Publications

2026

1. NCE

   Energy-aware spike budgeting for continual learning in spiking neural networks for neuromorphic vision

   Anika Tabassum Meem, Muntasir Hossain Nadid, and Md Zesun Ahmed Mia

   Neuromorphic Computing and Engineering, 2026

   Abs DOI arXiv Bib PDF

   Neuromorphic vision systems based on spiking neural networks (SNNs) offer event-driven sparse computation for event-based and frame-based cameras, yet catastrophic forgetting remains a critical barrier to deployment in continually evolving environments. Existing continual learning methods, developed primarily for artificial neural networks, seldom jointly optimize accuracy and activity-dependent spike cost, with particularly limited exploration on event-based datasets. We propose an energy-aware spike budgeting framework for continual SNN learning that integrates experience replay, learnable leaky integrate-and-fire neuron parameters, and an adaptive spike-budget controller to enforce dataset-specific spike-activity constraints during training. Our approach exhibits modality-dependent behavior: on frame-based datasets (MNIST, CIFAR-10), spike budgeting acts as a sparsity-inducing regularizer, improving accuracy while reducing spike rates by up to 47%; on event-based datasets (DVS-Gesture, N-MNIST, CIFAR-10-DVS), controlled budget relaxation enables accuracy gains up to 17.45% points with minimal computational overhead. Across five benchmarks spanning both modalities, our method demonstrates consistent performance improvements while keeping spike activity under explicit control, supporting practical continual learning under an activity-driven synaptic-event proxy rather than a direct hardware-energy measurement.

   @article{meem2026energy,
     title = {Energy-aware spike budgeting for continual learning in spiking neural networks for neuromorphic vision},
     author = {Meem, Anika Tabassum and Nadid, Muntasir Hossain and Mia, Md Zesun Ahmed},
     journal = {Neuromorphic Computing and Engineering},
     volume = {6},
     number = {3},
     pages = {034004},
     year = {2026},
     publisher = {IOP Publishing},
     doi = {10.1088/2634-4386/ae8627},
     url = {https://doi.org/10.1088/2634-4386/ae8627},
   }

2. arXiv

   Trilinear Compute-in-Memory Architecture for Energy-Efficient Transformer Acceleration

   Md Zesun Ahmed Mia, Jiahui Duan, Kai Ni, and 1 more author

   arXiv preprint arXiv:2604.07628, 2026

   Abs arXiv Bib PDF

   Transformer acceleration on Compute-in-Memory (CIM) hardware is bottlenecked by repeated runtime non-volatile memory (NVM) reprogramming of the dynamic attention operands (Q, K, V). We introduce TrilinearCIM, a Double-Gate FeFET (DG-FeFET)-based CIM architecture that performs the full attention dataflow in-memory without runtime ferroelectric reprogramming. A novel three-operand multiply-accumulate primitive (Y = A * B * C), enabled by DG-FeFET back-gate modulation, maps static projection weights to the non-volatile top gate while applying dynamic token activations through the volatile back gate. To our knowledge, this is the first architecture to perform complete Transformer attention computation exclusively in NVM cores without runtime reprogramming. TrilinearCIM reduces global buffer capacity requirements by  3x by eliminating intermediate Q/K matrix storage. Evaluated on BERT-base (GLUE) and ViT-base (ImageNet, CIFAR-10/100), it outperforms conventional FeFET CIM on 7 of 9 GLUE tasks, achieving up to 46.6% energy reduction and 20.4% latency improvement at 37.3% area overhead.

   @article{mia2026trilinear,
     title = {Trilinear Compute-in-Memory Architecture for Energy-Efficient Transformer Acceleration},
     author = {Mia, Md Zesun Ahmed and Duan, Jiahui and Ni, Kai and Sengupta, Abhronil},
     journal = {arXiv preprint arXiv:2604.07628},
     year = {2026},
     url = {https://arxiv.org/abs/2604.07628},
   }

3. ICLR

   RMAAT: Astrocyte-Inspired Memory Compression and Replay for Efficient Long-Context Transformers

