The semiconductor industry is undergoing a profound transformation, driven by the strategic integration of artificial intelligence. At SEMICON Taiwan 2025, experts from Microsoft, Advantech, and Nvidia shared insights on the sector’s rapid shift toward “Lights-Out” factories—fully automated, 24/7 operations powered by AI. This evolution is designed to tackle monumental challenges while unlocking new levels of efficiency and resilience.
AI Models Evolve for Deeper Insights
AI development has accelerated dramatically in recent years. From early domain-specific models for tasks such as entity recognition or computer vision, AI has advanced into sophisticated multi-domain models that integrate diverse functionalities within a single framework. Notable advancements include Retrieval-Augmented Generation (RAG), which constrains responses for greater accuracy, and multimodal models that seamlessly combine text, audio, and visual inputs.
A pivotal breakthrough has been the emergence of reasoning models, capable of delivering deeper insights for complex problems across scientific, mathematical, and manufacturing domains. These models provide richer and more comprehensive analyses compared with their predecessors. Beyond individual models, multi-agent systems are now automating entire workflows, enabling specialized AI agents to collaborate on intricate challenges. One example is Toyota’s “obeya agent,” which integrates multiple sub-agents to assist powertrain engineers in analyzing vast datasets.
According to Saj Kumar K, Senior Director of Manufacturing (APAC) at Microsoft, these advanced AI models are already being applied to manufacturing. The Production Copilot functions as a factory agent, analyzing data to identify root causes of downtime and optimize maintenance schedules. The Quality Advisor leverages multimodal AI for visual inspection and defect analysis, potentially rendering traditional image labeling obsolete. AI is also being used to optimize fab scheduling and Automated Material Handling Systems (AMHS), improving overall cycle times. Furthermore, large language models (LLMs) are training humanoid robots to perform complex tasks through single-shot teaching.
Advantech’s Edge AI Strategy Targets Speed and Precision
Semiconductor manufacturing faces unique pressures, particularly from the massive scale of data generated. A single wafer inspection can produce up to seven petabytes of data, requiring enormous computational power and real-time processing. The transition from single-beam to multi-beam inspection demands a tenfold increase in computational speed today, with projections of a hundredfold increase by 2030. Combined with the need for instantaneous data transfer and ultra-low latency to prevent defects, these requirements are driving a shift away from traditional data centers toward powerful edge computing solutions. Meanwhile, global labor shortages and skill gaps are accelerating the need for advanced automation.
Advantech is at the forefront of this transformation with a comprehensive Edge AI strategy focused on process optimization, inspection, and fab automation. "Inspection is the heart of semiconductor manufacturers," said Magic Pao, Vice President at Advantech. The company’s strategy is built on four pillars: high computing power, high throughput, low latency, and robust local storage for analytics.
Advantech is spearheading this transformation with a comprehensive Edge AI strategy focused on process optimization, inspection, and fab automation. “Inspection is the heart of semiconductor manufacturing,” said Magic Pao, Vice President at Advantech. The company’s approach is built on four pillars: high computing power, high throughput, low latency, and robust local storage for analytics.
To achieve this, Advantech embeds AI into every element of its solutions, including products such as SkyRack and the MIC-7000/7500 industrial servers with direct-liquid cooling, both carrying critical SEMI certifications for fab deployment. High-speed connectivity is ensured through Time-Sensitive Networking (TSN) and industrial-grade switches. AI-powered vision systems—incorporating smart cameras with Nvidia GPU acceleration and FPGA-based technology—are central to the precision required for inspection. Advantech also emphasizes ecosystem strength, partnering with Nvidia for advanced computing and Intel for reliable system integration.
Nvidia’s Physical AI Drives Industrial Autonomy
Nvidia is at the forefront of a new industrial revolution, evolving from a chipmaker into an “infrastructure company” advancing Physical AI. This strategy seeks to create “AI factories” that power advanced applications. Andrew Liu, Senior Manager at Nvidia, explained that this vision rests on two pillars: advanced foundation models for robotics and sophisticated simulation environments.
Nvidia is developing powerful Vision-Language-Action (VLA) models that enable robots to interpret high-level prompts and images, translating them into precise physical actions and learning from observation or single-shot teaching. To address the scarcity of real-world robotics data, Nvidia relies heavily on its Omniverse platform for digital twins and simulation. Omniverse provides physically accurate, photorealistic virtual environments where synthetic data can be generated and robots can be trained and validated. This approach helps alleviate labor shortages and enables rapid “factory cloning” to mitigate geographic supply chain risks. Nvidia’s integrated platform spans systems for training foundation models, servers for Omniverse digital twins, and embedded platforms for industrial robot deployment.
Toward the Lights-Out Future
The convergence of advanced AI models, strategic infrastructure, and targeted edge solutions is accelerating the path toward fully automated “Lights-Out” semiconductor manufacturing, promising significant gains in productivity, efficiency, and resilience that will redefine the industry’s future. At SEMICON Taiwan 2025, the AI Technology Zone (advised by the International Trade Administration) underscores this transformation by showcasing the full AI ecosystem—from chip manufacturing and IC design to dedicated hardware.
SEMICON Taiwan 2025 Smart Manufacturing Forum. SEMI
Article edited by Joseph Tsai
