Musaib Rafiq
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Musaib Rafiq has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
- Shubham Sahay
Department of Electrical Engineering, Indian Institute of Technology Kanpur, Kanpur, India
10 shared publications - Yogesh Singh Chauhan
Indian Institute of Technology (IIT) Kanpur
10 shared publications - Ateeb Naseer
Department of Electrical Engineering, Indian Institute of Technology, Kanpur, Kanpur 208016, India
4 shared publications - Amit Agarwal
Indian Institute of Technology Kanpur
3 shared publications - Somnath Bhowmick
Indian Institute of Technology Kanpur
3 shared publications - Mohammad Sajid Nazir
IIT Kanpur
2 shared publications
Latest Publications
- Open Access
- 23 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 11, 166-170
A SPICE-Compatible Compact Model of Ferroelectric Diode
In this work, for the first time, we present a SPICE-compatible compact model of ferroelectric (FE) diodes to enable their design exploration for diverse applications, including memory and unconventional computing paradigms. We propose modified Schottky barrier and hopping models for capturing the ON- and OFF-mode operations of the ferroelectric diode, respectively, in conjunction with the multi-domain Preisach model for incorporating the ferroelectric switching. Since the operating regime of the FE-diode is determined by the polarization-state and the sign of the applied voltage, directional dependence is also introduced in the proposed model to effectively capture the experimentally observed current-voltage characteristics. The proposed FE-diode model is validated by accurately reproducing two sets of asymmetric experimental current-voltage characteristics of BE/HZO/IGZO/TE FE-diodes.
- 23 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 11, 166-170
Open AccessA SPICE-Compatible Compact Model of Ferroelectric Diode
In this work, for the first time, we present a SPICE-compatible compact model of ferroelectric (FE) diodes to enable their design exploration for diverse applications, including memory and unconventional computing paradigms. We propose modified Schottky barrier and hopping models for capturing the ON- and OFF-mode operations of the ferroelectric diode, respectively, in conjunction with the multi-domain Preisach model for incorporating the ferroelectric switching. Since the operating regime of the FE-diode is determined by the polarization-state and the sign of the applied voltage, directional dependence is also introduced in the proposed model to effectively capture the experimentally observed current-voltage characteristics. The proposed FE-diode model is validated by accurately reproducing two sets of asymmetric experimental current-voltage characteristics of BE/HZO/IGZO/TE FE-diodes.
- Top Cited Papers
- 11 June 2025
- Published by Springer Nature in Nature
- Vol. 642 (8067) , 327-335
A complementary two-dimensional material-based one instruction set computer
Silicon has enabled advancements in semiconductor technology through miniaturization, but scaling challenges necessitate the exploration of new materials1. Two-dimensional (2D) materials, with their atomic thickness and high carrier mobility, offer a promising alternative2,3,4,5. Although significant progress has been made in wafer-scale growth6,7,8, high-performance field-effect transistors9,10,11,12,13,14,15,16,17,18,19,20 and circuits based on 2D materials21,22,23, achieving complementary metal–oxide–semiconductor (CMOS) integration remains a challenge. Here, we present a 2D one instruction set computer based on CMOS technology, leveraging the heterogeneous integration of large-area n-type MoS2 and p-type WSe2 field-effect transistors. By scaling the channel length, incorporating a high-κ gate dielectric and optimizing material growth and device postprocessing, we tailored the threshold voltages for both n- and p-type 2D field-effect transistors, achieving high drive currents and reduced subthreshold leakage. This enabled circuit operation below 3 V with an operating frequency of up to 25 kHz, which was constrained by parasitic capacitances, along with ultra-low power consumption in the picowatt range and a switching energy as low as approximately 100 pJ. Finally, we projected the performance of the one instruction set computer and benchmarked it against state-of-the-art silicon technology using an industry-standard SPICE-compatible BSIM-BULK model. This model was calibrated with experimental data that incorporate device-to-device variations. Although further advances are needed, this work marks a significant milestone in the application of 2D materials to microelectronics.
