Page 493 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Use Case 2: AI-based Semiconductor Design Automation and

Optimization 4.5: Manufacturing

Organization: AgileSoDA

Country: South Korea

Contact Person(s):

Steve Kim, steve.kim@ agilesoda .com
SeungYeol Baek, sybaek@ agilesoda .ai

1 Use Case Summary Table

Item Details
Category Manufacturing: Semiconductor Industry, AI Automation, EDA (Electronic
Design Automation)
Problem Rapidly increasing SoC (System-on-Chip) demand, shortage of design
Addressed professionals, and labor-intensive design processes pose global challenges
in sustainable industrial growth. Lengthy design cycles and manual verifica-
tion hinder productivity and competitiveness.
Key Aspects of Reinforcement Learning-based AI for automated chip placement optimization
Solution Seamless integration with existing tools and workflows
Multiple simulation capabilities
Expandable semiconductor design platform

Technology AI, Reinforcement Learning, EDA, Semiconductor Design, Physical Layout,
Keywords Macro Placement, Automation, ChipNSoDA[1]

Data Availability Private (confidential semiconductor design data in LEF/DEF formats)
Metadata (Type Semiconductor design files, physical layout data, macro cell placement data
of Data) (LEF/DEF format)
Model Training Design of Graph Neural Network with appropriate nodes and edges, and
and Fine-Tuning train the model using prepared data with reinforcement learning.
Test performance using proximate cost metrics, verify accuracy with commer-
cial EDA tools, and adjust parameters to improve performance.

Testbeds or Pilot An internal pilot deployment was carried out in collaboration with our partner
Deployments ASICLAND[2].

457