AI summary by TwinMind
The Next Big Thing in Computing Might Be Lasers
Sabine Hossenfelder
7m 3s
Sponsor Introduction - The Economist
- Video sponsored by The Economist
- Weekly news magazine covering global affairs, economics, science, technology
- Offers audio edition for listening on-the-go
- 20% discount available via economist.com/sabina
- Featured story: AI helping diplomats in peace negotiations
Computing Power and Moore's Law Challenges
- Computing progress driven by Moore's Law for half century
- Moore's Law: packing increasing transistors in microchips
- Declared dead multiple times due to cooling limitations
- Smaller transistors + closer packing = harder to cool
- Overheating = biggest hurdle to further miniaturization
Current Cooling Methods and Limitations
- Traditional: air blown over microchips
- Newer designs: small fluid channels around transistors
- Make chips larger again
- Limited miniaturization potential
- Berkeley Lab study: up to 40% of data center energy goes to cooling
- Supercomputing clusters now liquid cooled
- Liquid cooling has heat capacity limits
Laser Cooling Background
- Nobel Prize awarded 1997 for laser cooling
- Works by bouncing light off particles to carry away energy
- Tunes light to atomic/molecular transitions
- Light absorbed, re-emitted with higher energy
- Reduces average atomic energy = lower temperature
- Applications: quantum effects study, atomic clocks
- Cannot directly apply to microchips:
- Requires isolated atoms in vacuum
- Needs specific atomic transitions unavailable in semiconductors
Resonant Damping for Semiconductors
- Variant of laser cooling for semiconductors
- Analogy: slowing swing by accelerating opposite direction
- Direct laser light at hot spots in material
- Tune frequency to encourage more photon emission than absorption
- Heat carried away as light = cooling effect
Current Development and Performance Claims
- Sandia Labs + Maxwell Labs collaboration
- Prototype development targeting:
- 200% performance gains on latest Nvidia/AMD hardware
- 40x efficiency improvement vs liquid cooling (same net power)
- Selective targeting of localized hotspots
- Hotspots typically few hundred microns across
- No energy waste on areas not needing cooling
Research Timeline and Demonstrations
- Concept existed ~decade
- 2012: Neosbore Institute cooled gallium arsenide to 4 Kelvin
- 2020: Researchers demonstrated miniaturized systems integrated with microchips
- Estimated rollout: within decade
