The fast-paced nature of IC design often leads to ECO specifications being delivered as images accompanied by brief textual descriptions. This presents a challenge for implementation teams, who must manually interpret these visuals and translate them into scripts for ECO tools. This manual process is inefficient and susceptible to human error.
Our innovative solution leverages AI to automate this process. By directly processing image-based ECO specifications, we eliminate the need for manual interpretation. The system works as follows:
Input: The ECO specification is provided as an image embedded in a document (e.g., Markdown) with minimal supporting text.
Image Processing: The image is extracted and converted to base64 encoding. The accompanying text is also extracted.
AI-Driven Translation: The combined image and text data is fed to an AI model trained to generate ECO operations in plain text.
Execution: The generated plain text commands can be directly executed in GOF ECO Shell.
For instance, an ECO requiring the insertion of an AND gate can be specified with an image showing the desired connection in the Mark Down file.
Mark Down file for ECO spec:
The preview of the Mark Down file in VS code is shown in Figure 1.
Figure 1: ECO spec in Image
The AI generates the necessary ECO operations in plain text to complete the change, significantly accelerating ECO implementation and reducing errors.
Plain text ECO operation is applied in GOF Shell:
GOF completes the ECO process by calling internal APIs. The resulting schematic change is shown in Figure 2.
Figure 2: Image to ECO result
This work demonstrates the potential of AI to revolutionize the IC ECO process. By directly interpreting image-based specifications, our system eliminates manual translation, leading to faster turnaround times, fewer errors, and improved design efficiency. Future research will focus on expanding the types of ECO specifications supported and further integrating this technology into existing design flows.
