Natasa Samardzic
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Natasa Samardzic has a total of 78 co-authors. Here are the top co-authors by number of shared publications:
- Stanisa Dautovic
University of Novi Sad
5 shared publications - Goran M. Stojanovic
University of Novi Sad
4 shared publications - Alexey N. Mikhaylov
Research and Education Center “Physics of Solid-State Nanostructures”, National Research Lobachevsky State University of Nizhny Novgorod, 603022 Nizhny Novgorod, Russia
3 shared publications - Dalibor L. Sekulic
University of Novi Sad,Faculty of Technical Sciences,Department of Power, Electronic and Telecommunication Engineering,Novi Sad,Serbia
3 shared publications - Max Talanov
The Institute for Artificial Intelligence Research and Development of Serbia, 21000 Novi Sad, Serbia
3 shared publications - Vladimir V. Srdic
University of Novi Sad, Faculty of Technology, Department of Materials Engineering, 21000 Novi Sad, Serbia
3 shared publications
Latest Publications
- Open Access
- 22 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 4892
Memristive Neural Networks-Assisted Training Algorithms: Current State and Future Challenges
Existing challenges in modern AI architectures, including the von Neumann bottleneck and high energy demand, have led to increased research efforts towards the development of novel computing paradigms based on memristive technologies. This comprehensive review summarizes algorithmic approaches for training memristive neural networks (MNNs), including Memristive Spiking Neural Networks (MSNNs), integrating solutions implemented on hardware platforms with those developed in software. In contrast to existing reviews that consider algorithmic aspects within the broader context of hardware implementation, this review provides an algorithm-centric perspective, starting from training algorithms and analyzing different approaches to their adaptation to memristive hardware. The survey covers traditional machine learning algorithms such as backpropagation and its enhancements toward hardware-friendly solutions, as well as gradient-free methods, with particular emphasis on algorithms that address device non-idealities. We additionally provide an overview of bio-inspired learning rules, including Spike-Timing Dependent Plasticity (STDP) and reward-modulated STDP, which can be mimicked using memristive devices, along with reinforcement learning algorithms. Special attention is given to biologically plausible models of spiking neuron-astrocyte networks, and their direct implementation in memristive hardware, which significantly increase the efficiency of self-learning algorithms. Finally, this review presents an algorithm-centric perspective on open questions and challenges toward the large-scale implementation of memristive devices in AI systems.
- 22 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 4892
Open AccessMemristive Neural Networks-Assisted Training Algorithms: Current State and Future Challenges
Existing challenges in modern AI architectures, including the von Neumann bottleneck and high energy demand, have led to increased research efforts towards the development of novel computing paradigms based on memristive technologies. This comprehensive review summarizes algorithmic approaches for training memristive neural networks (MNNs), including Memristive Spiking Neural Networks (MSNNs), integrating solutions implemented on hardware platforms with those developed in software. In contrast to existing reviews that consider algorithmic aspects within the broader context of hardware implementation, this review provides an algorithm-centric perspective, starting from training algorithms and analyzing different approaches to their adaptation to memristive hardware. The survey covers traditional machine learning algorithms such as backpropagation and its enhancements toward hardware-friendly solutions, as well as gradient-free methods, with particular emphasis on algorithms that address device non-idealities. We additionally provide an overview of bio-inspired learning rules, including Spike-Timing Dependent Plasticity (STDP) and reward-modulated STDP, which can be mimicked using memristive devices, along with reinforcement learning algorithms. Special attention is given to biologically plausible models of spiking neuron-astrocyte networks, and their direct implementation in memristive hardware, which significantly increase the efficiency of self-learning algorithms. Finally, this review presents an algorithm-centric perspective on open questions and challenges toward the large-scale implementation of memristive devices in AI systems.
- Open Access
- 18 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (18) , 4267
In-Memory Computing Enabled by 32 × 8 1T1R Memristive Crossbar Arrays
Computing in memory represented by resistive switching (memristive) devices is a promising way to implement hardware for artificial neural networks. We see it as the way to create energy-efficient and high-performance systems for solving a wide range of tasks in artificial intelligence. Here, we present the structure and performance of our crossbar array based on Au/Ta/ZrO2(Y)/Pt/Ti memristive devices, integrated into a complementary metal–oxide–semiconductor fabrication process, and then demonstrate its applicability for the hardware implementation of artificial neural networks. The array contains 256 memristive devices combined into 32 columns and eight rows. Each memristive device is paired with an n-channel metal–oxide–semiconductor transistor. As the results of the simulations show, these devices enable running neural network algorithms for solving recognition problems using the Modified National Institute of Standards and Technology dataset with 98% accuracy and the Canadian Institute for Advanced Research (10 classes) dataset with at least 77.5% accuracy, even taking into account the errors of mapping weights onto the memristive crossbar. Fully hardware recognition of 4 bit images demonstrated in this work highlights the crucial role of selective transistors for such tasks. This paper explores the exciting potential of memristor-based components of neuroelectronics as the building blocks for a new era of artificial intelligence hardware, capable of surpassing current computational limitations, as well as their role in the ongoing transition to hybrid intelligence based on the symbiosis of electronic and biological systems.
