ASCENT: Wafer-scale Heterogeneous Integration of Lithium-Niobate-on-Silicon Optoelectronics for Ultralow-energy AI computing

Nontechnical Description: The increasing data volumes from artificial intelligence (AI), internet of things, and 5G/6G networks is challenging the processing power of CMOS-based computing hardware. To extend the computing power scaling and energy efficiency, we propose light-based photonic integrated computing circuits for computing at high clockrates and with ultralow-loss on-chip data movement. This project supports the national interest by significantly reducing the energy required for AI computations—potentially two orders of magnitude more efficient than current CMOS technologies—paving the way for powerful and sustainable computing systems. Its outcomes could revolutionize applications from autonomous vehicles and healthcare diagnostics to natural language processing and scientific discovery. The project will also strengthen U.S. competitiveness in semiconductor manufacturing by training a new generation of experts in chip design, photonics, and AI. Importantly, it will promote STEM participation through hands-on training and outreach. Technical Description: This project aims to build high efficiency (>100 TOPS/W), high-throughput photonic-electronic hybrid processors by leveraging wafer-scale heterogeneous integration of thin-film lithium niobate (TFLN) and silicon photonics/electronics. The key technical goals include: (1) developing space-time-wavelength hyperdimensional photonic circuits that can perform massive parallel tensor computations using scalable time-multiplexed data encoding; (2) enabling CMOS-compatible, high-speed electro-optic modulators via TFLN onto silicon to realize >50 GHz bandwidth and <10 fJ/conversion energy; and (3) co-designing hardware and software to, support large-scale models like LLMs and real-time decision-making in multi-agent systems. The proposed architecture achieves matrix-matrix multiplications with O(N) modulator scaling and energy efficiency—orders of magnitude beyond CMOS limits. The system will benchmark end-to-end AI performance in reinforcement learning with applications of autonomous driving. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.