NAND technology paved the way for a new breed of SSD that is able to emulate HDDs in most enterprise or consumer applications. These SSDs are far less expensive than DRAM-based SSDs and still offer several advantages over HDDs—particularly in terms of performance and reliability.

Because NAND-based SSDs are a solid state technology, they have no moving parts and offer much better performance than HDDs. When a command is issued to an HDD, the drive must seek with its actuator, spin its platter, and then transfer the data back to the host. But SSDs have no moving parts (requiring only the time it takes to process the command), and they have random access times as quick as 20µs [52].

The improved performance of new SSDs equates to 10,000 IOPS compared to less than 450 IOPS for the fastest HDDs [53]. When used in enterprise applications like Internet banking, SSDs might significantly boost information access.

With no moving parts to wear out or break, an SSD will

outlast almost any HDD, which typically has only a three- to five-year life expectancy if it is not bumped, banged, or dropped. By comparison, a modern SSD might last twice that long and do so without the sensitivities to mechanical shock and while consuming only a fraction of the power.

RAIDs, Connections, and the Next Step for SSDs

SSDs can go anywhere an HDD can, so for enterprise and consumer applications alike, SSDs are replacing HDDs—a trend that is sure to continue for the next decade or more. But using an SSD as a drop-in replacement for an HDD is not necessarily using SSDs to their fullest potential. RAID controllers, HDD interfaces, and storage subsystems have been optimized for the characteristics of rotating magnetic media and may be a bottleneck for solid state storage.

Due to the flexibility of NAND solid state storage, SSDs will once again change the picture of storage in computers. NAND-based storage will become more integrated into the computer and will enable new generations of applications. Productivity gains will be measurable and the power savings, dramatic.

Conclusion

Digital storage has come a long way since 1956, with the most recent innovation being SSDs. And now that SSDs are gaining new ground with the advancements made possible by NAND Flash technology, they represent the next evolutionary step for storage applications.

Notes:

[52] Wong, page 15.
[53] Justin Sykes, “SSDs to Boost Data Center Performance,” Micron Technology, Inc. Boise, Idaho, (July 30, 2008): page 3.

In 1978, StorageTek introduced the first modern SSD. This pioneering SSD had a maximum storage capacity of 90MB and sold for about $8,800 per megabyte. [42] “The SSD served the mainframe industry as a virtual memory extension for paging and swapping programs in and out of memory” [43]. That same year, Texas Memory Systems began marketing a 16KB RAM SSD to oil companies for a seismic data acquisition system. [44] SSDs were born, but didn’t take off. At least not right away.

As far as mainframes were concerned, “the arrival of expanded storage, a bus extension for additional main memory capacity, signaled the end of the SSD market—for a while,” explained Fred Moore, a one-time StorageTek director. [45]

“In the early 1990s, a few small companies were building SSDs for select applications running on Unix, but market visibility was low and price per megabyte was still high. During the 1990s, the popularity of Unix, NT, the Internet, and, later, Linux increased. They became the largest storage markets for databases, and the heavy I/O loads they generated created response time bottlenecks. Twenty-five years after their first appearance, SSDs are still a niche market but are becoming the new stealth weapon for system programmers and storage administrators who struggle to deliver the consistent response times necessary to meet service levels,” [46] Moore wrote in 2002.

“Based on high-density DRAM chips, rather than rotating disk media and moving heads, the variable and lengthy seek and rotational times for rotating disks are eliminated, leaving a very short access and data transfer time to complete an I/O operation. There are no cache misses or back-end data transfers on an SSD. Typical I/O operations on an SSD occur between 30 and 40 times faster than on a rotating disk. SSDs are a quick fix for severe I/O performance problems, and they don’t face the ongoing access density challenges of higher-capacity disks. These devices are fault-tolerant architectures and protect data from all types of device failures, not just from the loss of electrical power.” [47]

In terms of a storage evolution, the DRAM- or RAM-based SSD was almost too specialized to have a large impact.
NAND Flash Technology

Fujio Masuoka began working on Flash memory cells in the 1970s at Toshiba and received patents for his work in 1980. [48] Masuoka’s designs were perhaps the most important semiconductor innovation in the history of storage, but unfortunately, it went poorly for Masuoka. For his work Toshiba gave Masouka “a bonus worth a few hundred dollars”—and promptly let its archrival Intel take control of the market for his invention. Subsequently, Masuoka says, Toshiba tried repeatedly to move him from his senior post to a position where he could do no further research.” [49]