- 18 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (18) , 4267
Open AccessIn-Memory Computing Enabled by 32 × 8 1T1R Memristive Crossbar Arrays
Computing in memory represented by resistive switching (memristive) devices is a promising way to implement hardware for artificial neural networks. We see it as the way to create energy-efficient and high-performance systems for solving a wide range of tasks in artificial intelligence. Here, we present the structure and performance of our crossbar array based on Au/Ta/ZrO2(Y)/Pt/Ti memristive devices, integrated into a complementary metal–oxide–semiconductor fabrication process, and then demonstrate its applicability for the hardware implementation of artificial neural networks. The array contains 256 memristive devices combined into 32 columns and eight rows. Each memristive device is paired with an n-channel metal–oxide–semiconductor transistor. As the results of the simulations show, these devices enable running neural network algorithms for solving recognition problems using the Modified National Institute of Standards and Technology dataset with 98% accuracy and the Canadian Institute for Advanced Research (10 classes) dataset with at least 77.5% accuracy, even taking into account the errors of mapping weights onto the memristive crossbar. Fully hardware recognition of 4 bit images demonstrated in this work highlights the crucial role of selective transistors for such tasks. This paper explores the exciting potential of memristor-based components of neuroelectronics as the building blocks for a new era of artificial intelligence hardware, capable of surpassing current computational limitations, as well as their role in the ongoing transition to hybrid intelligence based on the symbiosis of electronic and biological systems.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 60535-60556
Critical Analysis of Energy Consumption in Neuro-Computational Systems
This article presents a unified benchmark of energy consumption per synaptic event across five classes of neuro-computational substrates: graphics processing units, neural processing units, field-programmable gate arrays, digital spiking processors, and in-memory/memristive devices, with biological synapses used as reference bands. The main contribution is methodological as well as empirical. Methodologically, we introduce a common event-level metric that makes heterogeneous systems directly comparable despite major differences in architecture, coding scheme, and learning dynamics. Empirically, we combine measurements obtained by the authors with carefully normalized literature data to map the present energy landscape of artificial and biological neural computation. The results reveal three robust regimes. Dense von Neumann ANN implementations on GPUs, NPUs, and FPGA operate mainly in the $10^{5}$ to $10^{8}$ fJ range per synaptic event. Digital spiking processors reduce this requirement to about $10^{3}$ to $10^{4}$ fJ per event. Memristive and transistor-based artificial synapses span a much broader interval, from array-level values near $10^{6}$ fJ down to a few femtojoules in the most efficient device-level realizations. In particular, the best organic PANI memristive samples approach ~2 fJ per synaptic event, entering the biological vicinity and, in some cases, surpassing the rat reference range. Taken together, the benchmarked landscape spans up to ten orders of magnitude across the full set of systems considered in this study. This result clarifies where current AI hardware stands relative to biological efficiency, identifies event-driven and in-memory computation as the most promising routes toward sustainable AI, and provides a quantitatively grounded reference framework for future neuromorphic benchmarking. INDEX TERMS Artificial neural networks, energy consumption, energy efficiency, memristive devices, neuromorphic computing, spiking neural networks, sustainable AI.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 60535-60556
Open AccessCritical Analysis of Energy Consumption in Neuro-Computational Systems
This article presents a unified benchmark of energy consumption per synaptic event across five classes of neuro-computational substrates: graphics processing units, neural processing units, field-programmable gate arrays, digital spiking processors, and in-memory/memristive devices, with biological synapses used as reference bands. The main contribution is methodological as well as empirical. Methodologically, we introduce a common event-level metric that makes heterogeneous systems directly comparable despite major differences in architecture, coding scheme, and learning dynamics. Empirically, we combine measurements obtained by the authors with carefully normalized literature data to map the present energy landscape of artificial and biological neural computation. The results reveal three robust regimes. Dense von Neumann ANN implementations on GPUs, NPUs, and FPGA operate mainly in the $10^{5}$ to $10^{8}$ fJ range per synaptic event. Digital spiking processors reduce this requirement to about $10^{3}$ to $10^{4}$ fJ per event. Memristive and transistor-based artificial synapses span a much broader interval, from array-level values near $10^{6}$ fJ down to a few femtojoules in the most efficient device-level realizations. In particular, the best organic PANI memristive samples approach ~2 fJ per synaptic event, entering the biological vicinity and, in some cases, surpassing the rat reference range. Taken together, the benchmarked landscape spans up to ten orders of magnitude across the full set of systems considered in this study. This result clarifies where current AI hardware stands relative to biological efficiency, identifies event-driven and in-memory computation as the most promising routes toward sustainable AI, and provides a quantitatively grounded reference framework for future neuromorphic benchmarking. INDEX TERMS Artificial neural networks, energy consumption, energy efficiency, memristive devices, neuromorphic computing, spiking neural networks, sustainable AI.
