• Thin substrate so that it takes up less space

• Light substrate so that it requires less power to spin

• High rigidity for low mechanical resonance and distortion under high rotational speed; needed for servo to accurately follow very narrow tracks

• Flat and smooth surface to allow the head to fly very low without ever making contact with the disk surface

• High coercivity (H_c) so that the magnetic recording is stable, even as areal density is increased

• High remanence (M_r) for good signal-to-noise ratio

• A square hysteresis loop for sharp transitions (to be discussed in Section 17.2.3)

Substrates

Up until the 1990s, aluminum was exclusively the material of choice for the hard disk substrate. More recently, electroless nickel-phosphorus plated aluminum has been used, which provides a surface capable of being polished to a high degree of smoothness. It is still the most common type of substrate in use today, mainly because of its low cost. However, there is a limit to how flat its surface can be polished, and the softness of aluminum makes it susceptible to accidental head slaps.

Glass and ceramics have been considered for many years as substrate materials, but only in recent years has glass been adopted in real products. The brittleness of glass has become less of a concern as disk diameter shrinks. Though costing more than aluminum, its ability to be polished to a very fine surface finish makes it an attractive alternative. Additionally, its hardness makes it less susceptible to head slap, and it is therefore a better choice for mobile applications. Its higher tensile strength also allows disks to be thinner.

Magnetic Layer

For the first 25 years of disk drives, what is known as particulate media was used exclusively in disk drives. With this type of magnetic coating, dispersions of magnetic particles in organic binders of polymer resins and solvents are essentially spray-painted onto a rotating substrate. Any excess “paint” is spun off by the rapid spinning of the disk, resulting in a thin film of polymer-magnetic particles. Before the film is dried, a magnetic field is applied to align the particles circumferentially to enhance their magnetic properties. Baking in an oven bonds the film onto the substrate. Finally, buffing and applying a coat of lubricant finish off the steps in creating a particulate media disk.

The most commonly used magnetic material in particulate media is particles of gamma ferric oxide, which are acicular in shape. Later developed particles include cobalt-modified gamma ferric oxide, chromium dioxide, metal particles, and barium ferrite. However, such newer particulate developments are applied more to flexible substrates, as thin film media made particulate media obsolete for hard disks.

With thin film media, a thin layer of magnetic metallic thin is deposited and bound directly onto the substrate without the use of polymer. The clear advantage of thin film media over particulate media is that more magnetic material is available as it is not being diluted by a non-magnetic binder. This enables a thinner layer of magnetic material to be used, which is a good thing because that allows narrower transitions to be written. Narrower transitions mean higher areal density.

Early methods of making thin film media include electroplating, chemically plating, and depositing onto the substrate by heating the desired material in a vacuum. The current method of production uses sputtering. Sputtering is performed by applying a high voltage across a low-pressure gas (argon is universally used), resulting in a plasma of electrons and argon ions in a high-energy state. The target magnetic material is bombarded by the energized argon ions accelerating toward it, displacing its surface atoms. The atoms are ejected with enough velocity to travel to the substrate and bond with it, resulting in the film of magnetic material on the substrate.

Disk Structure

A magnetic disk actually consists of more than just a substrate and a magnetic coating. Figure 17.9 shows a cross-section view of a typical thin film disk using an aluminum substrate. Starting from the bottom, above the substrate is a sublayer of nickel-phosphorus, which provides a much harder surface than aluminum, allowing it to be polished to a higher degree of smoothness and at the same time presenting a more damage-resistant surface. Next is an underlayer of chromium. Its purpose is to set up a desirable microstructure that the magnetic layer material is to replicate when deposited on this underlayer. The magnetic layer invariably uses some cobalt alloy, as cobalt alone does not have sufficient coercivity. Chromium is quite commonly used as one of the alloy metals. The thickness of this layer and grain size have a strong effect on coercivity and squareness. Grain size is influenced by rate of deposition, temperature of substrate, and other factors.

A layer of exposed magnetic material would leave it unprotected from potential scratches and corrosion. Therefore, a layer of wear-resistant overcoat is needed. The material for this layer should, naturally, be hard but not brittle, and it should be chemically inert but bond well to the magnetic layer. Today, the most common overcoat material is hard carbon, sputtered onto the disk. Finally, a very thin layer of lubricant is applied for reducing any wear and friction between the head and the disk.

• High reliability. If the motor does not spin properly, the disk drive is dead. Therefore, the motor must be able to run for many years and be able to go through tens of thousands or even hundreds of thousands of start/stop cycles.

• Low vibration. Vibration affects the ability of the head to stay in a stable position relative to the platter, thus impacting performance. This can also affect areal recording density.

