Content uploaded by Jeffrey P. Wilde
Author content
All content in this area was uploaded by Jeffrey P. Wilde on Jan 09, 2015
Content may be subject to copyright.
Magneto-optical disk drive technology using
multiple fiber-coupled flying optical heads.
Part I. System design and performance
Jeffrey P. Wilde, John F. Heanue, Alexander A. Tselikov, and Jerry E. Hurst, Jr.
A novel flying-optical-head data storage technology is described. It is based on a micro-optical recording
head that contains a silicon micromachined torsional mirror for high-bandwidth track following. Mul-
tiple heads and disks are contained in a Winchester-style rotating disk drive. Single-mode optical fibers
provide light delivery to and from the heads. Both polarization-maintaining and low-birefringence fiber
systems have been implemented for magneto-optical MO recording. A fixed optics module containing
a laser diode, MO detection optics, and a 1 N fiber bundle switch has been developed as an integral part
of this new recording architecture. A 5.25-in. 13.33-cm, half-height prototype drive design and its
performance are presented. © 2001 Optical Society of America
OCIS codes: 210.4680, 210.4770, 210.3810, 060.2370, 230.5440, 060.1810.
1. Introduction
Historically, one of the primary distinctions between
magnetic and optical recording products has been
fixed versus removable storage, respectively. Mag-
netic drives contain multiple disks with each record-
ing surface supporting a flying magnetic head.
Today’s fly heights are approximately 1 in 25 nm,
necessitating a sealed environment and a well-
engineered head–disk interface. In the future, fly
heights are expected to be reduced further, heading
toward contact recording. In contrast, mainstream
optical products support one disk at a time with a
single free-space recording head, meaning that only
one surface is available for on-line access. Tradi-
tionally, optical technology has been able to support
higher track density when compared with magnetic
recording because for many years optical technology
has taken advantage of two-stage coarse and fine
actuator systems that have a greater servo-loop
bandwidth than can be provided by the single rotary
voice-coil actuator typically employed in disk drives.
More recently the distinction between fixed mag-
netic drives and optical drives with removable media
has begun to blur with the availability of removable
magnetic cartridge products. However, they must
contend with contamination issues, making them in-
trinsically more expensive on a per gigabyte basis
than hard drives. On the other hand, a flying optical
head technology operating in the far field can offer
multiple heads in the same product and at the same
time provide a higher fly height and hence a more
robust head–disk interface. In addition, optical
technology allows for the possibility of multilayer re-
cording that can increase the effective areal density.
There exists a variety of previous research on min-
iature optical heads,
1–3
but none of these earlier
papers describes a complete solution that takes into
account the need for a second-stage microactuator.
In this paper we describe what is to our knowledge
the first demonstration of a flying optical head tech-
nology implemented in a complete prototype drive.
The target areal density is 3.5 Gbits兾in.
2
(1 in.
2.54 cm). Although this areal density is not at the
cutting edge, we believe that it is high enough to
provide a suitably interesting demonstration of a
new architecture. In general, the far-field technol-
ogy described here is limited in areal density in
much the same way that the more conventional
far-field magneto-optical MO recording is limited.
The primary distinction centers on use of multiple
When this research was performed all the authors were with
Quinta Corporation, a wholly owned subsidiary of Seagate Tech-
nology, Incorporated, 1870 Lundy Avenue, San Jose, California
95131. J. F. Heanue and J. E. Hurst, Jr., are now with Xros
Incorporated, 777 North Pastoria Avenue, Sunnyvale, California
94086. J. P. Wilde’s e-mail address is wilde@compuserve.com.
Received 27 July 2000; revised manuscript received 23 October
2000.
0003-6935兾01兾050691-16$15.00兾0
© 2001 Optical Society of America
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 691
heads and disks in a standard product form factor,
making for a more efficient volumetric capacity.
Specific issues regarding laser noise in our system
are discussed in detail in a companion paper.
4
Our drive architecture combines what we believe
are some of the best attributes of both magnetic and
optical recording and is referred to as optically as-
sisted Winchester OAW technology.
5–7
Our ap-
proach is based on MO recording
8,9
and utilizes
single-mode fiber optics for light delivery to and from
the heads. The optical head comprises a confocal
system, and as such it provides good depth
resolution—a feature that is highly desirable for a
multilayer recording device, although only single-
layer recording has been attempted to date. The
recording medium is based on the conventional rare-
earth transition metal alloy TbFeCo sputtered on a
patterned plastic substrate. The embossed pattern
defines the data tracks in a sampled servo scheme.
Hence the media is low in cost, but the prepatterning
tends to introduce significant radial runout, both re-
peatable and nonrepeatable, and thus requires a
high-bandwidth track-following servo system. For
this reason a small micromachined mirror is utilized
on the head as a second-stage fine actuator, and a
rotary voice-coil actuator serves as the first stage.
To support multiple heads, a novel and compact op-
tical module was developed that contains a 30-mW
red laser diode, beam-shaping optics, MO detection
optics, and a 1 N fiber-optic switch. In this way
only one laser diode with its driver electronics, along
with a single MO detection module, services all the
heads, which helps to keep the cost low. Use of
single-mode fiber provides for the delivery of a clean,
spatially filtered beam to the head. The fiber also
allows the functionality of the heads and the optical
module to be separated for easier fabrication at both
the subsystem and the drive levels.
Our nominal fly height is 15 in. 375 nm, some-
thing that is accomplished easily with modern air-
bearing technology. Although this height may be
large by today’s standards, it is still low enough to
allow for thin-film coil technology to be used to fab-
ricate a ring coil that is capable of generating a ver-
tical magnetic field of sufficient strength to support
high-data-rate MO recording 100 Mbits兾s.In
addition, the higher fly height provides an air bearing
that is sufficiently compliant to accommodate varia-
tions in pitch and roll stiffness associated with the
fiber attachment. To minimize the impact of fiber
attachment on the slider’s flying characteristics, the
fiber diameter should be chosen as small as is rea-
sonably possible, bearing in mind that small diame-
ters are also more fragile and difficult to handle. We
find that fiber with an 80-m cladding diameter rep-
resents a good compromise and is also economically
viable because this diameter is a standard size in the
fiber sensor industry.
When a flying MO head is implemented, significant
heating of the recording surface by the focused optical
spot 250 °C can create problems at the head–disk
interface. In the magnetic recording industry, a
thin film of liquid lubricant is typically used to min-
imize friction at the interface. However, in our case,
conventional liquid lubricants are susceptible to ther-
mal desorption, resulting in lubricant filling the coil
aperture and contaminating the surface of the objec-
tive lens. A great deal of research on this problem
has taken place at Quinta兾Seagate, and although the
details remain proprietary, suffice it to say that var-
ious techniques can be used to mitigate the problem
substantially. These include use of high-molecular-
weight liquid lubricants, solid lubricants, and a thick
protective dielectric overcoat to thermally buffer the
magnetic recording layer where absorption and
hence heating occur from the lubricated flying sur-
face.
We would be remiss if we did not discuss how the
OAW far-field scheme described here compares with
the recent research in the area of near-field
recording.
10–12
First, as was mentioned above, op-
eration in the far field allows for a higher fly height,
in the range from 10 to 20 in. 250 to 500 nm
versus the much more stringent requirement of 1–2
in. 25–50 nm for the near field. Therefore far-
field technology is perhaps better suited for remov-
able storage, and the confocal nature of our head can
be exploited for multilayer recording. Second, ad-
vanced magnetic superresolution MO media can be
employed to extend far-field recording beyond the
diffraction limit, making it more competitive with
near-field techniques.
13
Finally, our system can be
extended to the near field when, for example, one
inserts a solid immersion lens into the optical path of
the head and lowers the fly height. The lower fly
height makes near-field recording less attractive for
removable products, but multilayer near-field record-
ing by a solid immersion lens scheme with confocal
readout is still possible, making it more competitive
with magnetic disk drive technology.
The line between optical and magnetic recording
may continue to blur as magnetic recording contends
with superparamagnetic effects that lead to a ther-
mal decay of data. If and when this effect truly be-
comes intolerable is as yet unknown, but it certainly
has become a topic of concern in the magnetic record-
ing industry. High-coercivity recording media do
exist today that do not suffer from thermal decay;
however, at room temperature inductive magnetic
writers are not able to produce sufficient field
strength to alter the magnetization at the required
linear densities on such media. One potential solu-
tion involves thermomagnetic recording in which a
focused optical beam provides a localized heating
source to lower temporarily the media coercivity dur-
ing the writing process, much like that of conven-
tional MO recording. Data readout then occurs with
a magnetic flux reader at a nominal operating tem-
perature with no significant decay.
