MR AND PRML: TECHNOLOGIES IN SYNERGYHow Advanced Head And Read Channel A Quantum White Paper |
Sections:
Today’s desktop computer and workstation users face an unprecedented sea of data. Some have called it a tsunami of information. New multimedia playback and authoring capabilities, video applications, the downloading of massive volumes of informat ion from the Internet and other networks — they are all driving the need for higher-capacity, higher-performance hard disk drives. In response, the storage industry has begun a shift to new advanced drive technologies to meet these needs, while conti nuing to provide lower costs per megabyte (MB).
Magnetoresistive (MR) head technology and Partial Response Maximum Likelihood (PRML) read channels are the two most significant solutions now being employed to increase areal density (bits of data per square inch) and boost performance. Alone, each de livers substantial improvements in certain areas over traditional drive technologies, such as inductive heads and peak detection read channels. Together, they reduce the need for many of the capacity and performance trade-offs inherent to disk drive desig n, while accelerating the decrease in costs per MB.
MR heads and PRML read channels were first used together in 1990 in large-scale storage systems from IBM. Quantum was the first independent disk drive supplier to combine them in 2.5-inch drives for laptop computers earlier this year. They are now be ing implemented for the first time as complimentary technologies in the desktop computer arena, working together to make Quantum’s 2.5 GB Sirocco 2550 drive the highest-capacity desktop drive available.
Before looking at the synergy of MR heads and PRML read channels, it’s important to understand the advantages each technology contributes by itself.
The most economical and practical method for increasing hard disk drive capacities is to increase areal density — fit more bits of data onto the surface of the disk, as opposed to adding disks and heads to the drive. But as density increases, the bit patterns recorded on the disk necessarily grow smaller. This weakens t he signal generated by traditional inductive technology read heads, making it difficult to properly identify the patterns.
Several methods have been used to combat this. For example, the head can be made to fly closer to the surface of the disk or the disk can be made to spin faster to increase the strength of the signal. Turns, or coils of thin copper conductors, can also be added around the head to boost the read signal (which increases proportionally to the number of turns).
Each of these solutions, however, has its drawbacks. Flying the head closer increases the risk of crashes. Speeding the disk strengthens the signal, but also increases data frequencies; and today’s inductive heads cannot perform at very high frequencies.
Meanwhile, adding turns helps with the read process but hinders the write process by limiting the frequency with which current reversal can occur for write operations.
MR heads, on the other hand, employ independent read and write elements — using an inductive element (with few turns relative to inductive heads) for write operations, and an independent magnetoresistive element for read operations. The separate read element can also be made narrower to better read tightly spaced data tracks, thus sidestepping the dangers of misalignment.
MR heads also produce a strong signal when reading extremely closely spaced bits, regardless of linear disk speed. This means that disks do not have to spin faster in order to accommodate increased density. But, if they do (in order to maximize data rates in high-performance drives), the write head can be optimized for high-frequency write operations without degrading readback performance.
PRML read channels provide another means of obtaining areal density improvements, while at the same time improving performance by increasing data transfer rates. As bit density increases, so does the possibility of inter-symbol interference (ISI). IS I results from the overlap of analog signal “peaks” now streaming through the read/write head at higher and higher rates. This has been traditionally combated by encoding the data as a stream of “symbols” as it is written, in order to separate the peaks during read operations. The problem has been that the encoding requires more than one symbol per bit, exerting a negative effect on both disk capacity and drive performance.
PRML read channels do not have to separate the peaks during read operations, using instead advanced digital filtering techniques to manage ISI. They then employ digital processing and “maximum likelihood” data detection to determine the sequ ence of bits that were most likely written on the disk.
This means that drives using PRML channels can employ a more efficient coding scheme to facilitate accurate data readback. For example, drives using traditional peak detection typically experience a ratio of user data to stored symbols of 2-to-3. Qua ntum’s third-generation PRML uses an encoding scheme that increases that ratio to 16 to 17. This simple relationship “predicts” 40 percent more capacity on the disk for actual user data, and also has a positive effect on the internal data transfer rate.
As can be seen, both MR and PRML technologies deliver substantial advantages over traditional disk drive technologies. PRML with inductive heads can deliver a 20 to 30 percent increase in areal density at almost the same cost of older technologies. Ye t, when implemented together, MR heads and PRML read channels provide complementary benefits in the areas of achieving higher densities and performance, as well as ensuring data integrity. The following sections illustrate how the two technologies play t o each other’s considerable strengths.
