Approximate nearest neighbor search (ANNS) is a key retrieval technique for vector database and many data center applications, such as person re-identification and recommendation systems. It is also fundamental to retrieval augmented generation (RAG) for large language models (LLM) now. Among all the ANNS algorithms, graph-traversal-based ANNS achieves the highest recall rate. However, as the size of dataset increases, the graph may require hundreds of gigabytes of memory, exceeding the main memory capacity of a single workstation node. Although we can do partitioning and use solid-state drive (SSD) as the backing storage, the limited SSD I/O bandwidth severely degrades the performance of the system. To address this challenge, we present NDSEARCh, a hardware-software co-designed near-data processing (NDP) solution for ANNS processing. NDSeARCH consists of a novel in-storage computing architecture, namely, SEARSSD, that supports the ANNS kernels and leverages logic unit (LUN)-level parallelism inside the NAND flash chips. NDSEARCH also includes a processing model that is customized for NDP and cooperates with SearSSD. The processing model enables us to apply a two-level scheduling to improve the data locality and exploit the internal bandwidth in NDSearch, and a speculative searching mechanism to further accelerate the ANNS workload. Our results show that NDSEARCH improves the throughput by up to $31.7 \times, 14.6 \times, 7.4 \times 2.9 \times$ over CPU, GPU, a state-of-the-art SmartSSD-only design, and DeepStore, respectively. NDSEARCH also achieves two orders-of-magnitude higher energy efficiency than CPU and GPU.

Recent advances in large language models have demonstrated remarkable effectiveness in information retrieval (IR) tasks. While many neural IR systems encode queries and documents into single-vector representations, multi-vector models elevate the retrieval quality by producing multi-vector representations and facilitating similarity searches at the granularity of individual tokens. However, these models significantly amplify memory requirements for retrieval indices by an order of magnitude. This escalation in index size renders the scalability of multi-vector IR models progressively challenging due to their substantial memory demands. We introduce Embedding from Storage Pipelined Network (ESPN) where we offload the entire re-ranking embedding tables to SSDs and reduce the memory requirements by 5−16×. We design a flexible software prefetcher applicable to any hierarchical clustering based search, achieving hit rates exceeding 90%. ESPN improves SSD based retrieval up to 6.4× and end-to-end throughput by 68% to maintain near-memory levels of query latency even for large query batch sizes. The code is available at https://github.com/susavlsh10/ESPN-v1.

This paper uses the reconstruction-computation-quantization (RCQ)paradigm to decode low-density parity-check (LDPC) codes. RCQ facilitates dynamic non-uniform quantization to achieve good frame error rate (FER) performance with very low message precision. For message-passing according to a flooding schedule, the RCQ parameters are designed by discrete density evolution. Simulation results on an IEEE 802.11 LDPC code show that for 4-bit messages, a flooding Min Sum RCQ decoder outperforms table-lookup approaches such as information bottleneck (IB) or Min-IB decoding, with significantly fewer parameters to be stored. Additionally, this paper introduces layer-specific RCQ, an extension of RCQ decoding for layered architectures. Layer-specific RCQ uses layer-specific message representations to achieve the best possible FER performance. For layer-specific RCQ, this paper proposes using layered discrete density evolution featuring hierarchical dynamic quantization (HDQ) to design parameters efficiently. Finally, this paper studies field-programmable gate array (FPGA) implementations of RCQ decoders. Simulation results for a (9472, 8192) quasi-cyclic (QC) LDPC code show that a layered Min Sum RCQ decoder with 3-bit messages achieves more than a 10% reduction in LUTs and routed nets and more than a 6% decrease in register usage while maintaining comparable decoding performance, compared to a 5-bit offset Min Sum decoder.