Masouka’s Flash memory concepts have evolved, and today NAND Flash technology and SSDs have the potential to displace HDDs and force an evolutionary step in storage.
Like all semiconductor devices, NAND Flash memory relies on an electrical current to operate. Specifically, a voltage “is applied to the control gate to draw electrons from the substrate to tunnel through the gate oxide into a polysilicon floating gate layer. To store one bit, two charge levels in the floating gate layer can be stored to distinguish between a 1 and a 0.” [50]

“Single-level cell (SLC) NAND Flash memory

stores one bit of information per memory cell. This basic technology enables faster transfer speeds, lower power consumption, and increased endurance. For designs using mid-range densities, SLC NAND Flash will continue to be a good choice. Multiple-level cell (MLC) NAND, by comparison, stores two to four bits of information per memory cell, effectively doubling the amount of data that can be stored in a similar-size NAND Flash device. SLC NAND offers high performance and reliability, is supported by all controllers, and requires only 1-bit error correction code (ECC). SLC NAND is for applications like high-performance media cards, hybrid disk drives, solid state drives, and other embedded applications with processors, where it is used for code execution. MLC is supported only by controllers that include 4-bit or more ECC.” [51]

MLC is a low-cost file storage solution for consumer applications like media players, cell phones, and media cards (USB, SD/MMC, and CF cards) where density is more important than performance. MLC NAND has also emerged as the dominant Flash memory choice for SSDs targeted at the notebook PC market because they offer such a well balanced price-to-performance solution.

In fact, it is MLC NAND—for the most part—that

has powered so many of the recent advances in mobile computing and digital media convergence. MLC NAND has replaced the day planner with the BlackBerry, exchanged film for media cards in cameras, and enabled a musical revolution with the Apple® iPod® and other MP3s. Today, people can carry more memory around in a USB drive on their key chains than an entire room full of early HDDs could have stored.

Notes:

[42] Fred Moore, “Enterprise Storage Report for the 1990s,” StorageTek Corp., downloaded from http://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/19930003944_1993003944.pdf.
[43] Fred Moore, “What Goes Around Comes Around in Storage: Old Ideas Find New Applications for Today’s SANs,” Computer Technology Review, (September 2002), downloaded from http://findarticles.com/p/articles/mi_m0BRZ/is_/ai_101679012.
[44] Gregory Wong, “Solid State Drives: A Closer Look Report No. FI-NHL-SSD-1008,” Forward Insights, (October 2008): page 73.
[45] Fred Moore, “What Goes Around Comes Around in Storage: Old Ideas Find New Applications for Today’s SANs.”
[46] Fred Moore, “What Goes Around Comes Around in Storage: Old Ideas Find New Applications for Today’s SANs.”
[47] Fred Moore, “What Goes Around Comes Around in Storage: Old Ideas Find New Applications for Today’s SANs.”
[48] Benjamin Fulford, “Unsung Hero,” Forbes (June 24, 2002), downloaded from http://www.forbes.com/global/2002/0624/030.html.
[49] Fulford.
[50] Wong, page 13.
[51] “MLC vs. SLC Flash,” downloaded from http://www.micron.com/nandcom/.

In spite of new technologies like perpendicularly aligned bits and HAMR, HDDs are mechanical devices at heart and, as such, they face many performance challenges. Indications are that, ultimately, as storage systems continue to evolve, HDDs will be replaced.

Mechanical devices cannot improve as quickly as solid state technologies can. For example, “over the past 20 years, microprocessor technology—which plays a key role in data storage efficiency and function—has enabled CPU performance to nearly double every 18 months. Put another way, CPU performance has increased 16,800 times between 1988 and 2008, but HDD performance has increased by just 11 times.” [37]

Even leading HDD manufacturers recognize the

HDD performance problem. When Seagate Technology introduced faster, 15,000-RPM disk drives in 2004, it released a white paper describing the need for better HDD performance.

“Dramatic advances in processor speed, RAM size and RAM speed have combined to accelerate system performance to levels unthinkable just a few years ago. Such powerful hardware resources have made feasible software solutions with increasingly sophisticated and comprehensive capabilities, enabling business productivity to climb at a remarkable rate. Yet one aspect of system evolution has historically lagged behind: disc drive performance. While impressive advances in density have yielded exponential growth in disc drive capacity, disc drive speed has achieved only modest gains over the years,” [38] Seagate said.