   Md Zesun Ahmed Mia, Malyaban Bal, and Abhronil Sengupta

   In International Conference on Learning Representations (ICLR), 2026

   Abs arXiv Bib PDF

   Astrocytes, an essential component of the brain’s neural circuitry, demonstrate learning capabilities through bioplausible mechanisms such as presynaptic plasticity and hebbian plasticity. However, their integration into computational models remains underexplored. This paper advances astromorphic computing techniques to emulate transformer self-attention mechanisms, leveraging astrocytic nonlinearity and memory retention to improve long-range dependency processing in machine learning and natural language processing (NLP) tasks. Existing transformer models have difficulty handling lengthy contexts with thousands of tokens, even with substantial computational resources. We propose Recurrent Memory Augmented Astromorphic Transformers (RMAAT), integrating astrocytic memory and recurrent processing into self-attention, enabling longer context handling without quadratic complexity growth. Our bioplausible model has been found to outperform traditional transformers in experimental tests conducted on the Long Range Arena benchmark and IMDB dataset. Specifically, our model achieves a significant reduction in memory utilization and computational latency. This paves the way for biologically inspired AI models by illustrating how astrocytic characteristics may enhance the performance and efficiency of computational models.

   @inproceedings{mia2026rmaat,
     title = {{RMAAT}: Astrocyte-Inspired Memory Compression and Replay for Efficient Long-Context Transformers},
     author = {Mia, Md Zesun Ahmed and Bal, Malyaban and Sengupta, Abhronil},
     booktitle = {International Conference on Learning Representations (ICLR)},
     year = {2026},
     url = {https://openreview.net/forum?id=sTkJdbVxsI},
   }

4. EPEPS

   Bayesian Optimization of Crossbar-Based Compute-In-Memory System Design for Efficient DNN Inference

   Arnob Saha, Bibhas Manna, Nikhil Kotikalapudi, and 4 more authors

   In IEEE Conference on Electrical Performance of Electronic Packaging and Systems (EPEPS), 2026

   Abs arXiv Bib PDF

   Leveraging the high density and energy efficiency of Compute-In-Memory (CIM) crossbar-based Deep Neural Network (DNN) accelerators requires optimal Design Space Exploration (DSE), which becomes increasingly challenging as complex models for advanced AI workloads expand the highly non-convex design space. Among existing DSE approaches, multi-objective Bayesian Optimization (BO) is promising, as it explores high-quality design solutions while querying costly CIM simulators selectively. In this work, we propose a multi-objective BO framework that holistically co-optimizes hardware and algorithm parameters of a CIM crossbar-based hardware accelerator for various DNN inference tasks. Depending on NN model depth, our framework handles high-dimensional design spaces (with 26 and 50 dimensions) and extremely large search complexities on the order of O(10^12) and O(10^27) for VGG8/CIFAR-10 and VGG16/Tiny-ImageNet-200. Our method attains 91.72% and 57.2% accuracy, respectively, comparable to baseline designs, while improving chip area (65.52% and 50.7%), read latency (9.52% and 13.27%), read dynamic energy (31.23% and 52.07%) and increasing memory utilization (13.41% and 2.67%).

   @inproceedings{saha2026bayesiancim,
     title = {Bayesian Optimization of Crossbar-Based Compute-In-Memory System Design for Efficient {DNN} Inference},
     author = {Saha, Arnob and Manna, Bibhas and Kotikalapudi, Nikhil and Mia, Md Zesun Ahmed and Kumar, Rahul and Swaminathan, Madhavan and Sengupta, Abhronil},
     booktitle = {IEEE Conference on Electrical Performance of Electronic Packaging and Systems (EPEPS)},
     year = {2026},
     url = {https://arxiv.org/abs/2605.08461},
   }

5. MWSCAS

   Leveraging Intrinsic Physics of Spintronic Devices for Neuromorphic Computing

   Md Zesun Ahmed Mia and Abhronil Sengupta

   In 2026 IEEE 69th International Midwest Symposium on Circuits and Systems (MWSCAS), 2026

