Thanks for the Memories – Storage Media in Industrial Applications

Written by: Alan L. Koch, P.E., Advantech Corporation, eAutomation Group

When it comes to data storage for industrial applications, users no longer have to go around in circles. They now have a choice between rotating magnetic media – hard disk drives with spinning platters – and flash memory. Thanks to advances in size, speed, and reliability, solid state flash memory can now be a serious contender for the storage of operating systems, applications and vital process data.

Users, therefore, have to make a choice about which storage technology to implement. This selection will depend on a number of factors, including cost, speed, reliability, and ruggedness.

The environment also plays a role. The plant floor isn’t the office – or a living room. For example, offices and homes are heated and cooled, so that temperature swings are minimized. Thus, the commercial temperature range is from 0 to 50 degrees Celsius, while industrial operations take place over the -20 to +70 C span. What’s more, the industrial environment is dustier, possibly wetter, and more vibration prone than the office.

What this means is that the typical office computer might work in an industrial setting but it won’t do so for long. For that reason, the use of an office computer in an industrial environment isn’t recommended.

Similarly, some storage technologies that work fine in an office aren’t suited for industrial applications. To take one example, a location with large industrial stamping machines may shake so much that the normal protection in disk drives against sudden jolts – the parking of read/write heads at a safe location – may save the disk but play havoc with the collection of process data.

A look at the two competing storage technologies reveals the strengths and weaknesses of each. Based on this comparison, Advantech recommends the use of flash memory, if specific features and precautions are employed to ensure data integrity.

Going for a Spin: Rotating Media

Hard disks have been around decades and are a widely used storage technology. Consisting of rigid platters coated with magnetic media, hard disks store data by selective magnetization of the platter surface. The platters rotate on a spindle past a read/write head, which senses the magnetization in the media as it travels past. When writing data, the head imposes a particular magnetic orientation on the media, which translates into stored ones and zeros.

Typically, a hard disk will have some onboard solid state memory that buffers incoming data before it is written to the platter. Data transfer rates vary, but as of 2006, 50 megabytes per second was typical for a workstation. The transfer rate depends upon the platter rotation rate.

The magnetic regions that store the data are micron-sized or smaller and the read/write head flies on an air bearing nanometers above the surface. To protect the drive from dust and contaminants, it’s sealed within an enclosure.

When it comes to data storage, hard disks have various advantages. They’re inexpensive, costing less than $0.50 a gigabyte. They’re also getting cheaper, thanks to improvements in manufacturing and increases in data storage areal density. This density ultimately determines drive capacity and cost. While at one time the areal density was doubling every year, the rate of improvement dropped to about 20 percent a year as manufacturers ran up against physical limits that arose as magnetic regions shrank and were packed closer and closer together. The areal density improvement is expected to increase to about 40 percent a year starting in 2006 as manufacturers switch to perpendicular recording, which allows more data to be stored in a smaller area.

Disk drives have other advantages besides cost. For one thing, they’re fast with transfer rates in the tens of megabytes per second. Also, they’re readily available from many sources, ensuring easy access for new installations as well as replacements. Another advantage is that disk drives come in a variety of sizes and form factors, denoted by the size of the platter. The typical desktop computer will have a 3.5 inch drive with a capacity, as of 2006, of 100 gigabytes or more and a rotation speed of 7200 revolutions per minute. A laptop will have a 2.5 inch drive, with a rotation speed of 4200 or 5400 RPM. There are also 1.8 inch drives and others that measure an inch or less.

Size Doesn’t Matter

However, for all of these plusses, disk drives also have some disadvantages, particularly for industrial data storage applications. For one thing, disk drives consume power and therefore generate heat. In an industrial setting, the disk drive is often sealed inside a box with memory and a CPU. With no fan or even passive venting to pull in cool air from the outside, dissipating heat inside the box can be a challenge, one that a hot disk drive makes harder.

Disk drives are also mechanical devices, which makes them likely to fail sooner than other components constructed entirely out of integrated circuits. The mean time between failure (MTBF) for disk drives looks impressive, with as much as a million hours sometimes quoted. However, a close inspection of the specification reveals that MTBF figure assumes operation only a few hours a day and with a relatively low duty cycle. More importantly, disk drives are often only certified for a commercial temperature range. So the MTBF in an industrial setting with a wider temperature range and 24 hour operation will be significantly less than the quoted specification.

Finally, not all disk drives are created equal. The bigger sizes tend to have faster data transfer rates and be more reliable while the smaller sizes tend to use less power and so generate less heat. They also tend to have smaller MTBFs and in general are less rugged. The very smallest drives are, in general, the least reliable.

In one important parameter, though, size doesn’t matter. The cost to manufacture a disk drive is fixed to a degree. So a 2.5 inch drive costs as much to make as a 3.5 inch drive. Furthermore, the cost to make a disk doesn’t depend on its storage capacity until that capacity rises above a technology dependent threshold. Since most disk drives are manufactured for the laptop and desktop market, one consequence is that the smallest capacity drive available commercially keeps creeping up. In the late 1990s, drives were a few or perhaps 10 gigabytes in size. Less than a decade later, the smallest capacity still in production was about 50 gigabytes.