14,15
In this con-
text the compact light delivery system presented here
may serve as the basis for a future optically assisted
magnetic-storage technology.
16
692 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
2. Optical System Architecture
A. General Layout
The general optical system schematic is shown in Fig.
1. The laser diode source and detection optics are
similar to those used for conventional MO recording.
The new aspect here is use of a 1 N fiber-optic
switch that directs the laser light to any one of the
multiple recording surfaces. The particular system
we developed is a 5.25-in. 13.33-cm, half-height pro-
totype drive that supports 6 disks and 12 recording
surfaces. A top view of the system is shown in Fig.
2, where it can be seen how the fibers are routed down
the rotary actuator arms to the flying heads. As the
actuator moves the heads radially across the disk, the
fibers readily bend to accommodate the motion.
Because we are implementing thermomagnetic re-
cording, it is necessary to use a laser diode source
capable of outputting at least 30 mW of average
power. To extend laser lifetime, it is advantageous
to be able to run the laser below its maximum rated
power output during recording. Loss in the optical
path arises from the beam-shaping components 0.5
dB, the Faraday isolator 1.0 dB, the leaky beam
splitter 1.25 dB, fiber coupling 1.0 dB, and the
gold-coated mirror on the head 0.35 dB. We find
that with 20–25 mW of laser power, the beam arrives
at the disk surface with approximately 8–10 mW,
which is sufficient for writing. When focused with a
high-N.A. objective lens, this power level heats the
MO layer above its Curie temperature, typically in
the vicinity of 250 °C. The storage density is deter-
mined by the diffraction-limited spot size
0.55兾N.A., so a short wavelength is desirable.
Currently a red laser diode operating at 660 nm pro-
vides the best solution and is the type of laser used in
this research. Blue laser diodes at 410 nm may be-
come viable in the not too distant future.
The concept of optical fiber being used for light
delivery to a recording head is not new,
3,17,18
but to
our knowledge the research presented here repre-
sents the first complete system based on fiber light
delivery, with fully functional drives having been
demonstrated.
19
A MO-type recording system re-
quires the measurement of a small Kerr rotation of
the linear polarization upon reflection from the disk.
In conventional free-space MO drives, differential de-
tection is employed to yield sufficient signal detection
sensitivity while also providing common-mode inten-
sity noise rejection.
20
When a single-mode fiber is
used, however, problems associated with polarization
control and detection become more complicated. In
addition, there arise new ways in which significant
laser noise can manifest itself. A careful under-
standing of such laser noise is absolutely essential for
constructing a high signal-to-noise ratio SNR sys-
tem. We investigated both single-mode distributed
feedback DFB and multimode Fabry–Perot FP la-
sers and found their noise characteristics to be dra-
matically different.
4
In what follows, we describe
two separate fiber delivery systems, one based on
polarization-maintaining PM fiber and the other on
low-birefringence Lo-Bi fiber. A Jones matrix
analysis is employed to describe polarization propa-
gation through these systems.
B. Implementation Based on Polarization-Maintaining
Fiber
PM fiber is characterized by large linear birefrin-
gence, in the range from 10
4
to 10
3
. It is created
by the introduction of stress elements in the fiber
preform. The level of birefringence is sufficient to
produce adiabatic following, in which polarized light
launched along either the fast or the slow axis will
remain confined to the axis with little coupling to the
other axis.
21
The polarization cross talk is a mea-
sure of the coupling between the axes, and for high-
quality PM fiber it can be as low as 40 dB as long as
the fiber is not perturbed by excess bending or by
external forces because of mechanical clamping in a
test setup or epoxy shrinkage in a ferrule support.
Typically, PM fiber manufacturers specify a mode-
coupling parameter h ⱕ 5 10
5
per meter, which
converts to a best-case cross talk 10 logtanh hl
of 43 dB for l 1m.
22
In practice, though, we find
that the cross talk for short lengths of PM fiber of the
order of 1 m is dominated by the deployment condi-
tions and is generally better than 30 dB in our
system. Although the large birefringence does pro-
vide for polarization maintenence along a given axis,
it is much more difficult to control the phase shift
Fig. 1. General approach to the OAW drive architecture with
multiple flying optical heads based on single-mode fiber light de-
livery.
Fig. 2. Top view of the 5.25-in. 13.33-cm, half-height OAW drive
layout.
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 693
between the fast and the slow components because
the birefringence varies with temperature and exter-
nal bending. Potentially this can be a problem be-
cause, when the Kerr effect is detected, the phase
between the two orthogonal polarization components
must be held constant, ideally at either 0° or 90° for
in-phase or quadrature detection, respectively.
Various approaches for controlling the phase in a
PM fiber system for MO recording have been dis-
cussed previously in the literature.
18
We adopt the
configuration shown in Fig. 3, where the switching
function has been removed for simplicity. In the for-
ward direction, p-polarized light transmitted through
the leaky beam splitter LBS is launched at 45° into
the PM fiber so that half of the light propagates on
the fast axis, and the other half propagates on the
slow axis. This can be accomplished by a physical
rotation of the fiber axes at the launch point. A
quarter-wave plate QWP1 is positioned on the head,
oriented at 45° with respect to the fiber axes. This
arrangement converts the two linear polarization
states into two circular states. Because the two cir-
cular states have the same amplitude, they combine
to produce linear polarization at the disk with an
orientation determined by the phase shift encoun-
tered in the fiber. In the context of a Jones matrix
analysis, this situation in which the coordinate sys-
tem is assumed to be aligned with the principal fiber
axes is analyzed as follows:
E
input
⫽
1
冑
2
冉
1
1
冊
, (1)
S
fiber
⫽
冋
10
0
册
, (2)
S
QWP1
45° ⫽
1
冑
2
冋
1 i
i 1
册
. (3)
Here expi represents the fiber phase shift.
The phase term is given by 2nL兾, where n
is the fiber birefringence and L is the fiber length.
The optical field at the disk surface is then computed
as
冉
E
x
E
y
冊
disk
⫽ S
QWP1
45°S
fiber
E
input
⫽
1
2
冉
1 ⫺ i
i ⫹
冊
⫽
1
1 ⫹ i*
冋
1 ⫹ Im
Re
册
⫽
1
1 ⫹ i expi
冋
1 ⫹ sin
cos
册
. (4)
This output state represents linear polarization with
an orientation given by the value of . Having
linear polarization at the disk is important because
we chose to work with MO media that is tailored for
maximum Kerr rotation, as opposed to Kerr elliptic-
ity.
23
It is important to note that also depends on
the optical wavelength. Therefore when we use a
multimode FP laser, the optical field at the disk will
consist of a superposition of linear polarization
states, one for each longitudinal mode of the laser.
The Kerr rotation then acts on each wavelength sep-
arately. Standard differential detection is still pos-
sible as we show next.
After reflecting from the disk surface, the light
propagates back through QWP1 and the fiber and is
directed subsequently by the LBS to the differential
detection system. To understand how the nominal
polarization state propagates on the return path, it is
important to recognize that the two circular polariza-
tion states incident on the disk change their sense
upon reflection, so that right-hand circular polariza-
tion becomes left-hand circular polarization and vice
versa. The Kerr effect merely alters the phase be-
tween the two reflected circular states. After pass-
ing back through QWP1, the circular states are
converted back to linear, but now the axes are re-
versed. Therefore light that propagated down the
fiber on one axis in the forward direction returns on
the opposite axis. In this way the fiber becomes a
zero-phase optical path, similar to free space, if we
assume of course that the fiber possesses low-
polarization cross talk and QWP1 is fabricated and
aligned accurately.
The Jones matrix analysis of the return path pro-
ceeds by our first recognizing that reflection at the
disk changes the coordinate system. We choose to
retain a right-handed coordinate system, so that the
Jones matrix for the disk with a Kerr rotation angle
no ellipticity and an intensity reflectance R
d
is
given by
S
disk
⫽
冑
R
d
冋
cos sin
sin cos
册
. (5)
Likewise, the Jones matrix for a general optical com-
ponent transforms according to
S ⫽
冋
AB
CD
册
forward path
f S
˜
⫽
冋
A C
BD
册
return path
. (6)
Fig. 3. Optical path based on PM fiber. Various lenses are not
shown for simplicity. PBS, polarizing beam splitter.
