{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,8,2]],"date-time":"2025-08-02T19:41:24Z","timestamp":1754163684093,"version":"3.41.2"},"reference-count":50,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2025,7,31]],"date-time":"2025-07-31T00:00:00Z","timestamp":1753920000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Comput. Neurosci."],"abstract":"<jats:p>The energy demands of modern AI systems have reached unprecedented levels, driven by the rapid scaling of deep learning models, including large language models, and the inefficiencies of current computational architectures. In contrast, biological neural systems operate with remarkable energy efficiency, achieving complex computations while consuming orders of magnitude less power. A key mechanism enabling this efficiency is subthreshold processing, where neurons perform computations through graded, continuous signals below the spiking threshold, reducing energy costs. Despite its significance in biological systems, subthreshold processing remains largely overlooked in AI design. This perspective explores how principles of subthreshold dynamics can inspire the design of novel AI architectures and computational methods as a step toward advancing TinyAI. We propose pathways such as algorithmic analogs of subthreshold integration, including graded activation functions, dendritic-inspired hierarchical processing, and hybrid analog-digital systems to emulate the energy-efficient operations of biological neurons. We further explore neuromorphic and compute-in-memory hardware platforms that could support these operations, and propose a design stack aligned with the efficiency and adaptability of the brain. By integrating subthreshold dynamics into AI architecture, this work provides a roadmap toward sustainable, responsive, and accessible intelligence for resource-constrained environments.<\/jats:p>","DOI":"10.3389\/fncom.2025.1638782","type":"journal-article","created":{"date-parts":[[2025,7,31]],"date-time":"2025-07-31T10:05:48Z","timestamp":1753956348000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Exploring subthreshold processing for next-generation TinyAI"],"prefix":"10.3389","volume":"19","author":[{"given":"Farid","family":"Nakhle","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Antoine H.","family":"Harfouche","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Hani","family":"Karam","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Vasileios","family":"Tserolas","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"1965","published-online":{"date-parts":[[2025,7,31]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"7958","DOI":"10.1038\/s41467-024-51390-x","article-title":"Sub-threshold neuronal activity and the dynamical regime of cerebral cortex.","volume":"15","author":"Amsalem","year":"2024","journal-title":"Nat. Commun."},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.1038\/d41586-024-03408-z","article-title":"Fixing AI\u2019s energy crisis.","author":"Bourzac","year":"2024","journal-title":"Nature"},{"key":"B3","doi-asserted-by":"publisher","first-page":"e1002867","DOI":"10.1371\/journal.pcbi.1002867","article-title":"Passive dendrites enable single neurons to compute linearly non-separable functions.","volume":"9","author":"Caz\u00e9","year":"2013","journal-title":"PLoS Comput. Biol."},{"key":"B4","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2306.04615","article-title":"JOSS: Joint exploration of CPU-memory DVFS and task scheduling for energy efficiency.","author":"Chen","year":"2023","journal-title":"arXiv [Preprint]"},{"key":"B5","doi-asserted-by":"publisher","first-page":"8514","DOI":"10.1523\/JNEUROSCI.1503-22.2022","article-title":"Recent advances at the interface of neuroscience and artificial neural networks.","volume":"42","author":"Cohen","year":"2022","journal-title":"J. Neurosci."