Design Suite

High-sigma Integrated Circuit (IC) design allows only a few defects out of hundreds of millions to billions of units. For 6-σ verification, 1 billion Monte Carlo simulations are required. However, this is practically impossible. InfiniSigma uses neural networks to proactively predict minute variations that may occur during the manufacturing process and analyzes their impact on product performance with high-sigma (6-sigma) level accuracy. Traditionally, High-Sigma Monte Carlo (HSMC) has used a static modeling approach that samples from a predefined variation range and relies on a fixed model once created, which has made it difficult to identify rare defect cases that occur in actual industrial settings. However, InfiniSigma employs an AI-based intelligent active sampling methodology to learn in real-time from a small number of simulation results, intelligently identifying 'risk zones' where defects are most likely to occur and intensively exploring their surroundings to dramatically reduce unnecessary simulation counts.

AI technology is no longer an option but an essential element for the future survival of the EDA industry. The three major goals of chip design—power, performance, and area—present inherently conflicting trade-offs. AI comprehensively analyzes these complex interrelationships, helping designers discover optimal design points that are difficult to find through experience alone. InfiniOpt automatically explores multi-objective design spaces based on reinforcement learning. Applied to analog circuit optimization, it defines design parameters (transistor sizes, resistance values, etc.) as AI agent 'actions' and performance metrics as 'rewards', enabling the AI agent to autonomously learn through numerous simulations and discover optimal design combinations. The true value of InfiniOpt lies in achieving complex objectives simultaneously. AI automatically explores the perfect balance point among conflicting objectives such as performance, power, and linearity, significantly liberating designers from the burden of endless repetitive tuning. Even in complex design spaces where hundreds of variables are nonlinearly entangled, InfiniOpt presents solutions approaching the global optimum.

InfiniSim is a next-generation GPU-accelerated circuit simulator for large-scale analog and power circuits, based on Berkeley SPICE3f5 with full CUDA 12.0+ support, delivering up to 71x faster runtime compared to traditional CPU simulators (in inverter chains and power networks). Verified against CPU simulators across both linear and nonlinear device regions, it integrates a learned Neural Simulator achieving high prediction accuracy (R²≥0.8)—simultaneously enhancing both reliability and speed in design verification. Integration with the InfiniTree Design Suite (InfiniSigma·InfiniOpt) automates workflows from yield analysis to parameter optimization, dramatically shortening complex design cycles and reducing development costs, establishing a new industrial simulation standard.