Flash memory guide to architecture, types and products

Flash memory, also known as flash storage, demands a closer look.

The technology permeates a multitude of consumer products, from mobile phones to ubiquitous USB memory cards. Enterprise flash, meanwhile, plays an expanding role in data center storage, server and networking technologies. Flash, despite its name, isn't the most visible of technologies -- it's typically embedded in other products. But it's everywhere.

This flash memory guide covers uses for flash memory, the technology's history and its advantages and drawbacks. The guide also provides an overview of the different flavors of flash, from single-level cell chips to 3D NAND. We'll also take a look at the future of this far-reaching technology.

Flash memory is widely used for storage and data transfer. In the consumer sector, flash memory finds a home in a range of devices, including phones, cameras and tablets to name a few examples. Flash memory's small size and power consumption advantages make it well suited for use in on-the-go consumer devices. Indeed, consumer applications have helped propel the growth of the flash technology market.

Flash storage, a term often used interchangeably with flash memory, refers to any drive, repository or system using flash memory. At the consumer level, storage devices using flash include USB drives. In computer systems, flash-based SSDs, so called for their lack of moving parts, are prevalent in notebook computers and can also be found in some desktop PCs as a HDD option.

SSDs have also made their way into the data center in recent years. Here, flash storage adoption continues to expand in the form of enterprise-class all-flash arrays. Early flash storage deployments focused on accelerating I/O-intensive applications, which remain popular candidates for flash storage.

Amanda Regnerus, vice president of marketing and product development at U.S. Signal, a cloud and colocation services provider based in Grand Rapids, Mich., cited virtual desktop infrastructure, databases and applications requiring low latency -- such as electronic medical records systems -- as workloads that lend themselves to flash storage.

"Those are the primary applications that we run into" when considering flash as a storage choice, she noted.

"Flash will help with latency-sensitive applications, databases or, in general, any application with demanding-workload and fast-I/O requirements," added Sean McGrath, architect and field CTO at InterVision, an IT services provider with headquarters in Santa Clara, Calif., and St. Louis.

Declining costs and increasing density, however, have enabled organizations to extend their use of flash storage from specialized use cases to general enterprise workloads. Lower prices put data centers in a better position to buy flash in anticipation of future needs.

"There is still a use case for spinning disk if there is a smaller workload requirement," McGrath said. "But if the cost comparison is adequate, then the question becomes, 'Why not flash?' so organizations can future proof their workload."

"We will still see organizations deploying disk and hard drives," said Scott Sinclair, a senior analyst at Enterprise Strategy Group (ESG), a market research firm based in Milford, Mass. "I don't think disk is going away. We are just going to see flash become more and more the default choice."

The rise of digital businesses also contributes to flash storage adoption. Within such enterprises, machine learning workloads and high-level analytics are among the developments requiring faster data access.

"We are seeing a huge demand in data access," Sinclair said. "And that is fueling flash."

Sinclair said ESG's research has shown some correlation between companies that perceive themselves as digital businesses and higher usage of flash technology.For enterprises where the effective use of data has become more closely aligned with business success, being able to use data more quickly translates into more business success, he noted.

"They are attracted to higher-performance infrastructure solutions like flash," Sinclair said. "As companies become more digital, the value of flash performance actually increases, and they are willing to pay more for that performance."

But, at the same time, the price of flash has come down, making the storage technology feasible for a higher percentage of workloads in a higher percentage of companies, including digital business that were already inclined to invest in flash storage.

"It has a multiplier effect," Sinclair said of the convergence of market forces.

Speed is one of the more obvious advantages of flash memory when compared with magnetic media. Rich Castagna, freelance writer and storage commentator, said flash far outclasses HDD technology, particularly in areas such as IOPS. An SSD might offer performance measured in the tens of thousands of IOPS, while a fast HDD might be able to conjure up to 200 IOPS, according to Castagna.

Flash storage can also prove more durable than HDDs, mechanical devices with spinning platters and other components in motion. Flash reliability stands out in the mobile device market because there are no moving parts vulnerable to drops. SSDs also use less electricity and generate less heat compared with HDDs.

Cost has been, and continues to be, one drawback to using flash technology. Castagna said flash storage, on a capacity basis, costs five to eight times more than an HDD of comparable capacity. Although flash has closed the gap with traditional storage products in recent years, price lingers as a consideration for some applications.

