Abstract
Among various emerging memory technologies, the resistive random access memory (ReRAM) has gained significant attention due to its high data storage density, fast read and write speeds, and good compatibility with existing fabrication technologies. However, despite these advantages, the reliability of ReRAM is severely affected by the noise and interference. This thesis is dedicated to the design and analysis of novel channel coding and signal processing algorithms and techniques to mitigate various impairments and improve the reliability of ReRAM. First, to capture the main characteristics of ReRAM, this thesis proposes a cascaded channel model that incorporates two key factors that affect the reliability of ReRAM: the sneak path interference (SPI) in the memory crossbar array, and random variations of cell resistances caused by process variations. This thesis further develops a novel scheme to estimate the probability of each memory cell to be affected by the SPI, based on which the soft information of each channel bit can be generated for decoding the error-correcting codes (ECCs). Moreover, this thesis proposes a novel two-step acrossarray bit allocation scheme which distributes one ECC codeword to multiple memory arrays so as to minimize the correlation of the channel-coded bits for better decoding. As high-precision analog-to-digital converters are not applicable for high-speed ReRAM, the design of channel quantizers is critical to support error correction coding for ReRAM. This thesis first proposes a quantized channel model of ReRAM, based on which both the one-bit and multi-bit channel quantizers are designed by maximizing iv the mutual information of the channel. This thesis then proposes SPI-aware adaptive detection and decoding schemes, at both array level and column level. A channel decomposition method is further proposed that enables effective ways for theoretically analyzing the ReRAM channel. The proposed SPI-aware adaptive detection and decoding schemes can approach the ideal performance with three quantization bits, with only one decoding iteration. ECCs play a critical role to ensure the data recovery reliability of ReRAM. This thesis proposes to design protograph quasi-cyclic low-density parity-check (QC-LDPC) codes for the ReRAM channel. In particular, it first applies Gaussian mixture fitting and channel symmetrization to enable the theoretical analysis of the LDPC-coded ReRAM channel. It then utilizes a modified protograph extrinsic information transfer algorithm and the asymptotic weight distribution analysis to optimize the protographs. In addition, It also provides the QC progressive edge growth algorithm to obtain the QC parity check matrix which decreases the implementation complexity for encoding and decoding. To further combat the SPI of ReRAM, this thesis proposes an enumerative coding technique for constructing two-dimensional (2D) constrained codes that can prohibit the occurrence of the SPI within the codeword with code rates approaching the capacity. To reduce the error propagation (EP) induced by these codes during decoding, this thesis proposes to reversely concatenate the enumerative code with ECC. Moreover, to mitigate the SPI at the parity part of ECC, this thesis further proposes a novel 2D constrained code with limited EP, named the projective code. The EP of both the two types of constrained codes are analyzed and extensive simulation results demonstrate the effectiveness of the proposed 2D constrained codes in mitigating the SPI and improving the error rate performance of ReRAM arrays.