694 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
Therefore, with no Kerr effect 0°, the optical
field exiting the fiber at the original launch point is
given by
E
out
⫽ 0° ⫽ S
˜
fiber
S
˜
QWP1
45°S
disk
0E
disk
⫽
i
冑
R
d
冑
2
冉
1
1
冊
. (7)
In the new return-path coordinate system, this po-
larization state corresponds to that of the original
input to the fiber apart from an overall constant,
demonstrating that the round-trip path is indeed zero
phase, independent of the fiber birefringence and the
wavelength.
In the more general case, with a nonzero Kerr ro-
tation, the return fiber field is found to be
E
out
⫽
i
冑
R
d
冑
2
冋
expi
expi
册
. (8)
This result is easier to interpret if the coordinate
system is rotated back into alignment with the LBS
by use of the general rotation matrix
R ⫽
冋
cos sin
sin cos
册
. (9)
By use of 45°, the return field becomes
E
ˆ
out
⫽ R45°E
out
⫽ i
冑
R
d
冉
cos
i sin
冊
. (10)
Therefore it can be clearly seen that the Kerr-induced
component E
y
i sin is always in quadrature with
the primary component E
x
cos , independent of the
sign of . Because is typically quite small, in the
range from 0.5° to 1.0°, E
ˆ
out
represents a slight-
ly elliptical polarization state with a major axis along
the x axis. When changes sign, only the rotation
sense of the polarization changes, something that is
not detected readily. Recalling that changes sign
with magnetization reversal at the disk, we can see it
is not possible to read out the recorded data at this
point. Before we proceed with differential detection,
additional manipulation is required either 1 to
bring the two components back in phase or 2 to keep
the components in quadrature but make their rela-
tive amplitudes change as changes sign. Such ma-
nipulation is accomplished easily with QWP2
positioned after the LBS as shown in Fig. 3.
We further analyze the first of these two detection
possibilities by continuing to propagate the beam off
of the LBS and through QWP2 oriented at 0°. The
appropriate Jones matrices are given by
S
LBS
⫽
冋
冑
R
x
0
01
册
for reflection, (11)
S
QWP2
0° ⫽
冋
10
0 i
册
, (12)
where R
x
represents the intensity reflection factor for
the x-polarized component, typically in the range
from 0.2 to 0.3. It is assumed that the LBS does not
introduce any phase shift between the x and y com-
ponents. In this case the optical field following
QWP2 in the LBS coordinate system is given by
E
ˆ
QWP2
⫽ S
QWP2
0°S
LBS
E
ˆ
out
⫽ i
冑
R
d
冉
冑
R
x
cos
sin
冊
, (13)
which represents linear polarization having an angle
of tan
1
sin 兾
R
x
cos with respect to the x axis.
As changes sign, the corresponding rotation can
be sensed readily by a differential detection system
oriented at 45°. The field in the detection coordinate
system is
E
detect
⫽ R45°E
ˆ
QWP2
⫽
i
冑
R
d
冑
2
冉
冑
R
x
cos ⫹ sin
冑
R
x
cos ⫹ sin
冊
. (14)
When 0, the optical field is polarized linearly at
45° with respect to the polarizing beam splitter
PBS, and the differential detector is balanced.
When the Kerr effect is present, the linear output
polarization rotates accordingly to produce a differ-
ence signal:
V
diff
兩E
detect, y
兩
2
⫺ 兩E
detect, x
兩
2
⫽ 2R
d
冑
R
x
cos sin ⯝ 2R
d
冑
R
x
for small .
(15)
For a fixed laser power, the sum signal is essentially
a constant:
V
sum
兩E
detect, x
兩
2
⫹ 兩E
detect, y
兩
2
⫽ R
d
R
x
cos
2
⫹ sin
2
⯝ R
d
R
x
. (16)
In practice, physical rotation of the differential detec-
tion system may not be convenient, so alternatively
we can use a half-wave plate HWP to rotate the
polarization by 45° while keeping the detection sys-
tem oriented at 0° with respect to the LBS.
The second detection scheme places QWP2 in the
detection channel at an angle of 45°. In this case the
optical field seen by the differential detector is given
by
E
detect
⫽ S
QWP2
45°S
LBS
E
ˆ
out
⫽
i
冑
R
d
冑
2
冋
冑
R
x
cos ⫺ sin
i
冑
R
x
cos ⫹ sin
册
. (17)
When 0, the optical field is circularly polarized
and the differential detector is again balanced. For
nonzero , the circular state changes to elliptical and
the detector goes out of balance. The major axis of
the ellipse is along either the x or the y axis, depend-
ing on the sign of . This arrangement provides pre-
cisely the same signal sensitivity as the above case,
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 695
although in a way it is actually preferred because it
does not require the differential detector to be rotated
or a HWP to be utilized.
C. Implementation Based on Low-Birefringence Fiber
An alternative approach that circumvents issues as-
sociated with PM fiber birefringence centers on use of
Lo-Bi fiber; in particular, mode partition laser noise
can be mitigated with Lo-Bi fiber see the companion
paper
4
. This fiber is constructed by means of a
slightly birefringent preform that is rotated during
the fiber draw process.
24
For this reason it is some-
times referred to as spun fiber. Although this type
of fiber can display a modest level of circular birefrin-
gence, essentially it is free of linear birefringence in
the sense that linear input polarization yields linear
output as long as the fiber is not bent or twisted in
any significant way and external forces e.g., because
of pinching or clamping are minimized. In most
implementations, however, the fiber will experience
some degree of bending and external perturbation,
and as a result linear birefringence will be introduced
locally. The principal axes of the bend-induced bi-
refringence align with the local plane of bending, and
the magnitude of the birefringence is proportional to
r
f
兾r
b
2
, where r
f
is the fiber radius and r
b
is the bend
radius.
25
If the fiber is routed on an arbitrary path,
we can obtain the effective Jones matrix by treating
the fiber as a continuum of infinitesimal slices, each
slice having a particular magnitude and orientation
of retardation. The overall matrix is then computed
as the product of these matrices. Because each in-
dividual retardation matrix is unitary, the product is
also unitary and can therefore be diagonalized by a
suitable rotation transformation.
26
In other words,
an arbitrarily routed Lo-Bi fiber can be treated as a
retardation plate with an overall phase and orienta-
tion being determined by its physical path and by the
way in which it is mounted in place.
In the context of a MO recording system, Lo-Bi
fiber can be used as shown in Fig. 4. A voltage-
controlled polarization rotator, situated between the
LBS and the fiber, is used to launch linearly polarized
light along one of the effective principal axes of the
fiber. Linearly polarized light then exits the fiber, is
incident on the disk, and picks up a Kerr rotation
component upon reflection. The main polarization
component then propagates back through the fiber on
the same effective axis that it followed on the forward
path, and the orthogonally polarized Kerr signal
propagates along the other axis. The phase shift
between these two components on the return path is
relatively small 2 and can be compensated
for by means of a dynamic retarder in the detection
arm. In this way, the phase shift between the Kerr
and the non-Kerr components is actively controlled to
be zero. This scheme works well for a FP laser be-
cause the birefringence of the fiber is intrinsically
small, meaning that the phase 2nL兾 is
essentially constant over the 2–3-nm wavelength
range of interest. In contrast, active phase compen-
sation does not work well for PM fiber because the
birefringence is substantially larger, making the
phase vary significantly with temperature, stress,
and wavelength.
In the drive application described here, the fiber is
routed necessarily along a bent path and experiences
dynamic bending as the actuator arm moves. Our
measurements with Lo-Bi fiber indicate that with
suitable routing, the static bend-induced birefrin-
gence dominates. In fact, the mechanically induced
dynamic part can be made negligible, meaning that
the phase compensator need only offset the static or
quasi-static contribution. The temperature depen-
dence of the bend-induced fiber birefringence is also
small,
25
making the system fairly robust. Therefore
the dynamic rotator and phase compensator initially
need to be calibrated before the drive operation and
are only subsequently adjusted as needed if the sys-
tem drifts. On a related note, use of liquid-crystal
LC devices to compensate for variations in plastic
substrate birefringence has been proposed previous-
ly.
27
We can accomplish proper adjustment of both the
rotator and the phase compensator by utilizing the
photodiode signals. First, the rotator is adjusted to
minimize the total detection power i.e., minimizing
the sum signal, thereby ensuring that the return
light is predominantly p polarized with only the small
Kerr component being s polarized. Then the com-
pensator is subsequently adjusted to obtain the max-
imum MO signal strength in the difference channel.