},{"key":"B6","doi-asserted-by":"publisher","first-page":"693","DOI":"10.1038\/d41586-024-00478-x","article-title":"Generative AI\u2019s environmental costs are soaring \u2014 and mostly secret.","volume":"626","author":"Crawford","year":"2024","journal-title":"Nature"},{"key":"B7","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2107.12374","article-title":"Training energy-efficient deep spiking neural networks with single-spike hybrid input encoding.","author":"Datta","year":"2021","journal-title":"arXiv [Preprint]"},{"key":"B8","doi-asserted-by":"publisher","first-page":"82","DOI":"10.1109\/MM.2018.112130359","article-title":"Loihi: A neuromorphic manycore processor with on-chip learning.","volume":"38","author":"Davies","year":"2018","journal-title":"IEEE Micro"},{"key":"B9","doi-asserted-by":"publisher","first-page":"2191","DOI":"10.1016\/j.joule.2023.09.004","article-title":"The growing energy footprint of artificial intelligence.","volume":"7","author":"De Vries","year":"2023","journal-title":"Joule"},{"key":"B10","doi-asserted-by":"publisher","first-page":"1901408","DOI":"10.1002\/aelm.201901408","article-title":"Dual-gated MoS2 transistors for synaptic and programmable logic functions.","volume":"6","author":"Du","year":"2020","journal-title":"Adv Elect Materials"},{"key":"B11","doi-asserted-by":"publisher","first-page":"463","DOI":"10.1038\/s41928-023-01014-x","article-title":"AI hardware has an energy problem.","volume":"6","year":"2023","journal-title":"Nat. Electron."},{"key":"B12","doi-asserted-by":"publisher","first-page":"35","DOI":"10.1145\/3207719.3207727","article-title":"Less is more: Exploiting the standard compiler optimization levels for better performance and energy consumption","author":"Georgiou","year":"2018","journal-title":"Proceedings of the 21st International Workshop on Software and Compilers for Embedded Systems"},{"key":"B13","doi-asserted-by":"publisher","first-page":"10","DOI":"10.1109\/ISSCC.2014.6757323","article-title":"Computing\u2019s energy problem (and what we can do about it","author":"Horowitz","year":"2014","journal-title":"Proceedings of the 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC)"},{"key":"B14","doi-asserted-by":"publisher","first-page":"4449","DOI":"10.1523\/JNEUROSCI.1391-21.2022","article-title":"Visual stimulation induces distinct forms of sensitization of on-off direction-selective ganglion cell responses in the dorsal and ventral retina.","volume":"42","author":"Huang","year":"2022","journal-title":"J. Neurosci."},{"key":"B15","doi-asserted-by":"publisher","first-page":"216","DOI":"10.1016\/S0166-2236(00)01547-2","article-title":"Resonance, oscillation and the intrinsic frequency preferences of neurons.","volume":"23","author":"Hutcheon","year":"2000","journal-title":"Trends Neurosci."},{"key":"B16","doi-asserted-by":"publisher","first-page":"955","DOI":"10.1038\/nrn1790","article-title":"Exuberance in the development of cortical networks.","volume":"6","author":"Innocenti","year":"2005","journal-title":"Nat. Rev. Neurosci."},{"key":"B17","doi-asserted-by":"publisher","first-page":"1171","DOI":"10.1142\/S0218127400000840","article-title":"Neural excitability, spiking and bursting.","volume":"10","author":"Izhikevich","year":"2000","journal-title":"Int. J. Bifurcation Chaos"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2502.14023","article-title":"Dynamic activation with knowledge distillation for energy-efficient spiking NN ensembles.","author":"Konstantaropoulos","year":"2025","journal-title":"arXiv [Preprint]"},{"key":"B19","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2304.03271","article-title":"Making AI less \u201cThirsty\u201d: Uncovering and addressing the secret water footprint of AI models.","author":"Li","year":"2023","journal-title":"arXiv [Preprint]"},{"key":"B20","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2409.11416","article-title":"The unseen AI disruptions for power grids: LLM-induced transients.","author":"Li","year":"2024","journal-title":"arXiv [Preprint]"},{"key":"B21","doi-asserted-by":"publisher","first-page":"7140","DOI":"10.1038\/s41467-023-42981-1","article-title":"Monolithic three-dimensional integration of RRAM-based hybrid memory architecture for one-shot learning.","volume":"14","author":"Li","year":"2023","journal-title":"Nat. Commun."