Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2602.12962

TriGen: NPU Architecture for End-to-End Acceleration of Large Language Models based on SW-HW Co-Design

Jonghun Lee·Junghoon Lee·Hyeonjin Kim·Seoho Jeon·Jisup Yoon·Hyunbin Park·Meejeong Park·Heonjae Ha
View original source ↗Read source document ↗Export metadata ↓

Abstract

Recent studies have extensively explored NPU architectures for accelerating AI inference in on-device environments, which are inherently resource-constrained. Meanwhile, transformer-based large language models (LLMs) have become dominant, with rapidly increasing model sizes but low degree of parameter reuse compared to conventional CNNs, making end-to-end execution on resource-limited devices extremely challenging. To address these challenges, we propose TriGen, a novel NPU architecture tailored for resource-constrained environments through software-hardware co-design. Firstly, TriGen adopts low-precision computation using microscaling (MX) to enable additional optimization opportunities while preserving accuracy, and resolves the issues that arise by employing such precision. Secondly, to jointly optimize both nonlinear and linear operations, TriGen eliminates the need for specialized hardware for essential nonlinear operations by using fast and accurate LUT, thereby maximizing performance gains and reducing hardware-cost in on-device environments, and finally, by taking practical hardware constraints into account, further employs scheduling techniques to maximize computational utilization even under limited on-chip memory capacity. We evaluate the performance of TriGen on various LLMs and show that TriGen achieves an average 2.73x performance speedup and 52% less memory transfer over the baseline NPU design with negligible accuracy loss.

Explore related subjects

cs.ARcs.AI

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Jonghun Lee, Junghoon Lee, Hyeonjin Kim, Seoho Jeon, Jisup Yoon, Hyunbin Park, Meejeong Park, Heonjae Ha. 2026-02-13. TriGen: NPU Architecture for End-to-End Acceleration of Large Language Models based on SW-HW Co-Design. https://arxiv.org/abs/2602.12962

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

KEEP EXPLORING

Related papers

arXiv2026

HLSmith: An Expert-Guided Agentic Framework for C/C++-to-HLS Translation

Application-specific FPGA accelerators offer substantial performance and energy-efficiency gains across many application domains, but developing them is costly, often requiring months of specialized effort. Even with high-level synthesis (HLS), designers still need extensive hardware expertise to build high-performance accelerators. Although large language models (LLMs) have demonstrated strong software-generation capabilities, even frontier models lack the hardware intuition and procedural knowledge needed to reliably translate baseline C/C++ programs into high-performance HLS designs: they struggle to identify effective architectures, follow the optimization processes used by HLS experts, and apply hardware transformations consistently across diverse kernels. We present HLSmith, an expert-guided framework for translating C/C++ programs into optimized HLS accelerators. HLSmith combines three components: an HLS optimization expertise library that encodes guarded transformation recipes, their applicability and prerequisite conditions, and unsafe cases to avoid; a staged, feedback-driven orchestration flow modeled on expert HLS development practice that guides agents through synthesis, bottleneck analysis, and optimization; and a tool-grounded model-adaptation pipeline that converts optimization trajectories from commercial frontier models into training data for fine-tuning open-weight LLMs. We evaluate HLSmith on PolyBench against ChatHLS, a leading prior agent-orchestration framework for HLS accelerator development. HLSmith achieves a geometric mean speedup of 4.24x over ChatHLS while producing functionally correct designs, in both software and RTL simulation, for every benchmark, compared with ChatHLS's 57% valid-design rate. It further reaches speedups of up to 252x and 138x with commercial frontier models and open-weight models, respectively.

cs.AR↗
arXiv2026

Accelerating the Mitigation of LLM Inference Nondeterminism Across GPU Architectures

Large language model (LLM) outputs are expected to be reproducible under greedy decoding, yet in practice the same model, prompt, and software stack produce different outputs on different GPUs. The root cause is floating-point non-associativity combined with hardware-dependent kernel selection. Inference frameworks select different matrix-multiplication kernels on each architecture, with different parallel reduction orders and unspecified tensor-core arithmetic, and the resulting rounding differences can flip output tokens. Existing solutions have imperfect cross-architecture reproducibility and incur a significant performance penalty. We present a solution employing a set of fixed-configuration fused-upcast GEMM kernels that load 16-bit weights from memory, upcast them to FP32 in registers, and accumulate with IEEE-754 arithmetic in a reduction order that is a pure function of the problem shape and is therefore independent of the device, its SM count, or kernel scheduling. By fixing the floating-point reduction order as a function of problem shape alone, every GPU runs the same operation sequence, so cross-architecture reproducibility of the linear layers reduces to correct IEEE-754 arithmetic rather than to rounding differences staying below a tie-flip threshold. We confirm our solution's linear-layer outputs are bitwise identical across NVIDIA Ampere, Ada, and Hopper GPUs, while running $1.17$ to $3.1\times$ faster end-to-end than the state-of-the-art solution and cutting weight-memory traffic in half.

cs.AR↗
arXiv2026

Decoupling Logical Masks from GPU Execution for Dynamic Block-Sparse Attention

Attention computation makes inference expensive in video diffusion transformers (vDiTs), which generate videos through iterative denoising. Block-sparse attention (BSA) reduces this cost by computing only blocks selected by a logical mask, which specifies attention interactions to compute. However, coupling logical block geometry to execution choices limits adaptation to varying masks and graphics processing units (GPUs), while runtime kernel specialization can incur preparation overhead that outweighs execution time savings. We present Tessera, a specialized runtime for dynamic BSA that decouples logical masks from GPU execution while preserving specified attention interactions. Its physical mapping layer retains, combines, or subdivides logical attention blocks into physical tiles suited to different attention mask shapes and GPU architectures. Its task organization layer groups and schedules tiles within GPU tasks to reuse data, expose parallelism, and overlap data movement with computation. Finally, profile-guided regime selection enables low- overhead execution plan selection through a lookup table constructed from offline profiling. We implement Tessera with specialized CUDA kernels supporting four NVIDIA GPU generations. Evaluated on 2,315 real attention masks and industrial video diffusion models, Tessera achieves up to 6.79x BSA request speedup over baseline systems in the evaluated video diffusion models.

cs.AR↗