Design and FPGA Based Performance Analysis of Error Compensated Multiply Accumulate Architectures for Energy Efficient Image Processing
Computer and Decision Making: An International Journal · 01.11.2026 · 3 (6S) · 1123-1136
DOI: 10.1922/comdem.v3i6s.18889 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 01.11.2026. Verfasst von Hassan M Raza und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Student- This study investigates two eight-bit unsigned approximate multiply–accumulate architectures, AMACU-8 and AMACU-9, implemented for an Artix-7 FPGA target. Both architectures reduce the partial-product matrix and employ fixed error compensation; AMACU-8 uses a hybrid eight-bit carry-maskable/carry-propagate final addition scheme, whereas AMACU-9 uses a modified reduction arrangement followed by a 16-bit carry-propagate adder. The supplied Vivado implementation reports indicate that AMACU-8 uses 68 LUTs and 20 slices, with 9.622 W estimated total on-chip power and a longest reported unconstrained path of 15.863 ns. AMACU-9 uses 43 LUTs and 14 slices, with 7.333 W estimated power and a reported path of 10.971 ns. The supplied error evaluation reports mean error distances of 918.39 and 1710.46, respectively, while the supplied image-smoothing evaluation reports PSNR of 84.28 and 76.04 dB. These results reveal an accuracy–hardware-cost trade-off: AMACU-8 preserves better numerical fidelity while AMACU-9 reduces mapped resources and reported path delay. Because the power estimates have low confidence and the timing paths are unconstrained, the results are interpreted as implementation estimates rather than measured energy or validated maximum clock frequency. Further activity-controlled and constraint-complete evaluation is recommended before making general power-efficiency claims.
Zitieren
APA
Raza, H. M., & Varthamanan, Y. (2026). Design and FPGA Based Performance Analysis of Error Compensated Multiply Accumulate Architectures for Energy Efficient Image Processing. Computer and Decision Making: An International Journal, 3(6S), 1123-1136. https://doi.org/10.1922/comdem.v3i6s.18889
Vancouver
Raza HM, Varthamanan Y. Design and FPGA Based Performance Analysis of Error Compensated Multiply Accumulate Architectures for Energy Efficient Image Processing. Computer and Decision Making: An International Journal. 2026;3(6S):1123-1136. doi:10.1922/comdem.v3i6s.18889
BibTeX
@article{Raza2026,
author = {Raza, Hassan M and Varthamanan, Y.},
title = {Design and FPGA Based Performance Analysis of Error Compensated Multiply Accumulate Architectures for Energy Efficient Image Processing},
journal = {Computer and Decision Making: An International Journal},
year = {2026},
volume = {3},
number = {6S},
pages = {1123--1136},
doi = {10.1922/comdem.v3i6s.18889}
}
EUVIMED