Non-binary low-density parity-check codes are robust to various channel impairments. However, based on the existing decoding algorithms, the decoder implementations are expensive because of their excessive computational complexity and memory usage. Based on the combinatorial optimization, we present an approximation method for the check node processing. The simulation results demonstrate that our scheme has small performance loss over the additive white Gaussian noise channel and independent Rayleigh fading channel. Furthermore, the proposed reduced-complexity realization provides significant savings on hardware, so it yields a good performance-complexity tradeoff and can be efficiently implemented.

The presence of error floor in low density parity check (LDPC) codes is of great concern for potential applications of LDPC codes to data storage channels, which require the error correcting code (ECC) to maintain the near-capacity error correcting performance at frame error rate as low as 10-12. In order to investigate the error floor of LDPC codes under partial response channels used in data storage systems, we propose a new estimation method combining analytical tools and simulation, based on the concept of trapping sets. The definition of trapping sets is based on the dominant error patterns observed in the decoding process. The goal is to accurately estimate the error rate in the error floor region for certain types of LDPC codes under the partial response channel and further extend the frame error rate down to 10-14or lower. Towards this goal, we first use field programmable gate array (FPGA) hardware simulation to find the trapping sets that cause the decoding failure in the error floor region. For each trapping set, we extract the parameters which are key to the decoding failure rate caused by this trapping set. Then we use a much simpler in situ hardware simulation with these parameters to obtain the conditional decoding failure rate. By considering all the trapping sets we find, we obtain the overall frame error rate in the error floor region. The estimation results for a length -4623 QC-LDPC code under the EPR4 channel are within 0.3 dB of the direct simulation results. In addition, this method allows us to estimate the frame error rate of a LDPC code down to 10-14or lower.

Quasi-cyclic (QC) low-density parity-check (LDPC) codes form an important subclass of LDPC codes. These codes have encoding advantage over other types of LDPC codes. This paper addresses the issue of efficient encoding of QC-LDPC codes. Two methods are presented to find the generator matrices of QC-LDPC codes in systematic-circulant (SC) form from their parity-check matrices, given in circulant form. Based on the SC form of the generator matrix of a QC-LDPC code, various types of encoding circuits using simple shift registers are devised. It is shown that the encoding complexity of a QC-LDPC code is linearly proportional to the number of parity bits of the code for serial encoding, and to the length of the code for high-speed parallel encoding.

This paper presents methods for efficient encoding of quasi-cyclic LDPC codes. Based on these methods, encoding of quasi-cyclic LDPC codes can be implemented using simple shift-registers with complexity linearly proportional to the number of parity-check bits of a code for serial encoding and to the length of a code for parallel encoding. Various encoding circuits are devised and they provide a range of trade-offs between encoding complexity and speed.

Efficient Encoding of Quasi-Cyclic Low-Density Parity-Check Codes Quasi-cyclic (QC) low-density parity-check (LDPC) codes form an important subclass of LDPC codes. These codes have encoding advantage over other types of LDPC codes. This paper addresses the issue of efficient encoding of QC-LDPC codes. Two methods are presented to find the generator matrices of QC-LDPC codes in systematic-circulant form from their parity-check matrices given in circulant form. Based on the systematic-circulation form of the generator matrix of a QC-LDPC code, various types of encoding circuits using simple shift registers are devised. It is shown that the encoding complexity of a QC-LDPC code is linearly proportional to the number of parity bits of the code for serial encoding, and to the length of the code for high-speed parallel encoding.

We propose a novel iterative decoding algorithm based on partial combination of the parity check matrix for iterative soft decoding of Reed-Solomon product codes. It uses only the simple syndrome equations whose solutions are confined to parts of the columns of the parity check matrix. Compared to other algorithms, the proposed algorithm has lower complexity while offering better performance, which is demonstrated by simulations.

In this article, we present a new scheme that combines the turbo product codes with a simple single carrier frequency domain based equalization algorithm. By using transmit diversity in frequency selective Rayleigh fading channels, the system can resist intersymbol interference effectively. Simulation results show that the proposed system offers diversity gain of more than 6 dB.