Patterned media for future magnetic data storage

Abstract

One strategy to delay the onset of superparamagnetism and achieve magnetic storage densities approaching 1 Tb/in.² is the use of lithographically patterned magnetic media. While one of the main advantages enjoyed by magnetic recording is low cost due to use of featureless media, there are several advantages that may be realized by patterning the medium in a hard disk drive. The commercial success of patterned media will of course depend on the relative costs and gains. In particular, there are three main types of disk patterning proposed, each requiring a different length scale of patterned feature. Patterning of servo marks for maintaining the head position on-track, the fabrication of discrete tracks, and the fabrication of discrete bits have all been proposed and will be discussed. For discrete bit recording single domain magnetic islands are required, and one approach to fabricating these, by depositing magnetic films onto prepatterned substrates, is described. The switching characteristics of the islands as compared to those of the full film, along with initial recording results are presented.

1 Introduction

Current magnetic disk drives contain disks having approximately 100 Gb/in.² and technology demonstrations of 130 Gb/in.² have been published (Stoev et al. 2003). Very recently, companies have announced technology demonstrations of greater than 230 Gb/in.² using new, perpendicular, recording media. The achievable areal density is limited by thermally activated grain reversal, or the superparmagnetic effect, which occurs as grain sizes are scaled down to maintain signal to noise as the bit size is reduced (Moser et al. 2002). It is anticipated that by changing from longitudinal anisotropy media to perpendicular media disk drives will achieve densities approaching 500 Gb/in.² However, to progress beyond this density, new media architectures, such as those achieved by patterning, may be required.

While nanofabrication of recording media may provide some performance advantage, as discussed below, it will likely also entail an increase in cost. One of the key advantages to magnetic recording over other data storage technologies, such as DRAM, is the use of a continuous, featureless, recording medium. The magnetic recording head is fabricated using conventional lithography processes, but even here the number of processing steps is small as compared to semiconductor based storage. On a typical wafer of heads, there are roughly 40,000 heads, each requiring a few features at the minimum feature size, and thus on the order of 10⁵ features at the smallest dimension are required per wafer. This compares to 10¹⁰ minimum size features that are typically required on a semiconductor memory wafer. Therefore, to the extent that the device price will depend on fabrication costs, disk drives have a huge advantage over semiconductor memory. Given the inherent cost advantage of continuous, unstructured, media used in magnetic recording, nanofabrication must provide a compelling advantage to be economically feasible.

There are currently three types of patterning under consideration, patterned servo marks, patterned tracks and patterned bits, as shown schematically in Fig. 1. We introduce each of these below, in order of increasing advantage or gain, and likely also in order of increasing cost. We then focus on our recent work on the most challenging of these, patterned bits, and conclude with studies of the reversal mechanism and initial recording results on magnetic islands fabricated by etching the substrate.

Fig. 1

Schematic of patterned servo, discrete track and discrete bit media

2 Magnetic servo patterning

Currently, the magnetic servo pattern in a disk drive has to be recorded on all the disks, (drives contain between 1–6 disks) in a serial process that is both slow and expensive. This pattern is used to define the track positions enabling the head to repeatability write and read the data bits. The servo writing process consists of flying the head over the entire disk surface and recording the servo marks. If the throughput of this process could be improved, there is a potential for significant cost savings. In addition to servo write, two other applications would benefit from a fast process for copying magnetic data: drive formatting and distribution of digital content, such as software and music. However, the serial process of track-by-track recording in a hard disk drive is very slow and what is needed is a magnetic equivalent to the injection molding or stamping processes that are used in high volume distribution optical media like CDs and DVDs.

