Benchmarking Large Language Models for Automated Verilog RTL Code Generation