NAND-Based Approximate 8×8 Multiplier with Hybrid Reduction for Low-Power CMOS Applications
Abstract
This paper presents an 8×8 FA-MUX based approximate multiplier employing NAND-gate partial product generation and an optimized 4:2 compressor architecture to achieve low power consumption and high speed operation for digital signal processing applications. The proposed design is based on the multiplexer assisted carry selection logic and employs full adder circuitry to minimize switching activity, propagation delay and hardware complexity. An approximation of carry generation mechanism is incorporated to further reduce power consumption while preserving computational accuracy within acceptable limits for error-resilient applications. The proposed design is synthesized and simulated in Cadence Virtuoso (analog flow), NC LaunchSimVision and Cadence Genus (digital flow) in gpdk180 CMOS technology with a 1.8 V power supply. Experimental results show that the proposed multiplier can reduce the power consumption by about 35% and the delay by about 36% compared with a conventional exact Dadda multiplier. In addition, successful timing closure at 500 MHz is achieved for the proposed design. The proposed FA-MUX multiplier efficiently balances accuracy, power, speed and hardware optimization in DSP systems. Formal error characterization over all 65,536 input combinations yields a Mean Error Distance (MED) of 8.14 LSB, Normalized Error Distance (NED) of 1.24×10−4, Mean Relative Error Distance(MRED) of 0.83%, and Error Rate (ER) of 40.8%, confirming bounded-error behaviour. Furthermore, application-level PSNR evaluation in image-processing convolution confirms output image quality exceeding 37 dB, well above the 30 dB perceptual quality threshold.