How Much Quality Loss is Acceptable? Evaluating Semantic Acceptability and Energy Gains of Approximate Memory in a RISC-V Environment
Resumo
Approximate memories reduce energy consumption through non-reliable operation, but the resulting faults may compromise the usefulness of application outputs. This non-reliable operation may be tolerated up to a point that depends on application context and data types. However, uncontrolled errors may lead to execution crashes or unacceptable results, which reduce the energy gains. Defining whether a result is acceptable or not depends on the context of a complete application pipeline. This work proposes AxCept-Bench, a systematic interface for characterizing operating regions for different applications considering execution stability and output validation based on semantic acceptability. AxCept-Bench integrates the approximation source, the application pipeline and an acceptability validation of the results. Our evaluation results encompass a simulation based on memory approximation that defines acceptability datasets for the applications JPEG encoding and FFT. The JPEG application achieves energy savings of 37.01% with 90.3% more acceptable results on the pipeline than with simple error metrics, while FFT accepts 44.18% more results with 57.38% more energy savings. Therefore, energy gains must be interpreted jointly with execution validity and the semantic requirements of each application to potentialize the results with approximation.
Referências
Borgerding, M. (2003). KISS FFT. [link]. Software library, accessed July 2026.
Brumme, S. (2019). Toojpeg: A jpeg encoder in a single c++ file. Available at: [link]. Accessed: Jun. 2026.
Chandrasekar, K., Weis, C., Li, Y., Goossens, S., Jung, M., Naji, O., Akesson, B., Wehn, N., and Goossens, K. (n.d.). DRAMPower: Open-source DRAM power and energy estimation tool. [link]. Accessed: July 2026.
Chang, K. K., Yağlıkçı, A. G., Ghose, S., Agrawal, A., Chatterjee, N., Kashyap, A., Lee, D., O’Connor, M., Hassan, H., and Mutlu, O. (2017). Understanding reduced-voltage operation in modern dram devices: Experimental characterization, analysis, and mechanisms. Proc. ACM Meas. Anal. Comput. Syst., 1(1).
Fabrício Filho, J., Felzmann, I., and Wanner, L. (2022). Smartapprox: Learning-based configuration of approximate memories for energy-efficient execution. Sustainable Computing: Informatics and Systems, 34:100701.
Fabrício Filho, J., Felzmann, I. B., Azevedo, R. J., and Wanner, L. F. (2019). A resilient interface for approximate data access. In 2019 IX Brazilian Symposium on Computing Systems Engineering (SBESC), pages 1–8.
Felzmann, I., Filho, J. F., de Oliveira, J. R., and Wanner, L. (2021a). Special session: How much quality is enough quality? a case for acceptability in approximate designs. In 2021 IEEE 39th International Conference on Computer Design (ICCD), pages 5–8.
Felzmann, I., Filho, J. F., and Wanner, L. (2021b). AxPIKE: Instruction-level injection and evaluation of approximate computing. In 2021 Design, Automation Test in Europe Conference Exhibition (DATE), pages 491–494.
Fritts, L. (1997). University of iowa musical instrument samples. University of Iowa Electronic Music Studios, [link]. Accessed: July 2026.
He, K., Zhang, X., Ren, S., and Sun, J. (2016). Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778.
Khudia, D. S., Zamirai, B., Samadi, M., and Mahlke, S. (2015). Rumba: an online quality management system for approximate computing. SIGARCH Comput. Archit. News, 43(3S):554–566.
Kim, Y., Yang, W., and Mutlu, O. (2016). Ramulator: A fast and extensible dram simulator. IEEE Computer Architecture Letters, 15(1):45–49.
Mittal, S. (2016). A survey of techniques for approximate computing. ACM Comput. Surv., 48(4).
Serbai, G., Gonçalves, R., and Filho, J. F. (2025). Um estudo sobre aceitabilidade de resultados de sistemas aproximados em redes neurais artificiais. In Anais da XVI Escola Regional de Alto Desempenho de São Paulo, pages 50–53, Porto Alegre, RS, Brasil. SBC.
Yang, Y. and Newsam, S. (2010). Bag-of-visual-words and spatial extensions for land- use classification. In Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems, GIS ’10, pages 270–279, New York, NY, USA. Association for Computing Machinery.
Yarmand, R., Kamal, M., Afzali-Kusha, A., Esmaeli, P., and Pedram, M. (2021). Optima: An approach for online management of cache approximation levels in approximate processing systems. IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 29(2):434–446.
