ACCORDING to the SRC roadmap, the hard disk drive

Transcription

1 IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 3, MARCH Tbit/in 2 Reader Design Outlook Yonghua Chen, Dion Song, Jiaoming Qiu, Paul Kolbo, Lei Wang, Qing He, Mark Covington, Scott Stokes, Victor Sapozhnikov, Dimitar Dimitrov, Kaizhong Gao, and Bradley Miller Seagate Technologies, Bloomington, MN USA We review the 2 Tbit/in 2 reader design landscape based on existing knowledge and projection. We found that the reader signal-to-noise ratio (SNR) requirement will be highly challenging due to the rapid increase in noise and the additional requirements from assisted writing. An acceptable level of channel bit density can be achieved in spite of a slow head-to-media spacing (HMS) reduction provided that both the shield-to-shield (SS) spacing and the a parameter scale with the bit length. We expect the side reading control for high ktpi to be difficult, and potentially a reader side shield will be required. The reader will likely use a higher quality MgO tunneling giant magnetoresistance (TGMR) stack with improved permanent-magnet coercivity. Certain new structures such as the differential reader or the trilayer will likely be part of the solution. Index Terms Giant magnetoresistance, magnetic heads, tunneling. I. INTRODUCTION ACCORDING to the SRC roadmap, the hard disk drive (HDD) industry is expected to demonstrate and manufacture the 2 Tbit/in technology and products in years 2012 and 2015, respectively. While a large share of the challenges, attention, and efforts at the component level are on the write head, media, and head-disk interface (HDI), the read-back technology needs to make marked progress in order to meet the 2 Tbit/in requirements. Here we present an outlook to the 2 Tbit/in reader design landscape based on existing knowledge and projection. We seek to help direct the industrial and academic research/development efforts to a more constructive direction by highlighting the challenges and pointing out the gaps. Fig. 1. SRC roadmap as of May Original image simplified to highlight reader challenges. Courtesy of Y. Shiroishi. II. SIGNAL-TO-NOISE RATIO We expect the reader signal to noise ratio (SNR) to face double challenges above the 1 Tbit/in design point. The first challenge is from the increasing reader noise as the device volume decreases. By reader noise we refer to the intrinsic reader noise without the media contribution. The primary sources of the reader noise are 1/f noise, thermal magnetic noise, and spin momentum transfer (SMT) noise [1] [4]. The 1/f noise is not well understood, and the SMT noise is complex. We project the thermal magnetic noise for future products using Neil Smith s formula: Manuscript received August 28, 2009; accepted January 04, Current version published February 18, Corresponding author: D. Song ( dion.song@seagate.com). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /TMAG (1) where is the damping constant, the magnetic moment, the volume, and the effective hard-bias field, all of the free layer. Our noise projection contains both measurement and calculation (Fig. 2). The solid curve shows the thermal magnetic noise, which bends near 1 Tbit/in. This is because the noise density values below 1 Tbit/in are measurement, the values above 1 Tbit/in are calculation results based on (1), and the two regions do not exhibit exactly the same rate of change. The measured heads are selected from a few generations of Seagate products in various phases of development. They are representative of the respective population. A description of the noise measurement setup may be found in [4]. Specifically for the thermal magnetic noise, we measure the flat part of the noise spectra between the high frequency FMR peaks and the low frequency 1/f noise. Typically such measurement is taken between 100 MHz and 1 GHz. We then subtract the Johnson-Shot noise to obtain the thermal magnetic noise. The Johnson-Shot noise is computed based on the measured reader resistance and the bias voltage used. We found that the thermal magnetic noise approached and surpassed the Johnson-Shot noise in our recent products. This is a manifestation of the sharp one over volume /$ IEEE