   Abs Bib PDF

   Neuromorphic computing has made significant strides over the past few years by mapping functional primitives of neurons and synapses to nanoelectronic devices, yet most implementations do not tap into the temporal non-linear dynamics of the devices. We argue that spintronic devices, specifically magnetic tunnel junctions (MTJs), offer a palette of intrinsic physical phenomena that can serve as native computational primitives. Spin-transfer torque (STT), spin-orbit torque (SOT), voltage-controlled magnetic anisotropy (VCMA), magnetoelectric effect (ME) and stochastic superparamagnetic dynamics each enable functionalities difficult to replicate in deterministic CMOS. We underscore this perspective through two specific system-level instances focusing on the synapse and neuron: selectorless synaptic crossbar arrays that exploit a VCMA-driven switching-mechanism to eliminate per-cell selectors, and temporal information encoding through superparamagnetic neuronal MTJs that achieves significant network-level spiking sparsity via stochastic dynamics. Building on these exemplars, we envision how these primitives compose with the broader spintronic ecosystem — domain-wall synapses, oscillatory primitives, and stochastic neurons — toward neuromorphic systems that are simultaneously sparser, denser, and more functionally diverse.

   @inproceedings{mia2026spintronic,
     title = {Leveraging Intrinsic Physics of Spintronic Devices for Neuromorphic Computing},
     author = {Mia, Md Zesun Ahmed and Sengupta, Abhronil},
     booktitle = {2026 IEEE 69th International Midwest Symposium on Circuits and Systems (MWSCAS)},
     year = {2026},
     organization = {IEEE},
     url = {https://www.mwscas2026.org/},
   }

2025

1. ICONS

   Neuromorphic Cybersecurity with Semi-Supervised Lifelong Learning

   Md Zesun Ahmed Mia, Malyaban Bal, Sen Lu, and 4 more authors

   In Proceedings of the International Conference on Neuromorphic Systems (ICONS ’25), 2025

   DOI Bib PDF

   @inproceedings{mia2025neuromorphic,
     title = {Neuromorphic Cybersecurity with Semi-Supervised Lifelong Learning},
     author = {Mia, Md Zesun Ahmed and Bal, Malyaban and Lu, Sen and Nishibuchi, George M. and Chelian, Suhas and Vasan, Srini and Sengupta, Abhronil},
     booktitle = {Proceedings of the International Conference on Neuromorphic Systems (ICONS '25)},
     pages = {235--238},
     year = {2025},
     publisher = {IEEE Press},
     doi = {10.1109/ICONS69015.2025.00043},
     url = {https://doi.org/10.1109/ICONS69015.2025.00043},
   }

2. IEEE TCDS

   Delving deeper into astromorphic transformers

   Md Zesun Ahmed Mia, Malyaban Bal, and Abhronil Sengupta

   IEEE Transactions on Cognitive and Developmental Systems, 2025

   Abs Bib PDF

   Preliminary attempts at incorporating the critical role of astrocytes—cells that constitute more than 50% of human brain cells—in brain-inspired neuromorphic computing remain in infancy. This paper seeks to delve deeper into various key aspects of neuron-synapse-astrocyte interactions to mimic self-attention mechanisms in Transformers. The crosslayer perspective explored in this work involves bioplausible modeling of Hebbian and presynaptic plasticities in neuronastrocyte networks, incorporating effects of non-linearities and feedback along with algorithmic formulations to map the neuronastrocyte computations to self-attention mechanism and evaluating the impact of incorporating bio-realistic effects from the machine learning application side. Our analysis on sentiment and image classification tasks (IMDB and CIFAR10 datasets) highlights the advantages of Astromorphic Transformers, offering improved accuracy and learning speed. Furthermore, the model demonstrates strong natural language generation capabilities on the WikiText-2 dataset, achieving better perplexity compared to conventional models, thus showcasing enhanced generalization and stability across diverse machine learning tasks.

