161798 20251017144639.0 doi 10.1109/ICCD63220.2024.00024 sideral 144431 ART-2025-144431 eng Toca-Díaz, Yamilka Universidad de Zaragoza Ensuring the Accuracy of CNN Accelerators Supplied at Ultra-Low Voltage 2025 Access copy available to the general public Unrestricted Underscaling the supply voltage (Vdd) to ultra-low levels below the safe-operation threshold voltage (Vmin) brings significant energy savings in digital CMOS circuits but introduces reliability challenges due to increased risk of bitcell permanent faults. This work explores the impact of such faults on the accuracy of a CNN inference accelerator supplying on-chip activation memories at ultra-low Vdd. By examining fault pat-terns, activation values, and memory usage, this paper proposes two microarchitectural techniques exploiting activation outliers and activation memory underutilization. These approaches are cost-effective, do not require programmer intervention, and are application-independent. Experimental results show that the proposed approaches maintain the original CNN accuracy and achieve energy savings by 2.1 % and 8.2 % compared to the state-of-the-art technique and a conventional accelerator supplied at Vmin, respectively, with a negligible impact on the system performance (less than 0.25 %). info:eu-repo/grantAgreement/ES/AEI/PID2022-136454NB-C22 info:eu-repo/grantAgreement/ES/DGA/T58-23R info:eu-repo/semantics/openAccess All rights reserved http://www.europeana.eu/rights/rr-f/ info:eu-repo/semantics/article info:eu-repo/semantics/acceptedVersion Gran Tejero, Rubén Universidad de Zaragoza (orcid)0000-0002-4031-5651 Valero, Alejandro Universidad de Zaragoza (orcid)0000-0002-0824-5833 5007 035 Universidad de Zaragoza Dpto. Informát.Ingenie.Sistms. Área Arquit.Tecnología Comput. (2025), 92-95 Proceedings - IEEE International Conference on Computer Design 1063-6404 226527 http://zaguan.unizar.es/record/161798/files/texto_completo.pdf Postprint 3108595 http://zaguan.unizar.es/record/161798/files/texto_completo.jpg?subformat=icon icon Postprint oai:zaguan.unizar.es:161798 articulos driver 2025-10-17-14:31:10 ARTICLE