IEEE TRANSACTIONS ON MAGNETICS, VOL. 38, NO. 3, MAY 2002 1575
Evaluation of Test Procedures for Determining
Servo Compatibility of Heads and Media in
Magnetic Disk Drives
Priyadarshee D. Mathur and William C. Messner, Member, IEEE
Abstract—We evaluate a testing specification proposed by the shapes at track edges due to leakage fields at the sides of the thin
National Storage Industry Consortium’s (NSIC) Extremely High film inductive (TFI) write heads are relatively small compared to
Density Recording (EHDR) group for evaluating head and media the length of the TFI head; 2) even though the magnetoresistive
compatibility for servo performance in magnetic disk drives. These
tests use average amplitude and average noise profile measure- (MR) elements that are used for readback show signal strength
ments across isolated tracks to predict the shape, linearity, noise, asymmetries along the cross-track direction, the relatively large
and long-term stability of position error signal (PES) patterns. We width of the tracks still provides a monotonic and near-linear
compare the predictions from these tests to measurements from zone from where high-quality position error signals (PES) can
null and amplitude PES patterns written on a spin-stand. Results be derived. Lowering the BAR in the future will make these un-
show average PES-profile prediction errors of 1%–2% track width
and noise level prediction within a factor of 2. We present data from desired effects of the writer and the readback transducers more
tests for long-term stability of the magnetoresistive (MR) read el- pronounced.
ement after repeated write cycles by the inductive write head. In The most direct method to characterize the PES from a given
the set of heads we tested, the MR head’s center and effective width head, medium, PES pattern and demodulation scheme is by di-
changed only slightly. Although we evaluated the NSIC specifica- rect measurement on an experimental system. This requires that
tion for MR read elements, the specification should be equally valid
for other read head types also, as long as the PES patterns are sim- heads and media are available for testing and that a precision
ilar. setup capable of writing and demodulating PES patterns exists.
Index Terms—Disk drive, position signal. In contrast to this approach, micromagnetic simulation tools are
also very insightful and do not require a test system. For ex-
ample, these tools can simulate the writing and readback pro-
I. INTRODUCTION cesses, effects of thermal decay and PES generation [2]–[4].
The accuracy of the predictions depend on the models assumed
H IGH-DENSITY storage on magnetic disk drives requires
increasing both the linear bit density and the track den-
sity on the recording medium. Currently, the recorded transi-
for the individual components and the parameters of the sim-
ulation itself. While results of micromagnetic simulations cor-
tions, henceforth loosely referred to as the bits, are approxi- roborate observed data, these tools are computationally expen-
mately ten times wider in the radial (cross-track) direction than sive and, therefore, are generally not used for simulating large
they are in the circumferential (downtrack) direction. The Na- sections of media (patterns) or for gathering of statistical infor-
tional Storage Industry Consortium (NSIC) estimates that the mation. Moreover, these tools do not apply to per-component
bit aspect ratio (BAR) will approach one-to-one as the industry level testing that may determine, for example, which heads are
tries to maximize storage density in the future. Also, since the assembled into drives.
super-paramagnetic limit [1] dictates the minimum size of the In the middle are hybrid methods, such as the head and media
magnetic grains that can be used for reliable long-term storage, test specification developed by NSIC’s Extremely High Density
fewer grains would be available for storing each bit, resulting in Recording (EHDR) group, which rely partly on experimental
a decrease in the raw signal-to-noise ratio (SNR) for both servo data and partly on simulation. These tests use track average am-
and data channels. plitude (TAA) profiles and cross-track noise profiles from test
The task for servo engineers is to determine how these de- heads. Simulations can then approximate readback signals from
sign changes will affect the performance of the servo system. actual PES patterns using these data. Head and media manu-
The servo system derives position feedback from special mag- facturers can obtain TAA profiles easily on most commercial
netic patterns written on servo wedges that are embedded along testers and these tests automatically include the nonideal effects
with user data on the medium. Currently, the servo system ben- of the writer and reader. These are important factors since the
efits from the high BAR designs: 1) distortions in the transition NSIC-EHDR group developed these tests [5] so that head and
media manufacturers could evaluate head and media compati-
bility for servo performance on a per-component basis.
