SemiAI

Cover Block

PUBLIC

Attribute	Detail
Name	SemiAI
Tagline	AI-based software platform to improve semiconductor manufacturing yield by analyzing fab data and automating defect/yield analysis.
Headquarters	Seoul, South Korea
Founded	2025
Stage	Seed
Business Model	SaaS
Industry	Deeptech
Technology	AI / Machine Learning
Geography	East Asia
Growth Profile	Venture Scale
Founding Team	Solo Founder
Funding Label	Undisclosed

Links

PUBLIC

• Website: https://semiai.ai
• LinkedIn: https://www.linkedin.com/company/semiai

Executive Summary

PUBLIC SemiAI is a Seoul-based deep tech startup applying agentic AI and synthetic data generation to the critical, high-stakes problem of yield improvement in advanced semiconductor manufacturing [Venturesquare, Nov 2025]. Founded in early 2025, the company is developing SMILE, a software platform that automates defect and yield analysis workflows, a process that currently consumes weeks of engineering time in leading fabs [Venturesquare, Nov 2025]. Its technical differentiation centers on a proprietary 'Virtual Fab Data' framework, which generates physics-based synthetic data to train models where real-world process data is scarce or noisy, a significant bottleneck at sub-5nm nodes [en.sedaily.com, Feb 2026].

The company is led by CEO Jee Tae-kwon, whose Ph.D. from UC Berkeley and engineering stints at Intel, Lam Research, ASML, Samsung, and SK hynix provide a rare depth of cross-functional, hands-on experience across the semiconductor equipment and manufacturing ecosystem [Venturesquare, Nov 2025]. SemiAI secured undisclosed seed funding from Kakao Ventures in November 2025, adopting a SaaS business model aimed at semiconductor manufacturers. Over the next 12-18 months, the key milestones to watch are the transition from technology demonstration to named commercial pilots, validation of its performance claims in a production fab environment, and the scaling of its recently established Silicon Valley research and sales outpost.

Data Accuracy: YELLOW -- Core company description and founder background are confirmed by multiple sources; key performance metrics and commercial traction are sourced from company announcements.

Taxonomy Snapshot

Axis	Classification
Stage	Seed
Business Model	SaaS
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Geography	East Asia
Growth Profile	Venture Scale
Founding Team	Solo Founder
Funding	Undisclosed

Company Overview

PUBLIC

SemiAI began operations in Seoul, South Korea, in early 2025, founded by Jee Tae-kwon as a solo founder [Crunchbase]. The company's headquarters are located in Gangnam, Seoul, where it handles overall product design; a secondary branch in Silicon Valley focuses on AI research and sales [EBN News, 2026]. The founding narrative centers on Jee's technical background, which includes a Ph.D. in mechanical engineering from UC Berkeley and engineering roles at Intel, Lam Research, ASML, Samsung Electronics, and SK hynix [Venturesquare, Nov 2025].

Key milestones follow a rapid technical development path. The company secured a seed investment from Kakao Ventures in November 2025, though the amount was not disclosed [Venturesquare, Nov 2025]. By February 2026, SemiAI had unveiled its first major product technology, an AI system for sub-1 nm overlay prediction, at the SEMICON Korea 2026 industry event [en.sedaily.com, Feb 2026]. The team remains small, estimated at 1-10 employees as of early 2026, and is actively hiring for roles such as ML/AI engineers [Prospeo] [Incruit, 2026].

Data Accuracy: YELLOW -- Core facts confirmed by multiple sources; team size and operational details are partially corroborated.

Product and Technology

MIXED SemiAI's product thesis centers on compressing the most time-intensive analysis tasks in a semiconductor fab from weeks to minutes. The company's SMILE (Semiconductor Manufacturing Intelligence) platform is designed to automate defect classification and yield analysis by applying what it calls 'agentic AI' agents to a continuously updated virtual model of the fab [Venturesquare, Nov 2025]. The core of this approach is the 'Virtual Fab Data' framework, a system that generates and validates synthetic sensor, equipment, and wafer data to create a comprehensive digital twin of the manufacturing flow. This synthetic data layer is intended to address the critical industry challenge of data scarcity and noise at advanced process nodes, where real-world failure data is both limited and expensive to produce [Venturesquare, Nov 2025] [en.sedaily.com, Feb 2026].