• Minimal wobble. Wobble in the disk drive is known as Non-Repeatable Runout (NRRO). NRRO is a main contributor to track misregistration (TMR).¹ Besides impacting performance, it is also an inhibitor to increasing tracks per inch (tpi), as it prevents tracks written to the disk from being perfectly circular.

• Low power consumption. Power causes heat, which must be dissipated or else it will shorten the life of the disk drive. For mobile applications, low power consumption is also important for making a battery last longer.

• Low acoustic noise. This is especially important for disk drives used in consumer electronics.

• High shock tolerance. This is true of every component in the disk drive, especially one used in mobile applications.

Bearings

Bearings are needed to both support and separate the spindle hub from the stator shaft so that the spindle can rotate smoothly and quietly. For many years, time-tested metal ball bearings confined to raceways inside the spindle were used in spindle drive motors. However, there is limit as to how perfectly round these tiny balls can be manufactured. The raceway in which these ball bearings run is not 100% perfect either. As mentioned, NRRO is a significant contributor to TMR. Spindle motors with ball bearings have an NRRO in the 0.1-microinch range. Additionally, materials in high-speed contact are subject to invariable wear and tear and are dependent on the age and durability of the lubricant protecting them. As the rpm of spindle drive motor increases over the years, heat and noise issues plus these limitations of ball bearing motors become more and more apparent. Figure 17.11(a) is a cross-section drawing of a ball bearing motor.

In recent years, disk drives have started to transition to spindle motors using fluid dynamic bearing (FDB). With FDB, the ball bearings are eliminated and replaced by a very thin layer of high-viscosity lubrication oil trapped in a carefully machined housing. Without the ball bearings and with the damping effect of a lubricant film, the FDB spindle motors run more quietly, showing roughly a 4-dBA decrease in acoustic noise. Furthermore, wobble is also greatly minimized, with NRRO in the 0.01-microinch range, which is an order of magnitude improvement over ball bearing spindle motors. Non-operational shock resistance is also improved due to increased area of surface-to-surface contact, while the lubricant film provides additional damping also. Initially, FDB motors will cost more. However, as production volume increases and manufacturing techniques mature, they will become cheaper to build than ball bearing motors as fewer parts are required. Figure 17.11(b) is a cross-section drawing of an FDB motor.

Write Heads

The principle governing the operation of today’s write heads is exactly the same as that for the earliest recording heads [Robertson et al. 1997]. Nonetheless, the dimension, the geometry, the materials used, and the fabrication process have all evolved substantially over the years. While it is hard to tell by looking at the size and shape of today’s write heads and those of the RAMAC 350, all write heads are basically inductive ring heads.

The inductive write head basically consists of a ring core of magnetically soft material, such as ferrite, and a coil of wire wrapped around the core. There is a break in the core, forming a very short gap. The head “flies” very closely over the magnetic recording media, with the gap adjacent to the media. When an applied current passes through the coil, it induces a magnetic field inside the core, just like electromagnets, and hence the name inductive head. The direction of the magnetic field is dependent on the direction of the current in the coil. The two ends of the core at the gap form two magnetic poles of opposite polarities. At this gap, the magnetic flux leaks outside the core and fringes away from the gap. The leaked magnetic flux passes through the media, magnetizing the magnetically hard material of the media in accordance to its hysteresis loop characteristics and the amount of magnetic flux applied. This is illustrated in Figure 17.12. The space within which the magnetic field of the write head is strong enough for magnetic record to occur is sometimes called the write bubble. The magnetically hard media retains most of the magnetization after the magnetic field of the head is removed. This basic write head structure is used for both analog and digital recording.

In digital magnetic recording, the head/media/channel are designed such that when current is applied to the head, the resulting magnetic flux going through the media is of sufficient strength as to align the magnetic domains immediately adjacent to the head completely in the same direction as the applied magnetic field, regardless of the previous orientation. This is saturation magnetization. Since the magnetic media is moving under the write head due to the rotation of the disk platter, saturated magnetization occurs along the path (called track in recording terminology) that passes directly underneath the head, as long as the current is applied to the coil. When the current is switched off and then a current in the opposite direction is applied, the magnetic flux in the core reverses direction, and saturated magnetization in the opposite orientation as the previous magnetization now occurs in the magnetic media. This creates a transition in the media where magnetization changes from one orientation to the opposite orientation. This is shown in Figure 17.13. The transition is, of course, not a sharp line as drawn, but has a finite length to it, as the dipoles within it gradually change from predominantly pointing in one direction to predominantly pointing in the opposite direction. Media with a more square hysteresis loop will create sharper transitions. This transition length determines how closely the transitions can be spaced, which, in turn, determines the linear recording density. The sequence of pictorials illustrating the write process in Figure 17.6 can now be depicted with a write head as the source of the recording magnetic field, as shown in Figure 17.14. The size of the recorded bits relative to the head and its gap is not properly drawn to scale; in reality, the bit size is much smaller and closer to the dimension of the head’s gap. Writing, i.e., the creation of transitions, actually takes place at the trailing (left-hand side in Figure 17.14) side of the write bubble, as that is where the medium sees the final orientation of the magnetization field applied to it.

