IR choice dominates model choice in LLM hardware generation from natural language, with simulation pass rates ranging 3-88% across IRs but varying less than 1.25x across models.
High-level synthesis for FPGAs: From prototyp- ing to deployment,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
From Natural Language to Silicon: The Representation Bottleneck in LLM Hardware Design
IR choice dominates model choice in LLM hardware generation from natural language, with simulation pass rates ranging 3-88% across IRs but varying less than 1.25x across models.