Data Storage Systems Center

Most data storage systems employ Reed-Solomon (RS) codes for error correction because of their powerful burst error correction capability and fast hardware implementation. Towards the goal of achieving better error correction capability at higher densities, LDPC codes and soft-decision decoding of RS codes are receiving increasing research attention. However, the performance gains for magnetic recording channels is smaller than they are for AWGN channel. Another problem for the soft-decision decoding algorithms is their incompatibility with the standard hard decoding used by run-length limited (RLL) codes. We propose a new soft-decision decoding algorithm of RS codes based on Chase2 algorithm. The proposed method uses dominant error-patterns in order to maximize the performance at the given channel and avoids RLL hard decoder incompatibility problem by using soft-information before RLL decoder. Simulation results for perpendicular magnetic recording channels show attractive performance gains.

Recently, significant efforts are being made to increase the density and performance of flash memory and MRAM. However, increasing density by scaling down of the memory cell size is limited by the side effects that the cell becomes more vulnerable to the noise and disturbances and exhibits errors. In this preliminary effort, the errors from those side effects are modeled for multi-level cell (MLC) flash memory and spin transfer torque (STT)-based magnetic random access memory (MRAM). In the case of the MLC flash memory, increasing the number of levels of a cell can provide higher density, while the difficulty in detecting multiple levels can increase its bit error rate (BER) [1-2]. Consequently, to meet a target raw BER, the number of levels in the MLC cell is limited depending on the variability in cell responses. Similarly, for the STT-based MRAM, spin-polarized current is used to change the MRAM cell between two states, where the current needed to switch to one state is typically larger than that of the other [3]. When the maximum currents are limited, the errors are more frequent in one state and the errors can be modeled by a binary asymmetric channel. Thus, based on the information from error modeling, one can analyze the trade-off between density and BER performance in memory systems.

[1] Jim Cooke, Nand Flash Technology, http://www.denali.com/en/events /webcasts/open/micron/ [2] G. Atwood et. al., “Intel StrataFlash Memory Technology Overview,” Intel Tech. J., Q4, 1997 [3] T. M. Maffitt et al., “Design considerations for MRAM,” IBM J. Res. & Dev. Vol. 50, No. 1, 2006

In previous work, we investigated the error performance of bit-patterned media (BPM) channels in the presence of media noise and have shown that media noise has a significant impact on the error performance of BPM channels. Therefore having a realistic model of media noise is important in evaluating the error performance of the channel.

In this work, towards the goal of a more realistic media noise model, we extract the media noise characteristics from an image of a BPM template. The media noise is assumed to be mainly due to size and location fluctuations of magnetic islands. Using image processing techniques, we estimate the distributions of these fluctuations. The results show that Gaussian distribution is a good model for these fluctuations. These fluctuations are most likely due to slow fabrication variations and as a result, media noise can be correlated. To investigate if these fluctuations are correlated or not, we estimate their correlation coefficients. The results show that location fluctuations are highly correlated but size fluctuations are not correlated. We also investigate the error performance of BPM channels with noise whitening generalized partial response (GPR) equalizers and noise predictive maximum likelihood (NPML) detectors. Simulation results show that these equalizers and detectors improve the performance of the BPM channels in the presence of correlated media noise, but NPML detectors perform better.

Abstract and highlight