The desirable characteristics of the write head are the following:

The first heads for the RAMAC 350 used permalloy (an alloy of nickel and iron) as the core material, and the coils were hand wound and assembled by former watchmakers. Needless to say, they share little of the above list of desirable characteristics. The shape and dimension of write heads have greatly evolved since then. Ferrites, which are ceramic-like compounds of iron oxides and other metallic oxides, replaced permalloy as the core material during the mid-1960s. While having other desirable characteristics, ferrites suffer from having a much lower saturation flux density than permalloy. By the mid-1980s, they simply did not have enough magnetic flux to work on the high coercivity of thin film disks. The solution was to add (sputter on) a thin layer of metallic material, commonly sendust which is an alloy of iron, aluminum, and silicon, to the poles near the gap. Hence, this type of head was called a metal-in-gap (MIG) head. The result is a head that has the high saturation flux density of sendust and the other desirable characteristics of ferrites.

MIG heads were gainfully used from the late 1980s to the early 1990s. However, the way they were manufactured, which involved mechanically grinding and polishing the ferrite material and actual winding of the copper coil, placed a limit on how small the dimensions of the head could be reduced to. Miniaturization of the head was the key to increasing areal density. With research starting in the 1960s, by the early 1980s, a completely new technique for manufacturing the heads was perfected. This new fabrication method leveraged on the tools and thin film technology of semiconductor manufacturing. By using lithography to define the features of the head and thin film deposition to construct all its components, including the core, the gap, and the copper windings, a miniature head with well-controlled dimension could be easily manufactured. Thin film heads, as these heads came to be called, were introduced in the IBM enterprise class of disk drives in 1980 and eventually became economical for all disk drives by the mid-1990s.

While the fabrication process of thin film heads differs from that for the ferrite and MIG heads, not to mention their miniaturized dimensions, functionally, they are really just like ferrite heads. Structurally, they are essentially three-dimensional inductive heads collapsed front to back to become almost just two-dimensional. Even the core material has gone back to permalloy, the original material used in the RAMAC 350 heads. Figure 17.15 illustrates conceptually the typical construction of a thin film head, fabricated using semiconductor technology. While the front-to-back spacing is much more compressed, and the coil takes on a flattened arrangement, functionally, it is exactly an inductive head.

Read Heads

For reading in digital magnetic recording, the read head detects the transitions that are recorded in the media. Transitions detected in the sync field (to be described in more detail in Chapter 18, Section 18.1.3) are used to establish the clock (self-clocking) for data reading. In the data field, each transition is used to represent a binary “1,” while the absence of a transition in a clocked period represents a binary “0,” as discussed earlier in this chapter. The read channel samples at each clock interval the signal coming out of the read head to check for the presence or absence of transition.

Up until the introduction of the magnetoresistive head in 1991, the inductive head was used also for reading, in addition to its function as the write head. It operates on the well-known magnetic phenomenon that a change in magnetic flux inside a coil of wire would induce a voltage in that coil. In fact, Faraday’s Law provides that the induced voltage is

(EQ 17.3)

where N is the number of turns in the coil. While used as a read head, no current is applied to the coil by the drive’s electronics. As the media moves next to the head, its core collects the magnetic flux emanating from the recorded magnetization on the media. When it comes across a transition, a magnetic flux change occurs as the magnetic field reverses polarity, and this flux change induces a voltage in the coil which can then be detected by the read channel circuitry. Figure 17.16 shows the read voltage signal for the sample sequence of “101” of Figure 17.14. Equation 17.3 suggests that having more turns in the coil would increase the read signal; however, making N too large would increase the inductance of the head, limiting its high frequency, and hence high data rate, handling. Also, because the inductive head operates by detecting the rate of flux change, the higher the velocity is, the stronger the read signal. Thus, higher rpm and larger disk diameter are both favorable for inductive read heads.