2 698 IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 3, MARCH 2010 SRC 2 Tbit/in TABLE I READER MACROSPEC, TAKEN FROM SHINGLE WRITING MACROSPEC. COURTESY OF I. TAGAWA Fig. 2. Projected reader noise for a range of recording density. dependence in (1). However, we believe the less maturity in the higher density heads is also a contributing factor. Above about 1 Tbit/in, we project the thermal magnetic noise based on (1). The volume of the free layer is calculated based on areal density, tpi projection, and free layer thickness projection. We assume that, and remain constant for the lack of a better alternative treatment. For I and,we compute based on the projected bias voltage, reader resistance, and reader MR. The resulting trend showed a large increase in the thermal magnetic noise for high areal density above 1 Tbit/in, unless dramatically new structure or material are used. As a result, the shot noise becomes relatively less important. We also show pure Johnson noise in Fig. 2 as a reference. It is less than the shot noise term within the range of recording density. This highlights the fact that the shot noise is more important than Johnson noise, which is a direct consequence of the bias voltage being substantially above 50 mv, the thermal voltage at room temperature [5]. Since the thermal magnetic noise term poses such a severe challenge, we believe the magnetic recording industry can benefit from a more deliberate and focused effort on thermal magnetic noise reduction. According to (1), the levers for noise reduction includes, but not limited to, reducing the damping constant, increasing the magnetic moment or the volume, increasing the effective field. There has been extensive research in future stack technologies such as current confined path (CCP), Heusler current perpendicular to plane (CPP) GMR, etc. While these are well-justified investment intended for reducing reader resistance and the Johnson noise, they do not offer an apparent benefit in reducing the thermal magnetic noise according to (1). The second challenge for reader SNR is that there will likely be additional demand as assist-writing technologies become mainstream. For instance, both DTR and BPM reduce bit width relative to the track pitch. Most media designs propose a 1:1 bit to trench ratio due to process limit [6] [8]. This reduces the read-back amplitude by a factor of two or more, and results in a 6 db or more loss in the intrinsic reader SNR. Referring to Fig. 2, a 6 db loss is quite substantial in the roadmap. HAMR is another promising assist-writing technology [9]. The optically defined tracks contain more curvature due to the circular shape of thermal propagation in the media. The curvature first reduces the amount of available track width analogous to the patterned media. We expect the loss in intrinsic reader SNR to be comparable as well. However, the curved bits cause additional SNR loss than patterned media because the curved portion is not recessed. The media noise thus reaches the reader even when the flux does not contribute to read-back signal. III. LINEAR DENSITY According to the SRC macrospec (Table I), the 2 Tbit/in design point may be reached with 2400 kbpi. Using Bertram s analytical derivation for pulse width [10] one may conclude that the key success factor for bpi improvement and pulse width P reduction is in reducing reader shield to shield spacing SSS, the head to media spacing HMS, and the transition parameter a in the written bits. We have constructed a micro-magnetic model that calculates P for a range of linear densities (Fig. 3). The result is then converted to the channel bit density (CBD), and broken into contributions from each of the three terms in (2). The calculated P at 1600 kbpi is in good agreement with experimental data taken on Seagate commercial recording heads. The subsequent columns at higher linear density are calculations assuming SSS and a both scaling with the bit length, and HMS values that improve slightly for higher kbpi. It has been difficult to obtain large thickness reduction in lubricant and in the head/disk hard coating layers [11], [12]. Therefore, we believe an HMS projection with moderate improvement is appropriate. Our modeling shows that it is possible to meet a typical CBD requirement of about 2.2 up to 3000 kbpi without HMS scaling if SSS and a scale. Since the HMS improvement is less than scaling, the relative contribution of HMS to the CBD increases steadily. This highlights the need of continuous efforts in clearance reduction as well as developing robust and thin HDI coatings. On the other hand, the contribution from SSS and a to the CBD remains constant because of our scaling assumption. These assumptions are far from being underpinned and would require extensive development effort in media and head. In addition, there are data suggesting that the increased transition curvature would degrade CBD as well. In principle, there can be other choice of parameters that meet the CBD requirement. The very low magnetic spacing of 5 nm in the SRC macro-spec is much closer to scaling than the parameters we used in Fig. 3. The choice was made likely (2)

