Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.23406

GUMPIM: Unitary and Malleable Memory for Processing-in-Memory with Guaranteed PIM Pages

Yilong Zhao·Fangxin Liu·Yiwei Hu·Zongwu Wang·Mingyu Gao·Onur Mutlu·He Xian·Hanlin Dong·Naifeng Jing·Xiaoyao Liang·Haibing Guan·Qiang Xu·Chixiao Chen·Shuxue Quan·Tao Yang·Li Jiang
View original source ↗Read source document ↗Export metadata ↓

Abstract

DRAM-based Processing-in-Memory (PIM) addresses the "memory wall" by executing computations directly inside main memory. However, memory interleaving and virtual memory limit contiguous data size visible to PIM units, constraining PIM task granularity. To mitigate this, existing PIM systems drastically isolate PIM memory or disable memory interleaving. These design choices, however, decrease the CPU memory bandwidth and introduce extra data transfer, leading to an additional "system memory wall" that must be resolved to realize PIM's full potential. In this work, we propose GUMPIM, a PIM system that allows interleaved CPU pages and non-interleaved PIM pages to coexist in a Unitary and Malleable memory space with Guaranteed PIM page allocation. GUMPIM enables zero-copy during PIM task offloading and maintains CPU memory bandwidth while ensuring low PIM offloading overhead. First, we propose a dual-track memory management mechanism consisting of independent page allocation and address translation for CPU and PIM pages. Second, we design GUMPIM interface hardware on PIM-enabled DRAMs to provide a dynamic address mapping for the different data layouts of CPU and PIM pages. Third, we propose a PIM-assisted page migration mechanism that transparently migrates pages while preserving CPU access bandwidth, thereby enabling guaranteed and accelerated PIM page allocation. GUMPIM requires no changes to commodity DRAM standards; all hardware modifications are limited to the DRAM side, ensuring full compatibility with existing CPUs and enabling immediate deployment on current HBMx- and LPDDRx-based PIM platforms. Our results show only <0.1% performance degradation for CPU workloads on GUMPIM. For PIM workloads, GUMPIM reduces memory allocation and CPU-part computation times by 2.7x and 4.93x, respectively, yielding an end-to-end 2.3x speedup over a state-of-the-art baseline system.

Explore related subjects

cs.AR

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Yilong Zhao, Fangxin Liu, Yiwei Hu, Zongwu Wang, Mingyu Gao, Onur Mutlu, He Xian, Hanlin Dong, Naifeng Jing, Xiaoyao Liang, Haibing Guan, Qiang Xu, Chixiao Chen, Shuxue Quan, Tao Yang, Li Jiang. 2026-10-16. GUMPIM: Unitary and Malleable Memory for Processing-in-Memory with Guaranteed PIM Pages. https://doi.org/10.1145/3833426

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

VeriGRAG: Enhancing LLM-Based Verilog Code Generation with Structure-Aware Soft Prompts

Large language models (LLMs) have demonstrated remarkable capabilities in generating Verilog code from natural language descriptions. However, Verilog code inherently encodes the structural characteristics of hardware circuits, including data path organization and interconnections among components. Effectively leveraging these structural characteristics to enhance the functional and syntactic correctness of LLM-based Verilog code generation remains a significant challenge. To address this challenge, we propose VeriGRAG, a novel framework that converts Verilog code into data path graphs and employs graph neural networks (GNNs) to encode their structural information. A multimodal retriever, trained through knowledge distillation from a cross-attention encoder to a dual encoder, then selects the graph embeddings most relevant to the input hardware description. The VeriFormer module uses learnable query tokens and two-stage training to extract code-relevant structural features and align them with the LLM's input embedding space, producing structure-aware soft prompts. Experiments on VerilogEval and RTLLM show improvements in both syntactic and functional correctness. With Qwen2.5-Coder-32B as its backbone, VeriGRAG achieves the highest functional $pass@1$ and $pass@5$ scores among the compared methods on both benchmarks.

cs.AR↗
arXiv2026

Hardware-Efficient Accurate 4-bit Multiplier for Xilinx 7 Series FPGAs

As IoT and edge inference proliferate, there is a growing need for lookup-table (LUT)-based multipliers that combine a low LUT count with a high maximum operating frequency. This paper proposes a hardware-efficient exact unsigned 4 x 4-bit multiplier for AMD Xilinx 7-series FPGAs. By reorganizing the logic functions mapped to the LUTs, the proposed design uses 11 LUTs and two CARRY4 blocks and achieves a maximum operating frequency of 361.402 MHz. Among the eight evaluated configurations, it has the lowest LUT count and the secondhighest maximum operating frequency.

cs.AR↗
arXiv2026

SpikeDuet: CPU-Accelerator Co-Execution of Sparse SNNs on Commodity MCUs

Spiking neural networks (SNNs) expose sparse event activity, but commodity microcontrollers increasingly pair general-purpose processors with dense matrix-vector accelerators whose suitability for sparse inference is unclear. We investigate whether the on-chip Matrix Vector Processor (MVP) of the EFR32MG24 can accelerate SNN inference beyond optimized event-driven Cortex-M33 execution. Our automated snnTorch-to-C deployment flow combines hardware-calibrated engine selection, function-preserving neuron reordering that places co-active neurons adjacently for the accelerator, and asynchronous scheduling that overlaps CPU and MVP computation. On physical hardware, heterogeneous execution achieves 1.27--1.87$\times$ speedup and 11--34\% lower energy per inference than optimized sparse Cortex-M33 execution across five evaluated network configurations, while the smallest Braille network remains CPU-optimal. Reordering alone moves Braille-384's dominant hidden layer from the CPU to the MVP, and its benefit persists under asynchronous execution. These results show that activity locality and concurrency between CPU and MVP can jointly improve sparse SNN inference on commodity MCUs.

cs.AR↗