[ThD1] Heterogeneous Integration on Silicon Photonics (I)

9:30 AM - 10:00 AM JST(12:30 AM - 1:00 AM UTC)

[ThD1-01 Invited] High-Speed, Low-Energy Membrane Feedback Lasers for AI Applications

〇Nikolaos Panteleimon Diamantopoulos¹, Takuro Fujii¹, Hidetaka Nishi¹, Shinji Matsuo¹ (1. NTT, Inc. (Japan))

We review recent advances in membrane feedback lasers fabricated on SiO₂/Si substrates, targeting the escalating energy consumption and bandwidth demands of AI data centers and AI compute hardware. Directly modulated membrane lasers exploiting the bandwidth-enhancement photon-photon resonance (PPR) effect achieve 3-dB modulation bandwidths up to 60 GHz and data rates reaching 200 Gb/s using PAM-4 modulation. Leveraging this platform, we have demonstrated multi-channel, 1.6-Tb/s-class transmitter chiplets with sub-pJ/bit laser operating energies. In addition, we have realized spiking neuromorphic photonic hardware accelerators based on membrane feedback laser neurons, operating at modulation speeds up to 100 GBaud and enabling record-high image processing speeds. Ongoing efforts towards tighter co-integration of PPR-DMLs with electronic circuits, scalable multi-channel and multi-layer laser neural networks, and ultra-low-energy optical modulation for next-generation AI/ICT laser communications are expected to enable a new class of energy-efficient photonic hardware for next generation AI and ICT systems.