Bench4HLS: End-to-End Evaluation of LLMs in High-Level Synthesis Code Generation

In last two years, large language models (LLMs) have shown strong capabilities in code generation, including hardware design at register-transfer level (RTL). While their use in high-level synthesis (HLS) remains comparatively less mature, the ratio of HLS- to RTL-focused studies has shifted from 1:10 to 2:10 in the past six months, indicating growing interest in leveraging LLMs for high-level design entry while relying on downstream synthesis for optimization. This growing trend highlights the need for a comprehensive benchmarking and evaluation framework dedicated to LLM-based HLS. To address this, We present Bench4HLS for evaluating LLM-generated HLS designs. Bench4HLS comprises 170 manually drafted and validated case studies, spanning small kernels to complex accelerators, curated from widely used public repositories. The framework supports fully automated assessment of compilation success, functional correctness via simulation, and synthesis feasibility/optimization. Crucially, Bench4HLS integrates a pluggable API for power, performance, and area (PPA) analysis across various HLS toolchains and architectures, demonstrated here with Xilinx Vitis HLS and validated on Catapult HLS. By providing a structured, extensible, and plug-and-play testbed, Bench4HLS establishes a foundational methodology for benchmarking LLMs in HLS workflows.

DecoRTL: A Run-time Decoding Framework for RTL Code Generation with LLMs

As one of their many applications, large language models (LLMs) have recently shown promise in automating register transfer level (RTL) code generation. However, conventional LLM decoding strategies, originally designed for natural language, often fail to meet the structural and semantic demands of RTL, leading to hallucinated, repetitive, or invalid code outputs. In this paper, we first investigate the root causes of these decoding failures through an empirical analysis of token-level entropy during RTL generation. Our findings reveal that LLMs exhibit low confidence in regions of structural ambiguity or semantic complexity, showing that standard decoding strategies fail to differentiate between regions requiring determinism (syntax-critical regions) and those that benefit from creative exploratory variability (design-critical regions). Then, to overcome this, we introduce DecoRTL, a novel run-time decoding strategy, that is both syntax-aware and contrastive for RTL code generation. DecoRTL integrates two complementary components: (i) self-consistency sampling, which generates multiple candidates and re-ranks them based on token-level agreement to promote correctness while maintaining diversity; and (ii) syntax-aware temperature adaptation, which classifies tokens by their syntactical and functional roles and adjusts the sampling temperature accordingly, enforcing low temperature for syntax-critical tokens and higher temperature for exploratory ones. Our approach operates entirely at inference time without requiring any additional model fine-tuning. Through evaluations on multiple open-source LLMs using the VerilogEval benchmark, we demonstrate significant improvements in syntactic validity, functional correctness, and output diversity, while the execution overhead (performance overhead) is imperceptible.