In addition, flash storage devices accumulate wear and tear under heavy write loads -- repeated use over time will inevitably degrade an SSD. HDDs also decline with extensive use, but flash memory faces a different challenge in that regard. Flash drives typically used in enterprise storage devices must remove old data before new data can be added. This process slows writes and increases wear. HDDs, in contrast, create space for new writes, rather than deleting old data.

Hybrid flash arrays, which combine HDD and SSD storage, offer organizations the ability to harness the respective price and performance benefits of both technologies. The hybrid arrays let storage managers place frequently accessed data on speedier flash storage and house less frequently accessed data on HDDs. As a result, a data center can boost the performance of hot data, while avoiding the higher cost of flash for storing cold data.

Flash memory is a nonvolatile storage technology, which means it doesn't require power to retain data. There are two forms of flash memory: NOR and NAND. Both use floating gate transistors as the basis for memory cells that store data.

NOR flash memory was the first flash type to reach the commercial market, arriving in 1988. Parallel NOR links memory cells in parallel, providing random access. NOR flash memory is characterized by fast data reads and slower erase and write speeds. In general, NOR technology stores executable boot code and supports applications that demand frequent random reads of small data sets.

NOR flash is used in industrial robotics, medical devices, scientific instruments, IoT devices and portable consumer products, according to technical consultant Robert Sheldon.

NAND flash memory followed NOR to market about a year later. NAND is slower to read than NOR, but it takes less time to erase and write new data. NAND also offers higher storage capacity at a lower cost than NOR, so the technology's main function is data storage.

Indeed, a key objective of NAND development has been to boost chip capacity and reduce the cost per bit to make flash memory more competitive with magnetic storage devices. However, storage devices built on this technology face endurance limitations. NAND flash can support only so many program/erase (P/E) cycles, the process of erasing data before new data is written. Storage vendors use various methods to reduce P/E cycles and balance P/E loads with the goal of improving NAND flash memory durability.

NAND flash memory is broken down into several types, which are defined by the number of bits used in each flash memory cell. NAND flash memory types include single-level cell (SLC), which stores one bit in each cell; multi-level cell (MLC), which stores two bits; triple-level cell (TLC), which stores three bits; quad-level cell (QLC), which stores four bits; and penta-level cell (PLC), which stores five bits.

Each flash memory type has its own characteristics, strengths and weaknesses, which determine how and where it's used.

SLC: Performance at a priceSLC offers the highest performance, endurance and reliability compared to other NAND types. However, those benefits come with a higher price tag. Commercial and industrial applications rank among the top SLC adopters, as organizations in those sectors are more willing to pay a premium for the advantages of this NAND flash memory type.

In general, each additional bit added to a memory cell comes with a performance, endurance and reliability penalty, all along the SLC, MLC, TLC, QLC and PLC continuum. Performance takes a hit because each extra bit requires more time writing to and reading from a cell, said Marc Staimer, founder and president of Dragon Slayer Consulting in Beaverton, Ore. In addition, he noted that each additional bit reduces endurance by an order of magnitude: Although SLC tops the flash table at 10,000 write cycles, PLC is rated at about 10 write cycles.

MLC, TLC: Cheaper, denser toolsThe advantages of having additional bits in each memory cell are higher density and lower costs. MLC's price point makes it attractive to makers of consumer electronics devices, such as PCs. Enterprise MLC, however, provides more write cycles than consumer-grade MLC, making it an option for more write-intensive applications.

TLC, meanwhile, offers a still-higher storage density compared to MLC, but at the cost of lower performance, endurance and reliability. It also finds a niche in consumer electronics.

QLC: A fit for read-intensive workloadsLooking at QLC vs. TLC NAND as an either-or choice might be the wrong way to consider those technologies, which can prove complementary. Indeed, QLC technology's market positioning differs somewhat from TLC. QLC NAND focuses on read-intensive workloads, filling a niche between TLC flash and HDDs, according to consultant and technologist Kurt Marko, citing a Micron Technology technical brief.

QLC's read-intensive nature makes it suitable for enterprise applications such as data analytics and machine learning. Brien Posey, a writer and former CIO, noted that analytical workloads typically read enormous quantities of information, but don't modify that data. Other possible data center uses for QLC NAND include media streaming, where SSDs using the technology have the capacity and speed to host video files, and active archives, where data remains online and accessible, according to Posey.

But data stores of nearly any kind, provided they support read-dominant applications, can be suitable QLC workloads, noted Sheldon. He cited the example of NoSQL databases, which contain copious amounts of rich data and metadata, as a technology that could benefit from QLC NAND.

