Ternary-valued Associative Processor Design

Mira Hout1, Rouwaida Kanj2, Ahmed Eltawil3, Mohammed Fouda4
1ECE Dept., American University of Beirut,, 2Synopsys, 3King Abdullah University of Science and Technology, 4Rain Neuromorphics Inc.


Abstract

In-memory computing architectures have emerged as a promising solution to address the memory-wall bottleneck and enable efficient vectorized and parallel arithmetic operations. This paper proposes a hardware architecture and algorithms for ternary-valued associative processors (TAP) which is a critical step toward broadening their computational versatility. To facilitate the in-memory implementation of complex arithmetic and logic functions, we introduce an efficient methodology for the automatic generation of corresponding look-up tables (LUTs). A SPICE-MATLAB co-simulator is developed to evaluate the performance of the proposed TAP-based in-memory ternary adder in terms of energy consumption, computational delay, and area overhead. Comparative analysis demonstrates that the ternary AP adder achieves energy and area reductions of 11\% and 6.2\%, respectively, compared to its binary counterpart. Furthermore, compared to state-of-the-art ternary carry-lookahead adder, the proposed solution exhibits around 100x and 9.5x reduction in energy and delay, respectively. This work highlights the potential of in-memory and parallel computing processors to high-efficiency arithmetic operations, offering scalable and energy-efficient solutions for emerging computational workloads.