While inductive read heads operate by detecting flux change, magnetoresistive (MR) heads operate by sensing the flux directly [Tsang et al. 1997, 1994]. There is a class of materials, of which permalloy happens to be one, which exhibits the phenomenon of magnetoresistance. Basically, when a magnetic field H is applied to such a material, its electrical resistance changes as a function of the angle θ between the resultant magnetization M and the direction of flow of an applied electric current, as shown in Figure 17.17, in accordance to

(EQ 17.4)

where ΔR is the change in resistance, R is the sensor’s nominal resistance, and C_MR is the magnetoresistance coefficient of the material. For permalloy, C_MR is about 2–3%.

By applying a constant current I to the MR sensor, the presence of a magnetic field is detected by a change in the voltage across this sensor in accordance with Ohm’s Law,

(EQ 17.5)

as a result of the change in the direction of magnetization due to the applied field H. Because a magnetic field is applied during fabrication of the sensor to give it an easy axis parallel to the current direction (θ = 0°), the biggest change in resistance will be when the external magnetic field H is in a vertical direction, either up or down, perpendicular to the current direction (θ = 90°), as shown in Figure 17.17. The biggest resistance change generates the biggest signal. Such vertical magnetic fields occur on the media’s recorded surface only at the transitions, where magnetic fields emanating from both sides of the transition are both either going up or going down, as can be seen in Figure 17.7. Common practice is to bias² the MR sensor so that θ = 45° when no external magnetic field is present. As a result, the MR sensor generates up or down peak ΔV signals at transitions, with a waveform similar to the one shown in Figure 17.16 for inductive read heads. In order to prevent the MR sensor from detecting stray magnetic fields from adjacent transitions, the sensor is shielded front and back so that it can only “see” the magnetic field right underneath it.

The signal that can be created by an MR head is several times larger than that of an inductive read head. This is what drove the rapid increase in areal density after the introduction of the MR head in 1991. In fact, by increasing the sense current I, a bigger signal can be obtained, as Equation 17.5 indicates. However, heat and other heat-related issues limit the amount of current that can be practically applied. With no coil involved, the inductance of the MR head is so low that there is practically no limit to its high-frequency performance. Finally, since the MR head detects magnetic flux rather than the rate of change of flux, its signal is independent of the velocity of the media going by, unlike an inductive read head. This attribute is what makes small diameter drives with low rpm possible.

The MR sensor is constructed of a homogeneous ferromagnetic material such as permalloy. Around the early 1990s, researchers discovered that by constructing a composite sensor with multiple thin layers of different ferromagnetic and anti-ferromagnetic materials, made possible by the molecular beam epitaxy process, a sensor with giant magnetoresistance (GMR) results. An example of such a structure, as illustrated in Figure 17.18, consists of a “free” ferromagnetic layer and a “pinned” ferromagnetic layer separated by some non-magnetic material. The magnetization orientation of the pinned layer is fixed (pinned) by an adjacent layer³ of anti-ferromagnetic material (AFM). The GMR head was first introduced in a product in 1997.

Read/Write Heads

The transducers for reading and writing are oftentimes referred to collectively as the read/write head, or just simply the head. This is probably carried over from the earlier days when the inductive head was used for both reading and writing. Therefore, there is truly only one head. Requiring the same head to perform the read and the write functions means that compromise in design and performance would be necessary, as the optimum design for writing differs from that for reading.

Since the introduction of MR heads, the “head” is now really composed of an inductive write transducer and the MR (or GMR) read transducer. Having a separate read head and write head allows each design to be independently optimized for the function it needs to perform. The inductive head and the MR head are placed in tandem, with the write head going behind the read head, as illustrated in Figure 17.19. Standard practice today is for the write head to write a wider path (track) than the width of the read head, a technique called “write wide, read narrow,” by constructing the MR read sensor to be narrower than the width of the write pole tip, as shown in Figure 17.20. The advantage of such an approach is that the read head does not have to be perfectly centered over a track in order to have good read-back and not pick up noise from adjacent tracks. This also permits tracks to be placed closer together, thus increasing tpi.

In Figure 17.19, the width of a track W is determined by the width of the tip of the trailing pole of the thin film head. The distance between the center of two adjacent tracks is called the track pitch. It is different from the width of a track because standard recording technique places a guard band between two tracks. The guard band provides protection against stray magnetic fields fringing from the sides of the write head which can partially erase adjacent tracks. It is this track pitch, rather than the track width W, that determines the tpi:

tpi = 1/track pitch = 1/(W + guard band width) (EQ 17.6)

but track pitch is pretty much dominated by track width. The distance B between two back-to-back transitions defines the flux change density, fcpi:

fcpi = 1/B (EQ 17.7)

The actual linear bit density depends on the density ratio (DR) of data bits per flux change of the modulation code being used to encode the data bits (see Section 17.3.3):

bpi = DR × fcpi = DR/B (EQ 17.8)

However, many of the newer codes have a value quite close to 1 for DR, so it is not too far off to refer to B as the bit length. Historically, W has always been much larger than B. However, as areal density has increased over the years, tpi has been going up faster than bpi. As a result, the ratio of W to B has been dropping. Today, this ratio is close to 4:1.