Manuscript received October 25, 2000; revised December 12, 2001. This
work was supported in part by the National Science Foundation under Grant As part of a collaborative effort with the NSIC members,
ECD-8907068 and by the National Storage Industry Consortium. we evaluated the test specification at the Data Storage Systems
P. D. Mathur is with Seagate Technology, Shakopee, MN 55379-1863 USA. Center at Carnegie Mellon University. The NSIC members sup-
W. C. Messner is with the Carnegie Mellon University, Pittsburgh, PA
15213-3890 USA. plied test heads and media. This paper presents the results of
Publisher Item Identifier S 0018-9464(02)03645-2. the evaluation and is organized as follows. Section II describes
0018-9464/02$17.00 © 2002 IEEE
,1576 IEEE TRANSACTIONS ON MAGNETICS, VOL. 38, NO. 3, MAY 2002
(a) (b)
Fig. 1. Unaveraged (top) and averaged (bottom) pulse trains at (a) 2 MHz and (b) 10 MHz.
TAA profiling, while Section III summarizes1 the results of the high-frequency pulse trains. This latter may be inferred by con-
following subset of tests from the specification: 1) cross-track sidering the two unaveraged pulse trains shown in Fig. 1 (top).
write to read offset; 2) cross-track PES linearity; 3) cross-track The absolute value operation biases the estimate of the mean
SNR; and 4) PES shifts and effective MR head width variations when the signal is near zero. Since the 2 MHz pulse train is
after writing. Section IV compares the SNR and PES profile pre- near zero for a larger fraction of its cycle than the 10 MHz
dictions to measurements from actual PES patterns written on signal, additive noise rejection is better for the latter. The DSSC
a spin-stand. Concluding remarks are made in Section V. Sec- spin-stand provided an averaging feature to reduce the effect of
tion VI lists the nominal specifications of the head and media additive electronics noise in the measurements (Fig. 1, bottom).
samples used in this study. While this feature extended the range of the TAA measurements,
it was not used because most commercial testers cannot imple-
II. TRACK AVERAGE AMPLITUDE (TAA) PROFILING ment it.
A track average amplitude profile, , is the average A. Procedure to Obtain a TAA Profile
strength of the readback signal as a function of the cross-track
The NSIC specification defines untrimmed TAA and trimmed
position offset, , of the head. The offset is measured relative to
TAA profile tests. The procedure used for both was adapted as
the center of the test track. The test track is recorded with a long
follows: 1) DC erase a wide zone ( 10 tracks) around the test
constant frequency burst. The definition of the average strength
radius; 2) write a constant frequency burst around the track at
depends on the equipment used to measure this parameter.
the test radius (the head manufacturer specifies the burst fre-
Common choices are: 1) the amplitude of the signal at the
quency); 3) for trimmed TAA profiles, move the head toward
burst frequency, as measured by a spectrum analyzer; 2) the
the inner radius by a trim offset (50% track) and then DC erase
1-norm average of the signal, as measured by a rectify and
to reduce the width of the track; 4) record 50–100 cycles of the
integrate circuit; and 3) the average of the absolute value of the
readback waveform at each cross-track location using an oscil-
minimum and maximum peaks in the readback signal. TAA
loscope; and 5) calculate the TAA profiles using (1). The DSSC
data do not contain downtrack phase information because phase
tester, which uses air bearings on positioners, did not require
information is lost in calculating the average.
backlash avoidance techniques from the NSIC specification.
The tester used for this research calculated the signal strength
Fig. 2 shows trimmed and untrimmed TAA profiles collected
using software. We chose the 1-norm average. For a discrete
from head-A and head-D. The cross-track profiles from head-A
signal, , the definition of the 1-norm average, , is
(left) are more symmetric than from head-D. However, both
heads show that the centers of the written track, or the zero
(1) cross-track offsets on the plots, do not coincide with the location
of the peaks in the profiles. Manufacturing errors may cause off-
sets between the MR and TFI heads’ centers. For trimmed TAA
where is the total number of samples. Unlike method 1), profiles, the trimming operation, which is done on one side only,
contributions from all harmonics were of interest and, unlike shifts the center of the written track. Additionally, for both TAA
method 3), the 1-norm has better noise rejection properties for profiles, the MR head response to magnetic charge on the track
edges can cause asymmetry and peak shifting.
