Demand for larger areal density of storage and memory devices is growing exponentially with the emergence of the Internet, explosive growth of broadband communication, increasingly complex multi-media mobile devices, and the rapid expansion of on-demand databases serving multinational businesses [1]. Longitudinal recording has been the mainstream data storage technology for over four decades [2]. Today, the current state-of-the-art demonstrations by leading companies are in the range of beyond 200 Gbit/in2 [3]. The conventional technology has been abandoned in these recent demonstrations. Indeed, in agreement with earlier predictions, at such high areal densities, longitudinal recording is suppressed by thermal fluctuations in the media. This fundamental limit is often referred to as the superparamagnetic limit [4,5,6,7].

A number of technologies have been proposed to address the fundamental limit. The closest alternatives are perpendicular recording [8,9 ,10,11,12,13], patterned media [14,15,16], and heat-assisted magnetic recording (HAMR) [17, 18, 19, 20].

However, to defer the superparamagnetic limit substantially beyond 1 terabit/in2 density, it may be necessary to stack recording layers in a third dimension.  This vertical stacking underlies the concept of 3-D recording [21, 22]. If the annual areal density growth rate of 100 percent continues, 3-D recording may become the technology of choice within the next decade. 

Figure 1.  (a) Schematic illustrating recording across the thickness via continuous variation of the recording field. The boundary curve is the field profile when H = Hc, where Hc is the coercivity. (b) A diagram of a perpendicular recording system.  FIB images of FIB-fabricated (a) single pole writer with a 80-nm trackwidth and (b) a MFM nanoprobe for reading a component of the stray magnetic field.

In this most trivial implementation, a recording transducer above the media is used for both writing and reading information. During writing, variation of the current through the drive coil is used to vary the recording field across the media thickness. Recording is produced sequentially, “layer” by “layer” starting with the most bottom layer of the media (as seen by the head). First, the electric current is driven to a value sufficient to generate a large field (close to the coercivity, Hc) in the vicinity of the bottom layer (farthest from the head), as shown in Figure 2b (left). (The closer a layer is located to the head, the larger the field in the plane is.) Then, the current is reduced and its polarity is reversed or left the same depending on the information recorded in the “next” layer (closer to the head), as shown in Figure 2b (right), and so on. In this manner, the information in each layer could be recorded on a bit-by-bit basis or via an array of more than one bit depending on the encoding scheme. In summary, in this implementation, information would be recorded sequentially, starting with the bottom layer and ending with the top layer.

(a)(b)

Figure 2. (a) A diagram (not to scale) of a complete recording system with a SUL.  (b) Cross-section diagrams (not to scale) showing two consecutive time instances  illustrating how the field due to the electric current in the drive coil could be used to access each layer in a 3-D media (from bottom up) during the writing process.

Although the above-described implementation seems to be rather straightforward, it is not the most trivial one because of quite lengthy encoding schemes it might require. It has been found that the most trivial implementation naturally resulting from the conventional magnetic recording implementation would be multi-level magnetic recording, as described below.

In the multi-level mode, the recording process is similar to the process as used in the conventional magnetic technologies including longitudinal and perpendicular recording, with the principal difference in the number of useful signal levels. In the multi-level mode, more than one signal levels are used to identify information. It could be reminded that in the conventional cases, the information in each bit cell in the media is recorded in the form of the magnetization saturated in one of the two possible directions. In other words, the polarity of the media only carries the information. The latter naturally results in a binary form of the signal interpretation. In the multi-level mode, to increase the effective areal data density, it is chosen to identify the recorded information also as the degree of the polarization (magnetization) of the media rather than the polarity only. As described above, due to the weekly-exchange-coupled layers in 3-D media, more than one signal levels could be achieved via the control of the degree of the polarization across the thickness.  The adequately strong signal-to-noise ratio (SNR) to distinguish the signal multiplicity could be achieved because of the relatively strong anisotropy of each of the strongly coupled multilayers of which each separate layer is made. Before going in a more detailed description of multi-level magnetic recording, the concept of patterned 3-D media should be introduced.  As described below, the introduction of the concept is necessary to complete and clarify the description of the multi-level mode.

