microprocessor instructions and to execute codes used in portable electronic devices, almost exclusively of the consumer variety.

Serial NOR flash has a lower pin count and smaller packaging, making it less expensive than parallel NOR. Use cases for serial NOR include personal and ultra-thin computers, servers, HDDs, printers, digital cameras, modems and routers.

Leading NOR vendor products

Major manufacturers of NOR flash memory include Cypress Semiconductor Corp. -- through its acquisition of Spansion Inc. -- Macronix International Co. Ltd., Microchip Technology Inc., Micron Technology Inc. and Winbond Electronics Corp.

Cypress Semiconductor acquired NOR flash provider Spansion in 2015. The Cypress NOR portfolio includes FL-L, FL-S, FS-S and FL1-K products.

Macronix OctaFlash uses multiple banks to enable write access to one bank and read from another. Macronix MX25R Serial NOR is a low-power version that targets internet of things (IoT) applications.

Microchip NOR is branded as Serial SPI Flash and Serial Quad I/O Flash. The vendor's parallel NOR products include the Multi-Purpose Flash devices and Advanced Multi-Purpose Flash devices families.

Micron sells Serial NOR Flash and Parallel NOR Flash, as well as Micron Xccela high-performance NOR flash for automotive and IoT applications.

The Winbond serial NOR product line is branded as SpiFlash Memories and includes the W25X and W25Q SpiFlash Multi-I/O Memories. In 2017, Winbond expanded its line of Secure Flash NOR for additional uses, including system-on-a-chip design to support artificial intelligence, IoT and mobile applications.

The future of the flash memory market

The flash memory market continues to see advances in form factor and deployment options. Storage array vendors are adding support for the nonvolatile memory express (NVMe) controller interface, a protocol that accelerates data transfer between client systems and flash storage. An NVMe host controller exploits the high-speed performance of the PCIe bus.

Using PCIe enables an application to communicate directly with flash storage by reducing network hops that would occur with host bus adapters and routers. PCIe enables the emergence of drives built on the NVMe specification, providing an alternative that experts believe could supplant 2.5-inch and 3.5-inch form factors. NVMe SSDs plug into idle server slots on a computer, removing the cost and complexity associated with cabling.

Rising with NVMe and PCIe-connected SSDs are nonvolatile dual inline memory modules (NVDIMMs) that integrate DRAM, NAND flash and a dedicated backup power supply. NVDIMMs may be inserted in a standard DIMM connector on a memory bus. The flash memory on an NVDIMM card serves to back up and restore data kept in DRAM, while the onboard power supply ensures the storage remains nonvolatile.

Candidates to succeed NAND flash have emerged. In fact, some of these memories have been in development for a while. These alternative technologies are based on memory architectures that differ from NAND's floating gate transistor.

Phase change memory (PCM) stores data by changing the state of the bit cell using electrical pulses. These pulses alter the nonvolatile DRAM from an unstructured to a structured state. Potential advantages of PCM include lower latency on read data and accelerated write performance. In addition, PCM enables an individual byte to be written, whereas flash memory requires an entire block of data to be written.

Resistive random access memory, known also as ReRAM or RRAM, is nonvolatile storage that bears similarities to PCM. Instead of electrical pulses, ReRAM changes the resistance properties of the underlying dielectric, which can include various oxide metals and other solid materials. A ReRAM device includes a nonvolatile memory resistor, or memristor, which is a component that regulates a circuit's electricity flow and retains information on the amount of charge that has flowed through it historically.

In comparison with NAND, ReRAM is said to offer higher switching speed and lower power consumption. 4DS Memory Ltd., Crossbar Inc., Fujitsu Semiconductor, Israeli startup Weebit Nano and Western Digital -- through SanDisk ReRAM -- are all working to bring ReRAM-based storage products to market. Hewlett Packard Enterprise dropped its Memristor product in 2016, after experimenting with it in joint ventures with SanDisk and SK Hynix.

Magneto-resistive RAM (MRAM) stores data using magnetic states. A pair of ferromagnetic metal plates is used: one plate is a permanent magnet and a second plate is capable of being magnetized. Separating the two plates is a thin layer of insulating material. Binary bits are defined as one or zero according to the orientation of the two magnetic fields.

A related approach is spin-torque transfer, a method that sends an electric current to thin magnetic layers situated between non-magnetized materials. The current alters the spin of electrons to an up or down state.

This was last updated in November 2017

Continue Reading About flash memory

Dig Deeper on Solid-state storage

PRO+

Content

Find more PRO+ content and other member only offers, here.

Join the conversation

4 comments

Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.

As flash memory technology evolves, which enterprise storage technologies offer the most promise as NAND alternatives?
Cancel
How does NAND flash memory differ from NOR flash memory?
Cancel
NAND has more capacity over NOR but NOR has code execution while NAND does not. NOR performance is extremely slow with fast read and slow write while NAND has fast write and read capabilities. NOR reliability is standard that has bit-flipping issues whereas NAND has low performance and requires one bit for error management. NOR has full memory interface while NAND is I/O only. NOR’s hardware is easy to use while NAND is complicated.
Cancel
Thanks a lot for the explanation.
Cancel

-ADS BY GOOGLE

File Extensions and File Formats

Powered by:

SearchSolidStateStorage

SearchCloudStorage

SearchDisasterRecovery

SearchDataBackup

Close