Core Insights - The rapid development of generative artificial intelligence has accelerated the deployment of large-scale AI clusters globally, leading to a significant energy crisis due to increased energy consumption associated with data processing and transmission [1][3] - Photonics technology, particularly silicon photonics, has the potential to address energy consumption issues by enabling high-density interconnects and low-energy optical switching, which are essential for sustainable AI infrastructure [3][4] Group 1: Energy Consumption and Photonics - The energy supply required for data processing is expected to grow exponentially alongside data volume, necessitating the development of technologies that decouple energy growth from data growth [1] - Silicon photonics has evolved over the past two decades to provide an ideal platform for efficient optical interconnects, enabling high bandwidth and long-distance links while maintaining low energy consumption [3][4] Group 2: Optical Switches and ASICs - The energy efficiency of optical transceivers has matched the pace of Moore's Law, achieving over 5 pJ/bit, while the scalability of ASIC switches has lagged behind, indicating a bottleneck in the switch rather than the transceiver [4][6] - Optical switches maintain low power consumption even as throughput increases, contrasting with ASIC switches, which exceed 1000W at 100Tbps throughput [6] Group 3: System Applications and Development - Optical switches cannot directly replace ASIC switches due to their inability to process data packets, necessitating a complete system redesign and optimization for optical circuit switches (OCS) [8] - Google has pioneered the large-scale implementation of OCS in its data centers, leading to increased interest and development in optical switching technology [8][9] Group 4: Photonic Neural Networks (PNN) - Photonic neural networks leverage the high uniformity and yield of silicon photonic devices to perform matrix-vector multiplications at high speed and low energy, potentially alleviating the computational burden on high-energy digital processors [13][15] - New AI models based on electro-optic nonlinearity have been proposed to enhance the capabilities of PNNs, allowing for efficient computation without intermediate digital processing [15][21] Group 5: Future Directions - Significant advancements in silicon photonics have demonstrated its potential to enhance the sustainability of AI infrastructure through high-density I/O and optical AI accelerators [23] - Integrating photonic functional devices into traditional digital infrastructure presents challenges that require further research into overall system design and implementation [23]