SemiAI

SemiAI is a South Korean semiconductor AI company that builds fab-native software for chip manufacturers. Its product line, SMILE (Semiconductor Manufacturing Intelligence), is positioned as an agentic AI platform that forecasts process drift, predicts yield issues, accelerates root-cause analysis and supports process-control decisions. The platform is offered in tiers: SMILE Lite for real-time detection and multimodal analysis of process data, SMILE Pro for predictive redesign of advanced process control (APC) beyond reactive control, and SMILE Pro+ for domain-specific modeling that compresses wafer position and defect data into a single vector for instant similar-case retrieval.

A central element of the technology is Virtual Fab, a synthetic data generation and digital-twin approach that simulates sensor, equipment and wafer data across manufacturing stages. Models are pre-trained on these synthetic patterns so they can be used without customer data at the outset, with synthetic and real fab data cross-validating one another. The company states its agents operate under human-in-the-loop supervision and act on recommendations rather than stopping at observation, and that its approach shortens yield analysis and improvement cycles from more than seven days to under ten minutes.

The company reports overlay prediction accuracy of 0.3 nm, R² above 0.9 and a 3x RMSE improvement over XGBoost, and cites earlier core-model validation of 99% overlay defect prediction accuracy and 95% MVP RCA accuracy. It has published research accepted at SPIE Advanced Lithography 2025 and 2026 on proactive yield maximization in photolithography via human-in-the-loop AI, systematic optimization of overlay key positions, and domain knowledge-driven fusion machine learning for overlay prediction, and holds patents covering EPE budget component analysis and agent-based root-cause analysis in manufacturing in the US and Korea.

Founding story

SemiAI was founded in January 2025 by Taekwon Jee, who holds a Ph.D. in mechanical engineering from UC Berkeley and has more than 15 years of semiconductor industry experience at SK hynix, ASML, Samsung and Lam Research, including development of an Edge Placement Error improvement methodology for yield analysis and ML/AI-based yield improvement systems. He started the company with engineers who had worked on fab floors.

Business model

SemiAI licenses software to semiconductor manufacturers and fabs, entering customers through proof-of-concept engagements. It signed its first paid PoC contract with a semiconductor customer in early 2026 with two further PoCs running in parallel, and states an ambition to become an AI infrastructure company built on the data generated across semiconductor production.

Revenue is generated from paid proof-of-concept contracts with semiconductor customers, with the first such contract signed in March 2026.

Traction

SemiAI reports core model validation in April 2025 (99% overlay defect prediction, 95% MVP RCA), shipped SMILE Lite v1.0 and Pro v0.5 in December 2025 with operation in customer environments, and signed its first paid PoC in March 2026 with two more running in parallel. It has three papers accepted at SPIE Advanced Lithography 2026, a patent portfolio in process, metrology and RCA, and backing from ENSL Partners, Kakao Ventures, Newborn Ventures, Schmidt and the TIPS Deep Tech track.

Latest developments

In 2026 the company closed a Seed II round with Newborn Ventures, signed its first paid PoC, was named a Lam Capital Venture Competition Finalist 10, and raised a bridge round from Schmidt in July 2026 to expand US proof-of-concept work and hire locally. It presented three papers at SPIE Advanced Lithography 2026 and presented its agentic AI manufacturing technology at SEMICON Taiwan 2026.