},{"key":"B22","doi-asserted-by":"publisher","first-page":"1654","DOI":"10.1126\/science.3059497","article-title":"The intrinsic electrophysiological properties of mammalian neurons: Insights into central nervous system function.","volume":"242","author":"Llin\u00e1s","year":"1988","journal-title":"Science"},{"key":"B23","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1109\/NEWCAS50681.2021.9462779","article-title":"Subthreshold neuromorphic devices for Spiking Neural Networks applied to embedded A.I","author":"Loyez","year":"2021","journal-title":"Proceedings of the 2021 19th IEEE International New Circuits and Systems Conference (NEWCAS)"},{"key":"B24","doi-asserted-by":"publisher","first-page":"563","DOI":"10.1038\/nrn1949","article-title":"Variability, compensation and homeostasis in neuron and network function.","volume":"7","author":"Marder","year":"2006","journal-title":"Nat. Rev. Neurosci."},{"key":"B25","doi-asserted-by":"publisher","first-page":"11256","DOI":"10.1109\/CVPR.2019.01152","article-title":"Importance estimation for neural network pruning","author":"Molchanov","year":"2019","journal-title":"Proceedings of the 2019 IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR)"},{"key":"B26","doi-asserted-by":"publisher","first-page":"11","DOI":"10.1038\/s41545-021-00101-w","article-title":"Data centre water consumption.","volume":"4","author":"Mytton","year":"2021","journal-title":"NPJ Clean Water"},{"key":"B27","doi-asserted-by":"publisher","first-page":"100411","DOI":"10.1016\/j.device.2024.100411","article-title":"Shrinking the giants: Paving the way for TinyAI.","volume":"2","author":"Nakhle","year":"2024","journal-title":"Device"},{"key":"B28","doi-asserted-by":"publisher","first-page":"83","DOI":"10.1038\/s41699-024-00522-4","article-title":"Advancements in 2D layered material memristors: Unleashing their potential beyond memory.","volume":"8","author":"Nirmal","year":"2024","journal-title":"NPJ 2D Mater. Appl."},{"key":"B29","doi-asserted-by":"publisher","first-page":"280","DOI":"10.1016\/j.neuron.2021.10.024","article-title":"Neocortex saves energy by reducing coding precision during food scarcity.","volume":"110","author":"Padamsey","year":"2022","journal-title":"Neuron"},{"key":"B30","doi-asserted-by":"publisher","first-page":"227","DOI":"10.1038\/s41570-023-00480-3","article-title":"Aiming beyond slight increases in accuracy.","volume":"7","author":"Probst","year":"2023","journal-title":"Nat. Rev. Chem."},{"key":"B31","doi-asserted-by":"publisher","first-page":"579","DOI":"10.1084\/jem.76.6.579","article-title":"The metabolism of the central nervous system in experimental poliomyelitis.","volume":"76","author":"Racker","year":"1942","journal-title":"J. Exp. Med."},{"key":"B32","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1710.05941","article-title":"Searching for activation functions.","author":"Ramachandran","year":"2017","journal-title":"arXiv [Preprint]"},{"key":"B33","doi-asserted-by":"publisher","first-page":"413","DOI":"10.1007\/978-3-319-58667-0_22","article-title":"Metrics for energy-aware software optimisation","author":"Roberts","year":"2017","journal-title":"High performance computing"},{"key":"B34","doi-asserted-by":"publisher","first-page":"10","DOI":"10.1038\/s43588-021-00184-y","article-title":"Opportunities for neuromorphic computing algorithms and applications.","volume":"2","author":"Schuman","year":"2022","journal-title":"Nat. Comput. Sci."},{"key":"B35","doi-asserted-by":"publisher","first-page":"54","DOI":"10.1145\/3381831","article-title":"Green AI.","volume":"63","author":"Schwartz","year":"2020","journal-title":"Commun. ACM"},{"key":"B36","doi-asserted-by":"publisher","DOI":"10.21203\/rs.3.rs-2264132\/v1","article-title":"Adaptive programmable networks for in materia neuromorphic computing.","author":"Stenning","year":"2022","journal-title":"arXiv [Preprint]"},{"key":"B37","doi-asserted-by":"publisher","first-page":"7377","DOI":"10.1038\/s41467-024-50633-1","article-title":"Neuromorphic overparameterisation and few-shot learning in multilayer physical neural networks.","volume":"15","author":"Stenning","year":"2024","journal-title":"Nat. Commun."