3 CHEN et al.: 2 Tbit/in READER DESIGN OUTLOOK 699 Fig. 3. Channel bit density projection for higher linear densities. Fig. 4. Side reading for different reader width. The top two curves are for SSS of 26 nm. The bottom curve is for SSS of 20 nm. The horizontal guideline is the side reading control goal. to improve writability. Nevertheless, the low HMS relaxes the SSS and a requirement. For example, we note that the 20 nm SSS is significantly higher than our projection. The 20 nm shield-shield spacing was proposed to fit a conventional abutted junction MgO TGMR stack. Obviously, any major invention in the stack or reader structure that dramatically reduces the shield-shield spacing can potentially relax the demand on physical reader width reduction and HMS reduction. IV. SIDE READING In the cross track direction, the proposed physical reader width in Table I is 18 nm, narrower than the proposed shield-shield spacing. This makes it very difficult to control side reading. This can be seen in the following figure. Here we describe side reading by the ratio of magnetic track widths MT10/50. The numbers 10 and 50 refer to the percentage of the full peak height at which the track widths are measured. The approximate goal of controlling side reading is quantified as having an MT10/50 ratio of 1.6 or less. The figure contains both modeling and measurement. In the model, all magnetic layers including the stack and the permanent magnet are calculated micro-magnetically. The pinning field from the AFM layer is assumed to be uniform. All parameters are fixed except for the physical reader width RW and SSS. The written micro-track width in the media is fixed at 5 nm. RW is varied from 25 nm to 55 nm, the reader stripe height is fixed at 60 nm, and HMS at 12 nm. The measurement is taken on a typical group of HGAs in a Gb/in Seagate product. The reader has a nominal SSS of 26 nm, and the test was conducted at an HMS of approximately 12 nm. The solid line is the fitting result of the measurement. Fig. 4 shows that the side reading degrades rapidly for narrower reader. At a SSS of 26 nm, modeling and measurement showed consistent trend. The experimental result is wider than modeling. This is potentially due to a slightly wider SSS in the actual heads, or the HMS during the test is higher. It is also possible that the shield surface contains a magnetically dead layer that increases the real SSS. The calculated results for 20 nm SSS Fig. 5. FEM modeling result of flux vs. cross-track position of the micro-track for a range of reader side shield permeability. further indicates that reducing the spacing is an effective method of side reading control. However, the 20 nm SSS cannot support MT50 of much less than 1.6. In contrast, a physical RW of 18 nm corresponds to a MT50 of approximately 1.0 depending on HMS and SSS. This would lead to an MT10/50 ratio much larger than 1.6. As long as HMS does not deviate largely from scaling, a reader side shield can be an effective way of side reading control. We use FEM modeling to compute the amount of flux from a micro-track reaching the reader. The resulting profile (Fig. 5) may be used as a proxy for the micro-track profile of the read back signal. The calculation uses SSS of 30 nm, HMS 10 nm, physical RW 44 nm, and side shield to free layer spacing 7 nm. The side shield permeability varies from 1 to 100. At relative permeability of, the calculation represents the case without the side shield. For increased permeability, the profile becomes more compact. For, the result stabilizes and would represent the side-shielded case. This work implies that the side shield would reduce MT50 by 10% under the specified condition. The impact on MT10 should be more significant.

4 700 IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 3, MARCH 2010 Fig. 6. Spin-stand SNR measurement comparing readers made of CCP stacks and TGMR stacks. The inset shows TEM image near the reader barrier (horizontal white line) for a commercial recording head. V. COMMENTS ON STACK, PERMANENT MAGNET, AND ALTERNATIVE READER DESIGNS Alternative barrier technologies are important supplements for the existing MgO TGMR barrier [13] [15]. As an example, we compare our recent attempt in developing high quality CCP-GMR reader vs. TGMR control parts in Fig. 6. Both the CCP-GMR and the TGMR stacks make use of the conventional MgO barrier/nol and CoFeB electrodes. Details of the CCP-GMR structure, fabrication method, and stack performance will be disclosed in the future. Here we point out that the TGMR stack technology has been extended to lower resistance range below 0.5 using existing technology for the 380 Gbit/in node. The TGMR stack has not been optimized for low RA. Both types of reader elements have the same SSS and the same stripe height of 80 nm. Both types of heads contain only the read element and are tested on a spin-stand using pre-written tracks. The electronics SNR is calculated as