Challenges and Objectives

In summary, the general objective will be to utilize the physics of 3-D magnetic recording to answer open questions with regard to future device implementation (of a multi-level form) suitable for data densities beyond 10 Terabit/in2. The approach will be to move the research to the next level by taking advantage of the multidisciplinary background of the investigators to develop the guidelines necessary to build a 3-D magnetic device suitable for such ultra-high densities. The project will integrate the knowledge from several engineering disciplines including conventional magnetic data storage engineering, physics of nanomagnetism, state-of-the-art in nanofabrication, thermal measurements at nanoscale, data encoding, materials science and chemistry.

Among the key challenges to be addressed in the project are: 1) understanding of the basic physics associated with recording information in a 3-D media, 2) understanding and minimization of the symbol interference caused by the stray magnetic field from the bulk of a structure, i.e. maximization of SNR, 3) study of thermal effects in a 3-D device to defer the superparamagnetic limit, and 4) development of the basic guidelines to build a multi-level 3-D magnetic recording device.

Terminology

1. Multi-level versus Absolute 3-D:Two different modes of 3-D recording with respect to addressing separate magnetic layers will be referred to as multi-level and absolute 3-D modes, respectively.

 (a)      (b)
Figure 3.  Schematic diagrams illustrating how information is addressed in cases of (a) 3-D multi-level and (b) 3-D absolute recording modes, respectively.

In the multi-level mode, though a 3-D space is utilized for recording, it is not used efficiently. In a trivial case, the signal recorded and read back from a vertical stack is defined by the recording head above the stack. In this case, the information in all N layers within the stack simultaneously contributes to the signal (Figure 3a). The number of signal levels, L, is different from the number of layers, N, and determined by the ability of the head to generate a signal with an adequately large SNR for L levels to be distinguished from each other. In this case, the effective density increases by a factor of Log2L with respect to the density of the equivalent binary single-layer media. In contrast, in the absolute 3-D recording mode, each (n-th) layer in the N-layer recording media would be addressed separately (see Figure 3b) via some physical process (to be explored). In this case, the effective areal density would increase by a factor of N with respect to the areal density of the equivalent single-layer media. For example, if one thousand layers are used, assuming a bit cell cross-section of 160 x 160 nm2, 25 terabits of data could be stored in one square inch.

2. Recording Mode versus Memory Mode: By recording and memory systems, the investigators will refer to the systems with and without moving parts, respectively. It is possible that in the future any studied form of 3-D recording may gradually transition into some form of 3-D memory. For example, one of the feasible forms of 3-D memory may emerge as a 3-D version of the current form of magnetoresistive random access memory (MRAM). Nevertheless, at this phase of research, the group will focus on a study of 3-D recording in the form of a 3-D disk drive with one or two nanoscale transducers used to record and read information on and from the disk. In this case, only one or two active devices (heads) must have nanoscale feature dimensions. On the contrary, in the mode of memory, every bit element must have one or more active nanoscale components around it (for writing and reading information). Therefore, the mode of recording seems to be more economical and thus more appealing (especially, to the industry) at this initial stage of research

Why Is Patterning Necessary?   Here, it is assumed that the recording transducer is similar to the one used in perpendicular magnetic recording and includes a single pole head and a GMR element for writing and reading, respectively.  Throughout this development, micromagnetic simulations have been used to help choose optimal properties and the requirements for the recording media and the recording transducer [23 ].

(a)   (b)

Figure 4 .  Diagrams (not to scale) with magnetization patterns recorded on (a) a dc-erased media and (b) a media with information recorded in a adjacent bit cell. The overlapping side region is highlighted.

A potential issue with such a trivial data access is the relatively strong overlapping of the field profiles used for recording into adjacent bit cells. To illustrate this issue, Figures 4a and b show the magnetization pattern (a) right after information is recorded into both layers of a two-layer dc-saturated media and (b) after one of the adjacent cells has been preliminarily recorded, respectively. The highlighted overlapped region indicates the region in the media where the information is erased or, in other words, “wasted”.