Contact magnetic printing, or magnetic lithography, is one example of such a parallel magnetic patterning process. Here, a magnetic pattern on one substrate, the master or mask, is transferred using an applied magnetic field to a magnetic film on a second substrate, sometimes referred to as the slave. In a typical example of the process Bandic et al. (2003a, b) transferred a magnetic pattern from a “magnetic mask” or master using an applied in-plane magnetic field. The slave disk is first magnetized uniformly in one direction by an external magnetic field. Then, the magnetic mask, in this case a surface patterned with a soft magnetic material such as NiFe, is placed in close contact with the disk and the direction of the external magnetic field is reversed. The magnetic mask selectively shields the external field so that only the unshielded areas are affected by the applied field allowing the pattern to be transferred from mask to slave. This process requires intimate contact between the mask and disk, which has been demonstrated using flexible PET magnetic masks (Bandic et al. 2003a, b). Magnetic patterns have also been transferred when the master is fabricated by etching a substrate and subsequently coating the entire master surface with a soft magnetic film, so that the grooves as well as the lands are coated. In this case the additional gap between the master and disk due to the substrate etching prevents magnetic shielding from the grooves and the pattern is transferred (Sugita et al. 2003). While this technique of magnetic lithography has shown promise transferring magnetic features on 100 nm length scales, it remains to be demonstrated that it can achieve the tens of nanometer resolution required for future high density products (Driskill-Smith 2004).

3 Discrete track patterning

In conventional magnetic recording, both the bit length, along the track, and the bit width, across the track, are defined by the write field from the recording head. The track locations are defined by the prewritten servo marks, as discussed above. However, due to mechanical vibrations and non-perfect track following of the head, there is some misregistration of the head to the track locations from revolution to revolution. To ensure a maximum read back signal and minimum noise from poorly written edges of bits, the write head is wider than the read head, typically a read-to-write width ratio of ∼0.6 is used. The noise introduced by this misregistration between the write and read head locations can be reduced by physically defining the track positions on the disk. The regions between the tracks should be either nonmagnetic or recessed from the head such that the read element senses little or no magnetic response from them. This approach reduces the problem of reading poorly written and partially erased information at the edges of the track, which is a significant source of noise. Thus, by using discrete tracks, the read head can be made wider and can be comparable to the write head width. With the track width physically defined by the medium, the requirements on head tolerances may be eased, thus improving manufacturing yield.

The approaches for discrete track media which are potentially manufacturable, such as using embossing or nanoimprinting, have been demonstrated and in general involve patterning the substrate prior to the thin film deposition (Ishida et al. 1993; Wachenschwanz et al. 2005). Wachenschwanz et al. (2005) created discrete track media by first etching grooves into the electroplated NiP layer, which is a normal part of a Al-disk, and then depositing a CoPtCr-based longitudinal medium onto the topographically patterned NiP. The walls of the 60 nm deep grooves serve to isolate the tracks and the grooves are deep enough so as to reduce the read back signal amplitude in the head. Recording results at the highest densities, 500 kilo-flux-changes/in. (kfci), showed up to 35 dB reduction in trench read back signal, with a smaller reduction at lower linear density. These experiments were based on a master having a track pitch of 800 nm, whereas to exceed the projections for current technology track pitches below 100 nm will be required. It remains to demonstrated that recording on tracks of such dimensions will have the required signal-to-noise (SNR).

4 Discrete bit patterning

The most demanding, and potentially the most advantageous, application of nanofabricated media is discrete bit media (Terris and Thomson 2005; Ross 2001; Lodder 2004). As magnetic grain sizes in conventional media are reduced, the magnetic energy per grain becomes too small to prevent thermally activated reversals. In order to keep thermally activated reversal at an acceptable level, K _u V/k _B T, where K _u is the magnetic anisotropy, V the magnetic switching volume, k _B Boltzman’s constant, and T the temperature, must remain greater than approximately 60 for conventional longitudinal media (Weller and Moser 1999). However, to maintain sufficient SNR, it is desirable to maintain the number of grains per bit as the density is increased. Thus, the grain volume must be reduced. However, K _u cannot be increased without bound in order to maintain K _u V, as the required magnetic field to write a bit increases with K _u. The write field from a longitudinal recording head is limited by the saturation magnetization of the pole material, and current head designs cannot generate more than approximately 10 kOe of write field in the recording media. This limits the diameter of thermally stable grains which can be written with current heads to around 8 nm (for a grain of equal diameter and thickness) (Moser and Weller 2001).