The location of the outermost track that the write head can write is commonly referred to as the OD (outer diameter), while the innermost location is referred to as the ID (inner diameter).

• A smaller slider means the suspension can also be made lighter. This combined reduction in mass reduces the power required for positioning of the heads and improves seek time.

• Reduction in mass also improves shock resistance and head-disk reliability.

• Smaller slider size makes it easier to respond to unevenness of a disk’s surface, hence reducing the variation in flying height.

• Because a slider cannot fly outside of the disk’s surface, having a smaller slider means more of a disk’s surface area is usable, and thus, storage capacity is increased. This is especially important for smaller diameter disks since a larger percentage of the disk becomes available.

• Smaller size means more sliders can be produced from a same size wafer, and thus, production cost is reduced.

• Simpler design, and, hence, lower cost

• Smaller and lighter, which means faster seek performance and lower power consumption

• The seek stroke is longer, resulting in higher average seek time.

• The arms are longer, which, in turn, requires it to be thicker to provide the necessary stiffness. The result is an actuator with more mass, which means either the actuator will move slower (longer seek times) or more power will be needed to achieve the same performance as for a smaller diameter disk.

• A larger surface area means there is more air friction (windage). This translates to a greater power requirement to spin the disk at a given rotational speed.

• To attain the necessary stiffness so the disk does not sag, the thickness of the disk platter must increase proportionally to the increase in diameter. The total increase in weight from both a bigger diameter and more thickness means that even more power is needed to attain a given rotational speed.

• The seek stroke is not increased, and the added weight to an actuator for having additional HGAs is minimal. Thus, seek performance is not affected.

• While the added platters do increase the total weight that the spindle motor must spin, it is less than the increase from a larger diameter disk, as the smaller diameter platters are thinner while providing the same rigidity.

Contact Start/Stop

In this method, the head comes to rest on the disk surface as there is no longer an air bearing to support it [Bhushan et al. 1992]. When the disk starts spinning again, the head slides in contact over the disk surface until gradually an air bearing is formed under the slider and the head is lifted off the disk surface. With today’s high areal density, it is not a good idea to park the head on the portion of the disk where data is recorded, as the data with its tiny bit size can easily be damaged. Therefore, a designated landing zone is reserved on the disk surface for the head to park. While the landing zone can be placed outside either the OD or the ID, there are couple of reasons why placing it outside the OD is a bad design choice.

For the above reasons, the landing zone is placed at the inside of the disk. To further reduce the effect of stiction, the landing zone is textured so that the total surface area in actual contact is reduced. During a normal shutdown, the disk drive controller firmware would instruct the actuator to position itself over the landing zone before spinning down the disk. To guarantee that in the event of an unexpected power loss the head still gets properly stowed away in the landing zone, some sort of auto-park feature is needed. Some older drives use a weak spring to pull the actuator to the landing zone. Newer drives use the kinetic energy of the spinning platters to generate back-EMF (electro-magnetic force) power from the spindle motor to move the actuator to the landing zone.

Load/Unload

This technology was previously used in large diameter disk drives and is now being re-introduced into small form factor drives. Prior to spinning down, the actuator moves off the disk platters. A lift tab at the front of the suspension of each HGA slides up a ramp structure and comes to a rest at a detent, as shown in Figure 17.26. The heads are now safely off the disk surfaces, and only then are the disks spun down. To start up the disk drive, the disks are first spun up until a sufficient rotational speed is reached, and then the actuator is moved off the ramp; the heads will have enough air cushion for them to fly as soon as they reach the disk surfaces. With this technique, intentional contact between the head and disk surface is eliminated.

Just as in CSS, a self-parking mechanism is needed to ensure that the heads are safely unloaded up the ramp in the event of a power loss. Again, this can be accomplished by extracting energy from the spinning disks by using a circuit to apply current from the spindle motor back-EMF to the actuator.

Since there is no intentional contact of the head with the disk, reliability of the disk drive is improved as there is less likelihood of head or media damage. It also means the drive can handle more stop/start cycles in its lifetime than CSS. Additionally, because the heads are completely moved off the surfaces of the disks, the whole disk drive is more shock tolerant than drives using CSS. This makes load/unload especially attractive for mobile applications. The downside of load/unload is that it requires a loading zone on each disk surface outside the OD for the head to load onto the disk surface, and no data is stored in this loading zone. This loading zone takes up a lot more space than the landing zone of CSS, which is at the outside of the ID.

