Energy Efficient Binary Adders for Error Resilient Applications

Abstract

Next Generation portable systems need to be geared up to address the challenges of energy efficient processing. Approximate Computing is one of the promising methodologies that relies on captivating property of inherent error resilience of various multimedia applications. This paper proposes quality - energy optimal approximate adders based on systematic decomposition of full adder and higher dimensional adders are designed using the energy efficient and low error full adders. Novel approximate full adder with 87.5% accuracy is designed. The 8-bit, 16-bit binary adders are analyzed by incorporating the designed full adder. The proposed 8-bit approximate adders have the accuracy of 75.2%, 56.6% for 3 bits, 4 bits approximation respectively. 16-bit approximate adder with 8-bit approximation have an accuracy of 43% for an energy savings of nearly 50%. The designed adders when employed in Image Blending, Average PSNR, PSNR-HVS, PSNR-HVSM are found to be around 78dB, 37dB, 41dB respectively. The Application of Image brightness enhancement is analyzed with different constants (50,100,128) and different image sizes. The Image denoising is implemented and the Average MSE is found to be 0.06 and 0.2 for Gaussian, Salt & Pepper Noised image of size 1024 × 1024. The proposed energy efficient adders can sufficiently be used in multimedia applications without much loss of PSNR in real time.