Channel Modeling and Multi-Island Recording Scheme on Bit Patterned Media with Long-Range Island Orientation Fluctuations
Yang Wang,B. V. K. Vijaya Kumar,Yumei Wen,P. Li
DOI: https://doi.org/10.1109/intmag.2018.8508481
2018-01-01
Abstract:Usually for Bit patterned media recording (BPMR)[1], the media noise (island size and position fluctuations) exhibits short-range disordering[2]. However, recently some BPM fabricated with cost effectiveself-assembled technology exhibits long-range disordering, as shown in the TEM image (Fig. 1(a), courtesy of Jimmy Zhu and Vignesh at CMU). Specifically, the BPM consists of multiple small domains (rectangular frame in Fig.1(a)) with different island orientations (indicated by arrows). In each domain, the islands are relatively organized because islands in a domain have similar orientation, however, the islands’ orientation varies much more from one domain to another. Hence such BPM introduces long-range island orientation fluctuations among domains besides the usual local island size and position fluctuations. Since such BPM appears to be more attractive due to its potentially lower fabrication cost, an interesting question is whether it is possible to record the information reliably on such BPM. To investigate the feasibility of reliably storing and retrieving data from such BPM, we first need to develop a channel model. For channel modeling, we assume that the track pitch and bit length are 12.7nm and 12.7nm, the island size is 7nm x 7nm, corresponding to the channel density of 4Tb/in $^{2}$. Based on the TEM image in Fig. 1(a), the island position and size fluctuations are modeled as strongly and weakly 2-D correlated Gaussian noise, respectively. To model orientation fluctuations, a mask (size ${\mathrm{L}_{M}} \times {mathrm{W}_{M}})$ with size larger than the domain (size L x W) is assumed, where the aligned islands within each mask are rotated with some angle. Then after the rotation, only the islands remaining within each domain are retained (Fig.1 (b)). The orientation fluctuation among each domain is assumed to uniformly distributed on the interval [- a, a] i.e. a = the angle 0, 5° and 15° in Fig.2 (b), (c)). The modeled BPM is shown in Fig.1 (c). Then the write and read process is modeled as in[2]. A read head array with three elements (with free layer dimension of 15nm x 15nm x 4nm) and 13nm shield-to-shield spacing and 4nm magnetic fly height is used to readback three tracks. Then the multitrack signals are processed with 2D MMSE equalizer, detected with BCJR detector and the write/read errors are corrected with an LDPC code (Fig.1 (d)). For LDPC decoder, the number of internal and global iterations ${(\mathrm{N}_{turbo}})$ are 10 and 3. Considering that the significant media noise and the write in errors for such BPM, we have proposed a recording scheme using multi-island (2 by 2 dots) representation for one bit information compared to the regular recording scheme by recording one bit on one dot (Fig.2 (a)). For the LDPC encoding, the code rate for 4 dots/bit $(R=0.9)$ is about 4 times as high as 1 dot/bit $(R=0.225)$ to obtain the same user areal density (i.e., 0.9 Tb/in $^{2})$, considering there are 4 Teradots/in 2 . Then the raw BER performance is investigated (Fig.2(b)). Here we assume 10% island position and size fluctuations (1.27nm, 0.7nm), 5% switching field distribution, 5% track mis-registration and 5% write clock phase shift. It is found that the raw BER performances for both recording schemes degrade with increasing orientation fluctuations. However, the 1 dot/bit case degrades much more severely than the 4 dots/bit case because the redundancy in the 4 dots/bit representation reduces the impact of write errors and decreases the effect of the media noise. More importantly, we have investigated the decoded BER performance of both recording schemes at the same user areal density (Fig.2 (c)). It is observed that when the angle fluctuation is small (<=5 degrees), the decoded BER performance for 1 dot/bit is better than 4 dots/bit due to the much lower code rate (0.225) and hence the much higher error correction capability of the LDPC code in the former case. However, when the angle fluctuationis large (for example, 15°), the decoded BER performance of the 1 dot/bit case becomes worse than the 4 dots/bit one because the write in error becomes increasingly dominant compared to the readback error caused by the inter-track interference (ITI) when the orientation fluctuation increases. However, for the 4 dots/bit case, the redundancy in the multi-island representation reduces the impact of the write in errors and media noise because the magnetic flux picked by the reader is the superposition of the magnetic fluxes from all the 4 dots when it scans over them. Even though when one dot in the 4 dots/bit pattern is written in mistake, the picked up magnetic flux is still quite likely to be correct. Here the greatly increased write in error and media noises (i.e., dominant at large orientation fluctuation) are more detrimental to the LDPC decoding performance compared to the read error caused by AWGN and ITI (i.e., dominant at small orientation fluctuation) for the 1 dot/bit case[3]. In conclusion, A channel modeling and multi-island representation scheme (4 dots/bit) was used to investigate the BPM with orientation fluctuation, which is compared to standard 1 dot/bit scheme. For the LDPC code encoded at the same user density, it is found that the decoded BER performance of the proposed 4 dots/bit scheme is better than 1 dot/bit scheme when the orientation fluctuation is large.