Storing Data in a Flash

Flash memory is younger than disk drives by several decades, having been invented in the mid-1980s at Toshiba. Flash memory comes in two flavors, NOR and NAND. Both work by storing charge on floating gates in metal oxide semiconductor field effect transistors (MOSFETs). These gates are electrically isolated from other circuit elements but the charge on them can be sensed and therefore translated into ones and zeros. By using specific programming voltages, charge is placed on the floating gate or removed. Since the floating gate is isolated, the charge remains once power is removed, making flash memory nonvolatile.

NAND flash is the faster of the two, as well as offering higher density, lower cost per bit, and more endurance when it comes to repeated read/write cycles. It’s the type of flash memory found today in most applications.

For industrial use, flash memory offers a number of advantages. The available size has increased steadily, in part due to shrinking semiconductor feature sizes and in part due to improvements in flash memory layout and operation. As of 2006 four gigabit flash was readily available, with announcements by Toshiba of eight gigabit and Samsung of 32 gigabit chips. The four gigabit size is enough to store the latest operating system and specific industrial applications.

Another advantage is that flash memory is certified for industrial temperature ranges. There are even products capable of meeting military specifications of -40 to +85 C. Because there are no mechanical parts involved, flash memory is also largely immune to vibration and dust. The lack of any moving parts also makes mechanical failure a nonissue. Finally, flash memory inherently consumes little power and so generates little heat.

However, flash isn’t cheap. The cost runs in the dollars per gigabyte, not vice versa. What’s more, a flash memory can only be written to a certain number of times before it fails. The exact number varies depending upon the specific manufacturer, but a standard specification is for a million erase-write cycles. These two disadvantages – cost and a finite number of programming cycles – have kept flash memory from replacing hard disk drives in desktops and laptop computers.

Wear Leveling

While these are concerns, trends exist that promise to make both less of an issue. For one thing, the cost of flash memory is dropping steadily as the underlying semiconductor technology becomes less and less expensive. In late 2005, the technology analyst firm IDC predicted the cost of flash memory would decline by a 43 percent compound annual growth rate from 2004 to 2009. That decline, in turn, is fueling larger volume manufacturing, which leads to further price cutting. For example, the technology market analyst firm Gartner said that NAND flash memory-based portable media players would account for 80 percent of the market in 2005, with close to 188 million units predicted to ship in 2006. That nearly 40 percent growth rate from one year to the next is just one example of increases that are helping to push prices down.

What’s more, flash memory is more modular than disk drives for low storage needs. Thus, the cost of a flash memory solution for a given application may be equivalent to that of a disk drive because the hard disk comes with what amounts to excess capacity.

As for reliability issues, manufacturers are implementing techniques like wear leveling. In this approach, the amount of traffic a given bit undergoes is analyzed. When a high wear bit is nearing end of life, it’s electrically swapped with another that’s in a low usage area. By changing the mapping of external requests to internal memory locations, software can ensure that the wear on the bits is more uniform and extend the life of the chip. Likewise, with the cost dropping spare memory blocks can be built into the chip and brought into use when bits are close to wearing out.

Because of these improvements and the fact that hard disk capacities continue moving upward, flash memory is beginning to replace rotating media is various applications. In the portable media player market, for example, hard-drive based models accounted for 65 percent of the market in early 2004. That figure had dropped to 20 percent by late 2005. Flash memory is even being considered as complement to hard disk drives, particularly in applications were low power consumption is key. The idea is to write to flash memory until it’s full, keeping the disk spun down during that time.

Recommendations: Two Are Better Than One

Given the advantages and disadvantages of the two storage technologies, Advantech strongly recommends the use of flash memory – with the proper safeguards and correct implementation. With those steps taken, flash memory is the better choice because it is less expensive in the long run and more reliable.

For industrial applications, a best practice dictates that two flash memory sockets be used. There are several reasons for this. One has to do with reliability and overall system uptime. While the operating system will probably undergo little or no change, the same can’t be said for process data. That information will be logged on a regular basis, therefore modifying stored data at set intervals. The amount of data may be small but there will be continuous writing of bits. Applications, for the most part, will fall somewhere between these two extremes.

By having two flash memory sockets, the load can be split, with OS and applications on one and the process data on the other. The write load, therefore, will be confined to one, with that chip replaced on some preventive maintenance schedule.

There’s another reason for separating OS, applications, and data. Although rare, flash memory can be corrupted. One way the problem arises is when power fails during a write cycle, which cannot be completed without power. Corruption isn’t much of problem in commercial settings, but the loss of power – and the chance of corruption – is much more likely in industrial settings. Having two flash memory sockets with one dedicated to storing process information ensures that the writing of process data can’t scramble the OS or application.

For that reason, Advantech also recommends the use of the latest OS enhancements, such as the enhanced write filtering (EWF) found in Microsoft’s Windows XP and later operating systems. Improvements like EWF are specifically designed to increase the reliability of flash memory operation. Similarly, it is good idea to use flash that mimics a non-removable hard disk because this ensures that write operations complete and corruption chances are minimized.

Recommendation: Chose Suppliers Carefully

Another important point is that not all flash memory is created equal. For example, one way that suppliers boost capacity is by storing more than one bit per memory cell. This approach increases capacity but degrades reliability since the on-chip circuitry must now distinguish between multiple closely spaced charge values. A better solution for industrial applications is to go with a single bit per cell, but that will necessarily result in more expensive flash memory.