Second Green AI Summit at Harvard and Boston University Successfully Convened
Session Date: April 25, 2025 Location: Harvard University Gutman Conference Center
Overview:
Shifting focus to the fundamental building blocks of AI, Panel 3 explored the future of computing hardware beyond traditional silicon electronics. Moderated by Professor Gage Hills from Harvard University, this session brought together experts in electrical engineering, photonics, and quantum computing to critically examine emerging technologies, their integration possibilities, innovation challenges, and potential benefits in performance, energy efficiency, and environmental sustainability.
Key Discussion Points:
The panel compared and contrasted different future hardware paradigms:
Advanced Electronics: Discussions acknowledged the slowdown of traditional scaling (Moore's Law/Dennard scaling) and the resulting diminishing returns in energy efficiency for transistors alone. A key challenge highlighted was the communication bottleneck – moving data between chips and memory is often slower and more energy-intensive than computation itself. Innovations like 3D integration (stacking chips like High Bandwidth Memory or HBM) aim to mitigate this by physically bringing components closer together, though this adds manufacturing complexity and potential embodied carbon impacts.
Optical/Photonic Computing: Panelists discussed the growing potential of photonics. Initially explored decades ago but limited by practicality, advancements in photonic integrated circuits have renewed interest. Photonics is seen as particularly strong for specific operations common in AI, like matrix multiplications (GEMM operations). The vision presented was often electro-photonic computing, where photonics handles specific tasks (computation, high-bandwidth communication via wafer-scale networks, potentially even manipulating data stored in phase-change materials) while working alongside traditional electronics. Crucially, photonic chip manufacturing is potentially simpler and less carbon-intensive than complex CMOS processes, offering a potential "win-win" for both operational and embodied carbon footprints.
Quantum Computing: Described as a fundamentally different computing paradigm operating on quantum principles (using qubits like trapped atoms manipulated by lasers), quantum computing offers the potential to solve specific problems intractable for classical computers (e.g., code-breaking, materials simulation). However, panelists stressed it's still highly experimental and faces immense challenges: current systems are physically large ("clumsy") and much slower than classical computers for general tasks. Significant technical hurdles include achieving fault tolerance (requiring massive qubit overhead), managing diverse and complex hardware platforms (superconducting circuits, trapped ions, cold atoms, photonics), controlling the extensive classical systems needed (lasers, cryogenics, control electronics), and determining the actual net energy efficiency and carbon footprint, which remains an open question. The panel emphasized the opportunity to embed sustainability considerations early in quantum's development, given its nascent stage.
Panelists:
Gage Hills (Moderator): Assistant Professor, Electrical Engineering, Harvard University
Nate Gemelke: Chief Technology Strategist, QuEra Computing Inc.
Ajay Joshi: Professor, Electrical & Computer Engineering, Boston University & Architect at Lightmatter Inc.
Olivier Ezratty (Online): Co-founder, Quantum Energy Initiative
This panel provided a fascinating glimpse into the future hardware landscape, highlighting the ongoing innovations in electronics and the distinct potentials and hurdles faced by optical and quantum technologies as the world seeks more powerful and sustainable computing solutions for AI.