Magnetoresistive heads face new dangers
Concerns grow as thermal asperities affect dynamicgain and timingcontrol functions
By Erich Sawatzky
Magnetoresistive (MR) heads are rapidly becoming the preferred read sensors for harddisk drives. But, as with all technological innovations, MRhead technology brings its own set of challenges. Of growing concern are socalled “thermal asperities” arising from the heating of the MR stripe in the recording head. The head bumping against raised defects on the disk, debris on the media — anything that heats the MR stripe — can distort or mask the output signal of the head. Everyone involved with the design or test of read channels for MR heads has had to learn to cope with unrecoverable errors caused by thermal asperities.
A thermal asperity can be defined as a defect that thermally induces a resistance transient in the MR stripe. The simplest and most common are “contactpositive” thermal asperities — “positive” denoting an increase in stripe resistance — that cause the voltage across the stripe to increase. A contactpositive pulse occurs when the MR stripe contacts a disk asperity and frictional heating momentarily raises the stripe temperature. The thermal pulse produces a corresponding resistance transient, generating a positive voltage transient (i.e., a thermalasperity signal) in the output of the MR head. In addition to the MR stripe itself, other components of the head (for example, the top and bottom magnetic shields) play an important role — by helping to dissipate heat that is generated in the MR stripe by thermal asperities.
Why are thermal asperities bad? The magnetic and thermal signals are additive in an MR head because they both result from a change in the stripe resistance. The resulting thermalasperity signal can affect the dynamicgain and timingcontrol functions in a disk drive. Also, if large enough, the unwanted signal can exceed the dynamic range of the head preamplifier — causing device saturation and signal clipping. The magnitude and duration of those effects may be beyond the range of typical errorcorrection schemes.
Thermal signatures
A typical positive pulse produced by a “sharp” or “pointed” asperity on the disk surface is shown in Fig. 1(a). A fast rise time, large peak amplitude, and approximately exponential decay characterize the signal. The rise time depends on how long the asperity contacts the stripe. For example, an asperity with a 1micron diameter top would rub on the stripe for about 50 ns at a linear disk velocity of 20 m/s. That is the time during which heat is generated in the stripe. The signal shown in Fig. 1(a) was obtained from a 90mm diameter aluminum disk spinning at 7200 rpm, and the measured rise time was about 44 ns.
The peak thermalasperity, dVmax, equals the product, i ¥ dRmax, where I is the MR bias current and dRmax is the peak change in stripe resistance. In Fig. 1(a), the resistance of the read sensor changes by about 0.3 percent due to the thermal asperity. By comparison, the typical resistance variation due to readback of the recorded magnetic domain is about 0.2 percent. Thermalasperity signals corresponding to resistance changes as high as 1.2 percent have been observed. Since the resistance change in the stripe results from frictional heating, the TA signal voltage can also be written as:
where b is the temperature coefficient of resistance of the stripe material, R is the equilibrium head resistance, and dTmax is the maximum average temperature rise in the stripe.
For most conventional (anisotropic) MR heads, b @ 0.2 percent resistance change per degree Celsius. Since dT is generated by frictional heating, it is independent of the bias current. However, according to the above equation, the amplitude of a contactpositive thermal pulse also depends linearly on bias current. This equation can be used to calculate an estimated average rise in stripe temperature due to an asperity hit. The signal in Fig. 1(a) corresponds to an average temperature increase of about 1.5 degrees C. Of course, the instantaneous (flash) temperature of the edge of the stripe contacting the asperity may be as high as 50 to 100 degrees C.
Decay of the thermalasperity pulse provides a “signature” characterizing the thermal relaxation of the MR stripe. In Fig. 1(a), for comparison, we fitted an artificial exponentially decaying signal to the real signal. The artificial signal exhibited the same peak amplitude and the same characteristic decay timeconstant as the measured signal, but had a characteristic decay timeconstant of 1/e. As can be seen, the real signal deviates somewhat from a purely exponential decay.
In Fig. 1(a), the decay timeconstant is 0.5 microsecond (for an MR head designed to record data at 2 Gbit/square inch). As we shall see later, the decay timeconstant is closely related to the readgap thickness. (We have seen decay timeconstants of the order of 1 microsecond in heads with larger gaps that were designed for a density of 1 Gbits/square inch). Note that, to produce a contactpositive thermal pulse, the asperity must directly contact the MR stripe. Contacts elsewhere do not transfer sufficient heat to the stripe for production of thermalasperity signals.
Unfortunately, all thermal asperities do not have the short duration shown in Fig. 1(a). If the surface defect is a broad hillock, and the MR stripe “rides” on the defect, the thermalasperity signal persists — depending on how long the stripe contacts the defect. We can see an example of a more persistent thermal asperity in Fig. 1(b). Note that the signal persists for about 8 microseconds. A signal of that duration could cause an error of many bytes, and would be far beyond the depth of standard errorcorrection codes.