Tracks per Inch

As discussed in Section 17.2.3 and stated in Equation 17.6, tpi is the reciprocal of the track pitch, which is the distance between the center of two adjacent tracks. The track width accounts for much of the track pitch, and the track width is basically dictated by the width of the pole tip of the inductive write head. Hence, the width of the write head is a key factor in determining tpi.

While the head may be able to write very narrow tracks, other components of the disk drive must also be able to support a high tpi. The actuator and the servo system controlling it need to be able to both position the head accurately on the center of a track and maintain this centered position while the head is accessing the data on the track. In addition, while the guard bands on both sides of a track allow for a small amount of deviation, such deviation must be kept small. If a normal distribution is assumed for the position of the head with respect to the track center, a value of 3 sigma (representing a range with 99.7% probability) is designated as the TMR of the drive. Clearly, as track pitch decreases, TMR must decrease proportionally if the head is to write and read data properly. NRRO of the spindle motor (discussed previously), mechanical vibrations coming from sources both inside the disk drive and external to the disk drive (e.g., another neighboring disk drive), disk flutter, and even electronic noise in the servo control circuits all contribute to TMR.

Bits per Inch

The linear recording density bpi is predominantly determined by the head and the media. As given by Equation 17.8, bpi is inversely proportional to the distance B between two back-to-back transitions. Thus, bpi depends on how closely transitions can be placed one after the other. The ability to place two transitions closely without the detected pulse signals from them (Figure 17.16) overly interfering with each other⁶ depends on the width of each pulse—the narrower the pulse width the better. The standard measure of pulse width in magnetic recording is the half width, called PW50, which is measured at the 50% amplitude (see Figure 17.27).

Without derivation here, a number of analytical studies have shown that PW50 can be approximately represented by

(EQ 17.9)

where a is the transition parameter (transition length), d is the magnetic separation, g is the half gap in MR head (or pole gap in inductive head), and t is the thickness of media (as shown in Figure 17.28). It should be noted that the magnetic separation d is not the same as the flying height of the slider. Magnetic separation is the spacing between the tip of the sensor and the top of the recording media; so it includes the flying height, and the thickness of the lubricant, and the thickness of the carbon overcoat (on both the media and the sensor).

The transition parameter a can be approximately modelled with the following equation:

(EQ 17.10)

where M_r and H_c are the remanent magnetization and coercivity of the medium, respectively, and K is some constant. From Equation 17.9, it can be seen that PW50 can be decreased (which means bpi can be increased) by reducing the MR sensor gap g (miniaturizing the head), the magnetic separation d (lowering the flying height which is the major component), the thickness of the media t, and the transition parameter a. Furthermore, according to Equation 17.10, the transition length a itself can also be shortened by reducing d and t. Selecting a medium material with high coercivity will also help to reduce a, but it requires more current in the inductive head for writing. While a medium with low remanence will also help to make the transition parameter small, it must produce a sufficient magnetic field for the read sensor to detect.

Because reducing the magnetic separation is one of the most significant factors in increasing areal density, flying height has continuously been decreasing over the years. When the RAMAC 350 was first in troduced in 1956, the head-disk separation was 800 microinches. Today, the flying height of a typical disk drive is well below 1 microinch. To put things in perspective, the thickness of human hair is about 3000 microinches, and the size of a typical dust particle is about 1500 microinches.

Tribology

Tribology⁷ is the study of interaction of surfaces in relative motion, in this case between the head slider and the disk. While tribology is of great importance for CSS, even during the normal flying of the head, interactions between the slider and the disk surface can still occur. This is because asperities on a disk surface are inevitable and can result in contact with the head. Hence, tribology is an important aspect of disk drive design, even for drives using the load/unload technology which theoretically eliminates all intentional contact between the head and the disk. In addition to potentially causing wear due to physical abrasion (which, in turn, can also create contaminants), contact of the head with asperities on a disk surface raises the temperature of the MR head, resulting in temporarily changing its resistance. This phenomenon, known as thermal asperity, manifests itself as severe noise in the read signal. Thus, a high enough flying height is needed to avoid the surface asperities. Unfortunately, a higher flying height is not good for areal density, as discussed previously. This is why tribology is so important in disk drive design, as it holds the key to how low the head can fly over the media and hence dictates what areal density can be achieved. Surface texture, overcoat material, and lubricant material are all part of the tribology domain.