However, for discrete bit media, the SNR argument is very different. In this case, the number of grains, or more correctly the number of magnetic switching volumes, per bit is reduced to 1, and there is no statistical averaging over many grains to reduce noise. The transitions, both along and cross track, are now defined by the disk patterning and not by the head fields. The switching volume is now equal to the bit size, and thus islands as small as 10 nm and below will be thermally stable and still writeable with current recording heads.

High density recording requires thin media in order to achieve sharp bit transitions (Bertram 1994) and thin media will be required for patterned bit media as well. While the transitions are not defined by the head fields in patterned bit recording, the linear density of the bits will be limited by the media thickness in order to prevent the head from unintentionally writing neighboring islands with the tails of the head field gradient. The taller the pillar, the greater the head field a neighboring bit will experience, and hence the more likely it is to be inadvertently written. Thus, media with a thickness on the order of the bit spacing will be required. The symmetry of hard disk recording favors media with perpendicular anisotropy, as higher head write fields can be realized. Thus, given the media thickness limitation we conclude that it will be difficult to use tall pillars for high density recording. This implies that shape anisotropy cannot be used to achieve the necessary perpendicular anisotropy. The most likely perpendicular media are based on using interfacial anisotropy, such as Co/Pt or Co/Pd multilayers, or media similar to that proposed for perpendicular recording based on CoPtCr alloys grown with the c-axis normal to the substrate. To achieve single domain islands, it will be desirable to have high exchange coupling within the individual islands, rather than the low exchange desired in conventional magnetic recording.

5 Discrete bit fabrication and testing

Prototype patterned magnetic structures have been fabricated using focused ion beam (FIB) milling from a variety of magnetic thin films, including perpendicular granular media based on CoPtCr (Rettner et al. 2001, 2002). In this case, the film between the intended islands was not physically removed but rather poisoned by Ga⁺ and C-overcoat implantation. Islands below approximately 70 nm in diameter were observed to be single domain, as shown in Fig. 2, while larger islands could support a remanent state containing domain wall.

Fig. 2

AFM and MFM images of a 2×2 μm FIB patterned region are shown in a and b, respectively, along with higher magnification images of small regions in c and d, respectively. The pattern period is ∼100 nm. From reference (Rettner et al. 2001)

Coupons of thin CoPtCr-based perpendicular media were patterned into islands with periods as small as 56 nm by FIB and recorded on using a quasi-static read/write tester (Rettner et al. 2002; Albrecht et al. 2003, 2002a, b; Lohau et al. 2001a, b) as shown in Fig. 3. Results showed that, as expected, the write pulse needed to be synchronized with island position since transitions cannot be placed in the islands but must occur between islands. A simple model where the head field was integrated across the island was successful in predicting the island switching probability. The timing window for correct writing depended on both the write field gradient and the switching field distribution (SFD) of the island array (Fig. 4) (Albrecht et al. 2002a). In principle the timing window can be almost as large as the island period, but this will require sharp head field gradients and narrow island SFDs, in order to ensure that the desired island is written without disturbing neighboring islands. The effect of head field gradient can be seen by comparing patterned expitaxial longitudinal Co films, where the higher longitudinal head field gradient is utilized, to the patterned CoPtCr-based perpendicular films. A dramatic increase in the write margin was observed even though the SFDs of the two films were similar (Albrecht et al. 2003). Likewise, a narrower SFD would result in a wider write window for equal field gradients.

Fig. 3

a Longitudinal Co $(10\ifmmode\expandafter\bar\else\expandafter\=\fi{1}0)$ medium. Top: AFM image of an island row with a period of 87 nm. Bottom: MFM image after in-phase writing. b Perpendicular Co₇₀Cr₁₈Pt₁₂ medium. Top: AFM image of an island row with a period of 103 nm. Bottom: MFM image after in-phase writing. For the perpendicular case the islands in the MFM image are either black or white depending on their magnetization, while for the longitudinal medium each island appears as a dipole with a black and white contrast on either side of the island. From reference (Albrecht et al. 2003)