Fig. 3 shows TAA profiles for ensembles of two types of
1These tests used ensembles of five different heads and media combinations heads: head-A and head-E. Clearly, the profiles within each en-
(labeled head-A through head-E). To reduce the number of illustrations, plots semble have different peak amplitudes. However, the normal-
contain only data from selected heads and media. Where appropriate, only the
ensemble mean, maximum and minimum of the key results of the tests appear ized profiles are similar in both ensembles, except for their peak
on plots. locations. Profiles from head-E, which was the only head type
, MATHUR AND MESSNER: SERVO COMPATIBILITY OF HEADS AND MEDIA IN MAGNETIC DISK DRIVES 1577
(a) (b)
Fig. 2. Untrimmed and trimmed TAA profiles from (a) head-A and (b) head-D.
(a) (b)
Fig. 3. Ensemble of TAA profiles from (a) head-A and (b) head-E. The plots on the left for each case show unnormalized TAA profiles, while the plots on the
right for each case show normalized TAA profiles.
tested with a flex-on-suspension type wiring, illustrate that the
TAA value at the track edges is affected by the electrical cou-
pling between the head and the preamplifier. Compared to heads
A, B, C, and D, the noise floor at the preamplifier’s output in-
creased dramatically by merely connecting head-E to the am-
plifier’s input. Since all heads tested had similar MR element
resistance, it is unlikely that thermal noise [6]–[8] was a factor.
Fig. 4 shows the statistics of the maximum value of the
TAA profile for most of the heads tested. The variations in this
figure indicate that the characteristics of each head are unique.
Attempts to obtain absolute performance data entirely through
simulations or on the basis of data collected from one head in
an ensemble may at best only predict the ensemble mean.
III. HEAD AND MEDIA COMPATIBILITY TESTS
This section explains the tests from the NSIC-EHDR specifi- Fig. 4. Ensemble statistics of TAA measurement. The error bars span from the
cation [5] and their relevance to servo performance. ensemble minimum to the ensemble maximum. The number of heads tested in
an ensemble are indicated in parentheses.
A. Write-to-Read Offset Measurement keeping the write head on the data track center requires main-
The centers of the read and write elements are not co-located taining an MR head position offset with respect to the center
and it is the MR head that reads the servo bursts. Therefore, of the data track. This was unnecessary in drives that used one
Evaluation of Test Procedures for Determining
Servo Compatibility of Heads and Media in
Magnetic Disk Drives
Priyadarshee D. Mathur and William C. Messner, Member, IEEE
Abstract—We evaluate a testing specification proposed by the shapes at track edges due to leakage fields at the sides of the thin
National Storage Industry Consortium’s (NSIC) Extremely High film inductive (TFI) write heads are relatively small compared to
Density Recording (EHDR) group for evaluating head and media the length of the TFI head; 2) even though the magnetoresistive
compatibility for servo performance in magnetic disk drives. These
tests use average amplitude and average noise profile measure- (MR) elements that are used for readback show signal strength
ments across isolated tracks to predict the shape, linearity, noise, asymmetries along the cross-track direction, the relatively large
and long-term stability of position error signal (PES) patterns. We width of the tracks still provides a monotonic and near-linear
compare the predictions from these tests to measurements from zone from where high-quality position error signals (PES) can
null and amplitude PES patterns written on a spin-stand. Results be derived. Lowering the BAR in the future will make these un-
show average PES-profile prediction errors of 1%–2% track width
and noise level prediction within a factor of 2. We present data from desired effects of the writer and the readback transducers more
tests for long-term stability of the magnetoresistive (MR) read el- pronounced.
ement after repeated write cycles by the inductive write head. In The most direct method to characterize the PES from a given
the set of heads we tested, the MR head’s center and effective width head, medium, PES pattern and demodulation scheme is by di-
changed only slightly. Although we evaluated the NSIC specifica- rect measurement on an experimental system. This requires that
tion for MR read elements, the specification should be equally valid
for other read head types also, as long as the PES patterns are sim- heads and media are available for testing and that a precision
ilar. setup capable of writing and demodulating PES patterns exists.