- 27 September 2025
- Published by Wiley in International Journal of Circuit Theory and Applications
- Vol. 54 (5) , 2853-2863
Implementation of Direct Feedback Alignment Through Time Algorithm in Training of Shallow Fully Memristive Spiking Neural Networks
In this paper, we present shallow architectures of memristive spiking neural networks (MSNNs) with a single hidden layer and a simple memristive leaky integrate‐and‐fire (LIF) neuron. The SPICE‐based model is used to describe the response of a volatile memristor within LIF, whose intrinsic reset to the initial state eliminates the need for additional electronic components for discharging the membrane capacitance. The direct feedback alignment through time (DFATT) algorithm was implemented for the first time for the training of MSNNs. The deployment of the DFATT learning algorithm on two MSNN architectures resulted in average training times per epoch that were reduced by factors of 2.1 and 2.7 compared to equivalent architectures trained with the backpropagation through time (BPTT) algorithm, with comparable accuracy on MNIST and FMNIST datasets. The reported results indicate the possibility of combining memristive technology with the bioplausible DFATT algorithm toward building energy‐efficient hardware.
- 27 September 2025
- Published by Wiley in International Journal of Circuit Theory and Applications
- Vol. 54 (5) , 2853-2863
Implementation of Direct Feedback Alignment Through Time Algorithm in Training of Shallow Fully Memristive Spiking Neural Networks
In this paper, we present shallow architectures of memristive spiking neural networks (MSNNs) with a single hidden layer and a simple memristive leaky integrate‐and‐fire (LIF) neuron. The SPICE‐based model is used to describe the response of a volatile memristor within LIF, whose intrinsic reset to the initial state eliminates the need for additional electronic components for discharging the membrane capacitance. The direct feedback alignment through time (DFATT) algorithm was implemented for the first time for the training of MSNNs. The deployment of the DFATT learning algorithm on two MSNN architectures resulted in average training times per epoch that were reduced by factors of 2.1 and 2.7 compared to equivalent architectures trained with the backpropagation through time (BPTT) algorithm, with comparable accuracy on MNIST and FMNIST datasets. The reported results indicate the possibility of combining memristive technology with the bioplausible DFATT algorithm toward building energy‐efficient hardware.
- 10 December 2024
- Published by Springer Nature
- p. 402-410
Future of Neural Networks and Energy Consumption Aspects
- 10 December 2024
- Published by Springer Nature
- p. 402-410
Future of Neural Networks and Energy Consumption Aspects
- Open Access
- 8 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 94870-94884
Analytical Solutions for Charge and Flux in HP Ideal Generic Memristor Model With Joglekar and Prodromakis Window Functions
This paper presents, for the first time, analytical solutions for the charge and flux as a function of the state variable in the case of the HP memristor, modeled as an ideal generic memristor with Joglekar and Prodromakis window functions for a general value of the control parameter. The solutions are in the closed form of finite sums for an arbitrary positive integer values of the control parameter. Our approach is based on the decomposition of original problems using appropriate pairs of smooth hinge functions. The generalization of this approach allows us to define a novel window function based on smooth hinge functions, which brings simpler analytical expressions for charge and flux as a function of the state variable, and faster simulations when the model is used.
- 8 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 94870-94884
Open AccessAnalytical Solutions for Charge and Flux in HP Ideal Generic Memristor Model With Joglekar and Prodromakis Window Functions
This paper presents, for the first time, analytical solutions for the charge and flux as a function of the state variable in the case of the HP memristor, modeled as an ideal generic memristor with Joglekar and Prodromakis window functions for a general value of the control parameter. The solutions are in the closed form of finite sums for an arbitrary positive integer values of the control parameter. Our approach is based on the decomposition of original problems using appropriate pairs of smooth hinge functions. The generalization of this approach allows us to define a novel window function based on smooth hinge functions, which brings simpler analytical expressions for charge and flux as a function of the state variable, and faster simulations when the model is used.
