Using AI to Debug More Quickly and Accurately

It is widely acknowledged that functional verification consumes the greatest amount of time and resources on chip development projects, and that debug is the dominant task within the verification effort. It’s hard enough to define the right coverage points in chip design and testbench, and then develop the right constraints to generate tests that will steer stimulus to hit those points. Hitting coverage is meaningless unless the tests are self-checking, and the real trouble begins when some of these tests fail.

Numerous test failures are inevitable during verification. Human designers make mistakes, AI agents hallucinate, and even reused IP may not operate as expected. Verification engineers are human too, so errors in testbenches, tests, and constraints are also common. Whenever a test fails, the verification team must debug the failure to determine the root cause, which may be in the design or verification code, attempt to fix it, and verify that the fix is correct by rerunning the test successfully. This is a challenging process, and the engineers are always looking for help.

The Synopsys Verdi® debug and verification management platform is an essential component of any verification engineer’s toolkit. It offers many ways to explore the design and verification environment while analyzing simulation results. Users can move effortlessly among waveforms, source code, schematics, state machine diagrams, and other views while Verdi keeps all of them synchronized. Because there are so many options available, Verdi includes powerful AI technology to automate difficult and tedious debug steps.

Synopsys recently introduced Verdi Assistant, the next generation of AI debug automation. It includes specialized AI agents for debug and orchestrates them for optimal benefits. Each agent encapsulates domain expertise, turning repetitive debug muti-step debug tasks into one-call services. A diverse set of agents is important because verification is not a “one size fits all” task. Some users rely heavily on waveforms, some on logfiles, and some on source code navigation. Autonomous agents automate all these tasks, for much greater verification productivity.

Users communicate with Verdi Assistant using a natural language interface, reducing the barrier of writing scripts or tool commands. A Chatbot panel in the Verdi graphical user interface (GUI) enables natural language queries and requests. For example, a user might ask “How do I trace drivers in Verdi?” and receive a detailed response without having to explore menus or scroll through documentation. Users can also request Verdi to perform a task autonomously, such as typing “Trace the drivers for signal XYZ” into the chatbot panel.

Many users rely on integrated development environments (IDEs) to write and edit their design and verification code. In recent years, Microsoft Visual Studio Code (VS Code) has become the IDE platform of choice. Synopsys provides an integration between Verdi and VS Code to create a seamless experience. Users can edit their code within the VS Code environment, recompile for debug, and get easy access to Verdi Assistant and other Generative AI (GenAI) solutions from Synopsys. This synchronizes source code development and debug.

Verdi Assistant also enables connections to other AI applications using the Model Context Protocol (MCP) standard. MCP allows large language models (LLMs) and other AI models to access external databases, data, and tools with full security. Thus, an MCP client can request that Verdi provide debug service, and Verdi then becomes an MCP server.