Both the polarization rotator and the phase com-
pensator can be constructed with electro-optic mate-
rials. For example, we used nematic LC devices
that are attractive because of their compact size, low
voltage operation, and potential low cost. A nematic
LC cell possesses intrinsic linear retardation, typi-
cally approximately 兾2, that is reduced upon appli-
cation of an electric field. Our devices have a size of
approximately 6 mm 6mm 1.5 mm thick and
are driven with a 0–8-V
p.p.
square wave. Their tem-
perature sensitivity, however, necessitates either 1
feedforward correction based on a temperature mea-
surement or 2 feedback correction calculated from
the sum and difference signals. A dynamic polariza-
tion rotator is readily constructed by use of a variable
LC retarder in combination with a static QWP.
28
Alternatively, mechanically rotating wave plates
can be used for the rotator and compensator ele-
Fig. 4. Optical path based on Lo-Bi fiber. Various lenses are not
shown for simplicity.
696 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
ments. Obviously a rotating HWP can serve as a
dynamic polarization rotator. Perhaps somewhat
less obvious is the fact that a static QWP, when used
in combination with a rotating HWP, can function as
a dynamic phase compensator. The QWP is ori-
ented at 45° to convert the two linear states into
circular states. The HWP, set to an angle , then
alters the relative phase of the two circular states by
an amount 4. This can be seen by the following
analysis in which two equal-amplitude circular states
are incident on the HWP:
E
phase
⫽
冋
cos 2 sin 2
sin 2cos 2
册冋
1
冑
2
冉
1
i
冊
⫹
1
冑
2
冉
1
i
冊
册
⫽
expi2
冑
2
冉
1
i
冊
⫹
expi2
冑
2
冉
1
i
冊
. (18)
Miniature versions of mechanical rotary devices can
be envisioned based on silicon micromachined tech-
nology.
29,30
One last issue regarding implementation of a Lo-Bi
fiber system centers on the folding mirror at the head.
Because the mirror is typically metal coated, the com-
plex reflectivities differ for s and p polarizations. If
the magnitudes of the reflectivities are the same,
then any retardation introduced by the mirror can be
lumped together with that of the fiber. However, in
general, the magnitudes are not the same. For ex-
ample, a gold-coated mirror at a 45° angle of inci-
dence has intensity reflectivities of R
s
96% and
R
p
92% and a relative phase difference of approx-
imately 24°. The polarization-dependent loss R
s
R
p
can cause coupling between the bend-induced fi-
ber axes if these axes do not align with those of the
mirror. Such coupling is not desired but can be tol-
erated if sufficiently small. Our measurements
show that use of a gold mirror in the Lo-Bi fiber
system has a minimal impact on the readout SNR, so
this type of lossy element can in fact be utilized.
3. Optical Head
A. Head Design
One way in which optical storage can be made more
competitive in the marketplace is to supply multiple
heads and disks in a single drive for greater box
capacity. This is the premise behind OAW technol-
ogy. The fabrication of miniature flying optical
heads requires the use of micro-optics with sufficient
quality to produce a diffraction-limited optical spot.
One advantage of a far-field flying head is that a focus
servo can be eliminated; the head simply follows the
vertical runout of the disk. However, there still
needs to be a means for high-bandwidth track follow-
ing. In our case, tracking is facilitated by use of a
Winchester-style rotary voice-coil actuator in combi-
nation with a small micromachined torsional mirror
mounted on the head.
31
A miniature optical record-
ing head based predominantly on surface microma-
chining technology has been proposed previously.
32
Unlike our approach, this alternative scheme, at
least as described, does not provide the fine actuator
functionality. Our approach is also different in the
sense that several heads share the same laser and
fixed optics module FOM, which minimizes the com-
plexity of the head.
For MO recording, the head also needs a low-
inductance coil capable of generating at least 250 Oe
of magnetic field perpendicular to the recording layer
for writing. The field polarity must be capable of
switching at data rates. We utilize a 20-turn two
10-turn layers pancake coil having an inductance of
approximately 100 nH with leads, driven by up to 50
mA zero-to-peak. The coil turns are fabricated from
Cu, and the yoke is made from Permalloy. Thin-film
plating technology is used to deposit the coil on a flex
substrate for handling purposes.
33
A hole in the cen-
ter of the coil allows the spot to reach the disk. The
coil aperture is elliptical in shape and has dimensions
of approximately 15 m along the track direction and
20 m cross track, with the longer dimension allow-
ing for radial spot movement.
An optical ray trace for the head based on PM fiber
is shown in Fig. 5. It constitutes a relatively simple
finite-conjugate optical system, which in our case was
designed to operate at 660 nm. Light exits the single-
mode fiber, is reflected by the micromachined mirror
toward the disk, passes through a first-order quartz
QWP, is collected by a micro-objective lens, and is
brought to a focus on the disk surface. Rotation of the
tracking mirror about the axis shown in Fig. 5 leads to
spot motion in the radial direction. Upon reflection
from the disk, the light propagates along the return
path and couples back into the fiber, forming a confocal
system with the above-mentioned features and poten-
tial benefits. We use the terminology backcoupling
efficiency to denote the fraction of light exiting the fiber
on the head that actually couples back into the fiber for
the return path, taking into account apodization by the
microlens and any aberrations that might arise from
the mirror and lens but not including absorption by
the disk surface.
The torsional silicon micromirror is fabricated with
surface micromachining techniques and is contained
in a silicon chip, as can be seen in Fig. 6, that is
Fig. 5. Ray trace for the optical head with PM fiber. Rotation of
the torsional microelectromechanical system MEMS mirror pro-
duces beam steering of the focused spot into and out of the page
for track following.
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 697
bonded to the slider body on a 45° chamfer. It is
coated with a gold film to make it suitably reflective.
The micromirror is approximately 0.2 mm in length
and 0.15 mm wide and has a resonant frequency of
approximately 50 kHz, more than sufficient for high-
bandwidth track following. Its angular deflection
range of 2 deg allows spot motion of approximately
4 tracks 0.72-m track pitch without moving the
coarse rotary actuator. Mirror actuation derives
from an electrostatic force, so the change in mirror
angle goes as the square of the drive voltage, with a
deflection angle of 2.0 deg occurring at approximately
140 V. More details about the design and perfor-
mance characteristics of the mirror can be found in a
separate publication.
34
An exploded view of the prototype head is shown in
Fig. 7. The ceramic slider body has dimensions of
1.98 mm length1.60 mm width0.89 mm
height and is designed to fly at a nominal height of
15 in. 375 nm above the disk surface. Various
grooves and channels are machined in the slider body
to accommodate the components. The fiber has a
physical diameter of 80 m and an optical mode-field
diameter of 4.3 m when 660-nm light 0.1 N.A. is
used. The parts are assembled by precision tooling
and held in place with adhesives. The microlens is
mounted on a custom silicon lens holder having a
snap-away handle that can be broken off easily once
the lens is bonded at the proper working distance.
After the lens, mirror, and QWP are bonded, the
fiber is placed in a machined channel on the upper
surface of the slider and actively positioned to achieve
maximum backcoupling while the slider is sitting on a
reflective surface that mimics the disk. If the slider is
shimmed above the reflective surface by an amount
approximately equal to the fly height, then this fiber
alignment procedure in turn ensures optimum focus
when the head is in operation. In addition, this ap-
proach allows small errors in lens placement, namely,
a deviation in the target working distance, to be com-
pensated by fiber position, although doing so changes
the lateral magnification i.e., spot size and the lens
fill factor and hence the fiber backcoupling efficiency.
Therefore the amount of compensation is dictated by
acceptable values of spot size and backcoupling effi-
ciency. The final step in the assembly process is at-
tachment of the coil such that the optical spot is
centered in the coil aperture.
B. Optical Performance
Two different versions of the microlens were imple-
mented with image-side N.A.’s of 0.71 and 0.83, both
single-element biaspheric designs. They are fabri-
cated by a press-molding technique
35
and experimen-
tally found to be diffraction limited. The higher-
N.A. lens has tighter tolerances and is therefore
somewhat more difficult to manufacture. The re-
mainder of the discussion here focuses on the 0.71-
N.A. lens that has a clear aperture of 0.250 mm, an
effective focal length of 0.150 mm, and a working
distance of 0.056 mm. The fact that the focal length
must be relatively insensitive to both temperature
and wavelength variations precludes use of a plastic
lens. The predicted thermal focus shift of the 0.71-
N.A. glass lens head is approximately 4 nm兾deg.
The head maintains a head–disk separation of ap-
proximately 15 8 in. 375 200 nm when flying
on a preembossed plastic disk. Accordingly it is im-
portant to understand how the optical system per-
forms with this amount of focus offset. Figure 8
shows that the change in spot size is essentially neg-
ligible over this range. As can be seen in Fig. 9, the
backcoupling efficiency and Strehl ratio decrease by
6.7% and 3.0%, respectively, with 200 nm of defo-
cus. In relative terms, the backcoupling efficiency is
Fig. 6. Silicon mirror chip containing a surface micromachined
torsional mirror for track following and short seeks. Photograph
courtesy of J. Drake.