To try to close the HDD performance gap, manufacturers have increased the drive’s rotational speed, added more advanced heads, and used techniques like short stroking, which restricts data to 5%–30% of the platter to boost performance. Western Digital, for example, recently released a speedy 20,000 RPM HDD. But faster and faster disk rotation cannot be a lasting answer because these high-speed HDDs potentially make more noise, devour more power, and become increasingly less reliable. In addition, these higher-performance HDDs all sacrifice capacity. Each time the CPU issues a command “the hard drive’s mechanical system must then seek the requested data block or file by rotating its spinning platter and reaching out with its actuator.” [39]To be sure, HDD engineers have continued to improve these devices and thus, stave off their ultimate extinction.

HDD Mean Time Between Failures (MTBT)

“It is estimated that over 90% of all new information produced in the world is being stored on magnetic media, most of it on hard disk drives. Despite their importance, there is relatively little published work on the failure patterns of disk drives and the key factors that affect their lifetime. Most available data are either based on extrapolation from accelerated aging experiments or from relatively modest-sized field studies. Moreover, larger population studies rarely have the infrastructure in place to collect health signals from components in operation, which is critical information for detailed failure analysis.” [40]

This seeming lack of information about a modern HDD’s mean time between failures is a problem for large data centers and for the potential survival of HDDs. To try and shed light on the subject, Google created the first, large population HDD failure study in 2006 and released their findings at the 5th USENIX Conference on File and Storage Technologies in February 2007.

The Google research categorized dozens of failure types, found a handful of unexplained relationships, and generally showed that HDDs fail more often than manufacturers predict [41]. The study was an important first step since it provided users with foundational data for further research and it gave HDD manufacturers a sort of failure map. Solving some of these issues may result in better HDDs in the near future. If they go unaddressed, however, these failure issues could spell the end of HDDs.

Notes:

[37] Justin Sykes, “Performance Productivity for Enterprise Applications,” Micron Technology, Inc., Boise, Idaho, (July 30, 2008): page 1, downloaded from http://download.micron.com/pdf/whitepapers/performance_productivity_for_ent_apps.pdf.
[38] “Economies of Capacity and Speed: Choosing the most cost-effective disk drive size and RPM to meet IT requirements,” Seagate Technology LLC, Scotts Valley, CA, (May 2004): page 2, downloaded from http://www.seagate.com/docs/pdf/whitepaper/economies_capacity_spd_tp.pdf
[39] Sykes, “Performance Productivity for Enterprise Applications,” page 2.
[40] Eduardo Pinheiro, Wolf-Dietrich Weber, and Luiz Andre Barroso, “Failure Trends in a Large Disk Drive Population,” Google Inc., Mountain View, Calif. (February 2007): page 1.
[41] Pinheriro et al, page 12.

In 1980, Seagate Technology introduced the world’s first 5.25-inch hard drive, bringing HDDs to a broader audience; prior to 1980 only large and well funded companies could afford the technology.

HDD capacity grew as much as 30% each year in the 1980s before accelerating to more than 60% per year in the 1990s. By 1999, HDD capacity was doubling every nine months [26]

The SPE Barrier–HDD Innovation
To achieve the HDD’s nearly exponential density growth, scientists and engineers miniaturized the magnetic grains or bits on the platter’s surface, squeezing more bits into the same or even smaller physical space [27]. These same researchers also developed more sensitive read/write heads (the giant-magnetoresistive head introduced in 1997, for example), capable of detecting faint magnetic fields [28].

Since its inception, HDDs have faced a density-growth challenge in the form of the superparamagnetic effect (SPE). “Superparamagnetism occurs when the microscopic magnetic grains on the disk become so tiny that random thermal vibrations at room temperature cause them to lose their ability to hold their magnetic orientations. What results are ‘flipped bits’ – bits whose magnetic north and south poles suddenly and spontaneously reverse – that corrupt data, rendering it and the storage device unreliable.” [29]

Temperature plays a role in the SPE since another way to describe the effect is to say that when the ambient thermal energy equals the amount of energy needed to change a bit’s polarity, that bit can flip and lose the data it was storing.

As bits are compressed, they become more susceptible to SPE, meaning that larger and faster HDDs have the potential to become less reliable [30]. For several decades, HDD developers have searched for ways to stave off the eventuality of reaching the density and reliability limits of HDDs.

One of the chief ideas proffered was to align bits perpendicularly rather than longitudinally. Famed inventor Valdemar Poulsen, who is sometimes called the Danish Edison, was one of the first researchers to experiment with perpendicular recording nearly 100 years ago [31], but it took modern engineers at leading HDD makers to actually produce HDDs with perpendicular bits like Hitachi Global Storage Solutions first introduced in 2006.