• Receive commands (I/O requests) from the user, schedule the execution of the commands, and report to the user when a command is completed

• Manage the cache

• Interface with the HDA, including telling it where to seek to and which sectors to read or write

• Error recovery and fault management

• Power management

• Starting up and shutting down the disk drive

Processor—A microprocessor is commonly used to execute the above list of functions and to carry out any related computations. The speed with which those functions can be performed not only depends on the speed and power of the microprocessor, but also on how much of a particular function is automated with hardware by one of the other sub-components. For instance, cache function can be completely handled by the processor firmware, or it can be fully or partially automated as part of the memory controller.

ROM—The executable code of the microprocessor resides here. Optionally, only the boot code for the processor is stored here, with the remainder of the controller code read in from a reserved area in the disk. During development of the disk drive, E²PROM can be used instead. Executable code may be copied to SRAM for faster access by the processor.

Memory Controller—This can be as simple as just an interface to some DRAM, or it can include a cache manager performing cache space allocation and cache lookup. Cache is an important subject and will be covered in Chapter 22, “The Cache Layer.”

Host Interface—This is the physical link of the disk drive to the host system it is attached to. It provides the necessary interface hardware (such as specific registers), performs the handshakes of the interface’s protocol, and exchanges data with the user over this link. The host-disk interface will be discussed in greater detail in Chapter 20, “Drive Interface.”

Data Formatter—Its function is to move the data from the memory, partition it into sector size chunks, and route it to the ECC and CRC logic.

ECC and CRC Encoder/Decoder—Error correcting code (ECC) and cyclic redundancy checksum (CRC) are added to each sector of data to be stored together in the media. They are used to provide data integrity and fault tolerance. More detail will be covered in Chapter 18, “The Data Layer.”

1. Part of the memory is used as a scratch-pad for the controller. Each time the drive is powered up, it loads from protected areas in the disk operational tables and parameters onto the memory. Examples include the zoning table (discussed in Chapter 18) which is generally the same for all the drives of a given model, defect maps and no-ID tables (both discussed in Chapter 18) which are drive specific, and various parameters (many of them related to the operation of each individual head) that were self-calibrated during final manufacture testing. Memory is also used for run-time applications, such as for storing the queue of commands received from the host system.

2. The media data transfer rate is different from the data rate of the interface of the disk drive. Part of the memory is used for the purpose of speed-matching between these two different data rates. For instance, during a write operation, data from the host must be buffered in the memory and streamed out to the write channel at the media data rate. For a read operation, data from the media can be temporarily stored in the buffer memory even if the host or the drive interface happens to be busy. Without a buffer, write data may not be available, and read data may have no place to go right at the particular moment the head is ready to write or read that piece of data. The drive will then have to wait for one full disk revolution before it can attempt the same data access again, a condition commonly referred to as “missed rev.” The buffer is therefore essential in eliminating, or at least greatly reducing the number of, such performance-robbing missed revs.

3. Today’s disk drives all have some memory for cache. Caching is an important aspect of disk drive performance and will be treated separately in Chapter 22.

Write Channel

The write channel is the set of circuits that transforms the user data from its binary logic format into actual currents being sent to the write head. When writing data, the write channel supplies current to the write head, reversing the direction of current flow whenever a magnetic transition is to be created in the media. However, the 1’s and 0’s of user’s data are not directly recorded as is. Rather, some modulation code is employed to ensure that the sequence of bits recorded and later retrieved satisfies certain desirable properties. Such properties include the following:

The coding method of using a transition to represent a 1, as discussed in Section 17.1.5, is known as nonreturn to zero inverted (NRZI) encoding. NRZI encoding places no limit on the number of consecutive 0’s that can appear. One of the earliest methods introduced to prevent long runs of 0’s in recorded signal is the FM (frequency modulation) encoding. In this scheme, a 1 is simply inserted after every data bit. For example, the data sequence 10001 would be encoded as 1101010111. FM is a rather poor encoding technique as the coding efficiency is only 50%. A slightly improved modified frequency modulation (MFM) scheme is to encode 1 into 01, while a 0 is encoded into10 if the preceding bit is 0 and into 00 if the preceding bit is 1. Hence, the sample sequence of 10001 would be encoded as 0100101001. While the code rate for MFM is still only 50%, because two 1’s are separated by at least one 0, higher recording density can be achieved (assuming recording density is limited by the minimum distance between two transitions).