Fig. 7. Exploded view of the head assembly. Drawing courtesy
of M. Darling.
Fig. 8. Spot size full width at half-maximum versus focus offset
for the 0.71-N.A. head with 660 nm.
698 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
much more sensitive to focus error than is the spot
size. Hence the usable depth of focus for our fiber-
coupled head is less than that for a conventional op-
tical head of the same N.A. The mode-matching
requirement for efficient fiber backcoupling also
makes the system more sensitive to lens and mirror
aberrations to the extent that they exist when com-
pared with a conventional optical head. Fortunately
the small size of the optical components, combined
with the fact that we are using first-surface media
with no plastic overcoat, makes it easier to achieve
good wave-front quality with high N.A.
To use the micromachined tracking mirror effec-
tively, the optics have to be corrected for a sufficiently
wide field of view. The two-stage servo system re-
quires that the mirror be capable of steering the spot
by 3 m. Figure 10 shows the wave-front error
versus the radial spot displacement per the nominal
design. Clearly the field of view of the lens is more
than sufficient to support the tracking requirements.
C. Practical Considerations
Our flying optical head technology is aimed at appli-
cations currently being served by magnetic disk
drives. In this market there is enormous pressure to
make high-capacity drives as inexpensive as possible.
A head in a conventional magnetic drive costs only a
few dollars to produce. Therefore, to be competitive,
an advanced flying optical head must also be able to
be produced at low cost in high volume. The proto-
type head design shown in Fig. 7 involves too many
pick-and-place steps to meet this criterion. What is
needed is a wafer-based design that is amenable to
mass production by use of, for example, an array of
diffraction-limited microlenses.
36,37
One can then
envision fabricating the coil directly on the bottom
lens surface.
After the head is fabricated, it is attached to a
suspension, as shown in Fig. 11, to form a head-
gimbal assembly. At this point, the head-gimbal as-
sembly looks much like that of a standard disk drive,
so it can be tested and subsequently assembled into a
head stack with tooling and equipment similar to
that used in the disk drive industry. Custom mod-
ifications are obviously required to accommodate
handling of the fibers. In this regard, one of the
more significant issues centers on the manner in
which the fibers from the head stack are connected to
the fixed optics module. For Lo-Bi fiber, fusion splic-
ing of multiple fibers in parallel i.e., ribbon splicing
is a possibility. For PM fiber, fusion splicing is com-
plicated by the need to align the fiber axes accurately,
so in this case it may be more feasible to implement
a manufacturing process that allows the fibers to be
inserted into the fixed optics module without the need
for splicing. These types of issues remain to be re-
solved before proceeding with high-volume manufac-
turing.
4. Fixed Optics Module
As can be seen in Fig. 2, the FOM consists of three
functional blocks: the laser source assembly, the de-
tection module, and the switching function. In this
section we describe the design and implementation of
these functional blocks, with an emphasis on the fiber
switch because it is the most novel aspect.
A. Source and Detector Modules
The laser assembly provides a collimated Gaussian
beam with a diameter of approximately 0.450 mm at
660 nm. We achieved this by first circularizing the
edge-emitting laser output with a microlens attached
Fig. 9. Normalized fiber backcoupling and Strehl ratio versus
focus offset.
Fig. 10. Wave-front error versus radial spot displacement. A
3-m displacement corresponds to slightly more than four data
tracks.
Fig. 11. Illustration of the head-gimbal assembly showing the
head mounted on a suspension. Drawing courtesy of M. Darling.
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 699
by Blue Sky Research.
38
The circularized 0.1-N.A.
beam is then collimated by a 2.0-mm focal-length lens
Geltech 350150. To help provide low-noise opera-
tion, an option is to include a small Faraday isolator
supplied by Tokin America Inc.
39
The isolator pack-
age 10-mm diameter, 11-mm length contains a
CdMnTe crystal and permanent magnet; it has a
clear aperture of 1 mm and provides better than 25
dB of isolation with an insertion loss less than 1 dB.
Both DFB and FP lasers were utilized. Because
DFB lasers are more susceptible to feedback-induced
mode hopping, we find in general that Faraday iso-
lation is a requirement for consistent low-noise sys-
tem operation. It is well established that for FP
lasers, sinusoidal rf modulation of the drive current
can significantly suppress the feedback effects in the
absence of an isolator.
40
We also employ rf modula-
tion in the frequency range from 350 to 650 MHz to
suppress laser noise that is due to feedback as well as
to various interference effects.
4
The detection module in the PM fiber prototype
drive is comprised of a zero-phase LBS cube 3.2-mm
linear dimension, R
p
0.25, R
s
1.0,
ps
0°,
5兾4 and 3兾2 wave plates, a PBS cube 3.2-mm lin-
ear dimension, and two Si p-i-n photodiodes
Hamamatsu S6468-10, ⭋0.4-mm active area, each
with a small focusing lens f 2.5 mm, ⭋3mm.
The wave plates have transverse dimensions of 3.2
mm 3.2 mm and are mounted in such a way that
they can be rotated during an active alignment as-
sembly step. The 3兾2 wave plate is used to com-
pensate for any residual phase shift in the LBS or the
switch mirrors. The Lo-Bi fiber version of the detec-
tion module uses a LC variable retarder in place of
the wave plates.
B. Fiber Bundle Switch
We accomplished the switching function by altering
the
x
and
y
angles of the collimated beam entering a
gradient-index GRIN lens, thereby changing the x, y
location of the spot in the back focal plane. We placed
a fiber bundle in the back focal plane so that the var-
ious fibers can be accessed by appropriately adjusting
the beam angle. This principle is illustrated in one
dimension in Fig. 12. When a PM fiber is used, the
axes of the fibers in the bundle must be aligned relative
to one another, as can be seen in Fig. 13.
High-efficiency coupling requires that there be lit-
tle or no angular offset between the chief ray of the
focused beam and the fiber core. Because the fibers
in the bundle are parallel to the optical axis, the chief
ray for each fiber location should also be parallel to
the optical axis. Such a telecentric configuration is
possible if, in the front focal plane of the GRIN lens,
the center of the collimated beam always remains
coincident with the optical axis. This requirement is
essentially met if the pivot point of the steering mir-
ror resides in the front focal plane of the coupling
lens, as can be seen in the one-dimensional 1-D
layout of Fig. 14. The 1-D concept is readily ex-
tended to two dimensions, with the recognition that
the pivot point of the first mirror resides in what
would be the front focal plane location if the second
mirror were not there folding the optical path.
To make the switch faster and more robust to shock
and vibration, a servo system was developed. It is
an indirect servo in the sense that the positions of the
x, y steering mirrors not the fiber-coupling efficiency
per se are being sensing and servo controlled, with
an underlying assumption that a unique set of mirror
angles corresponds to maximum coupling of a partic-
ular fiber. The mapping between mirror angles and
fiber locations is determined by a calibration step
during which the fiber coupling is monitored and is
subsequently stored in a look-up table. Recalibra-
tion of the look-up table can occur on a slow time scale
as the system drifts because of, for example, temper-
ature changes. This approach requires a method for
precise measurement of the mirror positions as the
input to the servo loop. We opted for an optical
scheme that uses a portion of the waste light from the
LBS. This secondary collimated beam propagates
parallel to the main beam and reflects off the same
steering mirrors. As shown in Fig. 14, a 1-D position
sensing detector
41
PSD is placed at some distance
20 mm from the front focal plane of the GRIN lens
Fig. 12. Illustration of the fiber bundle switch principle. We
performed the switching by varying the incoming angle of a plane
wave at the front focal plane of a GRIN lens and placing the fiber
bundle in the back focal plane. For a 0.25-pitch GRIN lens, the
front and back focal planes correspond to the front and back sur-
faces of the lens.
Fig. 13. Photomicrograph of a PM fiber bundle polarization-
maintaining and absorption-reducing PANDA style with 14 fi-
bers. The fiber cladding diameter is 80 m, making the bundle
diameter approximately 300 m.
700 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
so that any motion of the steering mirror leads to
motion of the collimated sensing beam on the PSD.
The 1-D PSD has an active area of 3 mm 3mmand
generates a differential current that provides a mea-
sure of the spot centroid location on the active sur-
face. A normalization circuit that computes A
B兾A B makes the system relatively immune to
laser power fluctuations. Spatial resolution of 1
part in 1000 was achieved in a 200-Hz bandwidth.