Figure 5.  AFM image of a FIB-fabricated patterned media with an island diameter of approximately 45 nm.

3-D Patterned Media: To overcome the above described issue of the data erasure in the overlapped region, a media could be patterned. Today, it is not an issue to controllably fabricate a 2-D patterned media with a nanoscale cell. The nanofabrication tools, such as atomic lithography, electron-beam and FIB lithography, nanoimprint, X-Ray patterning, self-assembly, and many others are routinely used to define patterned media with cell side dimensions of less than 100 nm. For example, an atomic force microscopy (AFM) image of a FIB-fabricated patterned Co/Pd-based media with a square cell side of approximately 45 nm is shown in Figure 5. It could be noted that in the proposed implementation, a media is automatically patterned in a third (-z) dimension during the above described preliminary multi-layer deposition step. Finally, patterning of 3-D media additionally in the vertical and the two in-plane directions has the same advantage as patterning of conventional media - the media grains/clusters could be accessed by the recording head in a certain periodic and predictable manner rather than randomly as is the case with conventional continuous media. In other words, it is expected that patterning would further increase SNR in 3-D media. The simulations indicate the SNR increase by at least 6 dB for the patterned media adequate for 5 Terabit/in2.

(a) (b)
Figure 6. (a) Schematics of a helical 3-D pattern. (b) SEM image of a nanofabricated helical 3-D pattern (fabricated through the on-going collaboration with Prof. Toh-Ming Lu, Physics Department, RPI).

Helical Patterning: There may be certain advantages in more complex patterned structures compared to the described above patterned structures with vertical and straight walls. Helically patterned structures, as the one shown in Figure 6a, may provide larger SNR values because of the inherent simplicity in accessing the recorded information across the entire thickness.  For example, if each bit of information is recorded in a certain part of one helical turn (e.g. a half-turn as in the schematics above), one could devise many ways to identify the bit just via the physical shape of the helix. Today, with the advanced methods of oblique deposition, one could controllably manufacture such helical patterns with nanoscale dimensions.  As an example, a scanning electron microscopy (SEM) image of one such nanofabricated helical Si pattern with a period of one micron is shown in Figure 5b.  Although, in the future, such helical structures may appear to be advantageous, at this early research stage, it has been chosen to convey the discussion around the examples with the most trivial straight-wall patterned structures. Finally, it is proposed to use periodic arrays of vertically grown carbon nanotubes (CNT) with the adequate magnetic material plated or deposited via another method inside the nanotubes.

(a)    (b)

Figure 7.  (a) A diagram illustrating sequential recording of different levels in a 3-D patterned media. (b) The recording field profiles in the first, second, and third layers at a given value of the drive current in the recording transducer above the cell.

How Does Multi-Level Recording Work during Writing?    Figure 7a illustrates a micromagnetically simulated example of how each layer in a bit cell stack could be recorded in a patterned 3-D media in the multi-level mode. Initially, a bit cell is assumed to be saturated. This means that the entire bit cell stack is magnetized in one direction (blue color). According to a certain encoding scheme, this state may, for example, reflect a digital state “10”.  Then, the recording field would be increased in the opposite direction to reverse the magnetization (red color) in the top layer. This state could reflect a digital “9”. When the field is further increased to the value overcoming the coercivity of the second (counting from the top) layer, the magnetization in the second layer is reversed. This state would reflect a digital “8”, and so on. The simulated recording field profiles in the first, second, and third layers, at an arbitrary current value in the recording transducer above the cell are shown in the above Figure.

As a final remark, it could be mentioned that the above-described mechanisms are probably some of the most trivial mechanisms to explore. However, in the future, other mechanisms, such as variation of the coercivity value across the media thickness, localized heating of sub-layers or/and sub-clusters, ferromagnetic resonance, and many others, will be investigated.

(a)
(b)(c)

Figure 8. (a) The “charge” representation of different digital values in the bit cell consisting of many magnetic layers. (b) The stray field versus the digital level in the 3-d multi-level mode. (c) SNR versus the number of layers.

