ReFloat: Low-Cost Floating-Point Processing in ReRAM for Accelerating Iterative Linear Solvers. Song, L., <a href="https://homes.luddy.indiana.edu/fc7/" target="_bilank">Fan Chen</a></span>, Li, H., & Chen, Y. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC), pages 75:1–75:15, 2023. ACM. ![ReFloat: Low-Cost Floating-Point Processing in ReRAM for Accelerating Iterative Linear Solvers [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex 1 download Resistive random access memory (ReRAM) is a promising technology that can perform low-cost and in-situ matrix-vector multiplication (MVM) in analog domain. Scientific computing requires high-precision floating-point (FP) processing. However, performing floating-point computation in ReRAM is challenging because of high hardware cost and execution time due to the large FP value range. In this work we present ReFloat, a data format and an accelerator architecture, for low-cost and high-performance floating-point processing in ReRAM for iterative linear solvers. ReFloat matches the ReRAM crossbar hardware and represents a block of FP values with reduced bits and an optimized exponent base for a high range of dynamic representation. Thus, ReFloat achieves less ReRAM crossbar consumption and fewer processing cycles and overcomes the noncovergence issue in a prior work. The evaluation on the SuiteSparse matrices shows ReFloat achieves 5.02× to 84.28× improvement in terms of solver time compared to a state-of-the-art ReRAM based accelerator.
@inproceedings{SC2023,
author = {Linghao {Song} and {<a href="https://homes.luddy.indiana.edu/fc7/" target="_bilank">Fan Chen</a></span>} and Hai {Li} and Yiran Chen},
title = {ReFloat: Low-Cost Floating-Point Processing in ReRAM for Accelerating Iterative Linear Solvers},
booktitle = {Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC)},
pages = {75:1--75:15},
publisher = {{ACM}},
year = {2023},
url = {https://doi.org/10.1145/3581784.3607077},
doi = {10.1145/3581784.3607077},
abstract = {Resistive random access memory (ReRAM) is a promising technology that can perform low-cost and in-situ matrix-vector multiplication (MVM) in analog domain. Scientific computing requires high-precision floating-point (FP) processing. However, performing floating-point computation in ReRAM is challenging because of high hardware cost and execution time due to the large FP value range. In this work we present ReFloat, a data format and an accelerator architecture, for low-cost and high-performance floating-point processing in ReRAM for iterative linear solvers. ReFloat matches the ReRAM crossbar hardware and represents a block of FP values with reduced bits and an optimized exponent base for a high range of dynamic representation. Thus, ReFloat achieves less ReRAM crossbar consumption and fewer processing cycles and overcomes the noncovergence issue in a prior work. The evaluation on the SuiteSparse matrices shows ReFloat achieves 5.02\texttimes{} to 84.28\texttimes{} improvement in terms of solver time compared to a state-of-the-art ReRAM based accelerator.},
}
Downloads: 1
{"_id":"dLPHo9so5CRZekFXB","bibbaseid":"song-ahrefhttpshomesluddyindianaedufc7targetbilankfanchenaspan-li-chen-refloatlowcostfloatingpointprocessinginreramforacceleratingiterativelinearsolvers-2023","author_short":["Song, L.","<a href=\"https://homes.luddy.indiana.edu/fc7/\" target=\"_bilank\">Fan Chen</a></span>","Li, H.","Chen, Y."