Because two-dimensional 2-D beam steering is em-
ployed, a nonpolarizing beam splitter is used to divide
the position sensing beam between two separate PSD
chips. We find that use of two 1-D PSD’s provides
better resolution than a single 2-D PSD.
In short, switch operation occurs as follows:
1. Initially, the fiber plane is raster scanned to
locate individual fibers.
2. At each fiber location, the x and y PSD signals
are recorded and stored in memory.
3. Subsequent switching occurs when we engage
the servo control loop to lock to appropriate PSD
readings.
4. Fiber coupling is monitored during drive oper-
ation by a dither motion, and the PSD look-up table
is updated as needed to stay at peak coupling.
In our implementation, the GRIN lens is 1.8 mm in
diameter, is 4.45 mm in length, and has a pitch of
0.23. Use of a pitch slightly less than 0.25 provides
a small air gap between the GRIN lens and the fiber
bundle to allow for fabrication tolerances. The mir-
ror actuators consist of miniature voice-coil actuators
with flexure hinges. The mirrors themselves consist
of a dielectric stack designed for zero polarization
phase shift at a 45-deg angle of incidence. A mirror
angle change of 2.5 deg leads to 160 m of spot
motion at the fiber bundle plane along with 1.0 mm
of sensing beam motion on the corresponding PSD.
Keeping the PSD signals stable to 1 part in 1000
leads to a focal spot position error of less than 0.15
m in both the x and the y directions, much less than
the fiber core radius, meaning that the fiber-coupling
efficiency stays within 0.3% of its maximum value.
We demonstrated a worst-case switching time i.e.,
one side of the bundle to the other of 6 ms, which was
limited by laser intensity noise on the PSD detectors.
The mechanics, however, are capable of supporting
1–2-ms switching.
The calculated coupling efficiency across the back
focal plane is shown in Fig. 15 as a contour plot. The
area of the fiber bundle is denoted by a dashed circle
with a 150-m radius. It can be seen that the cou-
pling uniformity is quite good in the bundle region,
with less than a 3% variation. The calculated values
in Fig. 15 assume ideal components, leading to al-
most perfect mode matching with the fiber. In prac-
tice, however, slight wave-front aberrations and
mode mismatch cause the coupling to be reduced by
10–15%. We find that the average coupling effi-
ciency for all 14 fibers is routinely greater than 80%,
with a standard deviation less than 3%.
C. Fixed Optics Module Integration
Although the basic architecture and operating prin-
ciples of the FOM are relatively straightforward, its
implementation is tricky given the number of small
components that must be packaged in a compact
space. Figure 16 shows an exploded view of the en-
tire PM fiber FOM, with the various components and
beam paths depicted. The fully assembled FOM
with the mounting support structure is illustrated in
Fig. 17. The overall dimensions are 1 in. 2.54 cm
1 in. 2.54 cm2 in. 5cm.
In the construction of the forward path, the laser
module, LBS, mirror actuators, and PSD’s are first
positioned and locked in place. The fiber bundle
with the GRIN lens, having been fabricated as a col-
Fig. 14. Ray trace of the bundle switch in two different positions
showing how the position sensing device PSD is used to deter-
mine mirror orientation. Only a 1-D beam-steering configuration
is shown for simplicity. After initial calibration, a servo loop
drives the system to the specific PSD value that corresponds to the
active fiber. Recalibration takes place as needed.
Fig. 15. Contour plot of the fiber-coupling efficiency as a function
of the focused spot location in the back focal plane of the GRIN lens
for the optical path comprising ideal components in their nominal
positions. The spot center is defined by the intersection of the
chief ray with the focal plane. The fiber bundle area is indicated
with a dashed circle, showing good coupling uniformity in this
region.
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 701
limator subassembly, is then attached. Active
alignment of the fiber bundle collimator is automated
by use of computer-controlled positioning stages that
optimize the average fiber-coupling efficiency and
uniformity. The fact that the switch can be electron-
ically programmed to find the fiber locations and to
optimize coupling is a significant benefit, not only
during assembly but also while in operation as de-
scribed above. Figures 18 and 19 show the sensitiv-
ity of the average coupling efficiency to placement of
the bundle collimator. Transverse and longitudinal
tolerances of 25 m and 250 m, respectively,
ensure that the average efficiency stays within 2% of
the maximum value. Such alignment is obtained
readily with the computer-controlled positioning sys-
tem. To complete the assembly, we add the return
detection path and actively align the photodiodes by
backcoupling light from one of the fibers. A Faraday
modulator is used to simulate a MO signal, allowing
the wave plates to be adjusted for maximum SNR.
We end this section by mentioning that the FOM
should be relatively low in cost for this technology to
be commercially feasible. As with the head, the de-
sign described here is meant only to demonstrate
technical feasibility. A more compact and economi-
cally competitive FOM can be constructed with sili-
con optical bench technology,
42
including silicon
micromachined rotary actuators. The fact that our
servo-based approach allows for calibration after
component placement means that a significant num-
ber of the switch components can be placed passively
with the actuators accommodating reasonable place-
ment errors. Use of passive assembly techniques
can reduce the cost dramatically.
5. System Performance
In this section we briefly describe the overall system
performance of the 5.25-in. 13.33-cm, half-height
drive with 6 disks and 12 heads. An illustration of
the drive with the cover plate removed is shown in
Fig. 20. The rotational speed is 4500 rpm. The
areal density target for this proof-of-concept proto-
type drive is 3.5 Gbits兾in.
2
10
5
bits兾in. linear den-
sity, 3.5 10
4
tracks兾in. track density.
Fig. 16. Exploded view of the entire FOM. Drawing courtesy of
M. Darling.
Fig. 17. Fully assembled FOM measuring 2 in. 5cmlength by
1 in. 2.5 cmwidth by 1 in. 2.5 cmheight. Drawing courtesy
of M. Darling.
Fig. 18. Graph of the average coupling efficiency across all 14
fibers as a function of the fiber bundle collimator offset in the
transverse x direction. The calculated efficiency is scaled by 0.864
to fit the experimental data. The results show that the average
efficiency can be maintained within 2% of its peak value if the
collimator is positioned with an accuracy of 25 m in the trans-
verse direction. Experimental data were provided by H. Lee.
Fig. 19. Graph of the average coupling efficiency across all 14
fibers as a function of the fiber bundle collimator offset in the
longitudinal z direction. The calculated efficiency is scaled by
0.864 to fit the experimental data. The results show that the
average efficiency can be maintained within 2% of its peak value if
the collimator is positioned with an accuracy of 250 m along the
optical axis. Experimental data were provided by H. Lee.
702 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
A. Preformatted Sampled Servo System
In striving for a low-cost disk solution, we chose to
utilize plastic substrate technology with tracks being
defined by stamped patterns. Double-sided stamp-
ing allows both sides of the disk to be used. This
approach avoids time-consuming servo writing as is
done currently with magnetic drives. However, be-
cause our head employs a single-mode fiber, only the
backcoupled light intensity can be used for the servo
signal. Spatial information in the diffracted beam is
not retained through the fiber, so conventional optical
servo schemes such as push–pull tracking
43
are pre-
cluded. Instead, in our system, a sampled servo ap-
proach is taken whereby the servo signal is generated
by a change in the return light intensity as the fo-
cused spot moves over the embossed pits and radial
grooves. The return light intensity is given by the
sum signal of the two photodetectors. Diffraction by
the patterned features on the disk reduces the sum
signal for two reasons. First, a significant portion of
the diffracted light is simply not collected by the ob-
jective lens as is the case in a conventional optical
drive. Second, the diffracted light that is collected
by the lens is not efficiently coupled back into the
fiber because of an angular offset between the dif-
fracted beams and the fiber’s cone of acceptance.
The result is a high-fidelity servo signal.
With a high-N.A. head, vector diffraction effects
become important.
44
In our case they can lead to a
polarization-dependent backcoupled light intensity
as shown in Fig. 21.
45
When we use our head with
PM fiber and a FP laser, the light incident on the disk
consists of a distribution of linear polarization states,
so the average diffraction efficiency is always ob-
tained. However, with PM fiber and a DFB laser,
there is only one linear polarization state illuminat-
ing the disk with an orientation that depends on the
fiber retardation. Because the fiber retardation can
vary over several waves with only a slight tempera-
ture change or modest bending, the polarization ori-
entation is changing constantly. Thus it is crucial to
make the diffraction from the embossed pattern po-
larization independent. In our case, this is accom-
plished by our making all marks circular, with radial
bars being approximated by a linear chain of circular
marks.
