LLM adoption in EDA, Semiconductor

 Writing good HDL code from a specification targeting today’s advanced process nodes with stringent power, performance, and area (PPA) requirements would take a single engineer years to complete. A team might achieve it in months. Yet teams are becoming stretched thin as the semiconductor skills gap continues to grow. But design companies have lots of internal data, lots of experts, lots of experience. They are capturing this and using it on top of the LLMs and GenAI to provide the missing knowledge.

There are two main approaches when selecting a model for GenAI by the design companies:

1. Service-Based Models:  These models are provided by vendors on a pay-as-you-go plan. End users do not have access to the training data or model, and their interaction with the model is limited to queries.
2. Internally Hosted Models: If the company wants more control over its data and is uncertain about the security policies of external vendors, it can opt to host its LLMs internally.

Some of the most powerful LLMs available currently are service-based models.

• Design cleaning : Today, LLM proof of concept(POC) focuses on the design-cleaning process. Once the initial draft of the HDL description for the chip is created, engineers can use the LLM chatbot to examine the design, verify it against the specifications, identify and fix issues, perform analysis tasks, and receive explanations in their own language.

• Review and eliminate bugs : LLM can do spec reviews, code reviews, test reviews, and change management reviews. This can save hundreds of hours of individual engineering time, and hundreds of group meetings for specification and code reviews. And remove many bugs previously uncovered during the ramp-up to regression verification.

• Using Prompts : The essential usage of LLM is to load in the architecture specifications, design specifications, integration connection specifications, and the design itself. From there, the users can issue prompts to the LLM such as “list the name of irregular nets”, “list all possible irregular pins”, “automate hook up testbenches”, “tool script auto-completion”, and “RTL code auto-completion”. 

• Fault Detection and Quality Assurance: GenAI and LLM can play a crucial role in fault detection and quality assurance processes within semiconductor manufacturing. By analyzing large amounts of sensor data and historical records, these AI systems can detect potential defects or anomalies in real time, reducing production downtime and ensuring product reliability.

• Hallucination in GenAI : However, sometimes, the generative AI model or LLM might provide wrong information or fabricated answers to the user in a persuasive manner. This phenomenon is known as “hallucination” and is a obstacle when relying on generative AI. 

At times, training and fine-tuning methods can be excessive if the data and policies are constantly changing. This necessitates continuous retraining, which can render the process ineffective. These challenges highlight the obstacles companies must address before they can successfully utilize generative AI models trained on private data to deliver accurate responses for users.