How Does Multi-Level Recording Work during Reading?    It is believed that directly reading the stray magnetic field from a location above the media would accomplish the goal of the most trivial readback operation. Similar to conventional magnetic recording, a giant magnetoresistive (GMR) sensor could be used to read the magnetic field emanating from the media in the vicinity of each bit cell. In other words, this reading scheme reflects the 3-D multi-level recording mode, as described above (see Figures 8a-b).  In this case, the net magnetic signals corresponding to various multi-level magnetization patterns could be illustrated by the magnetic “charge” configurations, as shown in Figure 8a.  For example, for the totally saturated state (digital “10”), the net signal would consist of the magnetic field emanating from the “charge” on the top and bottom surfaces of the effective magnetic layer, respectively. For each intermediate state (“9”, “8”, etc), there is also contribution from the “charge” in the respective boundary plane, as shown in Figure 8a. According to this magnetic “charge” model, it is a fairly trivial task to calculate the net field for different magnetization patterns [17].  Thus calculated signal levels are shown in Figure 8b. It could be observed that the signal exponentially drops with the increase of the number of recording layers. The net signal-to-noise-ratio (SNR) strongly depends not only on the read sensor (GMR, and others) but also on the media transition and dc noise and the electronic noise in the pre-amplifiers. The calculated SNR versus the number of layers for a simplified case is shown in Figure 8b. The media was assumed to be ideally patterned in all three directions and thus the media noise could be neglected. Finally, it was assumed that there were two sources of the electronics noise: 10 Ohm GMR sensor and 0.2 nV/sqrt(Hz) preamp noise over a 500 MHz CTF bandwidth at 1 Gigabit/sec data rate.  It is believed that more advanced encoding software channels (e.g., turbo) could reduce the bit error rate (BER) to 10-9 even for a 10 dB SNR recording system.  According to the current model, even with such advanced encoding, only four to five layers could be distinguished in such a trivial implementation of 3-D recording.  Fortunately, this is not a fundamental limit. There are many solutions that could substantially improve the SNR characteristics of a 3-D system. For example, optimization of the 3-D media and the use of differential sensors could be exploited to further increase SNR. One of the powerful techniques used in this project to substantially improve the playback resolution relies on the integration of micromagnetic modeling with deconvolution based signal processing methods. Previously, it was shown that this technique could be used not only to substantially improve the resolution of the system but also to differentiate between different components of the magnetic field thus further improving the effective information density recorded in the system. An ion image of a specially fabricated “smart” nanoprobe (transducer) placed above the media to read information via reading various components of the magnetic field is shown in Figure 9a. Two MFM images, one (left) taken by a regular MFM nanoprobe and the other by the smart FIB-fabricated nanoprobe with further deconvolution, are shown in Figures 9b left and right, respectively. MFM images taken by the probe in its mode to read the perpendicular and along-the-track field components emanating from a regular ring head at its air bearing surface are shown in Figures 9c and d, respectively. The measured field profile clearly coincide with the expected bipolar and unipolar perpendicular and longitudinal field components, respectively.

(a) (b)

Figure 9.  (a)  An ion image of a FIB-fabricated Nanoscale probe (“smart” probe). (b) MFM patterns with a regular DI MFM probe (left) and the “smart” probe (right). MFM images of (c) perpendicular and (d) longitudinal components of the recording field measured by FIB-fabricated “smart” nanoprobes with adequate material choice. (e) Pre-recorded magnetization in the top and bottom layers of a 20-layer recording media. (g) Sets of signals read back at a SUL’s biasing current of 5.85 and 1.56 A-turn. 

How Patterning SUL Would Help Read across the Thickness of 3-D Media:  As a simulation input, two types of information were pre-recorded into the top and bottom layers of 3-D recording media with a net thickness of 50 nm (Figure 9e). The “softness” of a patterned SUL was controlled via continous variation of SUL’s bias current, as defined above . The respective signal profiles read back at biasing current values of 5.85 and 1.56 A-turn, respectively (Figure 9d) indeed strongly remind the prerecorded patterns. One could notice here that various deconvolution based signal encoding techniques could be implemented to further optimize SNR of 3-D system [24].