- Open Access
- 1 January 2023
- Published by Centre for Evaluation in Education and Science (CEON/CEES) in Biljni lekar
- Vol. 51 (5) , 690-709
The influence of different types of perennial mulch on the weed seed bank in an orchard
Mulching is an agrotechnical measure often used in orchards and it means covering the soil with the organic or inorganic material in order to prevent the germination of the weeds. The other positive effects imply: the improvement of the physical and chemical soil characteristics, decrease of soil erosion and water loss, the prevention of the soil crusting, etc. The research of the weed seed bank under the different types of mulch was conducted during the growing season of 2021. in a plum orchard located at Rimski Šančevi. The black and white agrotextile and white mulch film, unlike the other mulches, left the surfaces clean, without the weeds. The analysis of the weed seed bank revealed the seeds of 20 weed species, of which 12 were not found to be present in the orchard during the experimental year. The overall germination rate of the obtained weed seeds is low because in the three-year experiment with mulches, there was no influx of new weed seeds for several years, while the old ones lost their germination. The highest germination percentage was noted for the seeds of the following species: Amaranthus retroflexus, Stellaria media and Solanum nigrum, due to their ability to retain germination for a long period of time. The highest number of the seeds per m2 was found in the inter-row space due to the large influx of the seeds during the growing season. The highest germination rate was observed in case of the seeds found under the sawdust mulch. None of the ungerminated seeds showed the viability, as determined by the Crush test.
- 1 January 2023
- Published by Centre for Evaluation in Education and Science (CEON/CEES) in Biljni lekar
- Vol. 51 (5) , 690-709
Open AccessThe influence of different types of perennial mulch on the weed seed bank in an orchard
Mulching is an agrotechnical measure often used in orchards and it means covering the soil with the organic or inorganic material in order to prevent the germination of the weeds. The other positive effects imply: the improvement of the physical and chemical soil characteristics, decrease of soil erosion and water loss, the prevention of the soil crusting, etc. The research of the weed seed bank under the different types of mulch was conducted during the growing season of 2021. in a plum orchard located at Rimski Šančevi. The black and white agrotextile and white mulch film, unlike the other mulches, left the surfaces clean, without the weeds. The analysis of the weed seed bank revealed the seeds of 20 weed species, of which 12 were not found to be present in the orchard during the experimental year. The overall germination rate of the obtained weed seeds is low because in the three-year experiment with mulches, there was no influx of new weed seeds for several years, while the old ones lost their germination. The highest germination percentage was noted for the seeds of the following species: Amaranthus retroflexus, Stellaria media and Solanum nigrum, due to their ability to retain germination for a long period of time. The highest number of the seeds per m2 was found in the inter-row space due to the large influx of the seeds during the growing season. The highest germination rate was observed in case of the seeds found under the sawdust mulch. None of the ungerminated seeds showed the viability, as determined by the Crush test.
- 28 May 2022
- Published by Elsevier in Ceramics International
- Vol. 48 (18) , 26378-26386
Dielectric and ferroelectric properties of multilayer BaTiO3/NiFe2O4 thin films prepared by solution deposition technique
- 28 May 2022
- Published by Elsevier in Ceramics International
- Vol. 48 (18) , 26378-26386
Dielectric and ferroelectric properties of multilayer BaTiO3/NiFe2O4 thin films prepared by solution deposition technique
- Open Access
- 13 March 2022
- Published by MDPI AG in Electronics
- Vol. 11 (6) , 894
Volatile Memristor in Leaky Integrate-and-Fire Neurons: Circuit Simulation and Experimental Study
In this paper, circuit implementation of a leaky integrate-and-fire neuron model with a volatile memristor was proposed and simulated in the SPICE simulation environment. We demonstrate that simple leaky integrate-and-fire (LIF) neuron models composed of: volatile memristor, membrane capacitance and neuron resistance can mimic spatial and temporal integration, firing function and signal decay. The existing leaky term originates from the recovery of the initial resistive state in the memristor in the spontaneous reset cycle, which is essential for emulating the forgetting process in all-memristive neural networks (MNNs). Furthermore, a diffusive perovskite memristor was used to validate the model where intrinsic memristors’ capacitance acts as neuron membrane capacitance. Good agreement with experimental and simulation results was observed. Volatility, as an inherent property of specific memristors, eliminates the need for usage of an additional peripheral circuit which will reinitialize device state, thus allowing the development of energy-efficient, large scale complex memristive neural networks. The presented circuit level model of LIF neurons can facilitate the design of MNNs.
