Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 61
B Identifying Hubs of Research Activity in Key Areas of S&T Critical to this Study A s a data-gathering experiment, the committee identified hubs of research activity in semiconductor solicited input from experts on their sense of scaling, architecture, and parallel programming.3,4 A where innovation and engagement are taking visualization of the publication network for these place related to the power and performance challenges advanced research areas critical to the computing for sustaining growth in computing performance. In performance challenge is shown in Figure B-1. The goal particular, the committee asked experts to identify of this exercise is not to highlight individually-identified leading researchers around the globe focused on the researchers, but rather to present a methodology that challenges of sustaining growth in computer allows gleaning, at least in a rough, qualitative sense, of performance in (1) semiconductor device scaling, (2) potential insights from the connectedness between the power efficiency in computing hardware, (3) parallel hubs of research activity, as well as between U.S. and programming and models to leverage multicore and international research communities. other novel architecture, (4) chip architectures, and (5) Figure B-1 shows a map of highly connected circles. runtime and software infrastructure for power-efficient Each circle represents an individual researcher, and each and scalable computing. line between two circles represents a coauthored Approximately 170 leading researchers were publication. The size5 of each circle corresponds identified, based on input solicited from a dozen approximately to the total number of papers that computer scientists, engineers, and recommendations by researcher has authored, and the width of each line the committee.1 Approximately three-quarters of those corresponds to the number of coauthored papers shared identified were based in the United States. Individuals between two researchers. Hubs of research activity are who were identified by at least three people were colored in yellow and labeled with letters corresponding deemed to be "hubs" of concentrated research activity to their institutional affiliation.6 All other circles are for the purposes of the committee's analysis. Publication data (for the years 20012011) for each of these individuals, or hubs, was collected using SciVerse Scopus2 (resulting in a total number of 1,081 publications and 5,685 authors, 1,368 of which are 3 Coauthor publication network maps are not shown for ad- unique authors). vanced research in power efficiency in computing hardware or in Using this publication data, the committee generated runtime and software infrastructure for power-efficient and scala- a coauthor publication network map that includes all ble computing due to limited overlap in researcher nominations. 4 Coauthor publication network maps were generated using the Science of Science (Sci2) Tool, available at http://sci2.cns.iu.edu. 1 5 The following noncommittee members contributed to this data- The size of each node is calculated as a fraction of the largest collection exercise: Alex Aiken (Stanford University), Mark Bohr number of papers authored and/or coauthored by a single individ- (Intel), Robert Colwell (DARPA), Bob Doering (Texas In- ual. 6 struments), Bryan Ford (Yale University), David Patterson (Uni- Affiliations associated with each hotspot are as follows: (A) The versity of California-Berkeley); David Srolovitz (A*STAR), and University of California, Berkeley; (B) Massachusetts Institute of Dennis Sylvester (University of Michigan). Technology; (C) Stanford University; (D) University of Illinois at 2 See www.scopus.com. Last accessed on August 11, 2012. Urbana-Champaign; and (E) Advanced Micro Devices, Inc. 61
OCR for page 62
62 THE GLOBAL ECOSYSTEM IN ADVANCED COMPUTING FIGURE B-1 Coauthorship networks of hubs of research activity in three areas of advanced computing research. colored according to the geographic location of that research (e.g., semiconductor scaling, architecture, and researcher, as indicated in the figure.7 Three large circles parallel programming). marked by dotted lines are used to bin the hubs of It is worth noting that this mapping approach differs research activity by each of their associated areas of from traditional bibliometric analyses of coauthored publication data because the primary nodes (hubs) 7 Color coding by region (e.g., U.S., Asia, Europe, Other) was examined were identified based on the committee's data determined by using each authors' geographic location listed on input solicitations (as opposed to selecting hubs on the his or her conference publication. Addresses were not mapped basis of total number of publications or of most cited using Sci2 software.
OCR for page 63
HUBS OF RESEARCH ACTIVITY 63 publications). While the sampling size of the data Increasing circle size, increasing connectedness solicitations is small, Figure B-1 reveals several between researchers, and widening lines between interesting features that may be useful for subsequent researchers may all be useful indicators for identifying analysis. emerging hubs of research activity. For instance, a small For example, all hubs are located in the United circle with many connections might suggest an States, and all but one of these hubs are U.S. research individual who publishes less but collaborates frequently universities. The coauthorship network maps show that and is thereby more intimately connected to the global some areas of research (and some researchers knowledge network. In addition, a wide line between a specifically) tend to collaborate on a nation- or region- small circle and an established research hub might specific basis or both. For example, chip architecture and suggest a promising early-career researcher who hails parallel programming networks are primarily U.S. based from a strong research lineage. This analysis could also with limited participation by Europe and Asia. In be extended by observing how coauthored publication contrast, semiconductor device-scaling networks show a networks change over time. significant number of collaborations with Asia. In In summary, this methodology presents a unique particular, Taiwan holds the vast majority share of Asia's approach for identifying emerging, as well as representation, followed by Singapore and Japan (data established, hubs of research activity in three areas of not shown). science and technology. However, given the small While the individual hubs do not generally show a sampling size of the data solicitations, this experiment is significant degree of connectivity with one another (with not intended to provide any assessment or interpretation exception to two hubs in the semiconductor scaling of the hubs themselves (or of trends apparent in the networks), the semiconductor scaling and chip network maps). Rather, the goal of this experiment is to architecture networks appear to be highly interconnected. demonstrate an approach that could be extended and/or In fact, both of these networks share a common hub. In modified (e.g., to include statistically valid data- contrast, researchers within the parallel programming gathering methodologies) for subsequent in-depth networks display much less connectivity. exploration in any number of research areas.
OCR for page 64