Manufacturing at advanced nodes can no longer be achieved through human labor and experience alone.

Reasons:

・Number of process steps: thousands

・Equipment parameters: tens of thousands to hundreds of thousands

・Yield tolerance: extremely narrow (nm level)

・Variability in EUV exposure (stochastic effects)

・Strong interdependence between materials and equipment

・Production volumes increasing explosively due to demand for AI chips

For this reason, the world’s foundries (TSMC / Intel / Samsung) have entered an era of optimizing the factory itself with AI.

The smart fab centers on the following four technologies.

(1) Process Optimization AI

AI analyzes huge amounts of process data and automatically proposes:

・Optimal conditions

・Abnormal patterns

・Faster convergence

→ Shorter development periods (faster migration between nodes)

→ Lower ramp-up costs

→ Greater ability to absorb variation

(2) Fault Detection (FDC × AI)

AI is integrated into conventional FDC (Fault Detection & Classification).

With AI, abnormalities that humans or statistics cannot notice are detected, such as:

・Tiny changes in vibration

・Fluctuations in plasma light

・Deviations in motor current

・Subtle shifts in temperature waveforms

・Fluctuations in the EUV light source

This prevents yield loss before it happens.

(3) Digital Twin

A virtual copy of the entire factory is created, simulating everything from processes and equipment to transport and layout.

Examples:

・Optimizing the order in which lots are fed into equipment

・Maximizing throughput

・Checking for interference between equipment in advance

・Virtually running process recipes

TSMC and imec are leading the way, and it will become an essential technology in every advanced factory.

(4) Full Automation

・Unmanned transport (OHT/AGV)

・Automatic recipe changes

・Lot tracking

・Robotic wafer handling

The ultimate form is a factory where people no longer physically touch anything.

1.Equipment data acquisition (sensor data)

Temperature, pressure, RF power, flow rate, optical signals, etc.

2.Real-time analysis (AI × FDC)

Detects signs of abnormalities on a scale of seconds to milliseconds

3.Process optimization (Process Optimization AI)

Fine-tunes recipes to maximize the yield of good products

4.Digital twin verification

Verifies in advance how changes will affect yield

5.Fab scheduling optimization

Maximizes equipment utilization and transport efficiency

6.Feedback control

Applies the results to the next lot

→ An era in which humans are responsible only for supervision, not for judgment.

(1) EUV Variability (Stochastic Effects)

Because EUV produces stochastic defects, large-scale data analysis by AI is essential.

(2) Lower Yield in Multilayer Wiring

Wiring layers are a collection of different materials and different processes. AI analyzes correlations and optimizes them.

(3) Differences Between Equipment (Chamber Matching)

AI corrects differences in behavior among multiple tools to make them uniform.

(4) Shortage of Skilled Engineers

AI learns know-how, making it reproducible and transferable.

(5) Longer Development Periods

AI speeds up recipe exploration during node transitions.

TSMC

Has built its own “AI Manufacturing Platform.” EUV management, chamber matching, and yield management are driven by AI.

Intel

Under the banner of “Siliconomics,” it is investing heavily in AI-based fab optimization. Process optimization combined with PowerVia (backside PDN) in particular is accelerating.

Samsung

Strengthening automation through large-scale data analysis and equipment optimization.

imec

A leading-edge research institute for digital twins.

The smart fab has already become competitiveness itself.

Challenges

・Data formats have not yet been standardized

・Differences in specifications between equipment makers (black boxes)

・Enormous data volumes (several PB to EB)

・The problem of AI model drift

・Security and confidentiality

Future Outlook

・Self-evolving AI models

・Fully unmanned operation of entire factories

・Process development completed through simulation alone

・Autonomous operation of volume production lines

A semiconductor factory is neither an automobile factory nor an aircraft factory. The era of running the most complex factories in the world with AI has arrived.

・The smart fab is a foundational technology supporting advanced nodes

・Its four pillars are process optimization AI, fault detection AI, digital twins, and automation

・It addresses today’s challenges such as EUV, multilayer wiring, and chiplets

・All of the world’s top foundries are promoting AI-driven factories

・Semiconductor factories are evolving into factories that run on data

1.Explain the background behind the need for smart fabs.

2.In what ways is FDC × AI superior to conventional FDC?

3.List the benefits that digital twins bring to factory operations.

 

Column Supervisor: Koji Kakumoto (President, OTIS Co., Ltd.)

After studying abroad and working in planning and development at a trading company, he joined OTIS Co., Ltd. in 2011. While primarily working in the Corporate Planning Department, he has also served concurrently in manufacturing and technical divisions, and since 2018 has served as President, working to drive business growth and strengthen the organization.

 

This article is a general technical explanation intended for educational purposes and does not refer to any specific company, product, or technology.