=== PROJECT SUMMARY === ORGANIZATION: Massachusetts Institute of Technology SUBCONTRACTORS: none PRINCIPAL INVESTIGATOR: Norman Margolus, nhm@im.lcs.mit.edu, telephone (617) 253-2614, FAX (617) 258-8682. TITLE OF EFFORT: CAM-8: Moving Towards Ultimate Computation RELATED INFORMATION: http://www-im.lcs.mit.edu/cam8.html OBJECTIVE: Spatially-organized fine-grained computations are the best possible candidates for efficient and massively parallel implementation. The objective of this research is to significantly advance the state of the art of both the hardware and the applications for such computations. APPROACH: The approach has been to develop hardware that can efficiently simulate large cellular automata (CA) systems with great flexibility; emphasizing data analysis and visualization. This efficient hardware can then be used to stimulate and support collaborations with a range of scientists to develop physical simulations that demonstrate the power of spatially organized computation. It was necessary to develop new hardware because pre-existing computers were so poorly suited to the massive simulation of fine-grained spatial models, and so discouraged the development of these kinds of models. It was necessary to collaborate with researchers in a range of disciplines because of the range of expertise required to: identify appropriate problems; construct and verify new kinds of models; code and test applications for a new kind of computer; and to create appropriate new software tools, both for model development and analysis. PROGRESS: The current contract is a new effort that just began in April. It is a follow-on to a previous ARPA contract, under which the CAM-8 hardware was developed. Efforts under the new contract are focused primarily on issues related to disseminating the CAM-8 architecture; on developing languages, applications, and public domain CAM-8 simulators and environments; and generally on promoting CA modeling and methodology using CAM-8 as a vehicle and a stimulus. The MIT group already has about a dozen applications-oriented collaborations under way, some of which have already begun to yield significant new results. RECENT ACCOMPLISHMENTS: Collaborated on breakthroughs in materials modeling using the two Air Force CAM-8 machines, including efficient new discrete molecular dynamics codes for simulating elastic solids at the atomic level. Collaborated with University of Wisconsin group to develop chemical reaction codes that run (on their small, workstation-scale CAM-8 prototype) 17,000 times faster than their previous workstation codes for the same simulation (which ran on a Sun-3). Developed new CAM-8 visualization techniques that permit rapid 3-D rotations of volume bit-array data. Developed new wavefront simulation techniques suitable for the efficient simulation of large logic circuits on CAM-8, and for spinoff into new families of virtual-processor FPGA chips. FY-95 PLANS: Disseminate CAM-8 architecture and methodology through talks, scientific papers, collaborative research, production of public domain software, and by stimulating the commercial production of CAM-8 machines. Support collaborations by augmenting our CAM-8 software and simulators, and by continuing to develop languages and compilers. Continue to collaborate on developing and demonstrating promising applications such as the Air Force's new CA molecular dynamics simulation techniques. Support the Air Force's initiative to build a large CAM-8 machine through our work on software and applications, and by supplying architectural advice. Develop and exploit novel CA algorithms such as the new wavefront simulation technique mentioned above, which should be able to run multi-million-gate logic simulations at about 100 Hz on a workstation-scale 25 MHz CAM-8 prototype. TECHNOLOGY TRANSITION: Currently the MIT group is engaged in about a dozen active collaborations with outside groups, seven of which have (or are about to receive) copies of the CAM-8 prototype---the rest involve researchers coming to MIT to use machines. The most intensive collaboration has been with Jeff Yepez and his group at the U.S. Air Force's Phillips Laboratory, which currently has two CAM-8 prototypes, and will soon have several more. This Air Force group has developed important new models of gases, liquids, and solids. Last October, they started a new Air Force initiative to develop such models and to build a large CAM-8 machine in support of this work. This initiative has already produced a number of published papers, and the work has been well received at several conferences. Feedback from this collaboration has helped guide the CAM-8 software development, and has stimulated plans for a next generation of CAM machines that exploit modern high-bandwidth memory technology. Other collaborators are at universities and government laboratories, and most of the work has been on developing CA models of physical systems (mainly for academic publication and presentation). The prototype CAM-8 system consists of a workstation-scale 25 MHz implementation of an architecture optimized for large-scale CA simulations, with a demonstrated performance and capacity for lattice gas simulations comparable to that of a large CM-5. It includes an interface for Sun's SBus, used for controlling and communicating with the CAM-8 hardware, and prototype software and examples that run under SunOS. An efficient functional simulator of the CAM-8 hardware is also available, to allow software development in the absence of hardware. Contact the principal investigator. DATE PREPARED: July 26, 1994 === ADMINISTRATIVE DATA === 1. ARPA ORDER NUMBER: 6818 2. CONTRACT/GRANT NUMBER: N00014-94-1-0662 3. AGENT: ONR 4. CONTRACT TITLE: CAM-8: Moving Towards Ultimate Computation 5. CONTRACTOR/ORGANIZATION: Massachusetts Institute of Technology 6. SUBCONTRACTORS: none 7. PRINCIPAL INVESTIGATORS: Dr. Norman Margolus 545 Technology Square NE43-350 Cambridge, MA 02139 8. ACTUAL START DATE: 3/1/94 9. EXPECTED END DATE: 2/28/97 10. FUNDING PROFILE: based on data on or about 6/30/94 10.1. Current contract: TOTAL= 1,497,591 10.2. Options: Option 1= $596,001 Option 2= $601,590 10.3. Total funds provided to date. $225,000 10.4. Actual funds expended to date. $18,895 10.5. Date current funding will be expended: 12/31/94 10.6. Expected funding expenditures by quarter through 12/31/95: (Option years included in this projection) 1/95-3/95= $75,000 4/95-6/95= $149,000 7/95-9/95= $149,000 10/95-12/95= $149,000 10.7. Date. 6/30/94 11. ANYTHING ELSE YOU NEED: N/A === SIGNIFICANT EVENT === LATTICE-GAS MODELS OF CRYSTALIZATION AND ELASTIC SOLIDS In collaboration with the MIT CAM-8 group, Jeff Yepez of the U.S. Air Force has recently made a number of breakthroughs in lattice-gas modeling, including momentum conserving models of liquid/solid and gas/solid phase transitions, and the first CA models of elastic solids. All of these models implement a fully discrete molecular dynamics, in which only a few distinct velocities are used, and forces act only in a few discrete directions, at discrete distances. By adjusting the strength and direction of the forces at each distance, one can accurately tailor a desired inter-atomic potential. The kinds of patterns of spatial data-movement needed for implementing such models takes excellent advantage of the efficient spatial data-movement that is the most essential feature of the CAM-8 hardware. The figure here shows a time-history of a lattice-gas crystalizing into an elastic solid. Each of the "atoms" in the resulting crystal actually consists of several hundred lattice-gas particles. Note how defects in the crystal gradually anneal away. The sequence of configurations shown in the figure covers a span of about 10,000 updates of a $512x512$ space, and represents about 15 minutes of simulation time on an 8-node CAM-8 prototype. The figure on the quad chart shows a 3-D elastic solid, crystalized on a $512x512x512$ spatial lattice. === SIGNIFICANT EVENT: Vu-graph === This is a large PostScript file, and is available from the machine "im.lcs.mit.edu" by anonymous ftp, in the file "/arpa94/event.ps". This was produced on a Sun SPARCstation. === QUAD CHART === This is a large color-PostScript file, and is available from the machine "im.lcs.mit.edu" by anonymous ftp, in the file "/arpa94/quad.ps". A GIF version is also provided, as "/arpa94/quad.gif". Both were produced on a Sun SPARCstation.