An Agentic Python Framework for LLM-Driven UVM Testbench Synthesis with Closed-Loop Compiler Feedback

By Vikash Kumar, Arm

Abstract

Protocol bridge verification is a persistent bottleneck in SoC development: for every new bridge instance, engineers hand-craft a full UVM testbench from scratch, consuming 40-100 engineering hours of bring-up effort. We present an agentic Python framework that generates complete, protocol-aware UVM testbenches for arbitrary IP and subsystem configurations from a structured Verification Manifest—a JSON specification encoding interface topology, protocol semantics, and coverage intent. The framework employs a dual closed-loop architecture: an inner compiler feedback loop that injects categorised UVM diagnostics rather than raw tool output, and an outer coverage loop driven by intent-gap analysis against the Manifest. Validated on the AMBA AHB-to-APB bridge—a representative production protocol bridge with a 7-state control FSM, two-phase APB handshake, and non-trivial burst serialisation logic—the framework achieves 96.8% functional coverage, a 4.7× bring-up speedup, and a Protocol Verification Completeness score of 0.87, with only 2.1% of the generated 4,623-line testbench requiring human correction. The generated scoreboard independently identified a critical RTL functional fault that the manual baseline missed entirely. Generalisation across four protocol families confirms that speedup increases with protocol complexity, ranging from 3.1× (APB Subordinate) to 6.1× (CHI Subordinate).

Index Terms—UVM, testbench automation, agentic AI, large language models, closed-loop compiler feedback, protocol verification, verification manifest, chiplet, subsystem verification

To read the full article, click here

×
Semiconductor IP