Automated repair of hardware description language (HDL) designs faces two major hurdles: an enormous candidate space and strict syntactic‑semantic constraints. Generic mutation strategies tend to generate many syntactically invalid candidates, wasting compilation and simulation resources, while synthesis‑driven or template‑based methods limit generality and portability. This paper introduces a dictionary‑guided HDL repair system composed of two key components.
First, an ANTLR‑derived DUT‑specific mutation dictionary supplies category‑constrained token substitutions, insertions, and deletions. These operators are applied directly to Verilog source via regex matching, eliminating the need for abstract syntax tree manipulation or synthesis.
Second, a simulation‑divergence fault localization (FL) module runs a single simulation to capture diverging output wires, then scores source lines by structural proximity to those signals, steering the mutation search toward high‑suspicion regions.
The system exhaustively sweeps all dictionary mutations on the highest‑scored lines in a deterministic pass; if no passing repair is found, it falls back to a genetic programming (GP) search.
Evaluation on the CirFix benchmark suite across six DUT families shows correct oracle‑passing repairs for 14 bug variants, including a six‑edit multi‑bug case that CirFix cannot fix, and achieves roughly an 18× speedup on a two‑edit benchmark. These results demonstrate that dictionary‑constrained mutation operators combined with lightweight simulation‑divergence FL provide a practical, competitive approach to automated HDL repair without formal analysis or synthesis dependencies.
Review: The study highlights how language‑level dictionary constraints dramatically improve mutation efficiency and how a single‑run simulation can rapidly localize faults, offering a scalable path for HDL automatic repair.