arXiv subjects

Yunfei Gu

Publications and source records attributed to Yunfei Gu.

5 recordsLinked to original sources

arXiv2026

Akashic: A Low-Overhead LLM Inference Service with MemAttention

Recent LLM-based agent systems continuously accumulate context across multi-turn interactions, tool invocations, and cross-session workflows. Replaying the full history for every request quickly becomes impractical: long contexts increase prefill cost, may exceed context limits, and often bury task-relevant evidence in irrelevant content, degrading both serving efficiency and output quality. We propose Akashic, a low-overhead memory system built around MemAttention, which organizes context into bounded chunks and models semantic relationships across chunks, preserving cross-chunk evidence without repeatedly rewriting the full history. Akashic further applies hardware-software co-designed memory placement to co-locate likely co-retrieved chunks, reducing retrieval fragmentation and I/O overhead. Across four representative workloads and three model sizes, Akashic improves task accuracy by up to 10.2 points, throughput by up to 1.21x, and sustainable request rate by up to 1.88x over strong prior memory baselines.

cs.AI↗

arXiv2025

PS-WL: A Probability-Sensitive Wear Leveling scheme for SSD array scaling

As flash-based Solid State Drive (SSD) arrays become essential to modern data centers, scaling these arrays to meet explosive data growth is a frequent and critical operation. However, the conventional wear-leveling (WL) paradigm applied during scaling suffers from a fundamental flaw: it ignores the non-linear relationship between wear and failure probability, potentially pushing the most vulnerable, aged disks towards premature failure. To address this critical issue at its root, we propose the Probability-Sensitive Wear Leveling (PS-WL) scheme, which shifts the optimization goal from balancing wear to directly balancing failure risk. At its core, PS-WL introduces an "effective lifetime" model derived from a realistic failure probability to more accurately assess disk lifetime. This model guides a PID controller for wear leveling operation, with a conservative zone minimizes performance overhead by restricting warm data migration. Comprehensive simulations validate the superiority of PS-WL over state-of-the-art methods. The results demonstrate that our approach significantly reduces performance overhead while, most critically, consistently and effectively lowering the aggregated array failure risk across diverse system configurations and workloads. This proves that by directly optimizing for reliability, PS-WL builds a scalable storage system that is, by design, fundamentally safer, more efficient, and more stable.

cs.DC↗

arXiv2024

Gaze into the Pattern: Characterizing Spatial Patterns with Internal Temporal Correlations for Hardware Prefetching

Hardware prefetching is one of the most widely-used techniques for hiding long data access latency. To address the challenges faced by hardware prefetching, architects have proposed to detect and exploit the spatial locality at the granularity of spatial region. When a new region is activated, they try to find similar previously accessed regions for footprint prediction based on system-level environmental features such as the trigger instruction or data address. However, we find that such context-based prediction cannot capture the essential characteristics of access patterns, leading to limited flexibility, practicality and suboptimal prefetching performance. In this paper, inspired by the temporal property of memory accessing, we note that the temporal correlation exhibited within the spatial footprint is a key feature of spatial patterns. To this end, we propose Gaze, a simple and efficient hardware spatial prefetcher that skillfully utilizes footprint-internal temporal correlations to efficiently characterize spatial patterns. Meanwhile, we observe a unique unresolved challenge in utilizing spatial footprints generated by spatial streaming, which exhibit extremely high access density. Therefore, we further enhance Gaze with a dedicated two-stage approach that mitigates the over-prefetching problem commonly encountered in conventional schemes. Our comprehensive and diverse set of experiments show that Gaze can effectively enhance the performance across a wider range of scenarios. Specifically, Gaze improves performance by 5.7\% and 5.4\% at single-core, 11.4\% and 8.8\% at eight-core, compared to most recent low-cost solutions PMP and vBerti.

cs.AR↗

arXiv2018

Reservoir Computing based Neural Image Filters

Clean images are an important requirement for machine vision systems to recognize visual features correctly. However, the environment, optics, electronics of the physical imaging systems can introduce extreme distortions and noise in the acquired images. In this work, we explore the use of reservoir computing, a dynamical neural network model inspired from biological systems, in creating dynamic image filtering systems that extracts signal from noise using inverse modeling. We discuss the possibility of implementing these networks in hardware close to the sensors.

cs.CV↗

arXiv2018

Hardware based Spatio-Temporal Neural Processing Backend for Imaging Sensors: Towards a Smart Camera

In this work we show how we can build a technology platform for cognitive imaging sensors using recent advances in recurrent neural network architectures and training methods inspired from biology. We demonstrate learning and processing tasks specific to imaging sensors, including enhancement of sensitivity and signal-to-noise ratio (SNR) purely through neural filtering beyond the fundamental limits sensor materials, and inferencing and spatio-temporal pattern recognition capabilities of these networks with applications in object detection, motion tracking and prediction. We then show designs of unit hardware cells built using complementary metal-oxide semiconductor (CMOS) and emerging materials technologies for ultra-compact and energy-efficient embedded neural processors for smart cameras.

cs.CV↗