Sky-TCAM: Low-Power Skyrmion-Based Ternary Content Addressable Memory

Authors:

Abstract
Low-cost content-addressable memories (CAMs) are highly desirable for many applications where high-speed search is needed such as network routers and machine learning. We propose a low-power, compact, fast magnetic skyrmion-based ternary CAM (Sky-TCAM) cell design. Each cell comprises a 5T2R structure with five transistors and two magnetic tunnel junctions (MTJs). The search current polarity together with the stored bit can realize the XOR logic. When a mismatch happens, the skyrmion appears beneath one MTJ, leading to the discharge of the matchline (ML). While the search energy-delay-product (EDP) is comparable with that of non-volatile (NV) ternary CAMs (TCAMs), Sky-TCAM shows the lowest EDP of ${8.74} \times {10}^{-{25}}\,\,\text {J}\,\,\cdot \,\,\text {s}$ among all TCAMs we compare. Our research shows that Sky-TCAM is promising for building low-power and low-latency computation applications.
Funding Information
  • Alabama Graduate Research Scholars Program (GRSP) funded through the Alabama Commission for Higher Education and administered by the Alabama EPSCoR
  • AI Chip Center for Emerging Smart Systems
  • InnoHK funding, Hong Kong SAR
  • NSF EPM (DMR-2129879)
  • Auburn University Intramural Grants Program