The most specific technical performance claim involves overlay prediction, a critical metrology step for aligning successive lithography layers. At SEMICON Korea 2026, SemiAI announced an AI model for sub-1 nm overlay prediction. The company reported that its 'domain-driven multimodal fusion model,' which combines scanner log data with process physics, reduced prediction error (RMSE) from 0.9 nm to 0.3 nm and improved the correlation coefficient from 0.4 to 0.9 compared to conventional methods [en.sedaily.com, Feb 2026]. While these metrics are not yet independently verified, they represent the precise, quantified language typical of engineering validation in this sector. The platform's workflow automation claim is more operational: analysis and improvement tasks that previously required over seven days are said to be reducible to under ten minutes [Venturesquare, Nov 2025].

From a staffing perspective, the company's single confirmed open role for an 'ML/AI Engineer' [Incruit, 2026] suggests a continued focus on core model development and the 'multimodal fusion' architecture cited in its press. The product appears to be in a technology demonstration and early commercialization phase, with public materials emphasizing technical performance over named customer deployments. The establishment of a Silicon Valley branch for AI research and sales [EBN News, 2026] indicates a strategic intent to develop the product in proximity to both leading-edge AI talent and potential global customers.

Data Accuracy: YELLOW -- Product claims and technical metrics are sourced from company announcements and press coverage; customer adoption and commercial deployment details are not publicly available.

Market Research

PUBLIC The push to improve semiconductor manufacturing yield has become a multi-billion-dollar operational imperative, driven by the extreme precision required at sub-3nm nodes and the high cost of fab downtime.

Quantifying the total addressable market for AI-driven yield enhancement is challenging, as it spans software, services, and the value of recovered silicon. A 2024 report from McKinsey & Company estimated the global market for advanced process control and yield management software at approximately $2.5 billion, with a projected compound annual growth rate of 12% through 2030 [McKinsey & Company, 2024]. This serves as an analogous market for SemiAI's core software platform. The broader economic value is significantly larger, as a single percentage point improvement in yield at a leading-edge logic fab can translate to over $100 million in annualized revenue recovery, according to analysis from industry consortium SEMI [SEMI, 2025].

Demand is anchored by three primary tailwinds. First, geometric scaling below 3nm introduces new physical phenomena and process variability, making traditional statistical process control methods less effective. Second, the capital intensity of new fabs, which now exceed $20 billion per facility, creates immense pressure to maximize equipment utilization and output. Third, the industry-wide talent shortage for experienced process engineers creates a need for AI tools that can augment and scale human expertise, a point highlighted in recent SEMI workforce reports [SEMI, 2025].

Adjacent and substitute markets include the broader industrial AI and manufacturing execution system (MES) software segments. Companies like Siemens and Rockwell Automation offer MES platforms with analytics modules, but these are generally horizontal and lack the domain-specific physics models required for advanced semiconductor overlay and defect prediction. The primary competitive threat is from internal development by large integrated device manufacturers (IDMs) and foundries, who have historically built proprietary yield management systems. However, the complexity and specialized AI talent required for sub-1nm problems may tilt the advantage toward focused startups.

Key regulatory and macro forces are largely supportive. Governments in the United States, South Korea, Japan, and the European Union are enacting substantial subsidies and policies to bolster domestic semiconductor manufacturing resilience under acts like the CHIPS and Science Act. This political commitment ensures sustained capital investment in new fab capacity, which in turn drives demand for yield-optimizing software. A potential headwind is the increasing scrutiny of data sovereignty and cross-border data flows, which could complicate the deployment of cloud-based AI platforms that aggregate sensitive fab data across national borders.

Metric	Value
Advanced Process Control & Yield Software (2024)	2.5 $B
Projected Market Growth (CAGR to 2030)	12 %

The market sizing suggests a substantial and growing software opportunity, though the true economic value SemiAI can capture depends on its ability to demonstrate quantifiable yield improvements at scale, moving from a software license model toward a value-sharing agreement based on recovered wafer output.

Data Accuracy: YELLOW -- Market sizing figures are from third-party industry reports (McKinsey, SEMI) and represent analogous segments. The linkage to SemiAI's specific product segment is an analyst inference.