We also point out that, for the case of Lo-Bi fiber
and either style of laser, a fairly stable linear polar-
ization state is present at the disk surface. The po-
larization orientation depends on the fiber routing,
but with suitable routing it is possible to achieve
either in-track or cross-track polarization. It is con-
ceivable that the polarization orientation can be ac-
tively stabilized by a servo loop. The vector
diffraction response of Fig. 22 provides a means for
generating an error signal. The dynamic polariza-
tion rotator is employed to correct the error. In this
case small bar-type structures placed at 45° rela-
tive to the circumferential tracks would be needed to
bias the servo system in the linear region of the re-
sponse curve. With a stabilized polarization state,
land-groove recording becomes a possibility, allowing
for higher track density through cross-talk cancella-
tion.
46
Because the Lo-Bi fiber system already has a
phase compensator in the detection path, the system
can be made much more tolerant to groove depth
fabrication errors.
47
A photomicrograph of a typical servo wedge pat-
tern is displayed in Fig. 23, and the corresponding
servo signal is shown in Fig. 24. This pattern is a
unique design having high format efficiency. In the
5.25-in. 13.33-cm prototype drive, the rotary voice-
coil actuator provides a servo bandwidth of 700 Hz.
The second-stage fine-tracking mirror on the head is
capable of bandwidths greater than 15 kHz but is
limited by the servo sample rate. The overall loop
bandwidth is nearly 3 kHz, allowing for a short-seek
4 tracks and settle time of less than 0.5 ms. The
system seek-and-settle time, accounting for latency,
is less than 12 ms. This prototype drive has dem-
Fig. 20. Fully assembled 5.25-in. 13.33-cm, half-height OAW
drive without the top cover. Drawing courtesy of M. Darling.
Fig. 21. Plot of the modeled sum signal, including fiber backcou-
pling, as the 0.71-N.A. focused spot 660 nm traverses a
groove.
45
The signal is normalized by the power incident on the
disk, and the return path contains a LBS with a 30% reflectivity for
p-polarized light. The groove is assumed to be trapezoidal with a
width of 0.35 m at the top surface, a depth of 兾4, and a wall angle
of 65 deg. The surface is coated conformally with a quadrilayer
MO structure SiN 55 nm兾TbFeCo 20 nm兾SiN 10 nm兾Al 40
nm. The response for polarizations parallel and perpendicular
to the groove are seen to be very different.
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 703
onstrated a 3- track misregistration in the vicinity
of 70 nm. Some experimental drives displayed a
track misregistration as low as 32 nm, which extrap-
olates to a system capable of supporting track densi-
ties greater than 10
5
tracks兾in.
B. Magneto-Optical Recording
Data information in our drive is stored by means of
magnetic field modulation recording.
8
During re-
cording, the laser power is increased to provide ap-
proximately 8–10 mW of optical power at the disk,
which is sufficient to locally heat the recording layer
above its Curie temperature of 250 °C. The coil cur-
rent is then modulated so as to produce a series of
crescent-shaped magnetic domain marks. The
marks take on the polarity of the external field as the
trailing edge of the hot spot cools through the Curie
point. The medium used here consists of a TbFeCo
quadrilayer structure, designed for first-surface re-
cording. A liquid lubricant on the top surface of the
disk assists with flying, but it can become hot and
desorb during recording, thus contaminating the op-
tics on the head. Desorption can be minimized when
a media stack is designed with a thick top dielectric
100 nm and by use of a high molecular weight
lube. The thicker dielectric provides better thermal
isolation between the absorbing storage layer and the
top surface. Alternatively, a heat-resistant solid
lube could be developed for this application.
During readout, the optical power at the disk is
dropped to approximately 1.5 mW, a level that main-
tains data integrity. The total optical power during
readout that reaches the photodiodes is in the range
of 50–60 W, close to the media noise limit. The
recording performance is embodied in the roll-off
curve of Fig. 25 in which the SNR is plotted as a
function of mark length. These data correspond to a
spot size of 0.57 m, a media velocity of 15 m兾s, and
a bandwidth of 58 MHz. It can be seen that the
long-mark SNR is approximately 31 dB, falling to 25
dB at a mark length of 0.6 m. At an areal density
of 3.57 Gbits兾in.
2
and a data rate of 65 Mbits兾s, a raw
bit error rate of 9 10
5
is measured in a prototype
drive. At a spin stand, using our best heads and
media, we pushed the data rate to as high as 180
Mbits兾s at roughly the same areal density and bit
error rate.
48
6. Summary and Conclusions
We have described a new type of optical storage sys-
tem that allows for multiple heads and disks in a
Winchester-style drive, thereby yielding greater box
capacity. The technology is based on a flying micro-
optical recording head that contains a silicon micro-
machined torsional mirror for high-bandwidth track
following. Single-mode optical fibers provide light
delivery to and from the heads. Both PM and Lo-Bi
fiber systems have been implemented for MO record-
ing. A novel FOM with 1 N switching capability is
Fig. 22. Plot of the modeled sum signal, including fiber backcou-
pling, versus polarization angle when the focused spot is located at
the groove center. The signal is normalized by the power incident
on the disk. The optical system and media parameters are the
same as those for Fig. 21.
Fig. 23. Photomicrograph of the embossed servo wedge pattern.
Courtesy of N. Deeman.
Fig. 24. Sum signal corresponding to the servo pattern of Fig. 23.
Courtesy of A. Fennema and G. Szita. ID, identification.
Fig. 25. Roll-off curve showing the SNR versus the mark size for
the 0.71-N.A. OAW head. These data correspond to a spot size of
0.57 m, a media velocity of 15 m兾s, and a bandwidth of 58 MHz.
Data courtesy of S. Hallstein.
704 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
presented. Low-cost patterned media with em-
bossed features provide the basis for a sampled servo
system. The fact that the data signal is contained in
the optical polarization, whereas the servo signal is
carried by the intensity, means that cross talk be-
tween the two channels is minimized. It should be
obvious to the reader that this fiber-based technology
can also be applied to phase-change recording, which
in fact would be easier to implement because control
and detection of polarization is no longer needed.
We opted for MO recording, however, because it offers
virtually infinite cyclability at high data rates as well
as the prospect for enhanced areal density by ad-
vanced techniques such as magnetic superresolu-
tion.
13
Although the feasibility research centered on an
areal density of 3.5 Gbits兾in.
2
, higher densities can be
achieved through a near-field version of the head,
with or without multilayer recording capability. Al-
ternatively, an optically assisted thermomagnetic re-
cording head with flux detection may prove to be
more competitive. If so, the optical system de-
scribed here, or some variant of it, may form the basis
for such a technology. To make this architecture
commercially viable, however, the head and FOM
need to be designed in such a way that they can be
mass produced at low cost. In particular, wafer-
based manufacturing techniques should be employed
to minimize the number of discrete components that
are handled during fabrication.
The development of this new optical drive technol-
ogy involved the hard work of many talented staff
members at Quinta Corp., a wholly owned subsidiary
of Seagate Technology Inc. It was truly a team ef-
fort, led by Joseph E. Davis, who was vice president
of Product Development at Quinta during the devel-
opment. In addition, we thank M. Mansuripur at
the University of Arizona for his support and contri-
butions to this effort.
References and Notes
1. T. Suhara and H. Nishihara, “Integrated-optic disc pickup de-
vices using waveguide holographic components,” in Holo-
graphic Optics II: Principles and Applications, G. Morris, ed.,
Proc. SPIE 1136, 92–99 1989.
2. H. Ukita, Y. Katagiri, and S. Fujimori, “Supersmall flying
optical head for phase change recording media,” Appl. Opt. 28,
4360–4365 1989.
3. S. Renard and S. Valette, “Magneto optical reading and writing
integrated heads: a way to a multigigabyte multi-rigid-disk
drive,” in Optical Data Storage 1991, J. J. Burke, T. A. Shull,
and N. Imamura, eds., Proc. SPIE 1499, 238–247 1991.
4. J. P. Wilde, A. A. Tselikov, G. R. Gray, Y. Zhang, and S.
Gangopadhyay, “Magneto-optical disk drive technology using
multiple fiber-coupled flying optical heads. Part II. Laser
noise considerations,” to be submitted to Appl. Opt.
5. J. P. Wilde, J. E. Hurst, and J. F. Heanue, “System and method
using optical fibers in a data storage and retrieval system,”
U.S. patent 5,850,375 15 December 1998.