So far, we have dealt only with contact thermalasperity pulses caused by frictional heating. But noncontact thermalasperity signals can also occur. To understand this we must first examine the read/write element near the airbearing surface of a flying head. The bias current, I, heats the stripe. So the stripe`s equilibrium temperature is always higher than that of its surroundings — including the moving disk under the head. Heat generated in the stripe per unit time is given by the expression, i2R.
Heat generated by the bias current dissipates from the stripe to its surroundings. Most of the heat flows into shields on either side of the sensor stripe. A small amount of heat flows into the internal leads (neglected in this simple analysis). Another small amount of heat flows to the disk through the air gap separating the head and disk.
Most of the heat flows into the shields because that is the path of least resistance. The spacing between stripe and shields is small (about 0.15 micron in today`s heads), and the gap`s alumina dielectric is a reasonably good thermal conductor. In addition, the shields have high thermal conductivity and are very large compared to the stripe — hence, they act as nearly infinite heat sinks. In the absence of other heat sources (such as friction), equilibrium is established with the stripe temperature (at nominal bias current) some 10 or 20 degrees C higher than the ambient near the head (including the rotating disk).
When a disksurface asperity contacts the MR stripe, it generates heat and the added heat pulse dissipates from the stripe along the same heat transfer paths as the heat generated by the bias current. From basic heattransfer theory, we know that the temperature of a body decays exponentially with time after application of a finite heat pulse — and the characteristic 1/e decay timeconstant depends on the geometry and materials of the heattransfer path. That means the temperature spike in an MR stripe from an asperity hit decays exponentially with time along each heat transfer path — and the combined decay is the sum of the exponential decays due to each path.
In the case of a contact asperity, we can neglect all decay paths except those to the shields on either side of the stripe. The thermalasperity signal in Fig. 1(a) is then the sum of two exponential decays. The shields usually differ in thickness. Also, in many MR heads, the two shields are made of different materials with differing thermal properties. Typically, the alumina layers on the two sides of the stripe also differ in thickness. As a consequence of those differences, the two exponential decays have slightly different decay timeconstants.
Two exponential decays with differing timeconstants cannot be added to obtain a combined exponential with a unique timeconstant. That is why the thermalasperity signal in Fig. 1(a) does not exactly match a single exponential decay. Also, the rate of heat transfer is inversely proportional to the separation between stripe and shields; the decay time of thermalasperity signals is therefore closely related to readgap thickness.
Noncontact signals
Since the equilibrium stripe temperature of an MR head at nominal bias current is always higher than the temperature of the spinning disk, some heat is transferred from stripe to disk. Assuming pure conduction, the heat transferred from stripe to disk is given by
where Ts is the stripe temperature, Td the disk temperature, Wt the area of the edge of the stripe facing the disk, k an average thermal conductivity of air plus the carbon overcoats on stripe and disk, and d is the spacing between stripe and the magnetic medium, More heat is conducted away from the stripe when the spacing decreases, cooling it; less heat when the spacing increases, raising the temperature of the stripe. Fluctuations in stripe resistance correspond to the rising and falling temperatures of the stripe.
In Fig. 2(a), we see an asperity defect slightly lower in height than the nominal spacing (“d”) between the head and disk. As the defect approaches the stripe, the spacing rapidly decreases (Fig. 2a); then, as the defect passes the stripe, the spacing rapidly increases, causing a transient temperature decrease in the stripe, which is accompanied by a negative resistance transient. The corresponding output from the MR stripe is shown in Fig. 2(b). We call this a “noncontact negative” thermalasperity signal. As with contact asperities, this signal is additive with the magnetic signal and can lead to read errors. That also explains the dip prior to the positive signal in Fig. 1(a). The stripe is first cooled by the approaching defect just before it hits the asperity.
In Fig. 2(c), we see a pit in the disk surface passing under the MR stripe. Following similar reasoning to that for Fig 2a, the defect in Fig. 2(c) leads to a “non contact positive” thermal asperity. An example of a noncontact positive thermal asperity is shown in Fig. 2(d). The amplitude of noncontact thermal asperity signals varies as the cube of bias current (see story above). As in the case of contact thermal asperities, noncontact asperities must pass directly under the MR stripe. The majority of thermal asperities encountered on magnetic recording disks are contact positive types. However, significant numbers of noncontact asperities may occur — depending on the type of disk and the level of quality control.
Erich Sawatzky is a former principal engineer for head/media engineering at Seagate Technology in San Jose. He has a PhD in physics from the University of British Columbia.
References
1. F.W. Gorter and J.A.L. Potgiesser, “Magnetoresistive Reading of Information”, 1974 Intermag Conference, Paper 316.
2. R.D. Hempstead, “Thermally Induced Pulses in Magnetoresistive Heads”, IBM J. Res. Dev. 1974, pp.547550.
3. M.M. Dovek, J.S. Foster, D.K. Lam and E. Sawatzky, “Contact Magnetic Recording Disk File with a Magnetoresistive Read Sensor,” U.S. Patent No. 5,455,730 filed February 18, 1993.