   @article{mia2025delving,
     title = {Delving deeper into astromorphic transformers},
     author = {Mia, Md Zesun Ahmed and Bal, Malyaban and Sengupta, Abhronil},
     journal = {IEEE Transactions on Cognitive and Developmental Systems},
     year = {2025},
     publisher = {IEEE},
     url = {https://ieeexplore.ieee.org/document/10976578/},
   }

3. MWSCAS

   Toward Variation-Tolerant Ferroelectric Neural Computing: Special Session Paper

   Arnob Saha, Md Zesun Ahmed Mia, Jiahui Duan, and 2 more authors

   In 2025 IEEE 68th International Midwest Symposium on Circuits and Systems (MWSCAS), 2025

   Bib PDF

   @inproceedings{saha2025toward,
     title = {Toward Variation-Tolerant Ferroelectric Neural Computing: Special Session Paper},
     author = {Saha, Arnob and Mia, Md Zesun Ahmed and Duan, Jiahui and Ni, Kai and Sengupta, Abhronil},
     booktitle = {2025 IEEE 68th International Midwest Symposium on Circuits and Systems (MWSCAS)},
     pages = {583--587},
     year = {2025},
     organization = {IEEE},
   }

2024

1. Matter

   Self-sensitizable neuromorphic device based on adaptive hydrogen gradient

   Tao Zhang, Mingjie Hu, Md Zesun Ahmed Mia, and 8 more authors

   Matter, 2024

   Abs Bib PDF

   Neuromorphic computing aims to achieve superior efficiency by simulating the human brain to surpass traditional computers. It is crucial for neuromorphic devices to possess excellent biological characteristics, offering a vital way to address existing technical bottlenecks in artificial intelligence (AI). Here, we demonstrate adaptive self-sensitization based on a hydrogen gradient in artificial neurons, empowering neural networks to effectively tackle long-standing challenges of model failure in unknown situations beyond pre-defined boundaries. This advancement propels the design of adaptive neuromorphic devices to process unforeseeable signals autonomously. It significantly promotes the development of AI capable of interacting with complex environments, enhancing capabilities to perform tasks such as autonomous navigation, disaster rescue, and outer space exploration and opening up new possibilities for self-guided cognitive systems.

   @article{zhang2024self,
     title = {Self-sensitizable neuromorphic device based on adaptive hydrogen gradient},
     author = {Zhang, Tao and Hu, Mingjie and Mia, Md Zesun Ahmed and Zhang, Hao and Mao, Wei and Fukutani, Katsuyuki and Matsuzaki, Hiroyuki and Wen, Lingzhi and Wang, Cong and Zhao, Hongbo and others},
     journal = {Matter},
     volume = {7},
     number = {5},
     pages = {1799--1816},
     year = {2024},
     publisher = {Elsevier},
     url = {https://www.cell.com/matter/fulltext/S2590-2385(24)00108-5},
   }

2. iCACCESS

   Ultra Low Cost, Low Power, High Speed Electronic Braille Device for Visually Impaired People

   Md Zesun Ahmed Mia and Kazi Toukir Ahmed

   In 2024 International Conference on Advances in Computing, Communication, Electrical, and Smart Systems (iCACCESS), 2024

   Abs Bib PDF

   Highly advanced and sophisticated braille devices are always sought after, however, the expenses associated with them surpass the advantages for the majority of individuals with vision impairments. In this paper, we design an affordable, customized, and rechargeable braille pad utilizing solenoids that have the capability to read text quickly and with precise control. The device is constructed with great precision and compactness using 3D printing methodologies. A microcontroller is precisely installed within the gadget to control the entire process. The device can be easily handled in the palm of a hand and the sensitivity may be modified as well. In addition to its mobility, this device has a weight of only 338 grams and is capable of translating text into braille from both cellular devices and laptops. This specific structure serves as a prime example in reducing electricity usage and enhancing productivity.

   @inproceedings{mia2024ultra,
     title = {Ultra Low Cost, Low Power, High Speed Electronic Braille Device for Visually Impaired People},
     author = {Mia, Md Zesun Ahmed and Ahmed, Kazi Toukir},
     booktitle = {2024 International Conference on Advances in Computing, Communication, Electrical, and Smart Systems (iCACCESS)},
     pages = {1--6},
     year = {2024},
     organization = {IEEE},
     url = {https://ieeexplore.ieee.org/document/10499456/},
   }

3. ICECE

   Impact of Doping and Defects on Thermal Transport of Monolayer GaN Nanoribbons: A Molecular Dynamics Simulation Study