Posey, meanwhile, argued that QLC might be the right choice for enterprise desktops, providing a midlife performance boost. He pointed to the technology's modest price point and the use of integrated write caches to offset QLC's performance limitations. An organization can swap out a desktop PC's aging HDD for a QLC-based SSD, significantly improving performance. Enterprises taking this approach might also be able to delay their next desktop technology refreshment cycle, avoiding cost -- at least temporarily -- in the process.

In general, QLC NAND flash delivers lower-cost-per-gigabyte benefits that make the memory technology an affordable option compared to other flash varieties. And cost might be a key driver for organizations purchasing on an enterprise scale.

"QLC, in particular, allows for additional use cases to be served with higher density flash," said Scott Webb, global storage practice manager at World Wide Technology, based in Maryland Heights, Mo. "There is still a requirement with software and applications to address the reduction in endurance of the lower-tier flash, but we are starting to see it and believe it will only become more in demand."

PLC: One for the archivesFinally, PLC flash offers limited endurance, but low cost-per-gigabyte economy. As such, PLC targets archival applications and cool to cold data. PLC flash SSDs fall into the same class as other write-once-read-many technologies, according to Dragon Slayer Consulting's Staimer.

The shift from planar to 3D technology is another way flash manufacturers seek to improve flash's price and capacity characteristics. 3D NAND flash stacks memory cells vertically in multiple layers. This method of layering cells increases SSD capacities and lowers the per-gigabyte cost. 3D NAND is considered suitable for any business or consumer scenario that uses planar NAND.

Higher capacity in a smaller physical space is the main advantage of 3D flash memory. Higher manufacturing cost has been the primary drawback, but the technology's price has been dropping.

Another aspect of 3D NAND is charge trap technology. Many 3D flash drive manufacturers use the charge trap approach, which provides higher endurance rates than drives with floating-gate cells. Charge trap technology also supports faster read/write operations and lowers energy use, according to Sheldon.

A cursory look at flash memory might suggest the technology is similar to RAM. After all, Flash and RAM both employ solid-state chips and occupy the same solid-stage storage category.

But flash memory and RAM play different roles in a computer system, based on their performance, cost and manufacturing methods. Flash's main use, as noted above, is storage. RAM, on the other hand, makes calculations on data retrieved from storage.

In a price and performance comparison, RAM is faster than flash, but it costs more. Of the two types of RAM, static RAM is the fastest and most expensive option, typically used as the cache memory inside a computer's CPU. Slower and cheaper dynamic RAM (DRAM) serves as a computer's primary operational memory. From a historical perspective, it was the price crossover between DRAM and NAND in 2004 that sparked the wider adoption of flash memory. NAND prices fell below the cost of DRAM at that point in time.

Less expensive flash memory also differs from RAM in that it's non-volatile. Flash's lower power consumption, persistence and lower price point make it suitable for use as storage memory within SD cards, USB drives and SSDs, among other devices.

Flash memory standards go beyond the primary groupings of SLC, MLC, TLC and their successors. Internal and external connectivity standards are also worth noting, as the typical link-up approaches are beginning to shift.

As for internal connectivity, SAS remains the leading flash interconnect, but vendors are now gradually replacing SAS technology with NVMe, according to George Crump, former president of Storage Switzerland, an IT analyst firm focusing on storage and virtualization. NVMe offers the benefit of being optimized for interaction with memory storage.The flash drive, as a result, spends less time waiting for I/O requests, which paves the way for NVMe to supersede SAS as the connection method, Crump said.

Another key standard is NVMe-oF, versions of which support Fibre Channel and Ethernet networks. NVMe-oF targets external connectivity, providing the advantages of internal NVMe across the network.

Consultants advised organizations to avoid a technical evaluation in the early stages of mulling storage options.

"Before we ever start talking about any technology or specific products, it is important to understand the use case and the outcome the customer is trying to arrive at," said World Wide Technology's Webb. "Storage tiering remains extremely important, with multiple flash tiers available as well as traditional spinning disk and tape."

Organizations should also consider their data access habits.

"Roughly 80% of production data is not accessed on a regular basis," Webb said. "We want to understand customers' data access and performance requirements and architect from that standpoint."

In addition, IT buyers tend to consider the bigger storage picture, rather than fixate on flash as their primary focus.

"Typically, the storage solutions chosen really rely on a few other indicators and features, not so much the flash," InterVision's McGrath said.

Instead, storage decisions are made around considerations such as data management, hybrid cloud strategy and integration, workload handling, data protection features, multiprotocol use cases, snapshots, encryption, virtual integration and presentation, and application integration, he added.

