EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Progressive graph convolution triangulation masked topology aggregation attention network optimized multiply and accumulate unit in CNN accelerator for image processing applications

S. Saranya, K. Lakshmi Joshitha

AIP Advances · 01.11.2026 · 16 (11)

DOI: 10.1063/5.0345275 ↗

Kurz gefasst

Fachartikel in der Zeitschrift AIP Advances, online erschienen am 16.11.2026. Verfasst von S. Saranya und 1 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The increasing complexity of image processing necessitates advanced Multiplier–Accumulator (MAC) units, which assure high computational throughput with minimum power, area, and latency. Traditional architectures for Finite Impulse Response (FIR) filters are plagued by large hardware utilization, limited scalability, and high power consumption due to inefficient multiplier–accumulator operations and fixed structural constraints. A novel Progressive Graph Convolution Triangulation Masked Topology Aggregation Attention Network (PGAT-MT2AN) is therefore proposed for the optimization of MAC units integrated with the design of an FIR filter. The proposed system employs a fault-correcting adder design and Toom–Cook-5 decomposition-based Montgomery modular multiplication for efficient handling of large operand multiplications with enhanced precision. The PGAT-MT2AN framework employs a progressive graph convolution module for adaptive structural learning, masked attention to perform focused computations, and a triangulation topology aggregation optimizer for parameter tuning to enhance the speed and reduce power consumption in MAC operations. Experimental analysis is performed with 51 taps, 229 slices, 181 flip-flops, 1 block RAM/FIFO, and 1 with a dynamic power of 10 mW and leakage power of 36 mW. The results show outstanding efficiency with a gain of 6.82 dB, area reduction of 92.2%, and delay minimization of 65.7%. Therefore, the MAC-FIR system with optimization has achieved better hardware efficiency in terms of the area used, delay time, and power dissipation, thus making it suitable for real-time image processing and other signal-processing applications. The benefits include fewer hardware complexities and improved energy efficiency for next-generation signal processing hardware.

Zitieren

APA

Saranya, S., & Joshitha, K. L. (2026). Progressive graph convolution triangulation masked topology aggregation attention network optimized multiply and accumulate unit in CNN accelerator for image processing applications. AIP Advances, 16(11). https://doi.org/10.1063/5.0345275

Vancouver

Saranya S, Joshitha KL. Progressive graph convolution triangulation masked topology aggregation attention network optimized multiply and accumulate unit in CNN accelerator for image processing applications. AIP Advances. 2026;16(11). doi:10.1063/5.0345275

BibTeX
@article{Saranya2026,
  author = {Saranya, S. and Joshitha, K. Lakshmi},
  title = {Progressive graph convolution triangulation masked topology aggregation attention network optimized multiply and accumulate unit in CNN accelerator for image processing applications},
  journal = {AIP Advances},
  year = {2026},
  volume = {16},
  number = {11},
  doi = {10.1063/5.0345275}
}
Kostenlos starten →