   Md Zesun Ahmed Mia, Nazmus Saadat As-Saquib, Mehedi Hasan Himel, and 2 more authors

   In 2024 13th International Conference on Electrical and Computer Engineering (ICECE), 2024

   Abs Bib PDF

   Monolayer gallium nitride (GaN), with its graphenelike honeycomb structure, has emerged as a promising material for nanoelectronics and optoelectronic applications. In this study, Equilibrium Molecular Dynamics (EMD) simulations using the Stillinger-Weber (SW) potential were employed to investigate the impact of doping as well as defects on the thermal conductivity of monolayer zigzag GaN nanoribbon (GaN-NR). The thermal transport in aluminum (Al) and indium (In) doped GaN-NR is explored, considering the substitution of both Ga and nitrogen (N) atoms by the dopants. Ga replacement with In led to an increase in thermal conductivity from 11.51 W/m−K to 15.12 W/m−K at a 1% doping concentration. However, in all other cases, whether replacing Ga or N with Al or replacing N with In, it shows an opposite trend, i.e., the thermal conductivity decreases. Furthermore, defects lead to a substantial reduction in thermal conductivity, with a decrease ranging from 68.9% to 77.4% observed at a 2% vacancy concentration across various defect shapes, from point vacancies to hexagon vacancies. We further examine the effect of defect concentrations and defect types on its thermal transport. The results provide valuable insights into the tunability of thermal conductivity in GaN nanostructures, illuminating future optimization for applications in optoelectronic and thermoelectric devices.

   @inproceedings{mia2024impact,
     title = {Impact of Doping and Defects on Thermal Transport of Monolayer GaN Nanoribbons: A Molecular Dynamics Simulation Study},
     author = {Mia, Md Zesun Ahmed and As-Saquib, Nazmus Saadat and Himel, Mehedi Hasan and Limon, Mehedi Hossen and Subrina, Samia},
     booktitle = {2024 13th International Conference on Electrical and Computer Engineering (ICECE)},
     pages = {685--690},
     year = {2024},
     organization = {IEEE},
     url = {https://ieeexplore.ieee.org/document/11024945/},
   }

2022

1. ISIEA

   Study of 3-nm Cylindrical GAAFETs with Variations in High-k Dielectric Gate-oxide Materials

   KM Ashraful Hoque Fahim, Md Jubair Hasan Khalid, Md Zesun Ahmed Mia, and 1 more author

   In 2022 IEEE Symposium on Industrial Electronics & Applications (ISIEA), 2022

   Abs Bib PDF

   Semiconductor devices using high k dielectric materials are widely adopted in memory and amplifier applications. Among the semiconductor devices gate all around-FET (GAAFET) is now the latest trend being used instead of other field effect transistors to serve the purpose of reducing the short channel effects (SCE). In this work, we examine the performance of a circular cross-section gate all around-field effect transistor (GAA-FET) with varying gate dielectric characteristics with high-k dielectric oxide materials (Al2O3, HfO2, HfSiO4, SiO2, Ta2O5, TiO2) across the 3-nm channel length. These simulations showed that even though the dielectric constant over the channel increases in value, both ION-IOFF ratio and transconductance upsurge. The obtained results indicated that raising the dielectric constant in a gate oxide reduces subthreshold slope (SS), increases amplification rate, and reduces threshold voltage (VTH) roll-off as well. The Silvaco TCAD ATLAS simulation was calibrated against experimental data from different works of literature. The higher the dielectric constant, the lower the SCEs. It is also found that TiO2 is dominating over the other materials selected for the simulation for a higher value of dielectric constant.