In longitudinal magnetic recording, each bit is oriented horizontally on the platter, whereas perpendicular recording orients bits vertically on the platter and actually increases the number of bits that can be aligned on the disk [32]. Perpendicular recording is also inherently more stable across temperature ranges [33] because its poles are arranged south pole to south pole and north pole to north pole. In this way, bits naturally repel each other, reducing the likelihood of the SPE occurring [34]

Several of the world’s leading HDD makers now offer perpendicularly aligned HDDs.

Heat-Assisted Magnetic Recording
Heat-assisted magnetic recording (HAMR) is a hybrid of magnetic and optical technology that represents the latest innovation in HDD development. HAMR has the potential to increase HDD density by an order of magnitude while still avoiding the SPE ’s limitations [35].

With HAMR, engineers use a laser to briefly heat an area of an HDD’s platter. The heat lowers that area’s coercivity so it is below the coercivity of the magnetic field that the recording head is producing, essentially making it easier to flip a given bit’s magnetic orientation in a stable magnetic material and allowing for “smaller thermally stable grains.” [36]

Notes:

[26] Zeytinci, page 7.
[27] Kim Nguyen, “Perpendicular Recording: A Boon for Consumer Electronics,” Hitachi Global Storage Technologies, San Jose, Calif. (April 2005): Page 1.
[28] Zeytinci, page 10.
[29] Nguyen, page 1.
[30] “Hard Disk Superparamagnetics,” Dataclinic: downloaded from http://www.dataclinic.co.uk/hard-disk-superparamagnetic-effect.htm on October 12, 2008.
[31] Nguyen, page 3.
[32] Nguyen, page 3.
[33] Yochiro Tanaka, “Fundamental Features of Perpendicular Magnetic Recording and the Design Consideration for Future Portable HDD Integration,” IEEE (January 2005).
[34] John Best, “Perpendicular Recording: Opening the Boors for 10-fold Hard Drive Capacity Expansion,” Computer Technology Review (June 1, 2005).
[35] Mark H. Kryder, Edward C. Gage, Terry W. McDaniel, William A. Challener, Robert E. Rottmayer, Ganping Ju, Yiao-Tee Hsia, and M. Faith Erden, “Heat Assisted Magnetic Recording,” Proceedings of the IEEE, Volume 96, No. 11: (November 2008): page 1,810.
[36] Kryder et al: page 1,810.

The hard disk drive (HDD) is the workhorse of modern storage systems, from personal computers to enterprise networks. To record data, HDDs change the polarity of tiny sections (magnetic domains) of a magnetic platter. Flipped one way, a domain represents the binary 0; flipped the opposite way, it represents a 1. Domains are arranged in a circumferential fashion around the platters so that a read/write head driven by a servomechanical actuator can track the binary bits [21]. This sort of storage was nothing less than a modern marvel when IBM first introduced the 350 Disk Storage Unit in 1956.

“The 350 Disk Storage Unit consisted of the magnetic disk memory unit with its access mechanism, the electronic and pneumatic controls for the access mechanism, and a small air compressor. Assembled with covers, the 350 was 60 inches long, 68 inches high, and 29 inches deep. It was configured with 50 magnetic disks containing 50,000 sectors, each of which held 100 alphanumeric characters, for a total capacity of 5 million characters.”

“Disks rotated at 1,200rpm, tracks (20 to the inch) were recorded at up to 100 bits per inch, and typical head-to-disk spacing was 800 microinches. The execution of a ’seek’ instruction positioned a read-write head to the track that contained the desired sector and selected the sector for a later read or write operation. Seek time averaged about 600 milliseconds.” [22]

The 350 was one of six components in IBM’s 305 Random Access Memory Accounting (RAMAC) system, which also included an IBM 305 Processing Unit, an 80-position serial-output printer called the 370, a card punch, a console, and a huge power supply. Two years after it was introduced, IBM began offering the 305 RAMAC with an optional second 350 Disk Storage Unit, which doubled capacity. The 305 RAMAC originally leased for $3,200 [23].

As an interesting aside, each of the RAMAC’s 50 aluminum platters was coated with magnetic iron oxide, derived from the same chemical formula as the primer paint used on the Golden Gate Bridge. [24]

In 1973, IBM introduced the 3340 or Winchester Direct Access Storage Facility. Certainly IBM had not been inactive. The company developed several models between the 305 and the 3340, but the smaller and lighter 3340 marked the next real evolutionary step in hard disk storage.
“The 3340 featured a smaller, lighter read/write head that could ride closer to the disk surface—on an air film 18-millionths of an inch thick, and with a load of less than 20 grams. The Winchester disk file’s low-cost head slider structure made it feasible to use two heads per surface, cutting the stroke length in half. The disks, the disk spindle and bearings, the carriage, and the head-arm assemblies were incorporated into a removable, sealed cartridge called the IBM 3348 Data Module. A track density of 300 tracks per inch and an access time of 25 milliseconds were achieved.” [25]

At each step along this evolutionary cycle, storage enabled new applications and greatly increased productivity. Over the next several years, as storage memory continued to evolve, the HDD would emerge as the next new, more adaptable solution and would replace many of the earlier, groundbreaking storage technologies.