},{"key":"B38","doi-asserted-by":"publisher","DOI":"10.7551\/mitpress\/9780262028707.001.0001","author":"Sterling","year":"2015","journal-title":"Principles of neural design."},{"key":"B39","doi-asserted-by":"publisher","first-page":"13693","DOI":"10.1609\/aaai.v34i09.7123","article-title":"Energy and policy considerations for modern deep learning research.","volume":"34","author":"Strubell","year":"2020","journal-title":"AAAI"},{"key":"B40","doi-asserted-by":"publisher","first-page":"627","DOI":"10.1007\/s11265-024-01930-x","article-title":"Energy-aware register allocation for VLIW processors.","volume":"96","author":"Stuckmann","year":"2024","journal-title":"J. Sign. Process. Syst."},{"key":"B41","doi-asserted-by":"publisher","first-page":"10520","DOI":"10.1073\/pnas.0601974103","article-title":"Attentional modulation of sensorimotor processes in the absence of perceptual awareness.","volume":"103","author":"Sumner","year":"2006","journal-title":"Proc. Natl. Acad. Sci. U.S.A."},{"key":"B42","doi-asserted-by":"publisher","first-page":"68","DOI":"10.1038\/s41699-023-00422-z","article-title":"Ultra-low power neuromorphic obstacle detection using a two-dimensional materials-based subthreshold transistor.","volume":"7","author":"Thakar","year":"2023","journal-title":"NPJ 2D Mater. Appl."},{"key":"B43","doi-asserted-by":"publisher","first-page":"1043","DOI":"10.1057\/s41599-024-03520-5","article-title":"Ecological footprints, carbon emissions, and energy transitions: The impact of artificial intelligence (AI).","volume":"11","author":"Wang","year":"2024","journal-title":"Hum. Soc. Sci. Commun."},{"key":"B44","doi-asserted-by":"publisher","first-page":"4671","DOI":"10.1038\/s41467-024-48908-8","article-title":"Memristor-based adaptive neuromorphic perception in unstructured environments.","volume":"15","author":"Wang","year":"2024","journal-title":"Nat. Commun."},{"key":"B45","doi-asserted-by":"publisher","first-page":"123","DOI":"10.1038\/s41467-021-27774-8","article-title":"An optical neural network using less than 1 photon per multiplication.","volume":"13","author":"Wang","year":"2022","journal-title":"Nat. Commun."},{"key":"B46","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.2503.16183","article-title":"Variance-aware noisy training: Hardening DNNs against unstable analog computations.","author":"Wang","year":"2025","journal-title":"arXiv [Preprint]"},{"key":"B47","doi-asserted-by":"publisher","first-page":"641","DOI":"10.1038\/s41586-020-1942-4","article-title":"Fully hardware-implemented memristor convolutional neural network.","volume":"577","author":"Yao","year":"2020","journal-title":"Nature"},{"key":"B48","doi-asserted-by":"publisher","first-page":"1597","DOI":"10.1038\/s41467-023-37180-x","article-title":"Catalyzing next-generation artificial intelligence through NeuroAI.","volume":"14","author":"Zador","year":"2023","journal-title":"Nat. Commun."},{"key":"B49","doi-asserted-by":"publisher","first-page":"034002","DOI":"10.1088\/2634-4386\/ace64c","article-title":"Brain-inspired methods for achieving robust computation in heterogeneous mixed-signal neuromorphic processing systems.","volume":"3","author":"Zendrikov","year":"2023","journal-title":"Neuromorph. Comput. Eng."},{"key":"B50","doi-asserted-by":"publisher","first-page":"1867","DOI":"10.1109\/TCSI.2020.2971642","article-title":"A Robust 8-Bit non-volatile computing-in-memory core for low-power parallel MAC operations.","volume":"67","author":"Zhang","year":"2020","journal-title":"IEEE Trans. Circuits Syst. I"}],"container-title":["Frontiers in Computational Neuroscience"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fncom.2025.1638782\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,7,31]],"date-time":"2025-07-31T10:05:50Z","timestamp":1753956350000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fncom.2025.1638782\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,7,31]]},"references-count":50,"alternative-id":["10.3389\/fncom.2025.1638782"],"URL":"https:\/\/doi.org\/10.3389\/fncom.2025.1638782","relation":{},"ISSN":["1662-5188"],"issn-type":[{"type":"electronic","value":"1662-5188"}],"subject":[],"published":{"date-parts":[[2025,7,31]]},"article-number":"1638782"}}