Fig. 4

Probability of addressing islands correctly as a function of phase shift for patterned longitudinal (p=60 nm) and patterned perpendicular (p=103 nm) media. Corresponding results obtained from a simulation are indicated by dotted and dashed lines. The phase shift is the relative position of the island and the edge of the write field gradient, and is defined as zero when the head field reverses at the edge of the island

The promise of bit patterned media is increased areal density by the use of single domain islands having increased thermal stability. However, this can be realized only if the bit patterned media also has sufficient SNR. While single domain bits will eliminate traditional jitter noise due to grain irregularities in the transitions, lithography noise will result in a new form of jitter. When comparing jitter of patterned islands of CoPtCr-based media with continuous media Albrecht et al. (2002b) found that the jitter of the islands was independent of number of islands in the cross-track direction, as would be expected for correlated lithography noise. This is in contrast to the case of conventional continuous media, where the jitter decreases as the inverse square root of track width, as expected for grain averaging. In experiments where Co/Pt multilayer films were deposited onto SiO₂ islands, fabricated as described below, Moritz et al. (2002) found that the SNR did not increase with bit length, in contrast to continuous media. This is also consistent with SNR dominated by position jitter for the patterned islands. A jitter of 12 nm was estimated, in reasonable agreement with the value obtained for lithography jitter of 6 nm as measured by scanning electron microscope (SEM) (Moritz et al. 2004).

While FIB is a very convenient laboratory tool for fabricating test structures over small areas, it lacks the throughput and speed to be a manufacturing method for patterned media. Of the more manufacturable techniques being pursued, one of the most promising approaches is to create magnetically isolated single domain islands by depositing magnetic films onto pre-patterned substrates. Here electron beam lithography and nanoimprinting can be combined so that expensively produced electron beam masters are cheaply replicated using nanoimprinting. In one approach, the nanoimprinted pattern is then transferred to the substrate by etching, resulting in topographic pillars on the substrate. Magnetic films are then deposited onto the pre-patterned substrates resulting in single domain magnetic islands produced by a potentially manufacturable process (Moritz et al. 2002; McClelland et al. 2002).

In the work described here, we focus on the magnetic reversal properties of such islands. The magnetic single domain islands were made by DC magnetron sputtering Co/Pd multilayers onto SiO₂ substrates patterned by electron beam lithography and subsequent reactive ion etching (Hu et al. 2004). This is a simplified process as nanoimprinting is not required. The multilayers were sputtered at room temperature using an Ar pressure of 3 mTorr. In order to ensure multilayers with perpendicular anisotropy a Pd seed layer of thickness 30–50 Å was sputtered prior to the Co/Pd multilayer deposition. A thin capping layer, Pd=10 Å, was also included in the stack for corrosion protection.

Figure 5a shows a cross section SEM image of 100 nm wide electron beam patterned lines separated by 50 nm trenches and covered by a SiO₂/50 Å Pd/(3 Å Co/8 Å Pd)₁₀/10 Å Pd multilayer film. The image shows that both the top of the lines and the bottom of the trenches separating the lines were coated with the nominal film thickness, while a much thinner film was deposited on the trench side walls. To achieve domain isolation on the pillar tops, deep trenches and steep side walls are desirable. At a minimum, the trench depth must exceed the film thickness to preclude direct contact between the magnetic layers on pillar tops and trench bottoms. Thus, domain isolation can be controlled via either the trench depth or film thickness. It is difficult to fabricate trenches much deeper than the lateral dimension of the islands. On the other hand, thin films will lead to both reduced readback signals and reduced thermal stability. We have found a compromise design which meets these requirements with a trench depth of 24 nm and the multilayer film thickness of about 10 nm. Figure 5b shows atomic force microscope (AFM) and Fig. 5c magnetic force microscope (MFM) images of 30 nm islands arranged in a hexagonal array with a 60 nm periodicity. Figure 5c is an MFM image taken following AC demagnetization of the magnetic islands showing the islands reverse independently and appear single domain in the remanent state. The random distribution of the switched islands is a good indication that the islands are magnetically isolated.

