Date: Tue, 26 Nov 1996 00:30:28 GMT
Server: NCSA/1.5.2
Last-modified: Wed, 02 Aug 1995 19:04:34 GMT
Content-type: text/html
Content-length: 4230

<TITLE>Rutgers CAM Project </TITLE><h1>The Rutgers CAM Project</h1><!WA0><img src = "http://www.cs.rutgers.edu/cam/camlogo.gif"><p>The Rutgers CAM2000 Memory Architecture uses the concepts ofassociative processing to overcome the shortcomings of conventionalvon Neumann architecture in dealing with compute-bound applicationsinvolving massive amounts of data.  The architecture provides hardwaresupport for data parallel, reduction, broadcast, and parallelprefix/suffix operations within a memory architecture that is DRAMcompatible.  The CAM2000 architecture is particularly appropriate for applicationsthat require simple operations (e.g., counting, matching, etc.) overlarge amounts of data.  It is designed to replace the memory, not theprocessor, of a conventional computer and is a dual of cache.  Itameliorates the performance bottleneck imposed by the processor/memorybus by moving processing across into the memory.  The CAM2000 canemulate existing memory functions, and run existing softwareunchanged.  Furthermore, software can be upgraded incrementally, on astatement-by-statement basis, to use associative functions.<p>The CAM2000 applies the associative processing model by placing manyword-width, low performance processors in the memory.  The word-widthhardware designed for the CAM2000 provides significant speedup andreduces software complexity compared to approaches using single-bitdata paths.  In addition, the CAM2000 supports extended precision(functions on data wider than its hardware) allowing arbitraryprecision CAM operations to execute with minimal slowdown.This approach can produce significant speed-up even when the in-memoryprocessors run only at memory speed.  By 2000 a CAM architecture canprovide 5-10 giga-ops on a single chip that replaces DRAM,significantly increases system peak performance, and scales withmemory size.<p>The graphic above illustrates the architecture of the Rutgers CAM2000which replaces DRAM chips in an otherwise conventional architecture.Half the area of the CAM2000 chip is DRAM and the other half isprocessing elements.  The processing elements can provide enormouslyhigh parallelism whenever it is applicable; otherwise the CAM chipacts as ordinary memory.<h3>Personel</h3><ul><li><!WA1><a href="http://www.cs.rutgers.edu/~levy">Saul Levy </a> --   levy@cs.rutgers.edu, 908-445-3523<li><!WA2><a href="http://www.cs.rutgers.edu/~josh">Josh Hall</a> --  josh@cs.rutgers.edu, 908-445-3896<li><!WA3><a href="http://www.cs.rutgers.edu/~dsmith">Don Smith </a> -- dsmith@cs.rutgers.edu, 908-445-3628<li>Keith Miyake -- kmiyake@cs.rutgers.edu, 908-445-2558<li>Chung-Hsing Hsu -- chunghsu@cs.rutgers.edu, 908-445-4635</ul><p>This work is supported by ARPA and NASA under NAG 2-668<h3>Online Publications</h3><p>J. Storrs Hall,<!WA4><a href = "http://www.cs.rutgers.edu/cam/fornax.ps">"FORNAX"</a>, USENIX VHLL,April 1995 (FORNAX is a very high-level programming languagewe are developing to ameliorate the difficulties of programmingthe unconventional architecture of CAM (and other parallel machines)).<p>J. Storrs Hall and Don Smith,<!WA5><a href = "http://www.cs.rutgers.edu/cam/dbmm.ps">"Database Mining and Matching in Rutgers CAM"</a>, 2nd Associative Processing & Applications Workshop, Syracuse University, July 1993<p>J. Storrs Hall, Don Smith, and Keith Miyake,<!WA6><a href = "http://www.cs.rutgers.edu/cam/collective.ps">"Collective Functions in the Rutgers CAM"</a>, Associative Processing & Applications Workshop, Syracuse University, July 1992<p>Donald E. Smith, J. Storrs Hall, and  Keith M. Miyake, <!WA7><a href = "http://www.cs.rutgers.edu/cam/TR196.ps">"Rutgers CAM Chip Architecture"</a>, Technical Report No. LCSR-TR-196, Department of Computer Science, Rutgers University, 1992<p>Donald E. Smith, Keith M. Miyake, and J. Storrs Hall<!WA8><a href = "http://www.cs.rutgers.edu/cam/TRxxx.ps">"A Greedy Rescheduling Assembler for the CAM2000" </a>,Department of Computer Science, Rutgers University, 1994<p>Keith M. Miyake and Donald E. Smith, <!WA9><a href = "http://www.cs.rutgers.edu/cam/TR191.ps">"Circuit Design Tool User's Manual"</a>,  Technical Report No. LCSR-TR-191, Department of Computer Science, Rutgers University, 1992