arXiv subjects

Yihan Yin

Publications and source records attributed to Yihan Yin.

8 recordsLinked to original sources

arXiv2026

Potential Applications of HBF in LLM Serving Systems

LLM serving is increasingly constrained by memory capacity as model weights, KV caches, and the number of served model variants continue to grow. This report examines High-Bandwidth Flash (HBF) as a capacity-oriented extension to HBM-based serving systems. We first discuss how HBF can be integrated into the GPU memory hierarchy without undermining the bandwidth expected by the compute die. We then model the system-level value of added capacity as expanded residency for read-mostly model-state objects. Under this view, HBF can improve MoE serving by enabling more expert replicas and can improve multi-model serving by reducing model loading and supporting hot-model replication. Our simulation results show that these benefits depend on preserving the HBM-resident execution path while using HBF to expand the resident set of model weights.

cs.AR↗

arXiv2026

High-speed and high-gain graphene photovoltaic phototransistor gated by a van der Waals heterojunction

Two-dimensional (2D) material-based phototransistors offer a unique combination of optical sensing, signal amplification, and logic operation within a single device, yet fundamentally suffering from an inherent gain-speed trade-off. Here, we demonstrate a 2D photovoltaic phototransistor that overcomes this limitation using a MoS2/PtSe2 heterojunction to gate a graphene channel. The ultrafast photovoltaic effect in the heterojunction enables charge separation, yielding ultrahigh photoconductive gain (up to 10^8) in graphene channel via interfacial gating. Besides, the response time (below the instrumental resolution of 550 ns) is governed by carrier transit in graphene channel, enabling simultaneous high speed and high gain. Moreover, broadband photodetection from visible to near-infrared is enabled by the optical properties of the MoS2/PtSe2 heterojunction, with the detectivity exceeding 10^11 Jones. These results establish a new paradigm for high-performance 2D phototransistors by harnessing photovoltaic and photogating effects to overcome the classical gain-speed trade-off.

cond-mat.mtrl-sci↗

arXiv2026

CODA: Algorithm-Hardware Co-design for Edge Video Diffusion via NMP-Enabled Compute-Cache Operator Disaggregation

Deploying Video Diffusion Models (VDMs) on edge devices is appealing for localized and privacy-preserving generation, but their iterative Transformer-based denoising remains too slow for practical local inference. Cross-Timestep Caching (CTC) has emerged as a promising direction for reducing redundant computation, reusing activations across adjacent denoising steps rather than modifying model weights, while largely preserving generation fidelity. However, on memory-constrained edge GPUs, CTC requires a massive cache footprint that quickly exceeds on-device VRAM and forces the cache into host memory. More fundamentally, cache operators remain tightly interleaved and chain-dependent with native compute operators, so naive near-memory offloading still incurs repeated PCIe exchanges for residual and fusion computations, turning cache reuse into a communication- and serialization-bound execution flow. We therefore propose CODA, an algorithm-hardware co-designed architecture centered on Compute-Cache Operator Disaggregation. CODA separates dense compute paths and memory-bound cache paths across the xPU and a lightweight DIMM-side near-memory engine, reorganizes fragmented cache activity into hardware-friendly coalesced segments, and exploits Classifier-Free Guidance (CFG) branch independence to overlap xPU compute with cache-side execution. Experiments show that CODA achieves up to 1.80x end-to-end speedup and 1.74x higher energy efficiency, while preserving competitive generation quality compared with a state-of-the-art caching algorithm.

cs.AR↗

arXiv2026

Isolation-aware Scheduling Framework for DNN-based End-to-End Autonomous Driving System on Tile-based Accelerators

Level-4+ autonomous driving systems (ADS) must run dozens of heterogeneous deep neural networks (DNNs) as end-to-end (E2E) pipelines under a strict latency constraint (<=100 ms), even as execution time varies by up to 3.3x. Cost rules out dedicating isolated hardware to each function in mass-produced ADS, so these DNNs must be densely colocated on a single chip, which introduces shared-resource contention. Tile-based accelerators expose two scheduling opportunities that conventional ADS schedulers do not exploit. First, they provide a tunable degree of parallelism (DoP): assigning more tiles raises DoP and can shorten DNN execution time. Second, they provide hardware-native isolation: tiles can be physically partitioned among co-located DNNs. But using this flexibility is expensive: changing a task's DoP triggers a stop-migrate-restart reallocation of its weights and intermediate features. At ADS task rates of 10-240 Hz, these stalls accumulate along E2E chains and threaten deadlines. Reservation-based schedulers fix DoP and leave this flexibility unused; work-conserving schedulers exploit it but assume reallocation is cheap and treat deadlines as independent. We present ADS-Tile that combines configurable isolation and elastic reservation into a spatio-temporal isolation-sharing space that bounds where and when reallocation occurs; a probabilistic latency model and a DAG-aware runtime scheduler then use this space to decide task colocation and DoP under shared E2E deadlines. On an industry- and academia- derived ADS benchmark, ADS-Tile uses up to 32% fewer tiles than the work-conserving baseline in deadline-critical settings and cuts reallocation-induced wasted processing capacity from 17%-44% to below 1.2%. Controlled spatio-temporal sharing improves resource efficiency and latency predictability for tile-based ADS.