"In today's day and age, performance and feature sets can be very similar across the vendor landscape," Webb noted. "One of the most common features that customers are interested in now is cloud data management, meaning they have data in one -- or many -- public clouds, on prem and colocated at cloud-adjacent providers like Equinix. Customers need that management experience to be seamless and simple."

When it's finally time to drill down into the details of flash storage options, buyers have a few particulars to weigh. One such consideration: SSD form factors. Early in the evolution of enterprise flash, the 2.5-inch SATA form factor became popular among SSD vendors looking to ease the HDD-to-SSD transition. The SATA standard was created for HDD data transfer. The widely used SATA form factor let organizations adopt new drives more gradually, and the 2.5-inch size enabled an SSD to readily fit into a laptop or desktop drive bay, consultant Sheldon noted.

A smaller SATA SSD form factor, dubbed mSATA for mini SATA, targets laptops, notebooks, tablets and other power-constrained devices. The M.2 form factor followed mSATA, offering a still smaller SSD form factor that offers higher performance and greater storage capacity than mSATA drives, according to Sheldon.

SSDs can also take the form of add-in cards that plug into a computer's PCIe motherboard slot.

The market's move to NVMe-based high-performance SSD technology is another aspect of flash that storage buyers should evaluate. Consultant Marko said NVMe SSDs, which connect directly to the PCI system bus, are becoming more prevalent in the enterprise. SSDs adhering to HDD form factors and interfaces haven't taken advantage of the full potential of NAND flash.

NMVe SSDs offer an edge over SATA drives because the NVMe protocol was created for non-volatile semiconductor memory, such as NAND flash, Marko noted. And as NVMe grows in popularity as an interconnect for flash disks and arrays, NVMe-oF products have begun to provide a workable option for NVMe-based shared storage, he added. The NVMe-oF technology ecosystem is in a state of flux, but resources such as the University of New Hampshire InterOperability Laboratory can help evaluators identify standards-compliant products.

NAND flash memory vendors vary in what they offer the market. Some manufacturers provide general-purpose flash storage, while other companies specialize in specific market segments, according to former CIO Posey.

Some vendors offer a range of products, covering enterprise and consumer applications. Intel, for example, targets enterprise-class computing with its Optane Persistent Memory NAND flash product and Optane DC SSDs. But it also provides NAND storage devices oriented toward client devices and the consumer sector. Kioxia makes enterprise, client and data center SSDs, while Samsung's SSD portfolio is geared to the enterprise and consumer markets.

As for marketing, vendors might sell flash memory products to other vendors that embed the technology in their storage offerings or sell storage devices under their own brand. Vendors such as Micron Technology pursue both approaches.

The variability found in vendors' NAND flash offerings can also be seen in pricing. Although decreasing cost serves as the general rule for NAND flash pricing, the actual pattern at any given time is somewhat hard to pin down. Storage market analysts differ in their short- and long-term predictions for NAND pricing trends. Cyclical periods of shortage and oversupply cause NAND pricing to fluctuate. In addition, macroeconomic trends and global events -- such as the 2020 coronavirus pandemic -- can affect demand for flash memory and, therefore, pricing.

The future of flash memory is pointing toward greater capacity as vendors push on with plans to boost the number of 3D NAND flash layers and increase bit densities, Sheldon noted. Other anticipated trends include the increasing use of QLC drives in enterprise storage, at least for read-intensive workloads.

Developments such as NVMe and NVMe-oF, meanwhile, will continue to push flash storage toward fully realizing its higher-performance potential. The evolution of flash will also feed into other technologies, such as storage class memory (SCM). SCM represents a new storage/memory tier in the enterprise, existing between SSDs and DRAM, with the aim of supporting latency-sensitive applications, Sheldon noted.

ESG's Sinclair said adoption of NVMe-oF will "provide … the ability to leverage more storage class memory and lower-latency media types."

Flash storage advances, however, are revealing other performance obstacles, such as the CPU chokepoint. Dragon Slayer Consulting's Staimer contended that CPUs aren't keeping up with the performance gains of NVMe and NVMe-oF, noting that SCM will only exacerbate the issue because it places more load pressure on the CPU.

Sinclair said the evolution of flash suggests bottlenecks don't entirely disappear -- they just go somewhere else. Removing performance barriers from storage media and networks could eventually transfer the pressure to applications -- and software developers.

"Do they optimize their applications to take advantage of this higher-performing infrastructure?" Sinclair asked. "This is going to be the really interesting question as we move forward."