the ratio between the read-back amplitude for a track at 1660 kbpi and the off-disk rms noise. The fitted SNR measurement is plotted vs. MT10, and these heads have a typical MT10/50 ratio of approximately 1.7. The result indicates that the CCP readers have a slightly lower SNR compared to the TGMR reader using unoptimized stack. This is consistent with the experience of the industry, that the fast pace of improvement in MgO TGMR stack keeps pushing out the CCP interception. The above example highlights the fact that the alternative barriers have so far failed to produce sufficient breakthrough. And they do not address the key SNR concern at 2 Tbit/in as discussed in Section II. The inset in Fig. 6 shows typical barrier quality on the market for recently released products. The head comes from a well-respected source and represent one of the best quality barrier technologies in the industry. Still, the barrier typically contains many defects such as pinholes, grain boundaries etc., that are caused by polycrystalline film deposition, annealing, and stack composition. There is still much room of barrier quality improvement as demonstrated in the last four product generations of TGMR reader technologies in the industry. We expect the MgO barrier technology to be continuously improved to fulfill the industry s mid-term needs. Fig. 7. Calculated PM field in the reader for CoPt (800 emu/cc) and for FePt (1140 emu/cc). Permanent magnet (PM) provides the essential free layer bias for reader linearization and stability. The increasing linear density reduces SSS, thus the available PM thickness is also reduced. The resulting smaller PM grains are more susceptible to thermal fluctuation. In addition, most of the hard bias is provided by the few grains close to the reader stack, and the number of such grains becomes smaller for high areal density as the reader stripe height is reduced. Finally, the 5 nm magnetic spacing in Table I increases the risk of having a stronger media field on the PM. Considering the above, the stability of the magnet is a growing concern at higher areal density. Currently the hcp-phase CoPt magnet dominates the magnetic recording industry. The coercivity is generally around 3000 Oe, which could become a concern in future reader designs. Single-phase material such as L10 FePt, CoPt, or exchange-spring type magnet [16], [17] are potential candidates for more stable magnet. The thinner PM can potentially lead to insufficient bias. We note that the SSS at the PM includes the insulation layer, seed layer for PM growth control, permanent magnetic layer itself and the nonmagnetic cap layer to decouple the PM from the top shield. For the 20 nm SSS proposed in Table I, we estimated the PM thickness to be 10 nm. In Fig. 7, we compute the reader hard bias strength for the design point highlighted in Table I using a FEM model. We further compare the bias field for two different materials; CoPt and FePt. The PM remanence values are taken from L10 phase of CoPt and FePt and we assume squareness of one. The modeling result indicates a high level of hard bias at 2 Tbits/in using either of the material. We expect certain new structures such as the differential reader and the trilayer to be some of the driving forces behind the 2 Tbit/in reader. These structures have been widely studied by the industry and they alleviate certain design constrains in the conventional abutted junction structure. The differential reading concept has been with the industry for a long time [18], [19]. In certain embodiments, the pulse width P may be defined by the spacing between the sensing layers in the dual element. This would relax the SS spacing requirement and allow

5 CHEN et al.: 2 Tbit/in READER DESIGN OUTLOOK 701 possibility for alternative stacks that are thicker. At a minimal, the additional thickness may be used for the AFM layer to improve pinning, or for the pinned layer to improve noise, or for the reference layer and free layer to improve MR. On the other hand, the trilayer stacks are much thinner compared to a similar spin valve stack by removing the AFM pinning layer and the synthetic antiferromagnetic (SAF) structure [20], [21]. This enables a very narrow SS spacing. The removal of the AFM layer also improves reference layer roughness and thus the barrier quality. VI. SUMMARY For the 2 Tbit/in reader design, we expect reader SNR to face double challenge. The first challenge is from the increasing thermal magnetic noise and potentially the 1/f noise and SMT noise. The existing stack technologies in the pipeline such as CPP-GMR and Heuslor alloy are intended for MR increase or RA reduction, and does not address the noise issue explicitly. We believe the industry can benefit from a more deliberate and focused effort on