This proposal, "Optical Data Communication Technologies for Future AI Computing Systems: From Network architecture to optical devices and materials," is a research and development strategy concerning optical data communication technology that increases data communication speed and other aspects in data centers (DCs), whose use is increasing due to the development of artificial intelligence (AI). Focusing on the need for high-speed and high-capacity data communication networks, one of the technical challenges for DCs in response to the rapid advancement of AI, this proposal proposes a direction for research and development of optical data communication technology to meet the performance requirements for DCs from the 2030s onward. In this regard, it is crucial to promote research and development by leveraging Japan's strengths in high-speed optical device technology and materials technology.

With the advancement of digitalization, vast amounts of digital data such as images, videos, and audio are generated daily. High-performance DCs that store, analyze, and interpret this data to provide various services are being built worldwide, and the processing capacity and scale of DCs are expanding rapidly with the rapid advancement of AI technology. On the other hand, such DCs consume enormous amounts of electricity, leading to social challenges such as power supply, water resources, and site selection, as well as technical challenges such as scalability, power efficiency, cooling efficiency, operational efficiency, and security. Among the technical challenges concerning future data center systems, ensuring scalability, particularly increasing the speed and capacity of the data communication network within the data center, is a key issue. The emergence of large-scale language models and generative AI necessitates massive computational processing, requiring the interconnection of numerous GPUs (Graphics Processing Units) and servers for parallel processing. However, the network performance of this interconnected network is a bottleneck in improving the processing power of data centers. Overcoming this bottleneck requires faster and larger data communication networks, which is difficult with conventional electrical wiring technology; therefore, the introduction of optical wiring technology (optical data communication technology) is essential. However, in each technological layer constituting optical data communication technology within a data center―including network architecture, optical modules, optical devices, and materials/integration technologies―simply extending technologies used for long-distance optical communication will not realize future data centers. Network architecture needs to adapt to future advancements and trends in AI processing; optical modules need to achieve high reliability and ease of maintenance; optical devices need to be miniaturized and increased in density; and materials/integration need to overcome challenges such as the utilization of new optical properties and functions, and the integration of dissimilar materials.

Research and development efforts to overcome these challenges are becoming increasingly active globally. Currently, Japan holds a competitive advantage in research and development and related businesses for indium phosphide (InP)-based high-speed optical devices and materials. However, in low-speed optical devices and optical data communication module technologies using silicon photonics technology, which excels in integration, Japan lags behind Europe, the United States, and Taiwan. To maintain and expand its dominance in this field, Japan must proactively identify long-term research and development challenges in response to the evolving demands and challenges of data centers (DCs) and promote strategic R&D.

The following four points represent the research and development challenges that should be addressed for the use of optical data communication technology in DC networks, categorized by technology layer.

(1) Network Architecture Adapting to the Evolution of Data Centers
Data center network architectures need to adapt to various evolutionary scenarios of data centers, such as the scaling up of AI computations, the broadening of the purposes and applications of AI, and the increasing diversity of users. Therefore, in addition to architectural design that leverages innovations in components such as optical devices and modules, challenges include flexible, scalable, and automatic responses to various requirements, as well as support for distributed processing both inside and outside the data center, and real-time/ultra-high-resolution parallel processing.

(2) High-Speed, High-Capacity Optical Module Technology
Challenges include achieving high speed and high capacity through parallelization (wavelength division multiplexing, spatial division multiplexing) that leverages the properties of light, implementing optical modules that allow for high-density connection of more CPUs (Central Processing Units) and GPUs within a rack, ensuring high reliability and ease of maintenance, and miniaturizing and increasing the density of switches for switching data transmission and reception destinations.

(3) Innovative Optical Devices Realizing Orders of a Magnitude Performance Improvements and New Concepts and Functions
It is necessary to work on increasing the speed, efficiency, miniaturization, cost reduction, and reliability of optical devices such as semiconductor lasers, optical modulators, photodetectors, waveguides, multiplexers/demultiplexers, optical switches, and optical integrated circuits. Furthermore, from a longterm perspective, there is the challenge of creating devices that can be expected to achieve orders of a magnitude performance improvement through new concepts and functions.

(4) Materials and Integration Technologies Supporting Breakthroughs in Optical Devices
To create innovative optical devices with new concepts and functions, it is important to consider materials with different compositions and properties than those used so far. This presents challenges with new materials such as compound semiconductors like gallium arsenide (GaAs), Group IV semiconductors like germanium (Ge), thin-film ferroelectric materials, and materials with unique optical properties. Also, since it is difficult to achieve excellent functions such as light emission, modulation, and detection with only one material system, there are challenges with heterogeneous material integration technologies such as wafer bonding to enable the integration of devices with different material systems.

To efficiently advance the above research and development challenges, it is crucial to take a long-term perspective and foster a community of diverse researchers, create roadmaps through industry-academia-government consortia, promote new projects involving collaboration across technology layers, utilize large-scale research centers and shared facilities, leverage Japan's strengths for international collaboration, and promote human resource development.

First, it is effective for researchers and engineers involved in optical data communication technology to create a forum and new community where diverse personnel from industry, academia, and government can gather and discuss, such as by holding symposiums that transcend technology layers and fields at academic conferences of the Institute of Electronics, Information and Communication Engineers, the Japan Society of Applied Physics, and the Information Processing Society of Japan. Furthermore, it is desirable to create a roadmap for optical data communication technology, allowing researchers and engineers from different technology layers to share future research and development challenges and goals.

To enable researchers from different technology layers--network architecture, optical modules, optical devices, and materials/integration technology--to form teams and advance research and development towards common goals, it is essential to promote projects that lead and support this process. Depending on the level of technological maturity and application field, it is desirable for the Cabinet Office, the Ministry of Internal Affairs and Communications, the Ministry of Education, Culture, Sports, Science and Technology, the Ministry of Economy, Trade and Industry, and related funding agencies to collaborate and implement projects at appropriate times. For prototyping and evaluation of hardware such as optical modules, optical devices, materials, and integration technologies, it is crucial to develop large-scale research and development centers and shared facilities to meet the needs of researchers from both industry and academia.

Japan's greatest strength in business and research and development in this field lies in highspeed optical devices, primarily InP-based devices. It is important to continue this strength while actively promoting international collaboration and human resource development that maximizes its potential. For example, promoting international collaborative research and development projects that leverage the strengths of various countries, centered on academia, with the aim of creating innovative optical devices, would be effective. Furthermore, promoting international standardization through collaboration with overseas companies and research institutions strong in system integration, etc., through technologies that integrate InP-based devices and silicon photonics, is also noteworthy. Such international collaboration is expected to foster the development of personnel with broad knowledge beyond their own specialized fields, and to facilitate the exchange of excellent researchers and engineers both domestically and internationally.

• (in Japanese) Optical Data Communication Technologies for Future AI Computing Systems ~ From network architecture to optical devices and materials ~