Sarcouncil Journal of Engineering and Computer Sciences

Abstract: Semiconductor die size prediction during early design phases represents a fundamental challenge affecting product costs, manufacturing feasibility, and market competitiveness. Traditional estimation techniques rely heavily on simplified scaling factors and linear approximations that fail to capture the complex interdependencies present in modern system-on-chip architectures. The emergence of artificial intelligence-based prediction frameworks offers transformative potential for addressing these challenges through sophisticated machine learning algorithms capable of modeling non-linear relationships between design parameters and final silicon dimensions. Machine learning applications in VLSI design demonstrate superior performance across multiple technology nodes, with neural network-based approaches achieving verification coverage rates exceeding ninety-five percent compared to conventional formal verification tools that typically reach seventy to eighty percent coverage within practical time constraints. Advanced feature engineering techniques enable the extraction of complex spatial and connectivity patterns from circuit layouts, while ensemble learning methodologies provide robust prediction capabilities through the combination of multiple weak learners. The integration of human-centric AI approaches in multi-objective optimization shows remarkable effectiveness in balancing area minimization, power efficiency, and performance maximization requirements. Contemporary frameworks successfully model technology node transitions with high accuracy, enabling designers to evaluate architectural alternatives across different manufacturing processes while maintaining prediction reliability for both adjacent and multi-generation technology migrations.