Considering that Ultra Low Power (ULP)system technology will become one of the most important technologies, the Ikoma Group of the Center for Research and Development Strategy (of the Japan Science and Technology Agency)held a workshop in October 2004 to identify the specific challenges involved and the research approaches that should be taken pursuant to this. The proposal from the workshop was that the technologies adaptively manage the supply of electric power required by the information systems in advanced network society to ensure quality of service. To this end, R&D efforts should be made for the comprehensive management of energy/electricity consumption at every level of the information system in order to achieve reduction in energy/electricity consumption of three orders of magnitude within 10 years’time frame.
In order to gather information that would be helpful in planning the research strategy to be pursued, a team of experts (eight members led by Professor Takashi Nanya of the University of Tokyo) went to the United States in March/April 2005 to visit research institutes and funding bodies which are active in the area of “ultra -low power system technology.”
The institutions visited, the individuals met, the main themes of discussion and the major findings are listed in the following. (Institutions Visited / Individuals me / Themes of Discussion / Major findings)
NSF / Dr. H.Gill Program Director, CISE / Progress of ULP project supported by NSF, activities of researchers / - 20% of VLSI design projects (about $6M)are related to ULP. - Also interested in ULP of embedded systems
DARPA / Mr. R. Graybill Project Manager, DARPA/IPTO / PAC/C project status and achievements / - Ended in 1Q 2005 - Basic technology in phase 1;5 prototypes made in phase 2
IBM Watson Res. Ctr. / Dr.G. Chiu Senior Manager, Advanced Server Hardware Systems/ Low electricity consumption technology of BlueGene/L / - Design philosophy of “small and many” - Scalability and packaging technology are important
MIT / Prof. A. Chandrakasan Dept. of EECS / Progress of research on sensor network project, etc. / - Extensive ULP research promoted - Targeting Sensor network with 10 μW nodes within 5 years
UC Berkeley, CITRIS / Dr.G. Baldwin Executive Director, CITRIS / Progress of research on sensor network, HPC, etc. / - Development of “invisible”electronics – Application oriented research on power supply, wireless communication, and reliability
Intel / Shekhar Borkar Director, Circuit Res. Lab / Progress of research on low electricity consumption technology / - Research strategy device generations - Power consumption/ performance should be minimized under an assumption that transistor cost will be zero in the future.
IEEE IPDPS / / Progress of research on low electricity consumption technology in HPC area / - Many research developments regarding low power HPC and power-aware HPC - Importance of ULP re confirmed
Through these visits and discussions, the team was able to reconfirm or discover the following.
While the reduction of power consumption is already a major challenge in the fields of electronic devices and circuits, a further leap ahead in power reduction would require a design that accommodates trade-offs in the areas of power consumption, performance and reliability at every level-including system specifications, and algorithms and communication protocols-and dynamic energy management at the software and architecture levels. Therefore, research will be needed on power-saving technologies at all levels (including algorithms, protocols, software, architecture, circuitry and electron devices), together with an adaptive energy management technology that integrates all these aspects, creates a research theme of vital importance.
In the field of high-performance computing, since an increase in speed using a small number of high-performance processors is limited by high power consumption and heating problems, the future direction is likely to be the pursuit of higher speed using a large number of small, low-speed processors. IBM’s successful BlueGene/L and Intel’s multi-core strategy seem to indicate this direction.
In the area of semiconductors, challenges are posed by an increase in current leakage, PVT (Process/source Voltage/Temperature) variations and a lower reliability of components due to the ever-increasing reduction in size.
Technical trends, such as the shift toward slow and many processors for high-performance computing, and an increasing number of parts which results in the reliability degradation clearly indicate the importance of achieving greater fault tolerance and technological reliability. They indicate that system designs that involve trade-offs in performance, power consumption and reliability-all of which are closely related-will pose major technological challenges in the future.
Based on the findings of the trip, the following is proposed regarding the research strategy for the Ultra Low Power system.
1) Field of research
Information systems and equipments can be classified by their power consumption range, as follows :over 1W (e.g. PCs, servers and HPCs);the milliwatt class from 1mW to 1W (e.g. embedded systems and mobile terminals);and the microwatt class of 1mW and below (e.g. ubiquitous computing and sensor networks). Research should be emphasized mainly on Ultra Low Power technology in the milliwatt class. The research results can be applied to the over the W class and the microwatt class.
2) Research strategy
A major innovative leap forward is needed in specific technologies at every level of systems research, including service specifications, protocols, algorithms, systems software, architecture, circuitry and electron devices. Also needed is the development of a comprehensive Ultra Low Power technology for adaptive energy management to integrate all these elements.
3) Research Organization
A virtual laboratory needs to be set up to promote the collaboration of existing research organizations (industry, academia and the government)under the strong leadership of a leader. A research organization at the same time, the education of key researchers should be carried out throng such research activities.