Can AI Create Missing Models?

Entities

Companies:TSMC NVIDIA Synopsys Keysight EDA Siemens EDA ChipAgents Normal Computing

People:Jensen Huang Arvind Srinivasan Ramesh Narayanaswamy Tom Demuer Sathishkumar Balasubramanian Abhi Kolpekwar Thomas Ahle Mehir Arora Alexander Petr

Technologies:3nm EUV AI Behavioral models Analog mixed-signal Machine learning Reinforcement learning Simulation acceleration Model ord

Tags

AI in semiconductor EDA flow Model creation AI modeling Chip design automation Simulation acceleration Behavioral models Analog mixed-signal Machine learning Chip verification Model cost reduction Semiconductor toolchain

News Summary

As artificial intelligence continues to permeate the semiconductor industry, the efficiency and cost of model creation have become central concerns. In traditional electronic design automation (EDA) f... Read original →

Industry Analysis

AI-driven model generation is evolving from an auxiliary feature to the core engine of EDA flows, especially below the 3nm node where traditional behavioral modeling struggles to balance accuracy and speed. Foundries like TSMC will likely mandate AI-embedded validation loops in PDK deliveries from EDA vendors such as Synopsys and Siemens EDA—shifting yield risk away from themselves. From a compliance standpoint, AI-generated models lacking physically traceable foundations could fail automotive or medical certification, inflating quality assurance costs. NVIDIA leverages its CUDA ecosystem alongside startups like ChipAgents to bypass legacy simulation bottlenecks, while Keysight bets on reinforcement learning for RF modeling differentiation. Within 18 months, a 'trusted AI modeling' certification framework will emerge; purely data-driven approaches without domain knowledge integration will be sidelined. The automation race is shifting from tool efficiency to control over knowledge-AI hybrid architectures.