- 13 March 2022
- Published by MDPI AG in Electronics
- Vol. 11 (6) , 894
Open AccessVolatile Memristor in Leaky Integrate-and-Fire Neurons: Circuit Simulation and Experimental Study
In this paper, circuit implementation of a leaky integrate-and-fire neuron model with a volatile memristor was proposed and simulated in the SPICE simulation environment. We demonstrate that simple leaky integrate-and-fire (LIF) neuron models composed of: volatile memristor, membrane capacitance and neuron resistance can mimic spatial and temporal integration, firing function and signal decay. The existing leaky term originates from the recovery of the initial resistive state in the memristor in the spontaneous reset cycle, which is essential for emulating the forgetting process in all-memristive neural networks (MNNs). Furthermore, a diffusive perovskite memristor was used to validate the model where intrinsic memristors’ capacitance acts as neuron membrane capacitance. Good agreement with experimental and simulation results was observed. Volatility, as an inherent property of specific memristors, eliminates the need for usage of an additional peripheral circuit which will reinitialize device state, thus allowing the development of energy-efficient, large scale complex memristive neural networks. The presented circuit level model of LIF neurons can facilitate the design of MNNs.
- 10 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Education
- Vol. 65 (4) , 562-567
Introducing Innovative Prototypes in Course Materials and Fabrication Technologies in Medical Devices for Electrical and Biomedical Engineering Students
Contribution: This article presents the innovative prototypes used in materials and fabrication technologies in medical devices (MFTIMD) laboratory course, under the curriculum of Biomedical (BME) and Electrical Engineering (EE) undergraduate studies. A special part of the course is devoted to state-of-the-art topics, such as flexible and wearable electronic devices and microfluidic devices. The article contains a brief overview of the laboratory course content as well as an assessment analysis with a satisfactory review and an academic outcomes review. Background: The course should address the properties and the applications of new types of materials used in medical electronics as well as corresponding fabrication technologies and devices. The course content and literature provide a multidisciplinary approach to contemporary engineering problems, specifically in a fast-developing field of biomedical electronics. Intended Outcomes: After successfully completing the MFTIMD course, undergraduate Biomedical (BME) and EE students are trained to independently fabricate and characterized various materials and components for medical devices. They are also capable of creating original ideas and realizing them according to gained knowledge. Application Design: The assessment and evaluation study consist of satisfactory test with positive statements and gradual multiple-choice options for student to rate the statements as well as teachers and the literature. An additional survey of acquired knowledge was conducted before the June exam period and a short prior/post-test focused on main course topics. Findings: The satisfactory survey results indicated a very positive students' attitude toward the course material and the teaching methods. The pass rate was above 85% in first (June) exam period with an average mark of 8.63 (on a scale from 5 to 10). The prior/post test revealed a significant improvement in the overall knowledge of the course subject, while the preparation test before the exam period resulted in 84% of correct test answers.
- 10 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Education
- Vol. 65 (4) , 562-567
Introducing Innovative Prototypes in Course Materials and Fabrication Technologies in Medical Devices for Electrical and Biomedical Engineering Students
Contribution: This article presents the innovative prototypes used in materials and fabrication technologies in medical devices (MFTIMD) laboratory course, under the curriculum of Biomedical (BME) and Electrical Engineering (EE) undergraduate studies. A special part of the course is devoted to state-of-the-art topics, such as flexible and wearable electronic devices and microfluidic devices. The article contains a brief overview of the laboratory course content as well as an assessment analysis with a satisfactory review and an academic outcomes review. Background: The course should address the properties and the applications of new types of materials used in medical electronics as well as corresponding fabrication technologies and devices. The course content and literature provide a multidisciplinary approach to contemporary engineering problems, specifically in a fast-developing field of biomedical electronics. Intended Outcomes: After successfully completing the MFTIMD course, undergraduate Biomedical (BME) and EE students are trained to independently fabricate and characterized various materials and components for medical devices. They are also capable of creating original ideas and realizing them according to gained knowledge. Application Design: The assessment and evaluation study consist of satisfactory test with positive statements and gradual multiple-choice options for student to rate the statements as well as teachers and the literature. An additional survey of acquired knowledge was conducted before the June exam period and a short prior/post-test focused on main course topics. Findings: The satisfactory survey results indicated a very positive students' attitude toward the course material and the teaching methods. The pass rate was above 85% in first (June) exam period with an average mark of 8.63 (on a scale from 5 to 10). The prior/post test revealed a significant improvement in the overall knowledge of the course subject, while the preparation test before the exam period resulted in 84% of correct test answers.