Study of New Concept: Patterned Soft Underlayers and Interlayers

Traditionally, SUL has a flat surface boundary with the recording layer. The image model could be used to describe the effect of SUL on the recording field within the recording layer (Figure 10a). Although, ideally a factor-of-2 increase in the field is expected, from the same image model one could observe that the image is located further away from the center of the recording layer as compared to the real head. The difference between the separations of the real and image heads from the center of the recording layer (the spacing loss) is ~ equal to the thickness of the recording layer. The closer to the recording media the head is placed, the stronger and more localized the recording field in the media. Because the image head is located further away compared to the real head, the net effect will result in a deteriorated signal compared to the signal due to two equally separated heads. The Reciprocity principle states that the detrimental effect exists for both write and read processes. The following new concept is proposed to resolve the issue (of the spacing loss).

(a) (b)

(c)
Figure 10 .  (a) A diagram of an image model to illustrate the spacing loss due to the offset in the separation between the real and image heads with respect to the central plane of the recording layer.  (b) A schematic illustrating the “move” of the image closer to the recording layer if a convex SUL boundary is used. (c) Patterned SUL with rectangular islands.

One could recall that magnetic imaging is very much like regular mirror imaging. Therefore, exactly as in the case of a convex mirror, one could use a SUL with a convex boundary to move the image closer to the “mirror’s” (SUL’s) boundary with the recording layer. To illustrate this effect, a diagram comparing the locations of the images in the two cases is shown in Figure 10b. Using a convex SUL instead of a flat SUL could make the image head be “closer” to the recording layer and thus increase the areal density. Based on the above described concept, it is proposed to use a patterned SUL with each patterned island having a convex shape. Patterning is necessary to implement the novel concept on the physical scale of one bit. The cross-sectional dimensions of each patterned island correspond to the cross-sectional dimensions of the targeted bit cell. For example, if an areal density of 1 terabit/in2 must be achieved, assuming a square symmetry in the plane, each of the (x and y) period values, T, should be approximately 26 nm.  Via numerical simulations, it has been discovered that this favorable effect exists not only for patterns with convex islands but also for patterns with rectangular islands (Figure 10c). This was attributed to the combined effect of patterning and periodicity. Methods similar to the ones for patterning hard layers will be used to pattern also SULs.

c)

Figure 11.  Drawings illustrating integration of a patterned SUL with (a) a continuous recording layer and (b) a patterned recording layer.  (c) A schematic diagram of the cross-section of a convex patterned SUL in X-direction.

Both continuous (in conventional sense) and patterned recording layers will be integrated with a patterned SUL, as shown in Figure 11a and b, respectively. Patterning in the latter case combines the described below advantages of patterned SULs with the well-known advantages of patterned recording layers.

Key Advantages: Some of the key practical advantages due to the use of a convex patterned SUL are the following and described below in detail:

The period of the grid, T, in each direction corresponds the bit cell dimension in this direction and thus defines the effective areal density, D, as 1/T2 (bit/in2) (Figure 11c). Below, it is shown that at a given grid period, the thickness of the island, L, could relatively sensitively control the advantageous effects of the patterned SUL. To simplify the description, below the ratio of the grid period to the island thickness, L/T, is arbitrarily defined as the pattern “curvature”.

Figure 12.  (a) The perpendicular recording field versus the distance along the line from the top surface of the SUL to the air bearing surface of the head for three different values of the curvature of the semispherical island in the SUL type under study. For comparison, an equivalent dependence for a regular flat SUL is also shown. Diagrams showing the location of the recording head with respect to an island in the convex patterned SUL (b) with the head and the island perfectly centered and (c) with a half-period offset line between the head and the island.