Fig. 5

a SEM image of lines (100 nm wide) covered by a multilayer of SiO2/50 ÅPd/(3 ÅCo/8 ÅPd)₁₀/10 ÅPd. b AFM image of a 30 nm island array with 60 nm periodicity; c MFM image of the same array after AC demagnetization

6 Magnetization reversal in patterned media island arrays

To investigate the reversal properties of the islands, MFM and the magneto-optical kerr effect (MOKE) were used to measure remanent and angle dependent remanence curves. Figure 6 shows remanence curves for island arrays (diamonds and solid circles) fabricated under the same sputtering conditions as the continuous films. Note that under the different sputtering conditions, the islands show very similar magnetic reversal behaviors, whereas those for the continuous films are markedly different. This observation indicates that the reversal properties of the continuous film do not have an obvious correlation with the switching behavior of the small islands. The data show that the coercivity of the islands is significantly greater than that of the continuous film, with the coercivity of these 30 nm islands more than an order of magnitude higher than that of the continuous film. Moreover, the SFD of the islands is clearly very wide, and is much greater than the calculated width assuming just a change in the magneto-static fields due to reversal of neighboring islands (∼400 Oe). These two effects become more pronounced as the island size gets smaller and have been seen for islands over sizes ranging from 5 μm to 30 nm (Hu et al. 2005a). Based on modeling results of patterned media recording (Hughes 2003), island SFDs having a normalized Gaussian width of less the 10% will be required, clearly much sharper than those shown in Fig. 6. Achieving such narrow switching distributions remains one the challenges to implementing patterned media.

Fig. 6

Normalized remanence curve of a continuous film (squares) measured by MOKE, together with 50 nm (diamonds) and 30 nm islands (circles) measured by counting islands in MFM images. The 5×5 μm MFM images were taken in the AC demagnetized states of the continuous films. a Both the seed layer and the multilayer were grown at 3 mTorr Ar pressure. b The seed layer and the multilayer were grown at 3 and 10 mTorr Ar pressure, respectively. c The deposition pressure was 10 mTorr for both seed layer and multilayer

In order to understand the origin of these differences we investigated the reversal mechanism of the film and the island arrays. The continuous film exhibits a very abrupt switching behavior as indicated by the solid squares in Fig. 6 and has a coercivity much less than the anisotropy field of approximately 6 kOe as measured using a hard axis hystersis loop (Hu et al. 2005b). These observations are consistent with a model whereby continuous films reverse via nucleation at a low field followed by rapid domain wall propagation.

In contrast to continuous films, the reversal of islands is quite different. Experimentally, we cannot obtain a multidomain remanent state in islands with a diameter of less than 200 nm, due to the high exchange coupling in Co/Pd multilayers. We expect that at some critical island diameter the magnetization will always be collinear and hence the island will be a single domain particle reversing by coherent rotation. In this case, in the limit of negligible magnetostatic fields, the zero temperature island switching field is determined by its anisotropy and magnetic moment. The sensitivity of the anisotropy of Co/Pd multilayers to Co thickness allows the coercivity or switching field of the island arrays to be investigated as a function of anisotropy. The island anisotropy was varied by changing the Co layer thickness from 2 to 6Å in 1Å steps while keeping the saturation magnetization (M _s) constant at 511±37 emu/cm³ by maintaining the Co/Pd thickness ratio at 3/8.

The calculated anisotropy field H _k, based on K _u measurements for the full film and corrected for island shape (Hu et al. 2005b), is plotted in Fig. 7a and b for the 50 nm and 30 nm islands (open circles), respectively. In the case of coherent rotation this is the switching field at T=0 K when the field is applied along the anisotropy axis. Figure 7 clearly shows that the anisotropy field calculated in this way gives a better measure of the island switching field trends than of the continuous film but is not in quantitative agreement with the experimental data. The model used does not include thermal activation effects which, since the measurements were conducted at room temperature and on a time scale of 3 s, are significant. The coercivity (H_c) of the island arrays at room temperature can be estimated using Sharrock’s equation (Sharrock and McKinney 1981). The solid circles in Fig. 7a and b are the calculated coercivities of 50 and 30 nm island arrays at T=300 K, respectively. Measured coercivities for the island arrays are shown as solid squares whilst coercivities of the continuous films are the solid triangles. The Stoner–Wohlfarth (S–W) model modified to include thermal activation gives qualitative agreement with the measured coercivities for the island arrays and clearly fails in the case of the continuous films, as expected. The agreement between model and measured data is closer for the 30 nm island array suggesting that the assumption of coherent reversal is more closely met for this island size. These results are consistent with the micromagnetic modeling where coherent rotation was found for island diameter smaller than 25 nm (Dittrich et al. 2005).

