EE 292P: Atoms, Bits, and National Interest (Jan 6)

Attendees: Dustin J Ross Date: January 6, 2026 Type: Class Session

Summary

Course Overview: EE 292P - Atoms, Bits, and National Interest

  • 13-week course combining semiconductor technology with policy implications
  • Teaching team: Professor Ali, Professor Shravanti Chowdhury (EE), Tom Lee (circuit design), Joe Mata (VC/computer scientist)
  • Guest lecturer: Steve Blank (entrepreneurship, national security policy)
  • Format: lectures, invited talks, panel discussions (110-minute sessions)
  • Structure: Start with semiconductors/materials → build up to computing → applications → economy/society/law → globalization/competition

The Perfect Storm in Semiconductors

  • Moore’s Law significantly slowed - only handful of companies (TSMC, Intel, Samsung) can innovate at leading edge
  • Dennard scaling ended - voltage scaling no longer viable
  • Von Neumann architecture limitations - memory/processor separation creates energy-intensive data movement
  • Most chips now power-limited rather than technology-limited
  • Solution focus: Efficiency metrics (performance per watt, operations per watt)

AI Compute Explosion and Sustainability Crisis

  • Training compute doubling every 3-4 months (vs Moore’s Law 2-year cycle)
  • GPT-4 training: ~50,000 megawatt hours
  • Claude 4: Even higher compute requirements
  • Exponential compute growth could surpass US GDP if trends continue
  • Economic unsustainability: Training costs reaching $100M+ per model

Innovation and Economic Impact

  • 92% of US GDP growth (first half 2025) came from information processing and software
  • Debate: Economic growth measurement challenges in digital age
  • Company lifespan in S&P 500: 35+ years (1970s) → 10-12 years (now)
  • 40%+ of public companies are venture-funded
  • Hardware/software investment ratio shifted dramatically since 2012

Semiconductor Industry Structure (Steve Blank)

  • Semiconductor manufacturing most complex in world - atomic-level engineering
  • Leading-edge fab cost: $20-30B for 3-5nm, potentially $50B for 2nm
  • Single wafer processing: 2,500+ steps, 2-3,000 gallons water per wafer
  • 50-year evolution: 3M transistors (4-inch wafer) → 15-30 trillion transistors (12-inch wafer)
  • 5 million times increase in transistor density over 50 years

Supply Chain Dependencies and Vulnerabilities

  • Five companies globally make chip manufacturing equipment
    • ASML (Netherlands): photolithography
    • KLA, Lam Research, Applied Materials (US): wafer processing
    • Tokyo Electron (Japan): wafer processing
  • Materials supply chain highly specialized by country
  • Ukraine: neon supply; Israel: RF components; various countries control critical materials
  • TSMC: ~2/3 of most advanced chips globally

China-Taiwan Geopolitical Tensions

  • “Semiconductors are the oil of the 21st century” - Steve Blank
  • China’s largest import: semiconductors (more than oil)
  • US 6th largest export: semiconductors
  • Taiwan invasion timeline: China aims for “reunification” by 2027 (Communist Party centennial)
  • US export controls (October 2022): Blocked semiconductor equipment exports to China
  • China building parallel semiconductor ecosystem - currently 1-2 generations behind

US Policy Response: CHIPS Act

  • $53B in incentives for domestic fab construction
  • Not full funding - matching incentives to offset cost differential vs Taiwan
  • TSMC Arizona fabs: ~2% of TSMC total capacity
  • Workforce challenges: Skills gap, cultural differences (24/7 fab operations vs work-life balance)
  • Missing elements: Insufficient workforce development, compensation incentives

Chinese Semiconductor Strategy

  • “Big Fund” national investment program (many executives arrested for fraud by 2022)
  • Shanghai Stock Exchange created for semiconductor startups
  • More fabs under construction in China than anywhere else
  • Developing alternative technologies (synchrotron-based lithography vs EUV)
  • National imperative with “gun to their head” - will likely succeed eventually

Technology Innovation Ecosystem

  • Fabless model dominance: Apple, Nvidia, AMD design chips, TSMC manufactures
  • IP core industry: ARM supplies 40% of chip cores globally
  • EDA software duopoly: Synopsys and Cadence control chip design tools
  • Packaging evolution: Chiplets enabling performance gains as Moore’s Law slows
  • Critical chokepoints throughout supply chain vulnerable to disruption