Increased SNR: Figure 12 shows the simulated perpendicular recording field versus the distance along the line from the top surface of the patterned SUL to the ABS of the head (from point s to point e in Figure 10a) at saturation for three “curvature” values of the patterned SUL. The three sets of three parameters, T, P, and L, are 1) 98, 2, and 200 nm, 2) 98, 2, and 100 nm, and 3) 98, 2, and 25 nm, for curvature values of 1, 2, and 4, respectively.  Curvature values 1 and 4 correspond to the sharpest and flattest surfaces of the island, respectively. In all these cases, the head is centered with respect to an island in the patterned SUL, as shown in Figure 12b. For comparison, the solid black line in the Figure shows an equivalent field line for the case of a conventional flat-surface continuous SUL. Especially for a curvature of 1, a pronounced increase of the recording field near the top surface of SUL could be observed. While in the case of the conventional flat SUL, the field reaches its maximum at the air bearing surface of the head, in the case of the patterned SUL, the field reaches its maximum in the vicinity of SUL. The recording field at SUL exceeds 30000 Oe, while at the air bearing surface of the head it barely reaches 15000 Oe. For the conventional flat SUL, the field is less than approximately 12500 Oe when it reaches the surface of SUL. Such a more than a factor of 2 increase of the recording field in the region of the recording layer means that a recording media with substantially higher anisotropy could be used and thus substantially higher recording densities could be achieved. The fact that the field reaches its maximum at the side of the recording layer farthest away from the ABS is especially favorable for 3-D recording. As described above, the most trivial form of recording across the thickness of 3-D media could be achieved via sequential recording from bottom up via continuous variation of the current in the drive coil. Patterning of SUL provides another knob to control the recording field across the thickness. As seen from Figure 12a, at a given value of the drive current, the field maximum could be shifted from the bottom to top side of the recording layer via variation of the softness of SUL, as described above.  Moreover, the investigators also explore the possibility of employing soft interlayers (SIL) to separate magnetic layers across the thickness. As illustrated below, the use of SILs is expected to substantially facilitate recording across the thickness.

Profiles of the simulated recording field along a track line in the recording layer 2.5 nm away from the top surface of SUL (Figures 10b-c) for a patterned SUL and a conventional (flat) SUL are shown in Figure 13a.  Although for the patterned SUL the recording field drops with the distance away from SUL, it is still larger compared to the field in the conventional case. The graphs also indicate that patterning of SUL would result in a more localized field. The latter effect will be discussed below in more detail.

Study of the Localization of Recording and Sensitivity Field: Patterning of SUL is fundamentally different from patterning of the recording layer with respect to addressing information during writing and reading. Here, it could be reminded that SUL is an indispensable part not only of the media but also of the head.  The latter is not true for the recording layer. This major difference is due to the fact that SUL is made of a “soft” magnetic material. This is in contrast to the recording layer, which is made of a “hard” magnetic material. As a result, patterning of SUL effectively “patterns” also the field generated by the head.  This field is the recording magnetic field or the sensitivity field depending on the writing or reading processes, respectively.

(a)(b)

(c)(d)
Figure 13. (a) Field profiles inside the recording layer for the cases of patterned and conventional SUL. (b) The field profiles along same line, 5 nm away from the surface of SUL, along a patterned direction along or across the track. The recording field profiles for two systems, (c) with a patterned SUL and (d) without a SUL, respectively.

To illustrate “patterning” of the recording field because of patterning of SUL, two different locations of the recording head with respect to the patterned SUL are considered, as shown in Figures 12b and c, respectively. In the first case, the recording head is centered with respect to an island of the patterned SUL. In the second case, the recording head is centered with respect to a groove between two adjacent islands. In other words, in the second case, the head is a half-period offset with respect to an island.  The respective field profiles along same line, 5 nm away from the surface of SUL, along a patterned direction are shown in Figure 13b.  The fact that the field profiles are so different from each other clearly indicates the above mentioned process of “patterning” the recording field. One could observe the field in the centered case is more than by a factor of two larger and substantially more localized compared to the field in the off-centered case. This means that the head itself is going to substantially improve its recording quality when it is being recording information in the favorite position compared to any off-centered positions. This is in contrast to the conventional recording process in which the recording field profile does not depend on the location of the head with respect to the media, irrespective of whether or not the media being patterned. 

REFERENCES