Notes

[21] Zeytinci, page 6.
[22] “IBM 350 Disk Storage Unit,” IBM, Armonk, N.Y. downloaded from http://www-03.ibm.com/ibm/history/exhibits/storage/storage_350.html on October 11, 2008.
[23] “650 RAMAC Announcement,” IBM, Armonk, N.Y. (September 14, 1956) downloaded from http://www-03.ibm.com/ibm/history/exhibits/650/650_pr2.html on October 11, 2008.
[24] Zeytinci, page 7.
[25] “IBM 3340 Direct Access Storage Facility,” IBM, Armonk, N.Y. downloaded from http://www-03.ibm.com/ibm/history/exhibits/storage/storage_3340.html on October 28, 2008.

Almost in parallel to tape drive development, magnetic drum memory was finding use as a data storage media.

While working under a contract with IBM (The Tabulating Machine Company) in 1928 [14], Austrian engineer, Gustav Tauschek, who was self-taught [15], developed the first electromagnetic drum storage device. He received a U.S. patent for his work on drum storage in 1932, but his invention would not become generally popular until the 1950s and 1960s.

In its most basic form, magnetic drum memory is simply a metal drum
or cylinder coated with a ferromagnetic material. Stationary write heads emit an electrical pulse, changing the magnetic orientation of a particle at a given position on the drum. The read heads, which are also stationary, recognize a particle’s orientation as either a binary 1 or 0. Tauschek’s prototype could store 500,000 bits across the drum’s total surface for a capacity of about 62.5KB.[16]

The Workhorse of Modern Industry

In the 1950s, the world of computers was changing, and while it would be decades before the personal computer completely revolutionized the business world, companies like IBM were making huge strides in electronic data processing. It was against this backdrop that the engineers at IBM’s Endicott, New York, laboratory launched the 650 Magnetic Drum Data Processing Machine in 1953 [17].

Originally, IBM believed that the total market for 650’s might be 50 installed units. But in less than two years, 75 of the drum-based machines had been installed and the company expected to install more than 700 more units over the next few years [18].

“The development requirement underlying the 650 was for a small, reliable machine offering the versatility of a stored-program computer that could operate within the traditional punched card environment. IBM—and the industry—wanted a machine capable of performing arithmetic, storing data, processing instructions and providing suitable read-write speeds at reasonable cost. The magnetic drum concept was seen as the answer to the speed and storage problems.”[19]

By 1962, when IBM stopped manufacturing the 650, more than 2,000 units had been sold, making it the most popular computing machine of the era. [20]

The principles at work in magnetic drum memory would help to lead researchers to another and perhaps even more important innovation: the hard disk drive.

Notes

[14]Gustav Tauschek, “Setting Device for Calculating Machines and the Like,” U.S. Patent 1880523 (October 4, 1932).
[15] Brian Randell, “The History of Digital Computers,” University of Newcastle upon Tyne (1974):page 4
[16] “People Behind Informatics,” Universitat Klagenfurt, downloaded from http://cs-exhibitions.uni-klu.ac.at/index.php?id=222 and “Lecture 16: The First Modern Computers,” School of Analytic Studies & Information Technology, York University, downloaded from http://www.yorku.ca/lbianchi/sts3700b/lecture16a.html.
[17] “The IBM 650, Workhorse of Modern Industry,” IBM, Armonk, N.Y. downloaded from http://www-03.ibm.com/ibm/history/exhibits/650/650_intro.html.
[18] “The IBM 650, Workhorse of Modern Industry,” IBM, Armonk, N.Y. downloaded from http://www-03.ibm.com/ibm/history/exhibits/650/650_intro.html.
[19] “The IBM 650, Workhorse of Modern Industry,” IBM, Armonk, N.Y. downloaded from http://www-03.ibm.com/ibm/history/exhibits/650/650_intro2.html.
[20] “The IBM 650, Workhorse of Modern Industry,” IBM, Armonk, N.Y. downloaded from http://www-03.ibm.com/ibm/history/exhibits/650/650_intro.html.