SEMIKONG - A large-scale language model designed specifically for the semiconductor industry
SEMIKONG is a large-scale language model (LLM) specifically designed for the semiconductor industry, jointly developed by Aitomatic, FPT Software, and Tokyo Electron. Based on in-depth domain knowledge, it solves problems related to semiconductor manufacturing and design...
What is SEMIKONG?
SEMIKONG is a large-scale language model (LLM) specifically tailored for the semiconductor industry, jointly developed by Aitomatic, FPT Software, and Tokyo Electron. Based on deep domain knowledge, it addresses unique challenges in semiconductor manufacturing and design, such as complex physical and chemical problems. SEMIKONG integrates expert knowledge and optimizes the pre-training process, providing a foundational model capable of expert-level understanding of etching problems. Compared to general-purpose LLMs, SemiKong demonstrates superior performance in semiconductor manufacturing tasks, laying the foundation for company- or tool-specific proprietary model development and driving further research and application of domain-specific AI models.
SEMIKONG's main functions
- Understanding the expertise in the semiconductor fieldTo understand the complexities of semiconductor manufacturing and design, especially in the field of etching.
- Optimize manufacturing processBased on learning from a large amount of semiconductor-related data, it assists in optimizing semiconductor manufacturing processes, such as parameter optimization, anomaly detection, and predictive maintenance.
- Auxiliary IC DesignSEMIKONG can assist in integrated circuit (IC) design tasks, including design rule checking, layout generation, and design space exploration.
- Improve the performance of AI solutionsImprove the performance of AI-driven semiconductor manufacturing tasks by fine-tuning pre-trained large language models.
- Expert knowledge integrationIntroducing a framework that integrates expert knowledge to advance the evaluation process of domain-specific AI models.
SEMIKONG's technical principles
- Data planningSEMIKONG's development began with the planning of large-scale, high-quality text datasets specific to the semiconductor field, including technical books, papers, and patents.
- Pre-training and fine-tuningThe model is trained using a pre-training and fine-tuning approach. The pre-training phase uses domain-specific data to enhance the model's domain knowledge, while the fine-tuning phase trains the model to perform specific tasks.
- Domain OntologyCollaborating with semiconductor experts to build a systematic ontology of semiconductor manufacturing processes helps AI researchers develop domain-specific AI models more effectively.
- Expert feedback loopBased on expert feedback loops, the system uses expert evaluation models to generate answers, producing evaluation criteria and high-quality benchmark tests.
- Model quantization and adaptationAfter pre-training and fine-tuning, model quantization and adaptation are performed to facilitate model deployment.
SEMIKONG's project address
- Project official website:semikong.ai
- GitHub repository:https://github.com/aitomatic/semikong
- HuggingFace model library:https://huggingface.co/pentagoniac
- arXiv technical paper:https://arxiv.org/pdf/2411.13802
Application scenarios of SEMIKONG
- Process parameter optimizationAdjusting parameters in the semiconductor manufacturing process improves production efficiency and product quality.
- Anomaly detection systemIt helps identify anomalies in the manufacturing process, reducing defects and improving reliability.
- IC design auxiliary toolsIt helps engineers adhere to design rules and generate optimized layouts in integrated circuit design.
- Expert decision supportAs an expert system, it provides solutions and decision support for complex technical problems.
- Technology Education PlatformUsed as an educational tool to help students understand the complex concepts of semiconductor manufacturing and design.