   @inproceedings{fahim2022study,
     title = {Study of 3-nm Cylindrical GAAFETs with Variations in High-k Dielectric Gate-oxide Materials},
     author = {Fahim, KM Ashraful Hoque and Khalid, Md Jubair Hasan and Mia, Md Zesun Ahmed and Rasheduzzaman, Mirza},
     booktitle = {2022 IEEE Symposium on Industrial Electronics \& Applications (ISIEA)},
     pages = {1--5},
     year = {2022},
     organization = {IEEE},
     url = {https://ieeexplore.ieee.org/document/9873651/},
   }

2. ICO

   DCNN-LSTM based audio classification combining multiple feature engineering and data augmentation techniques

   Md Moinul Islam, Monjurul Haque, Saiful Islam, and 2 more authors

   In Intelligent Computing & Optimization: Proceedings of the 4th International Conference on Intelligent Computing and Optimization 2021 (ICO2021) 3, 2022

   Abs Bib PDF

   Everything we know is based on our brain’s ability to process sensory data. Hearing is a crucial sense for our ability to learn. Sound is essential for a wide range of activities such as exchanging information, interacting with others, and so on. To convert the sound electrically, the role of the audio signal comes into play. Because of the countless essential applications, audio signal & their classification poses an important value. However, in this day and age, classifying audio signals remains a difficult task. To classify audio signals more accurately and effectively, we have proposed a new model. In this study, we’ve applied a brand-new method for audio classification that combines the strengths of Deep Convolutional Neural Network (DCNN) and Long-Short Term Memory (LSTM) models with a unique combination of feature engineering to get the best possible outcome. Here, we have integrated data augmentation and feature extraction together before fitting it into the model to evaluate the performance. There is a higher degree of accuracy observed after the experiment. To validate the efficacy of our model, a comparative analysis has been made with the latest conducted reference works.

   @inproceedings{islam2022dcnn,
     title = {DCNN-LSTM based audio classification combining multiple feature engineering and data augmentation techniques},
     author = {Islam, Md Moinul and Haque, Monjurul and Islam, Saiful and Mia, Md Zesun Ahmed and Rahman, SMA Mohaiminur},
     booktitle = {Intelligent Computing \& Optimization: Proceedings of the 4th International Conference on Intelligent Computing and Optimization 2021 (ICO2021) 3},
     pages = {227--236},
     year = {2022},
     organization = {Springer},
     url = {https://link.springer.com/chapter/10.1007/978-3-030-93247-3_23},
   }

2021

1. ICCIT

   Irfd: A feature engineering based ensemble classification for detecting electricity fraud in traditional meters

   Md Zesun Ahmed Mia, Md Moinul Islam, Monjurul Haque, and 2 more authors

   In 2021 24th International Conference on Computer and Information Technology (ICCIT), 2021

   Abs Bib PDF

   Nations have suffered significant economic losses as a result of non-technical electric losses resulting from power fraud. It is a criminal act of stealing electricity by applying various mechanisms that incorporate unauthorized tapping to the power line, bypassing the smart meter, etc. Electricity theft is a significant concern for not only developing countries but also developed countries as well. However, for most developing countries, the implications are catastrophic, given that their usage is always less than their demands. Electricity theft must be detected precisely and quickly in order to be mitigated. In our study, we have proposed a method of predictive ensemble machine learning techniques (IRFD) with a novel combination of feature distinction methods to detect electricity theft. In our proposed model, we have combined feature selection technique, Recursive Feature Elimination with Stratified 10-Fold cross-validation (RFECV) and Isolation Forest (IF), to identify and remove outliers along with several machine learning classifiers to forecast the theft of electricity. This study additionally enhances the management of highly imbalanced fraudulent data with Borderline-SMOTE with SVM (SVMSMOTE) and feature scaling with StandardScaler. Following the study, the Random Forest classifier observed a higher degree of accuracy (97.06%) with higher precision, recall, and F1-Score. To evaluate the efficacy of our proposed model, comparative analysis of the classification metrics is also assessed with several machine learning classifiers like Logistic Regression, Gradient Boosting, XGBoost, AdaBoost, KNN, ANN, along with Random Forest before and after fitting our proposed feature engineering techniques.

   @inproceedings{mia2021irfd,
     title = {Irfd: A feature engineering based ensemble classification for detecting electricity fraud in traditional meters},
     author = {Mia, Md Zesun Ahmed and Islam, Md Moinul and Haque, Monjurul and Islam, Saiful and Rahman, SMA Mohaiminur},
     booktitle = {2021 24th International Conference on Computer and Information Technology (ICCIT)},
     pages = {1--6},
     year = {2021},
     organization = {IEEE},
     url = {https://ieeexplore.ieee.org/document/9689842/},
   }