Modulation coding has gone through several major changes and many evolutionary changes over the years. FM and MFM are actually special cases of run-length-limited (RLL) codes. RLL code is specified as m/n(d,k), where m is the number of data bits encoded into n code bits, d is the minimum number of 0’s required between two 1’s, and k is the maximum number of 0’s allowed. Thus, FM is a 1/2(0,1) RLL code, and MFM is a 1/2(1,2) RLL code. The DR is given by

(EQ 17.11)

For example, the 8/9(0,4) RLL code has a DR of 8/9, which makes the achievable bpi with that code 0.89 that of the fcpi of what the head-disk technology can deliver. Newer and more efficient codes have been developed in recent years, and multiple choices are oftentimes offered by suppliers of channel chips for disk drive designers to select the one that meets their needs.

The k parameter of RLL code is determined by how long the clocking can go without resync. The d parameter is a function of how well the channel can handle inter-symbol interference. Inter-symbol interference (ISI) occurs when the transitions are packed so close together that the detectable signal from a transition is distorted by those of its neighboring transitions. The pattern of ISI depends on the particular recorded bit sequence.

The simplest example would be of two 1’s in the midst of a string of 0’s. Figure 17.31(a) shows that when the two transitions are placed far enough apart, there is no interference of one pulse with the other. When the two transitions are packed closer together to increase bpi, the two pulses start to cancel each other. As a result, the amplitudes of the two pulses are reduced, and, more importantly, their peaks are shifted, as shown in Figure 17.31(b). The first peak arrives early, while the second peak arrives late. As the read process of today’s disk drives is done synchronously based on a fixed clock rate, a shift causes the peak to be out of step with the clock signal and can potentially lead to read error. Since ISI is the unavoidable result of packing transitions more densely in order to increase bpi, multiple techniques have been developed to deal with ISI. One such technique is to precompensate the write signal delivered to the write driver. For the simple peak shift example of the double transitions, precompensation would delay the first transition and move up the second transition. Write precompensation is sometimes referred to as write equalization.

Figure 17.32 is a functional block diagram of the sub-components of a write channel. The write driver, by means of a simple flip-flop circuit, uses a current source to apply a voltage of the right polarity, in accordance to the write signal provided by the write precompensation circuit, to the inductive write head.

However, before it can do that, the write driver must be enabled by three control signals:

All three conditions must be met all the time during the write for the write driver to apply current to the head.

Read Channel

The read channel is a series of circuits that converts the analog signals sensed by the read head back into the original user data in their binary logic format. Figure 17.33 shows a block diagram of the major components of a typical read channel. With today’s microscopic bit size, the signal picked up by the GMR head is very small, well below 1 mV. The first order of business is to greatly amplify this signal as soon as possible before it picks up more noise. This is handled by the preamp which is a differential amplifier located in the AEM. For the preamp of a read head to go active, the controller must first signal that it wants to do a read to the read/write selector, and the head must also be selected by the decoder based on the controller’s selection of which head to use. When enabled, the selected head’s AEM module applies a sensing current to its MR sensor.

The amplified signal from the AEM is passed on to an automatic gain control (AGC) circuit to adjust the peak voltages of the signal to within the range required by subsequent circuitries. A low-pass filter following the AGC reduces high-frequency noise. The equalizer is a circuit which slims the pulses, that is, makes the shape of the pulses narrower. This is another technique employed to combat ISI, in addition to write precompensation described previously.

The conditioned signal is now ready for the detection circuit. The detector extracts servo information (see Section 17.3.4 and Chapter 18), as well as the original recorded binary information. Servo information is fed to the actuator servo control circuit. Encoded user data is written to the disk synchronously using a clocking signal. Hence, to extract data accurately from the read-back signal, the detector must also extract the clocking information embedded with the data. The extracted clock signal is fed to a phase-locked loop (PLL) circuit to regulate the clock that the detector uses to recover data. While the frequency of the clock is adjusted gradually based on the average of several bits, adjustment for individual data pulse timing is immediate and based only on the current detected clock pulse.

Earlier disk drives used peak detection to directly look for pulses created by transitions in the media. With this technique, the pulse signal is differentiated, and the detection logic looks for zero-crossing of the differentiated output as where a peak has occurred. In 1990, the IBM 0681 disk drive introduced PRML (partial response maximum likelihood) as a new method for data recovery. While known to the data communication field for a long time, PRML was a breakthrough technology for magnetic recording. It is a very effective technique for dealing with ISI and permits denser packing of transitions than peak detection. There are two key aspects to PRML. First, the signal is sampled, which is called a partial response signal. Second, instead of decoding each pulse individually as in peak detection, it decodes a short sequence of bits in a sliding window. For each sequence of sampled signal, it looks for the bit sequence which is most likely (hence, maximum likelihood) to generate that sampled sequence.

Regardless of the data detection method used, the last stage of a read channel is to run the data output of the detector through a modulation code decoder to recover the original user data and ECC/CRC information.