Index Terms—Disk drive, position signal. In contrast to this approach, micromagnetic simulation tools are
also very insightful and do not require a test system. For ex-
ample, these tools can simulate the writing and readback pro-
I. INTRODUCTION cesses, effects of thermal decay and PES generation [2]–[4].
The accuracy of the predictions depend on the models assumed
H IGH-DENSITY storage on magnetic disk drives requires
increasing both the linear bit density and the track den-
sity on the recording medium. Currently, the recorded transi-
for the individual components and the parameters of the sim-
ulation itself. While results of micromagnetic simulations cor-
tions, henceforth loosely referred to as the bits, are approxi- roborate observed data, these tools are computationally expen-
mately ten times wider in the radial (cross-track) direction than sive and, therefore, are generally not used for simulating large
they are in the circumferential (downtrack) direction. The Na- sections of media (patterns) or for gathering of statistical infor-
tional Storage Industry Consortium (NSIC) estimates that the mation. Moreover, these tools do not apply to per-component
bit aspect ratio (BAR) will approach one-to-one as the industry level testing that may determine, for example, which heads are
tries to maximize storage density in the future. Also, since the assembled into drives.
super-paramagnetic limit [1] dictates the minimum size of the In the middle are hybrid methods, such as the head and media
magnetic grains that can be used for reliable long-term storage, test specification developed by NSIC’s Extremely High Density
fewer grains would be available for storing each bit, resulting in Recording (EHDR) group, which rely partly on experimental
a decrease in the raw signal-to-noise ratio (SNR) for both servo data and partly on simulation. These tests use track average am-
and data channels. plitude (TAA) profiles and cross-track noise profiles from test
The task for servo engineers is to determine how these de- heads. Simulations can then approximate readback signals from
sign changes will affect the performance of the servo system. actual PES patterns using these data. Head and media manu-
The servo system derives position feedback from special mag- facturers can obtain TAA profiles easily on most commercial
netic patterns written on servo wedges that are embedded along testers and these tests automatically include the nonideal effects
with user data on the medium. Currently, the servo system ben- of the writer and reader. These are important factors since the
efits from the high BAR designs: 1) distortions in the transition NSIC-EHDR group developed these tests [5] so that head and
media manufacturers could evaluate head and media compati-
bility for servo performance on a per-component basis.
Manuscript received October 25, 2000; revised December 12, 2001. This
work was supported in part by the National Science Foundation under Grant As part of a collaborative effort with the NSIC members,
ECD-8907068 and by the National Storage Industry Consortium. we evaluated the test specification at the Data Storage Systems
P. D. Mathur is with Seagate Technology, Shakopee, MN 55379-1863 USA. Center at Carnegie Mellon University. The NSIC members sup-
W. C. Messner is with the Carnegie Mellon University, Pittsburgh, PA
15213-3890 USA. plied test heads and media. This paper presents the results of
Publisher Item Identifier S 0018-9464(02)03645-2. the evaluation and is organized as follows. Section II describes
0018-9464/02$17.00 © 2002 IEEE
,1576 IEEE TRANSACTIONS ON MAGNETICS, VOL. 38, NO. 3, MAY 2002
(a) (b)
Fig. 1. Unaveraged (top) and averaged (bottom) pulse trains at (a) 2 MHz and (b) 10 MHz.
TAA profiling, while Section III summarizes1 the results of the high-frequency pulse trains. This latter may be inferred by con-
following subset of tests from the specification: 1) cross-track sidering the two unaveraged pulse trains shown in Fig. 1 (top).
write to read offset; 2) cross-track PES linearity; 3) cross-track The absolute value operation biases the estimate of the mean
SNR; and 4) PES shifts and effective MR head width variations when the signal is near zero. Since the 2 MHz pulse train is
after writing. Section IV compares the SNR and PES profile pre- near zero for a larger fraction of its cycle than the 10 MHz
dictions to measurements from actual PES patterns written on signal, additive noise rejection is better for the latter. The DSSC
a spin-stand. Concluding remarks are made in Section V. Sec- spin-stand provided an averaging feature to reduce the effect of
tion VI lists the nominal specifications of the head and media additive electronics noise in the measurements (Fig. 1, bottom).
samples used in this study. While this feature extended the range of the TAA measurements,
it was not used because most commercial testers cannot imple-
II. TRACK AVERAGE AMPLITUDE (TAA) PROFILING ment it.
A track average amplitude profile, , is the average A. Procedure to Obtain a TAA Profile
strength of the readback signal as a function of the cross-track
The NSIC specification defines untrimmed TAA and trimmed
position offset, , of the head. The offset is measured relative to
TAA profile tests. The procedure used for both was adapted as
the center of the test track. The test track is recorded with a long
follows: 1) DC erase a wide zone ( 10 tracks) around the test
constant frequency burst. The definition of the average strength
radius; 2) write a constant frequency burst around the track at
depends on the equipment used to measure this parameter.
