Hooman Jarollahi

Yongyuan Zang

Tohoku University

Japan

PhD student Hooman Jarollahi recently undertook an intense research internship at Tohoku University, Japan, where he made important advances in the area of associative memories, by proposing and fabricating an integrated content-driven search accelerator based on a variation of the sparse clustered networks (SCNs). The proposed architecture, called SCN-CDS, eliminates the brute-force search operations while requiring a significantly lower number of physical memory bits compared to that of competing Content Addressable Memories (CAM). In addition, the SCN-CDS links the contents related to multiple search fields and stores them within low-complexity logic-in-memory (LIM) elements instead of storing the actual search data that would occupy large amounts of memory. In the implementation of SCN-CDS, the three-dimensional stack-ability and leakage-power elimination benefits of magnetic-tunnel-junction (MTJ) devices are exploited to construct non-volatile LIM cells. The SCN-CDS provides a compact solution that can be used with complex software based CDS engines to improve energy-efficiency and accelerate the exhaustive search operations.

A joint manuscript describing this work has been written by Mr. Jarollahi with his colleagues at Tohoku University.