cs.AR↗

arXiv2026

AccelCIM: Systematic Dataflow Exploration for SRAM Compute-in-Memory Accelerator

SRAM-based compute-in-memory (CIM) offers high computational density and energy efficiency for deep neural network (DNN) accelerators, but its limited capacity causes on/off-chip data movement overhead for large DNN models. Existing CIM accelerator studies typically assume that DNN models fit entirely on-chip, leaving efficient dataflow design largely untapped. This paper introduces AccelCIM, a systematic dataflow exploration framework for SRAM CIM accelerator, which addresses two key limitations of prior work. (1) It formulates a systematic dataflow design space spanning CIM macro configurations and macro-array organizations. (2) It introduces rigorous design evaluation using cycle-accurate architectural simulation and post-layout PPA analysis. We conduct an extensive design space exploration and apply AccelCIM to representative LLM applications, providing practical insights for the principled design of CIM accelerators.

cs.AR↗

arXiv2026

Hardware-Software Co-design for 3D-DRAM-based LLM Serving Accelerator

Large language models (LLMs) have been widely deployed for online generative services, where numerous LLM instances jointly handle workloads with fluctuating request arrival rates and variable request lengths. To efficiently execute coexisting compute-intensive and memory-intensive operators, near-memory processing (NMP) based computing paradigm has been extensively proposed. However, existing NMP designs adopt coarse-grained KV cache management and inflexible attention execution flow. Such limitations hinder these proposals from efficiently handling \textit{highly dynamic} LLM serving workloads, limiting their ability to accelerate LLM serving. To tackle these problems, we propose Helios, a Hybrid-bonding-based \uline{L}LM \uline{S}erving accelerator. Helios aims to bridge the fundamental gap between the dynamic nature of KV cache management in LLM serving and the distributed, non-uniform memory abstraction among NMP processing engines (PEs). To this end, we design both the intra-PE execution flow and the inter-PE communication primitives for distributed tiled attention execution. We further propose \textit{spatially-aware} KV cache allocation mechanism to balance the attention workload distribution while minimizing the inter-PE data transfer overhead. Compared with existing GPU/NMP designs, Helios achieves 3.25 times (geomean) speedup and 3.36 times (geomean) better energy efficiency, along with up to 72%/76% P50/P99 time-between-tokens degradation.

cs.AR↗

arXiv2023

Synchrotron Radiation Dominates the Extremely Bright GRB 221009A

The brightest Gamma-ray burst, GRB 221009A, has spurred numerous theoretical investigations, with particular attention paid to the origins of ultra-high energy TeV photons during the prompt phase. However, analyzing the mechanism of radiation of photons in the $\sim$MeV range has been difficult because the high flux causes pile-up and saturation effects in most GRB detectors. In this letter, we present systematic modeling of the time-resolved spectra of the GRB using unsaturated data obtained from Fermi/GBM (precursor) and SATech-01/GECAM-C (main emission and flare). Our approach incorporates the synchrotron radiation model, which assumes an expanding emission region with relativistic speed and a global magnetic field that decays with radius, and successfully fits such a model to the observational data. Our results indicate that the spectra of the burst are fully in accordance with a synchrotron origin from relativistic electrons accelerated at a large emission radius. The lack of thermal emission in the prompt emission spectra supports a Poynting-flux-dominated jet composition.

astro-ph.HE↗

arXiv2023

GRB 220408B: A Three-Episode Burst from a Precessing Jet

Jet precession has previously been proposed to explain the apparently repeating features in the light curves of a few gamma-ray bursts (GRBs). In this {\it Letter}, we further apply the precession model to a bright GRB 220408B by examining both its temporal and spectral consistency with the predictions of the model. As one of the recently confirmed GRBs observed by our GRID CubeSat mission, GRB 220408B is noteworthy as it exhibits three apparently similar emission episodes. Furthermore, the similarities are reinforced by their strong temporal correlations and similar features in terms of spectral evolution and spectral lags. Our analysis demonstrates that these features can be well explained by the modulated emission of a Fast-Rise-Exponential-Decay (FRED) shape light curve intrinsically produced by a precessing jet with a precession period of $18.4 \pm 0.2$ seconds, a nutation period of $11.1 \pm 0.2$ seconds and viewed off-axis. This study provides a straightforward explanation for the complex yet similar multi-episode GRB light curves.

astro-ph.HE↗