001050452 001__ 1050452
001050452 005__ 20260220104305.0
001050452 0247_ $$2doi$$a10.48550/ARXIV.2507.06079
001050452 0247_ $$2datacite_doi$$a10.34734/FZJ-2026-00222
001050452 037__ $$aFZJ-2026-00222
001050452 1001_ $$0P:(DE-Juel1)174486$$aSiegel, Sebastian$$b0$$eCorresponding author
001050452 245__ $$aQS4D: Quantization-aware training for efficient hardware deployment of structured state-space sequential models
001050452 260__ $$barXiv$$c2025
001050452 3367_ $$0PUB:(DE-HGF)25$$2PUB:(DE-HGF)$$aPreprint$$bpreprint$$mpreprint$$s1768997506_10779
001050452 3367_ $$2ORCID$$aWORKING_PAPER
001050452 3367_ $$028$$2EndNote$$aElectronic Article
001050452 3367_ $$2DRIVER$$apreprint
001050452 3367_ $$2BibTeX$$aARTICLE
001050452 3367_ $$2DataCite$$aOutput Types/Working Paper
001050452 520__ $$aStructured State Space models (SSM) have recently emerged as a new class of deep learning models, particularly well-suited for processing long sequences. Their constant memory footprint, in contrast to the linearly scaling memory demands of Transformers, makes them attractive candidates for deployment on resource-constrained edge-computing devices. While recent works have explored the effect of quantization-aware training (QAT) on SSMs, they typically do not address its implications for specialized edge hardware, for example, analog in-memory computing (AIMC) chips. In this work, we demonstrate that QAT can significantly reduce the complexity of SSMs by up to two orders of magnitude across various performance metrics. We analyze the relation between model size and numerical precision, and show that QAT enhances robustness to analog noise and enables structural pruning. Finally, we integrate these techniques to deploy SSMs on a memristive analog in-memory computing substrate and highlight the resulting benefits in terms of computational efficiency.
001050452 536__ $$0G:(DE-HGF)POF4-5234$$a5234 - Emerging NC Architectures (POF4-523)$$cPOF4-523$$fPOF IV$$x0
001050452 536__ $$0G:(DE-Juel1)BMBF-03ZU1106CB$$aBMBF 03ZU1106CB - NeuroSys: Algorithm-Hardware Co-Design (Projekt C) - B (BMBF-03ZU1106CB)$$cBMBF-03ZU1106CB$$x1
001050452 588__ $$aDataset connected to DataCite
001050452 650_7 $$2Other$$aMachine Learning (cs.LG)
001050452 650_7 $$2Other$$aArtificial Intelligence (cs.AI)
001050452 650_7 $$2Other$$aFOS: Computer and information sciences
001050452 7001_ $$0P:(DE-Juel1)192385$$aYang, Ming-Jay$$b1$$ufzj
001050452 7001_ $$0P:(DE-Juel1)176778$$aBouhadjar, Younes$$b2$$ufzj
001050452 7001_ $$0P:(DE-Juel1)201205$$aFabre, Maxime$$b3$$ufzj
001050452 7001_ $$0P:(DE-Juel1)188273$$aNeftci, Emre$$b4$$ufzj
001050452 7001_ $$0P:(DE-Juel1)188145$$aStrachan, John Paul$$b5$$ufzj
001050452 773__ $$a10.48550/ARXIV.2507.06079
001050452 8564_ $$uhttps://juser.fz-juelich.de/record/1050452/files/2507.06079v1.pdf$$yOpenAccess
001050452 909CO $$ooai:juser.fz-juelich.de:1050452$$pdnbdelivery$$pdriver$$pVDB$$popen_access$$popenaire
001050452 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)174486$$aForschungszentrum Jülich$$b0$$kFZJ
001050452 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)192385$$aForschungszentrum Jülich$$b1$$kFZJ
001050452 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)176778$$aForschungszentrum Jülich$$b2$$kFZJ
001050452 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)201205$$aForschungszentrum Jülich$$b3$$kFZJ
001050452 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)188273$$aForschungszentrum Jülich$$b4$$kFZJ
001050452 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)188145$$aForschungszentrum Jülich$$b5$$kFZJ
001050452 9131_ $$0G:(DE-HGF)POF4-523$$1G:(DE-HGF)POF4-520$$2G:(DE-HGF)POF4-500$$3G:(DE-HGF)POF4$$4G:(DE-HGF)POF$$9G:(DE-HGF)POF4-5234$$aDE-HGF$$bKey Technologies$$lNatural, Artificial and Cognitive Information Processing$$vNeuromorphic Computing and Network Dynamics$$x0
001050452 9141_ $$y2025
001050452 915__ $$0StatID:(DE-HGF)0510$$2StatID$$aOpenAccess
001050452 9201_ $$0I:(DE-Juel1)PGI-14-20210412$$kPGI-14$$lNeuromorphic Compute Nodes$$x0
001050452 9201_ $$0I:(DE-Juel1)PGI-15-20210701$$kPGI-15$$lNeuromorphic Software Eco System$$x1
001050452 9801_ $$aFullTexts
001050452 980__ $$apreprint
001050452 980__ $$aVDB
001050452 980__ $$aUNRESTRICTED
001050452 980__ $$aI:(DE-Juel1)PGI-14-20210412
001050452 980__ $$aI:(DE-Juel1)PGI-15-20210701