thermal magnetic noise and other noise reduction. The second challenge for reader SNR is that there will likely be additional demand as assist-writing technologies become mainstream. For example, both BPM and HAMR require additional SNR of 6 db or more. This will likely lead to a new balance in optimizing the reader performance metrics. According to the SRC macrospec, the 2 Tbit/in design point may be reached with 2400 kbpi and 860 ktpi. A very low magnetic spacing of 5 nm is essential in making this a viable design point. Specifically, the low clearance is a must for reaching 2400 kbpi with a relatively wide shield-to-shield spacing of 20 nm. In the cross track direction, the proposed physical reader width is 18 nm, narrower than the shield-shield spacing. This makes it very difficult to control side reading, and a reader side shield may be necessary. The 20 nm shield-shield spacing was proposed to fit a conventional abutted junction MgO TGMR stack. Obviously, any major invention in the stack or reader structure that dramatically reduces the shield-shield spacing can potentially relax the demand on physical reader width reduction and HMS reduction. We expect the MgO barrier technology to be continuously improved to fulfill the industry s mid-term needs. Alternative barriers have so far failed to produce sufficient breakthrough. High coercivity magnet may be necessary to improve PM stability. We expect certain new structures such as the differential reader and the trilayer to be some of the driving forces behind the 2 Tbit/in reader. REFERENCES [1] A. Ozbay, A. Gokce, T. Flanagan, R. A. Stearrett, E. R. Nowak, and C. Nordman, Low frequency magnetoresistive noise in spin-valve structures, Appl. Phys. Lett., vol. 94, no. 20, [2] Y. Zhou, Simulation study on CPP-GMR head of sub-50 nm dimension, in Proc. Intermag, [3] N. Smith and P. Arnett, White-noise magnetization fluctuations in magnetoresistive heads, Appl. Phys. Lett., vol. 78, pp , [4] A. Nazarov, K. Nikolaev, Z. Gao, H. Cho, and D. Song, Microwave generation in MgO magnetic tunnel junctions due to spin transfer effects, J. Appl. Phys., vol. 103, no. 7, [5] P. Horowitz and W. Hill, The Art of Electronics. Cambridge, U.K.: Cambridge Univ. Press, [6] A. Kikitsu, Prospects for bit patterned media for high-density magnetic recording, J. Magn. Magn. Mater., vol. 321, no. 6, pp , Mar [7] X. Yang, Y. Xu, K. Lee, S. Xiao, D. Kuo, and D. Weller, Advanced lithography for bit patterned media, IEEE Trans. Magn., vol. 45, no. 2, pp , Feb [8] M. Hashimoto, K. Miura, H. Muraoka, H. Aoi, R. Wood, M. Salo, and Y. Ikeda, Influence of patterning fluctuation on read/write characteristics in discrete track and bit patterned media, J. Magn. Magn. Mater., vol. 320, no. 22, pp , Nov [9] M. Kryder, E. Gage, T. Mcdaniel, W. Challener, R. Rottmayer, G. Ju, Y. Hsia, and M. F. Erden, Heat assisted magnetic recording, Proc. IEEE, vol. 96, no. 11, pp , Nov [10] H. N. Bertram, Theory of Magnetic Recording. Cambridge, U.K.: Cambridge Univ. Press, 1994, p [11] C. Yeo, A. A. Polycarpou, J. D. Kiely, and Y. Hsia, Nanomechanical properties of sub-10 nm carbon film overcoats using the nanoindentation technique, J. Mater. Res., vol. 22, no. 1, pp , Jan [12] N. Yasui, H. Inaba, K. Furusawa, M. Saito, and N. Ohtake, Characterization of head overcoat for 1 Tb/in magnetic recording, IEEE Trans. Magn., vol. 45, no. 2, pp , Feb [13] K. Nakamoto, H. Hoshiya, H. Katada, and K. Hoshino et al., CPP-GMR heads with a current screen layer for 300 Gb/in recording, IEEE Trans. Magn., vol. 44, no. 1, pp , Jan [14] X. Peng, P. Kolbo, K. Nikolaev, and S. Chen et al., Current-confinedpath (CCP) giant magnetoresistive (GMR) sensors, J. Magn. Magn. Mater., vol. 321, no. 12, pp , Jun [15] L. Jiang, H. Naganuma, D. Mikihiko, and Y. Ando, Large tunnel magnetoresistance of 1056% at room temperature in MgO based double barrier magnetic tunnel junction, Appl. Phys. Exp., vol. 2, p , [16] D. Weller, A. Moser, L. Folks, and M. E. Best et al., High Ku materials approach to 100 Gbits/in, IEEE Trans. Magn., vol. 36, no. 1, pp , [17] E. F. Kneller and R. Hawig, The exchange-spring magnet: A new material principle for permanent magnets, IEEE Trans. Magn., vol. 27, no. 4, pp , [18] H. N. Bertram, Theory of Magnetic Recording. Cambridge, U.K.: Cambridge Univ. Press, 1994, p [19] S. Mao, E. Linville, N. Amin, J. van Ek, and O. Heinonen, Differential CPP Reader for Perpendicular Recording, U.S. Patent [20] R. Lamberton, M. Seigler, K. Pelhos, and H. Zhou et al., Current-inplane GMR trilayer head design for hard-disk drives: Characterization and extendibility, IEEE Trans. Magn., vol. 43, no. 2, pp , Feb [21] S. Mao, L. Wang, C. Hou, and E. Murdock, Vertical GMR recording heads for 100 Gb/in, IEEE Trans. Magn., vol. 39, no. 5, pp , Sep