This proposal "Disruptive Computing: A Computing-Domain-Oriented Approach" addresses challenges in R&D on disruptive computing technologies capable of performing the advanced data processing required for cyber-physical systems (CPSs), which closely integrate the physical world (physical space) with cyber space, and the method of implementing such R&D.

Realizing the advanced data processing systems needed for CPSs will require faster speeds and lower power consumption than what current computing can achieve, artificial intelligence (AI) technologies for real-time and robust recognition and decision-making in edge devices, and advanced data processing technologies such as data compression and encryption for safely transferring information of value to the cloud.

In the meantime, semiconductor integrated circuits, which are responsible for today's advances in data processing performance, are up against limits in miniaturization. Any further improvements in performance will be difficult with a von Neumann computing technology that employs conventional CMOS digital circuits. Consequently, it is desirable to conduct R&D on new computing technologies free from conventional computing algorithms and architectures.

Owing to the expansion of networks and websites, the popularization of smartphones and other mobile devices, and dramatic improvements in the performance of computers in recent years, there is a growing movement to create new services related to purchase analyses and material searches enabled by big data processing and to image recognition, disease diagnosis, and other areas that utilize AI technologies such as deep learning. While the current standard for these services is to perform data processing in the cloud, the increasing prevalence of the Internet of Things (IoT) will make it more important to perform such data processing on the edge, including analysis of raw data from sensors and other devices, conversion of this raw data to data with value, safe data transfers (protecting privacy and ensuring security), and recognition and decision-making (image recognition, speech recognition, danger avoidance, etc.). Since such data processing on the edge requires different attributes from processing in the cloud, including a need for ultra-low power consumption, real-time processing, and robustness, there is much anticipation for new computing technologies capable of implementing these qualities.

Overseas there has been a growing interest in and increased R&D activity on developing new computing technologies, including the development of neuromorphic chips in America's DARPA SyNAPSE Program, research on the brain and the development of new computing technologies simulating the brain through Europe's Human Brain Project, the establishment of the IEEE Rebooting Computing initiative to conduct R&D on new computing technologies, and R&D carried out by such corporations as Google, Microsoft, and IBM on accelerators for deep learning and quantum computing. Japan must take this opportunity to chart a new course for computing technologies by promoting R&D on disruptive computing.

Future R&D challenges in computing technologies that must be addressed involve all technology layers, including algorithms/software, circuits/architecture, and devices/ materials. However, rather than including all technologies comprehensively, it will be important from the perspective of social implementation and efficiency to select only technologies suitable for important applications from each technology layer specific to the computing domain and with consideration for the timeline, and to develop technologies through a vertical integration approach. New knowledge and newly developed technologies should be accumulated to eventually serve as Japan's assets. For these purposes, the following two areas of R&D are proposed.

• (1) Technological development and performance verification through vertical integration of software and hardware
  

  The development of component technologies in specific technology layers does not always lead to performance improvements or optimization of data processing as a whole. Therefore, technological development must be conducted with a view toward vertically integrating all technology layers of computing. Computing domains such as real-time image recognition and autonomous control are particularly important on the edge, and it is important to focus on each computing function to develop an optimal computing technology for these domains. This type of computing-domain-oriented approach must be utilized to emphasize ultra-low power consumption, real-time processing, and robustness; to develop integrated technologies leveraging Japan's skill in hardware technology and system creation; and to verify performance advantages. Computing-domain-oriented technology development must also be applied in the cloud in order to develop accelerator technologies for achieving ultrafast speeds and high-efficiency in data searches, deep learning, and route optimization.

• (2) Organizing new common fundamental technologies and strengthening each technology layer
  

  New core technologies in each technology layer developed through a computing-domain-oriented approach are likely applicable to more than just one specific computing domain. It is essential that these core technologies be organized as new common fundamental technologies to strengthen each technology layer.

  Within algorithms and software, it will be necessary to define the scope of application for new algorithms in non-von-Neumann computing, such as approximate computing, neural nets, neuromorphic computing, and quantum computing, to workloads on the edge and in the cloud, and to codify this knowledge.

  For circuits and architecture, it will be important to accumulate architecture-related technologies suitable for implementing new algorithms, and to refine and organize these technologies for use in other computing domains. This technology layer is an important research area for linking software to hardware. As such, it will be necessary to organize methods of utilizing new packaging and interconnect technologies and new devices. User-friendly design tools must also be simultaneously developed with accessible algorithms and open-source software.

  In devices and materials, it is essential to organize the performance, functions, and reliability of new devices and materials, accurately learn the range of their applications, and clarify guidelines for their design. Japan is strong in this technological area and is only expected to become stronger from the perspective of computing.

Conducting this R&D will require construction of a vertically integrated system of R&D for each computing domain. Here, project managers (domain-specific architects) that assemble researchers and research organizations from different technology layers will play an important role in ensuring the best team composition and facilitating integrated, top-down R&D with close cooperation among researchers from each technology layer. Establishing a research environment that facilitates researchers and engineers from industry, academia, and government in developing software, designing functions and circuits, developing tools, and manufacturing hardware prototypes will also be important. Use of such a research environment is hoped to improved efficiency in R&D, promote collaboration among industry, academia, and government, and foster human resources with an interest in R&D in various technology layers.

Promoting this style of R&D and setting up a suitable research environment will require strategic cross-ministerial funding with considerations for the innovativeness and risk in new computing technologies, the technology readiness level (TRL), and feasible time frames. The relevant ministries and agencies must draw up a complete portfolio covering short-to-long term plans for R&D and what topics are to be addressed, necessitating open sharing of strategies and an effective division of roles. With the cooperation of industry, an ecosystem for industry, academia, and government collaboration should be constructed to implement strategic efforts on acquiring intellectual property rights, sharing and making available technologies, and establishing international standards.

Since R&D on new computing technologies is expected to continue along its upward trend, it will be desirable to utilize knowledge and human resources from around the world to create new fields of research and to form new communities through collaboration with societies involved in algorithms/software, circuits/architectures, and materials/devices.