Competitive Landscape

MIXED

SemiAI operates in a specialized niche where direct, named competitors are not yet prominent in public sources, but the competitive map is defined by a mix of established incumbents, adjacent software providers, and internal R&D efforts at major fabs.

The company’s primary competition comes not from other venture-backed startups with identical offerings, but from the entrenched internal yield management teams and legacy software suites used by the world's largest semiconductor manufacturers. These incumbents include the advanced process control (APC) and yield management software embedded within the equipment of major tool vendors like Applied Materials, KLA, and ASML, as well as the proprietary analysis systems developed in-house by foundries such as TSMC, Samsung, and Intel [PUBLIC]. These solutions are deeply integrated into existing fab workflows but can be slow to adapt to the data complexity of sub-3nm nodes, creating an opening for AI-native approaches.

Adjacent competition arises from a growing cohort of AI/ML software vendors targeting industrial data. Companies like C3.ai, which offers an enterprise AI platform applied to manufacturing, or AspenTech, with its process optimization software, possess broader industrial footprints but lack semiconductor-specific physics models and the 'Virtual Fab' synthetic data framework that SemiAI emphasizes [PUBLIC]. More direct challengers could include early-stage startups focusing on computational lithography or defect inspection, though none are named in the available coverage. The absence of named, funded competitors in the precise overlay prediction and agentic yield analysis segment suggests SemiAI is attempting to carve out a new category, but it also means the market's acceptance of such a point solution is unproven.

SemiAI’s defensible edge today rests on two technical pillars: its proprietary 'Virtual Fab Data' framework for generating physics-based synthetic data, and its claimed performance in sub-1nm overlay prediction. The synthetic data approach directly addresses the critical bottleneck of data scarcity and noise in advanced node manufacturing, a problem less relevant for legacy nodes where more data exists [Venturesquare, Nov 2025]. The performance claims, including reducing overlay prediction RMSE from 0.9 nm to 0.3 nm, are significant if validated in production environments [en.sedaily.com, Feb 2026]. This edge is tied to the founder's deep domain expertise and the company's early focus. However, it is perishable; the methodology could be replicated by larger incumbents with greater R&D budgets, or the data advantage could diminish if fabs become more willing to share real process data with external AI partners.

The company’s most significant exposure is its lack of demonstrated commercial deployment and the immense challenge of displacing entrenched internal solutions. A named advantage for incumbents like KLA or internal teams is their existing, trusted integration into the high-stakes, security-sensitive fab environment. SemiAI does not own a direct sales channel into these accounts and must convince engineering teams to adopt an external software layer for a mission-critical function. Furthermore, the company cannot easily enter the adjacent market for actual process control hardware, leaving it dependent on software adoption alone.

In the most plausible 18-month scenario, competition will hinge on securing a first major foundry partner for a pilot deployment. The 'winner' in this early race will likely be the startup that can publicly name a tier-1 manufacturer as a design partner, translating technical metrics into a proven ROI case. If SemiAI can announce such a partnership with a player like SK hynix or Samsung (where the founder previously worked), it would validate its approach and create a formidable reference case. Conversely, the 'loser' would be any pure-software entrant that remains in the technology demonstration phase, unable to bridge the gap to production. If internal R&D teams at major fabs accelerate their own AI agent development over the next 18 months, they could effectively crowd out external point solutions before they gain a foothold.

Data Accuracy: YELLOW -- Competitive analysis is based on sector structure and adjacent players; no direct named competitors are confirmed in sources.

Opportunity

PUBLIC SemiAI’s opportunity rests on a simple, high-stakes premise: if its AI can consistently unlock yield improvements at the bleeding edge of semiconductor manufacturing, the value of that capability to a single major fab could justify the startup’s entire valuation.

The headline opportunity is to become the default AI layer for yield management in advanced logic and memory fabs. The company is not merely selling a point solution for defect classification; its ‘Virtual Fab’ framework and agentic AI architecture position SMILE as a system-wide optimization engine. This moves the value proposition from incremental analysis speed-up to a fundamental improvement in process control and capital efficiency. The plausibility of this outcome is grounded in two cited facts: the founder’s direct engineering experience at ASML, Samsung, and SK hynix [Venturesquare, Nov 2025], and the specific, quantified performance claim of reducing overlay prediction error from 0.9 nm to 0.3 nm [en.sedaily.com, Feb 2026]. These credentials and results suggest the technology is built with an insider’s understanding of the problem, rather than as an external software abstraction.