Fig. 7

The measured coercivity (solid triangles) of the continuous film, the calculated anisotropy field (open circles), calculated coercive field (solid circles) and measured coercivity (solid squares) of (a) 50 nm island arrays (b) and 30 nm island arrays as a function of Co layer thickness in a Co/Pd multilayer film

Angle dependent switching of the island arrays has been measured by MOKE and compared to that of the continuous film. Figure 8 shows the angle dependent remanent coercivity for the continuous film and for the 50 and 30 nm island arrays. H_c for the continuous film increases monotonically with angle as 1/cosθ, as expected for a domain wall motion controlled reversal mechanism (Coffey et al. 2003). While for the 50 nm and 30 nm islands arrays a S–W like angle dependent switching behavior is observed. The minimum in H_c at 45° is ideally 0.5 H_c (0°), which is not quite achieved. However, a small dispersion of the anisotropy axes of less than 3° is most likely the reason for this non-ideal behavior. Surprisingly, the angle dependent reversal behavior of large islands (1 μm) is very similar to that of the sub-100 nm island arrays. Since these islands are clearly too large to switch via coherent rotation, the result indicates that a S–W like angle dependence itself does not prove a coherent rotation reversal mechanism of the islands. One possible explanation of the reversal behavior observed in large islands is that the island reversal is governed by a nucleation event which has S–W characteristics, followed by domain wall motion which causes the entire island to reverse. Further study is underway to investigate the relationship between the nucleation and wall propagation field in the large islands (Hu et al. 2005a).

Fig. 8

Remanent switching field of the 30 nm, 50 nm and 1 μm island arrays and continuous films as a function of angle θ between the applied field and film normal

7 Recording on patterned media island arrays

We conclude by briefly mentioning recording experiments which have been performed on these single domain magnetic island arrays. The experiments were all done using a quasi-static read/write tester where a conventional recording head is scanned at low velocity over the patterned island arrays. A typical readback signal along an island row after DC magnetizing the sample is shown in Fig. 9a, which clearly reveals the island topographic pattern. It is worth noting that a DC magnetized continuous film only gives rise to a DC signal in readback, while the signal on pattern media reveals the trench locations between islands. This signal is a result of the increase in head/media spacing over the trench compared to that over an island. This information can be used to position the write element quite accurately with respect to the topographic pattern. A magnetic pattern was written with the write element of the recording head at 10 kfc/mm. The width of the written pattern was determined by the width of the write element, and for the head employed here resulted in a pattern comprised of columns two islands wide. Since the patterned array has a 100 nm period which is exactly half that of the write waveform, the adjacent island columns were magnetized up and down alternatively. The readback wave form was collected after writing, by the GMR read sensor as shown in Fig. 9b. This reveals a magnetization pattern with alternate up and down magnetic domains. An MFM image of the written pattern corresponding to a linear recording density equivalent to 70 Gbit/in.² is shown in Fig. 9c.

Fig. 9

a GMR readback signal from patterned media after DC magnetizing the sample and b after writing a square wave pattern. c Corresponding MFM image of the written columns of islands. From reference (Hu et al. 2004)

8 Summary

Nanofabricated magnetic recording media are one route to increasing density or providing additional function to magnetic hard disk drives. Patterning of servo marks, discrete tracks, and discrete bits have all been demonstrated in the laboratory. Recording results on all of these approaches show promise, and we have discussed in some detail the status of one approach for the most challenging of these, namely discrete bit media on topographically patterned substrates. However, while some promising manufacturing approaches have been proposed, it has not yet been demonstrated that the benefit of any of these approaches will surpass the costs, and this will likely be the largest hurdle to overcome for these to be successful technologies.