Energy-efficient Buffer-Based Ternary SRAM Cell with Application to Image Processing

This paper presents a design of a variation-resilient and energy-efficient ternary memory cell (TSRAM) suited for power-demanding IoT applications that run on batteries. The TSRAM cell utilizes a latch composed of an efficient ternary buffer (TBUF) with positive feedback, a single-bitline, and a transmission gate for switching access, with an overall area only about 39% more than binary 6T SRAM. The threshold voltage (V_th) tuning of CNTFET devices has been explored to achieve the three storage levels. Simulations were conducted using the Standard 32-nm CNTFET model file in the Synopsis HSPICE simulator. The projected design offers substantial reductions of 54.94% in read power, 67.06% in write power, and 21.59% in area compared to the best buffer-based TSRAM designs. These power savings are achieved by minimizing the transistor count and eliminating any direct current path between VDD and ground in the TBUF design for getting logic ‘1’. Furthermore, the proposed design demonstrates the highest logic ‘1’ static noise margin (SNM1) and shows resilience to process, voltage, and temperature (PVT) variations. The TSRAM electrical quality matrix (TEQM), a crucial figure of merit, indicates the superior performance of the proposed design for IoT applications. The study was further extended to conduct simulations and report the performance metrics of the proposed TSRAM array. Ultimately, to evaluate the real-world application of the triple memory structures, the pixel-by-pixel storage process of a grayscale image with three-value data content is performed based on a hardware algorithm. The obtained results demonstrate that the proposed TSRAM architecture has about a 26.3% improvement in hardware performance compared to its highest performing counterpart scheme.