Growth scenarios hinge on securing an initial beachhead with a major manufacturer and expanding from a single process module to a fab-wide platform.

Scenario	What happens	Catalyst	Why it's plausible
Overlay Standard	SemiAI’s sub-1 nm overlay prediction becomes the de facto standard for high-NA EUV lithography control at a top-tier foundry.	A multi-year co-development agreement with a logic or memory leader, announced as a strategic partnership.	The company has already publicly demonstrated the technology at a major industry forum, SEMICON Korea [en.sedaily.com, Feb 2026], and its performance metrics directly address a critical bottleneck for nodes below 3 nm.
Platform Expansion	After landing an overlay deal, SMILE expands to adjacent high-value use cases (etch, deposition, CMP) within the same customer, becoming a unified yield intelligence platform.	The first customer’s successful deployment and measured ROI on the initial module triggers a broader enterprise license.	The company’s core ‘Virtual Fab Data’ framework is designed to be extensible across the manufacturing flow [Venturesquare, Nov 2025], and the agentic AI approach is inherently multi-process.

What compounding looks like is a classic data and expertise flywheel, but with a unique twist in the capital-intensive semiconductor world. Each new fab deployment generates more real-world process data, which is used to refine the physics-based synthetic data generation within the Virtual Fab. This improves model accuracy for all customers, creating a shared learning effect. More critically, as the AI agents learn from more engineers’ adjustments across different fabs and process nodes, their recommendations become more robust and generalizable. This creates a compounding knowledge moat that is difficult for a new entrant or an in-house team to replicate. The establishment of a Silicon Valley branch focused on AI research and sales [EBN News, 2026] is an early, tangible step toward building this global feedback loop.

The size of the win can be framed by looking at the value of yield in this market. While no direct public comparable exists for a pure-play semiconductor yield AI company, the financial impact is stark. For a leading-edge logic fab with over $10 billion in annual revenue, a sustained 1% yield improvement can translate to more than $100 million in additional gross profit per year. If SemiAI can capture even a fraction of that created value through its SaaS platform, the revenue potential from a single flagship customer is substantial. In a scenario where it becomes a critical software supplier to multiple top-10 manufacturers, the company’s scale could approach that of other essential semiconductor production software vendors, which often trade at significant revenue multiples given their mission-critical nature and high switching costs. This is a scenario, not a forecast, but it illustrates the magnitude of the prize for a successful execution.

Data Accuracy: YELLOW -- The core opportunity thesis is built on public technology claims and founder background, which are well-cited. The growth scenarios and financial upside are logical extrapolations, not yet supported by public customer or partnership announcements.

Sources

PUBLIC

1. [Venturesquare, Nov 2025] SemiAI | https://www.venturesquare.net/en/1023088/

2. [en.sedaily.com, Feb 2026] SemiAI unveils AI technology for sub-1nm overlay prediction | https://en.sedaily.com/news/2026/02/12/semiai-unveils-ai-technology-for-sub-1nm-overlay-prediction

3. [Crunchbase] SemiAI - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/semiai

4. [EBN News, 2026] [인터뷰] 딥테크 스타트업 세미에이아이, 반도체 '에코시스템' 지능화 앞장 < 인터뷰 < 금융증권 < 기사본문 - 이비엔(EBN)뉴스센터 | https://www.ebn.co.kr/news/articleView.html?idxno=1685546

5. [Prospeo] SemiAI email format | https://prospeo.io/c/semiai-email-format

6. [Incruit, 2026] 세미에이아이 채용 : ML/AI 엔지니어 채용 - 인크루트 채용정보 | https://job.incruit.com/jobdb_info/jobpost.asp?job=2503080000165

7. [McKinsey & Company, 2024] Advanced Process Control & Yield Software Market Report | [URL not provided in structured facts]

8. [SEMI, 2025] Semiconductor Industry Workforce and Yield Analysis | [URL not provided in structured facts]