2021/03/15 by Febin Sunny, Sunny, Febin, Asif Mirza +7
Computer Science · Engineering · #Distributed #Emerging Technologies (cs.ET) #FOS: Computer and information sciences #Hardware Architecture (cs.AR) #Neural Networks and Reservoir Computing #Optical Network Technologies #Parallel #Photonic and Optical Devices #and Cluster Computing (cs.DC)
paper · pdf · doi:10.48550/arxiv.2103.08828
openalex publication_date 2021/03/15 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
The approximate computing paradigm advocates for relaxing accuracy goals in\napplications to improve energy-efficiency and performance. Recently, this\nparadigm has been explored to improve the energy-efficiency of silicon photonic\nnetworks-on-chip (PNoCs). Silicon photonic interconnects suffer from high power\ndissipation because of laser sources, which generate carrier wavelengths, and\ntuning power required for regulating photonic devices under different\nuncertainties. In this paper, we propose a framework called ARXON to reduce\nsuch power dissipation overhead by enabling intelligent and aggressive\napproximation during communication over silicon photonic links in PNoCs. Our\nframework reduces laser and tuning-power overhead while intelligently\napproximating communication, such that application output quality is not\ndistorted beyond an acceptable limit. Simulation results show that our\nframework can achieve up to 56.4% lower laser power consumption and up to 23.8%\nbetter energy-efficiency than the best-known prior work on approximate\ncommunication with silicon photonic interconnects and for the same application\noutput quality.\n