the test radius (the head manufacturer specifies the burst fre-
Common choices are: 1) the amplitude of the signal at the
quency); 3) for trimmed TAA profiles, move the head toward
burst frequency, as measured by a spectrum analyzer; 2) the
the inner radius by a trim offset (50% track) and then DC erase
1-norm average of the signal, as measured by a rectify and
to reduce the width of the track; 4) record 50–100 cycles of the
integrate circuit; and 3) the average of the absolute value of the
readback waveform at each cross-track location using an oscil-
minimum and maximum peaks in the readback signal. TAA
loscope; and 5) calculate the TAA profiles using (1). The DSSC
data do not contain downtrack phase information because phase
tester, which uses air bearings on positioners, did not require
information is lost in calculating the average.
backlash avoidance techniques from the NSIC specification.
The tester used for this research calculated the signal strength
Fig. 2 shows trimmed and untrimmed TAA profiles collected
using software. We chose the 1-norm average. For a discrete
from head-A and head-D. The cross-track profiles from head-A
signal, , the definition of the 1-norm average, , is
(left) are more symmetric than from head-D. However, both
heads show that the centers of the written track, or the zero
(1) cross-track offsets on the plots, do not coincide with the location
of the peaks in the profiles. Manufacturing errors may cause off-
sets between the MR and TFI heads’ centers. For trimmed TAA
where is the total number of samples. Unlike method 1), profiles, the trimming operation, which is done on one side only,
contributions from all harmonics were of interest and, unlike shifts the center of the written track. Additionally, for both TAA
method 3), the 1-norm has better noise rejection properties for profiles, the MR head response to magnetic charge on the track
edges can cause asymmetry and peak shifting.
Fig. 3 shows TAA profiles for ensembles of two types of
1These tests used ensembles of five different heads and media combinations heads: head-A and head-E. Clearly, the profiles within each en-
(labeled head-A through head-E). To reduce the number of illustrations, plots semble have different peak amplitudes. However, the normal-
contain only data from selected heads and media. Where appropriate, only the
ensemble mean, maximum and minimum of the key results of the tests appear ized profiles are similar in both ensembles, except for their peak
on plots. locations. Profiles from head-E, which was the only head type
, MATHUR AND MESSNER: SERVO COMPATIBILITY OF HEADS AND MEDIA IN MAGNETIC DISK DRIVES 1577
(a) (b)
Fig. 2. Untrimmed and trimmed TAA profiles from (a) head-A and (b) head-D.
(a) (b)
Fig. 3. Ensemble of TAA profiles from (a) head-A and (b) head-E. The plots on the left for each case show unnormalized TAA profiles, while the plots on the
right for each case show normalized TAA profiles.
tested with a flex-on-suspension type wiring, illustrate that the
TAA value at the track edges is affected by the electrical cou-
pling between the head and the preamplifier. Compared to heads
A, B, C, and D, the noise floor at the preamplifier’s output in-
creased dramatically by merely connecting head-E to the am-
plifier’s input. Since all heads tested had similar MR element
resistance, it is unlikely that thermal noise [6]–[8] was a factor.
Fig. 4 shows the statistics of the maximum value of the
TAA profile for most of the heads tested. The variations in this
figure indicate that the characteristics of each head are unique.
Attempts to obtain absolute performance data entirely through
simulations or on the basis of data collected from one head in
an ensemble may at best only predict the ensemble mean.
III. HEAD AND MEDIA COMPATIBILITY TESTS
This section explains the tests from the NSIC-EHDR specifi- Fig. 4. Ensemble statistics of TAA measurement. The error bars span from the
cation [5] and their relevance to servo performance. ensemble minimum to the ensemble maximum. The number of heads tested in
an ensemble are indicated in parentheses.
A. Write-to-Read Offset Measurement keeping the write head on the data track center requires main-
The centers of the read and write elements are not co-located taining an MR head position offset with respect to the center
and it is the MR head that reads the servo bursts. Therefore, of the data track. This was unnecessary in drives that used one