As we have seen, the problem with inductive head technology lies in supplying enough read signal. Adding more turns to the head and/or spinning the disk faster have proven to be limited solutions. MR technology, however, provides greater sensitivity and actually requires drive designers to decrease the applied magnetic field in order to stay within the linear range of the head. PRML technology argues for doing exactly the same — reducing the magnetic strength of the disk to avoid non-linear tra nsition shifts.
Magnetic strength is expressed by Mrt, itself the product of two parameters. The “t” represents the thickness of the media while Mr represents the media’s magnetic properties or “remanence.” Inductive head technology might re quire an Mrt of 3 units to get a good signal, while an MR head requires an Mrt of only 1. While a high Mrt provides a stronger signal, writing bits closer together in such an environment causes severe non-linear effects. Reducing the Mrt permits much sm aller spacing between magnetic transitions, enabling higher areal density.
In the past, the industry has had to continually compromise between signal strength and transition density. For example, PRML technology implemented alone can benefit from reduced Mrt. Nevertheless, Mrt has still been kept high to accommodate inducti ve heads. That compromise is now removed. An MR head intrinsically supplies more output, can fly higher and will handle higher frequencies. Meanwhile, tracks can be written closer together. With its more efficient coding, PRML technology in turn enabl es significantly higher bit densities that benefit from reduced Mrt.
MR and PRML technologies also work together to allow higher data rates, a major measure of disk drive performance. The separate read/write elements of an MR head open the door to these higher data rates. Because the write element is not burdened with extra turns (necessary for improved read sensitivity), it can write faster and higher frequencies. The independent read element possesses the sensitivity to handle smaller bits.
PRML’s contribution to faster data rates revolves around its more efficient RLL (Run Length Limited) coding scheme. The (1,7) RLL coding used by traditional peak detection read channels is slow. The “1” means that there must be at least one digital 0 between every pair of digital 1s in the data sequence, meaning at least two “symbol” periods between every pair of magnetic transitions on the disk. The “7” means there can be no more than eight symbol periods between transitions. These constraints can be met with the relatively bulky 2 to 3 ratio of user bits-to-stored symbols.
The more efficient (0,k) PRML coding used in Quantum’s Sirocco drives results, for example, in a more compact 16 to 17 ratio, with data rates of up to 72 megabits (Mb) per second. This high data rate is achieved at a clock speed of 76.5 MHz. In contrast, with a 2-to-3 code ratio, symbols would have to be written at a 108 MHz clock rate to achieve the same 72 Mb per second data rate. Not being forced to go to a higher clock rate makes the data easier to handle downstream in the read circuitry.
While PRML technology supports higher areal densities and higher data rates, it must also ensure the same low bit error rate as a peak detection read channel. PRML does this by applying digital filtering techniques to shape the analog readback signal so it can be accurately processed by the channel’s advanced sequence-detection algorithm (the Viterbi algorithm).
In a very simplified explanation, samples are taken at various points of the analog waveform, converted to digital data, and submitted for Viterbi detection. The Viterbi algorithm determines the most likely sequence of data bits written on the disk by comparing sequences of samples with sequences of possible samples.
Limitations arise when using inductive heads with PRML. The problem comes with undershoots to the waveform. These undershoots result from the outside edges of an inductive head as it passes over the disk and reads a transition. A PRML read channel requires that these undershoots be removed in one way or another. Typically, electronic equalization (in the form of undershoot filters) is applied to the problem, but it adds to the expense and complexity of the PRML channel. Alternatively, rounding the sharp outside edges of the pole tips can help minimize the size of the undershoots, but this may also add complexity and cost.
MR head technology best complements PRML in this specific regard. Responses from an MR head are rough in shape, but are relatively easy to shape for sampling. As a result of combining MR and PRML technologies, the overhead associated with undershoot filtering is eliminated — an overhead that can be expected to consume approximately five percent of a PRML chip’s area — and a more elegant, simplified solution is presented for ensuring optimal data integrity.
For many years, the disk drive industry has been able to provide a 60 percent annual improvement in areal density, thus continually delivering decreased costs per MB. Maintaining this curve will become more difficult in the future.
It is clear that traditional technologies, such as inductive head and peak detection channels, will no longer continue to meet the challenge. Without new technologies such as MR heads, maintaining the 60 percent annual improvement in capacity is unthinkable. Still, it is critical that these new technologies be implemented in such a way as to maximize their individual contributions.