6. J. P. Wilde, J. E. Davis, J. E. Hurst, J. F. Heanue, and J.
Drazan, “Optical system and method using optical fibers for
storage and retrieval of information,” U.S. patent 5,940,549 17
August 1999.
7. J. Davis, “Beyond the superparamagnetic limit. II: Far-
field recording,” in Data Storage PennWell, Amsterdam, The
Netherlands, 1998, pp. 33–36.
8. M. Mansuripur, The Physical Principles of Magneto-optical
Recording Cambridge U. Press, Cambridge, UK, 1995.
9. T. W. McDaniel and R. H. Victora, eds., Handbook of Magneto-
Optical Data Recording Noyes, Westwood, N.J., 1997.
10. B. D. Terris, H. J. Mamin, D. Rugar, W. R. Studenmund, and
G. S. Kino, “Near-field optical data storage using a solid im-
mersion lens,” Appl. Phys. Lett. 65, 388–390 1994.
11. I. Ichimura, S. Hayashi, and G. S. Kino, “High-density optical
recording using a solid immersion lens,” Appl. Opt. 36, 4339–
4348 1997.
12. H. Yoshikawa, T. Ohkubo, K. Fukuzawa, L. Bouet, and M.
Yamamoto, “Readout characteristics of a near-field optical
probe as a data-storage readout device: submicrometer scan
height and resolution,” Appl. Opt. 38, 863–867 1999.
13. Y. Tanaka, M. Kurebayashi, Y. Murakami, and S. Yonezawa,
“Short-marks recording characteristics of laser-pumped
magnetic-field-modulation recording in a narrow track pitch
and on magnetically induced superresolution disks,” Appl.
Opt. 37, 2699–2707 1998.
14. H. Saga, H. Nemoto, H. Sukeda, and M. Takahashi, “New
recording method combining thermo-magnetic writing and
flux detection,” Jpn. J. Appl. Phys. 38, 1839–1840 1999.
15. H. Katayama, M. Hamamoto, J. Sato, Y. Murakami, and K.
Kojima, “New developments in laser-assisted magnetic record-
ing,” IEEE Trans. Magn. 36, 195–199 2000.
16. K. A. Belser, T. McDaniel, J. E. Davis, and J. E. Hurst,
“Magneto-resistive magneto-optical head,” U.S. patent
5,889,641 30 March 1999.
17. F. S. Barnes, K. S. Lee, and A. W. Smith, “Use of optical fiber
heads for optical disks,” Appl. Opt. 25, 4010–4012 1986.
18. M. N. Opsasnick, D. D. Stancil, S. T. White, and M. Tsai,
“Optical fibers for magneto-optical recording,” in Optical Data
Storage 1991, J. J. Burke, T. A. Shull, and N. Imamura, eds.,
Proc. SPIE 1499, 276–280 1991.
19. Quinta兾Seagate demonstrated various versions of the OAW
drive technology at COMDEX 1998, Las Vegas, Nev., Novem-
ber 1998.
20. G. T. Sincerbox, “Miniature optics for optical recording,” in
Gradient-Index Optics and Miniature Optics, Vol. 935 of SPIE
Critical Review Series, D. C. Lerner and J. D. Rees, eds. SPIE,
Bellingham, Wash., 1988, 63–76.
21. J. Noda, K. Okamoto, and Y. Sasaki, “Polarization-
maintaining fibers and their applications,” J. Lightwave Tech-
nol. LT-4, 1071–1089 1986.
22. See, for example, the 3M Specialty Single-Mode Fiber: 3M™
Single-Polarization Fibers, Product Data Application Note
3M Specialty Optical Fibers, West Haven, Conn., 1995.
23. A. B. Marchant, Optical Recording: A Technical Overview
Addison-Wesley, Reading, Mass., 1990,p.83.
24. R. E. Schuh, X. Shan, and A. S. Siddiqui, “Polarization mode
dispersion in spun fibers with different linear birefringence
and spinning parameters,” J. Lightwave Technol. 16, 1583–
1588 1998.
25. Z. B. Ren, Ph. Robert, and P.-A. Paratte, “Temperature depen-
dence of bend- and twist-induced birefringence in low-
birefringence fiber,” Opt. Lett. 13, 62–64 1988.
26. See, for example, C. Cohen-Tannoudji, B. Diu, and F. Laloe¨,
Quantum Mechanics Wiley, New York, 1977, Chap. 2, pp.
176–181.
27. E. C. Gage, “Apparatus and method for optimizing perfor-
mance in an optical storage system read兾write head,” U.S.
patent 5,347,297 13 September 1994.
28. See, for example, the product guide for Meadowlark Optics,
Frederick, Colo. www.meadowlark.com.
29. D. A. Horsley, A. Singh, A. P. Pisano, and R. Horowitz, “An-
10 February 2001 兾 Vol. 40, No. 5 兾 APPLIED OPTICS 705
gular micropositioner for disk drives,” in Proceedings of the
Tenth Annual International Workshop on MEMS Asian Tech-
nology Information Program, Tokyo, 1997, pp. 454–459.
30. W. C. Tang, T. H. Nguyen, and R. T. Howe, “Laterally driven
polysilicon resonant microstructures,” Sens. Actuators 20,
25–32 1989.
31. J. P. Wilde, J. E. Davis, J. E. Hurst, Jr., J. F. Heanue, K.
Petersen, T. McDaniel, and J. Drazan, “Flying optical head
with dynamic mirror,” U.S. patent 6,044,056 28 March 2000.
32. L. Y. Lin, J. L. Shen, S. S. Lee, and M. C. Wu, “Realization of
novel monolithic free-space optical disk pickup heads by sur-
face micromachining,” Opt. Lett. 21, 155–157 1996.
33. T. McDaniel and Y. Wang, “Coil for use with magneto-optical
head,” U.S. patent 5,903,525 11 May 1999.
34. J. Drake and H. Jerman, “A micromachined torsional mirror
for track following in magneto-optical disk drives,” in 2000
Solid-State Sensor and Actuator Workshop, Technical Digest
Transducers Research Foundation, Inc., Cleveland, Ohio,
2000,p.10.
35. The microlenses were fabricated by Geltech Inc., Orlando, Fla.
www.geltech.com.
36. See, for example, the various microlens arrays made by
MEMS Optical, LLC Huntsville, Ala. on their website www.
memsoptical.com.
37. J. F. Heanue and M. A. Wardas, “High numerical aperture
objective lens manufacturable in wafer form,” U.S. patent
6,049,430 11 April 2000.
38. The CircuLaser diode is manufactured by Blue Sky Research,
San Jose, Calif. www.blueskyresearch.com.
39. Tokin America Inc., San Jose, Calif. www.tokin.com.
40. G. R. Gray, A. T. Ryan, G. P. Agrawal, and E. C. Gage, “Control
of optical-feedback-induced laser intensity noise in optical data
recording,” Opt. Eng. 32, 739–745 1993.
41. See, for example, PSD devices made by Hamamatsu Photonics
www.hamamatsu.com. The PSD used here is a custom part
made for Quinta兾Seagate by Hamamatsu Corp., San Jose,
Calif.
42. J. F. Heanue, J. P. Wilde, J. E. Hurst, Jr., and J. H. Jerman,
“Data storage system having an optical processing flying
head,” U.S. patent 6,034,938 7 March 2000.
43. M. Mansuripur, “Analysis of astigmatic focusing and push–
pull tracking error signals in magnetooptical disk systems,”
Appl. Opt. 26, 3981–3986 1987.
44. M. Mansuripur, “Certain computational aspects for vector dif-
fraction problems,” J. Opt. Soc. Am. A 6, 786–805 1989.
45. The vector diffraction modeling is carried out with DIFFRACT
a product of MM Research, Tucson, Ariz. in combination with
Delta supplied by L. Li, University of Arizona, Tucson, Ariz..
46. T. D. Goodman and M. Mansuripur, “Optimization of groove
depth for cross-talk cancellation in the scheme of land-groove
recording in magneto-optic disk systems,” Appl. Opt. 35, 1107–
1119 1996.
47. A. Fukumoto, S. Kai, S. Masuhara, and K. Aratani, “Magneto-
optical detection using an optical phase shifter in higher track
density land兾groove recording,” Jpn. J. Appl. Phys. 37, 2144–
2149 1998.
48. W. D. Huber and D. A. Schmid, “High data rate magneto-
optical recording,” in Optical Data Storage 2000, D. G. Stinson
and R. Katayama, eds., Proc. SPIE 4090, 252–257 2000.
706 APPLIED OPTICS 兾 Vol. 40, No. 5 兾 10 February 2001