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","author":[{"firstnames":["Linghao"],"propositions":[],"lastnames":["Song"],"suffixes":[]},{"firstnames":[],"propositions":[],"lastnames":["<a href=\"https://homes.luddy.indiana.edu/fc7/\" target=\"_bilank\">Fan Chen</a></span>"],"suffixes":[]},{"firstnames":["Hai"],"propositions":[],"lastnames":["Li"],"suffixes":[]},{"firstnames":["Yiran"],"propositions":[],"lastnames":["Chen"],"suffixes":[]}],"title":"ReFloat: Low-Cost Floating-Point Processing in ReRAM for Accelerating Iterative Linear Solvers","booktitle":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC)","pages":"75:1–75:15","publisher":"ACM","year":"2023","url":"https://doi.org/10.1145/3581784.3607077","doi":"10.1145/3581784.3607077","abstract":"Resistive random access memory (ReRAM) is a promising technology that can perform low-cost and in-situ matrix-vector multiplication (MVM) in analog domain. Scientific computing requires high-precision floating-point (FP) processing. However, performing floating-point computation in ReRAM is challenging because of high hardware cost and execution time due to the large FP value range. In this work we present ReFloat, a data format and an accelerator architecture, for low-cost and high-performance floating-point processing in ReRAM for iterative linear solvers. ReFloat matches the ReRAM crossbar hardware and represents a block of FP values with reduced bits and an optimized exponent base for a high range of dynamic representation. Thus, ReFloat achieves less ReRAM crossbar consumption and fewer processing cycles and overcomes the noncovergence issue in a prior work. The evaluation on the SuiteSparse matrices shows ReFloat achieves 5.02× to 84.28× improvement in terms of solver time compared to a state-of-the-art ReRAM based accelerator.","bibtex":"@inproceedings{SC2023,\n author = {Linghao {Song} and {<a href=\"https://homes.luddy.indiana.edu/fc7/\" target=\"_bilank\">Fan Chen</a></span>} and Hai {Li} and Yiran Chen},\n title = {ReFloat: Low-Cost Floating-Point Processing in ReRAM for Accelerating Iterative Linear Solvers},\n booktitle = {Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC)},\n pages = {75:1--75:15},\n publisher = {{ACM}},\n year = {2023},\n url = {https://doi.org/10.1145/3581784.3607077},\n doi = {10.1145/3581784.3607077},\n abstract = {Resistive random access memory (ReRAM) is a promising technology that can perform low-cost and in-situ matrix-vector multiplication (MVM) in analog domain. Scientific computing requires high-precision floating-point (FP) processing. However, performing floating-point computation in ReRAM is challenging because of high hardware cost and execution time due to the large FP value range. In this work we present ReFloat, a data format and an accelerator architecture, for low-cost and high-performance floating-point processing in ReRAM for iterative linear solvers. ReFloat matches the ReRAM crossbar hardware and represents a block of FP values with reduced bits and an optimized exponent base for a high range of dynamic representation. Thus, ReFloat achieves less ReRAM crossbar consumption and fewer processing cycles and overcomes the noncovergence issue in a prior work. The evaluation on the SuiteSparse matrices shows ReFloat achieves 5.02\\texttimes{} to 84.28\\texttimes{} improvement in terms of solver time compared to a state-of-the-art ReRAM based accelerator.},\n}\n\n","author_short":["Song, L.","<a href=\"https://homes.luddy.indiana.edu/fc7/\" target=\"_bilank\">Fan Chen</a></span>","Li, H.","Chen, Y."],"key":"SC2023","id":"SC2023","bibbaseid":"song-ahrefhttpshomesluddyindianaedufc7targetbilankfanchenaspan-li-chen-refloatlowcostfloatingpointprocessinginreramforacceleratingiterativelinearsolvers-2023","role":"author","urls":{"Paper":"https://doi.org/10.1145/3581784.3607077"},"metadata":{"authorlinks":{}},"downloads":1},"bibtype":"inproceedings","biburl":"https://homes.luddy.indiana.edu/fc7/fan-publication.bib","dataSources":["SzsArXknrJasD2EPT"],"keywords":[],"search_terms":["refloat","low","cost","floating","point","processing","reram","accelerating","iterative","linear","solvers","song","<a href=\"https://homes.luddy.indiana.edu/fc7/\" target=\"_bilank\">fan chen</a></span>","li","chen"],"title":"ReFloat: Low-Cost Floating-Point Processing in ReRAM for Accelerating Iterative Linear Solvers","year":2023,"downloads":1}