Fig. 1. Contactpositive thermal asperities: (top) Signal from pointed surface defect (with an artificial exponential decay superimposed for comparison); (bottom) thermalasperity caused by a wider and more rounded surface defect.

Fig. 3. (above) (a) Wavy disk passing under MR stripe; (b) baseline modulation (caused by waviness), with superimposed magnetic signal.

Fig.2. (right) (a) A raised defect (lower than the gap flying height) passing under the MR stripe; (b) a negative, noncontact thermalasperity signal caused by a raised defect; (c) a depression or pit defect passing under the MR stripe; and (d) a positive, noncontact thermalasperity caused by a pit defect.
Designing future MR heads
All magnetic disks exhibit some roughness or waviness. This is illustrated in Fig. 3(a), where the crests of the waves are not quite high enough to contact the stripe. As the head flies over the disk waves, the stripe temperature alternately rises and falls resulting in corresponding fluctuations in stripe resistance and output signal of the MR stripe, called “baseline modulation.” [3]
An example of baseline modulation, together with a magnetic signal, is shown in Fig.3(b). Clearly, baseline modulation forms a signature of disk waviness at any spot on the disk, and the frequency of the modulation signal is related to the waviness and the linear velocity of the disk surface. Fortunately, disk waviness and velocity are such that baselinemodulation frequencies typically fall in the range from around 100 kHz to 200 kHz and can be managed through filtering.
That the modulated baseline amplitude varies as the cube of MR bias current can be demonstrated as follows: The roughness of the disk spinning under the head varies as a function of time. This analysis, assumes that the peaktopeak amplitude of the “waves” on the disk (denoted as “2x” in Fig. 3(a)) is always less than average spacing between the MR stripe and the disk (denoted as “d”); in other words, no contact occurs between head and disk. At equilibrium, the heat flowing into the stripe must equal the total heat flowing out of it. That relationship can be expressed as:
Qin=12R=Qo+Qsd (1)
where Qin is the heat generated in the stripe per unit time, Qsd is the heat flowing from stripe to disk as before, and Qo is the remainder of the heat dissipated into the slider. Assuming that the shields and the rest of the slider are at temperature Td, Eq. 1 can be written as:
i2R=(TsTd)Go+(TsTd)G (2)
where Go (= Aoko/go) is a composite thermal conductance representing heat flowing into the slider, and G [= Wtk/(d±x)] is the thermal conductance between stripe and disk. Since Go is much larger than G, Eq. 2 can be rewritten, yielding the following approximation:
TsTd@12R(GoG)/Go2 (3)
Expanding Eq. 3 and substituting for G, we obtain:
Ts@Td+i2R/Goi2RWtk/
(d±x)Go2 (4)
Assuming x is much less than d, the following approximate expression for Ts results:
Ts@(Td+i2R)/Go(i2RWtkx)/dGo2
± (i2RWtkx)/d2Go2 (5)
The first three terms on the right side of Eq. 5 are independent of disk waviness and represent the average temperature of the stripe. They produce an average stripe resistance, which is not of immediate interest to us. The last term in Eq. 5 represents the variation in stripe temperature due to disk waviness. Let us designate this term as dTBLM. Then the striperesistance variation dlRBLM is approximately bR b dTBLM. We can multiply by the bias current to get the approximate fluctuation in the baseline of the MR output voltage as IbR b dTBLM. Substituting for dTBLM yields:
d VBLM@i3R2b Wt kx/d2Go2 (6)
Eq. 6 shows that the amplitude of baseline modulation is proportional to the cube of bias current, and inversely proportional to the square of average stripetodisk spacing. Of course, this analysis can also be applied to a single asperity with a height less than d, or to a pit in the disk surface. Thus we have proved that the amplitudes of noncontact thermal asperities vary as the cube of bias current — as contrasted to contactpositive thermal asperities where the relationship with bias current is linear. That fact can always be used to distinguish between contact and noncontact thermal asperity signals.
Since most of the heat flows from the stripe to the shields, Go can be estimated as approximately 4DWkal/g, where D is the stripe height, kal is the thermal conductivity of the alumina gaps, and g is the read gap thickness. Making these substitutions in Eq. 6 yields:
d VBLM@[(i3R2b)(tkxg2)]/4Wd2D2kal2 (7)
Eq. 7 can be used in the design of MR heads and headdisk interfaces to minimize baselinemodulation sensitivity. Strategies such as reducing gap thickness and using gap materials with higher thermal conductivity are clearly advantageous. Stripe dimensions (hence R) and bias current should be optimized for performance as a first priority.
Reducing disk waviness, x (i.e., using smoother disks), also reduces dVBLM. However, that is offset by the fact that baseline modulation is proportional to 1/d2, and smoother disks are usually designed for closer flying. Disk waviness is a natural consequence of the disklapping process. Since the wavelength of disk waviness increases as disks are made smoother, migrating to supersmooth disks in magnetic recording results in lower baselinemodulation frequencies — which makes filtering easier. Clearly, baseline modulation must be considered in the design of future recording heads and head/disk interfaces. — ES