- 11 June 2025
- Published by Springer Nature in Nature
- Vol. 642 (8067) , 327-335
Top Cited PapersA complementary two-dimensional material-based one instruction set computer
Silicon has enabled advancements in semiconductor technology through miniaturization, but scaling challenges necessitate the exploration of new materials1. Two-dimensional (2D) materials, with their atomic thickness and high carrier mobility, offer a promising alternative2,3,4,5. Although significant progress has been made in wafer-scale growth6,7,8, high-performance field-effect transistors9,10,11,12,13,14,15,16,17,18,19,20 and circuits based on 2D materials21,22,23, achieving complementary metal–oxide–semiconductor (CMOS) integration remains a challenge. Here, we present a 2D one instruction set computer based on CMOS technology, leveraging the heterogeneous integration of large-area n-type MoS2 and p-type WSe2 field-effect transistors. By scaling the channel length, incorporating a high-κ gate dielectric and optimizing material growth and device postprocessing, we tailored the threshold voltages for both n- and p-type 2D field-effect transistors, achieving high drive currents and reduced subthreshold leakage. This enabled circuit operation below 3 V with an operating frequency of up to 25 kHz, which was constrained by parasitic capacitances, along with ultra-low power consumption in the picowatt range and a switching energy as low as approximately 100 pJ. Finally, we projected the performance of the one instruction set computer and benchmarked it against state-of-the-art silicon technology using an industry-standard SPICE-compatible BSIM-BULK model. This model was calibrated with experimental data that incorporate device-to-device variations. Although further advances are needed, this work marks a significant milestone in the application of 2D materials to microelectronics.
- Open Access
- 6 May 2025
- Published by IOP Publishing in Neuromorphic Computing and Engineering
- Vol. 5 (2) , 024007
Exploiting drain-erase scheme in ferroelectric FETs for logic-in-memory
The conventional computing platforms based on von-Neumann architecture are highly space- and energy-intensive while handling the emerging applications such as AI, ML, and big data. To overcome the von Neumann bottleneck, compact and light-weight logic-in-memory (LiM) implementations of Boolean logic gates based on emerging non-volatile memory (e-NVM) such as RRAMs, PCM, STT-MRAMs, etc were proposed recently. However, these e-NVMs not only exhibit significant temporal and spatial variability, but their large-scale integration with CMOS process is also a technological challenge. To overcome these issues with the emerging non-volatile memories, ferroelectric FETs based on CMOS-compatible doped hafnium oxide with the capability of large-scale CMOS integration in the advanced logic nodes were proposed. Considering the high scalability and CMOS-compatibility of the FeFETs, in this work, for the first time, we propose a LiM implementation utilizing a single ferroelectric fully-depleted-silicon-on-insulator (Fe-FDSOI) FET exploiting the unique drain erase phenomenon. In our proposed LiM implementation, inputs are applied at the gate and drain terminals using a novel input-to-voltage mapping scheme, and output is obtained as the current flowing through the Fe-FDSOI FET. We utilize an experimentally calibrated compact model of the ferroelectric capacitor connected to the baseline industry standard BSIM-IMG compact model for the FDSOI transistor for proof of concept demonstration. We also perform a comprehensive analysis of the performance metrics of the proposed LiM implementation. Our results indicate that we can realize at least 10 Boolean logic gates with high energy and area-efficiency utilizing the proposed scheme.
- 6 May 2025
- Published by IOP Publishing in Neuromorphic Computing and Engineering
- Vol. 5 (2) , 024007
Open AccessExploiting drain-erase scheme in ferroelectric FETs for logic-in-memory
The conventional computing platforms based on von-Neumann architecture are highly space- and energy-intensive while handling the emerging applications such as AI, ML, and big data. To overcome the von Neumann bottleneck, compact and light-weight logic-in-memory (LiM) implementations of Boolean logic gates based on emerging non-volatile memory (e-NVM) such as RRAMs, PCM, STT-MRAMs, etc were proposed recently. However, these e-NVMs not only exhibit significant temporal and spatial variability, but their large-scale integration with CMOS process is also a technological challenge. To overcome these issues with the emerging non-volatile memories, ferroelectric FETs based on CMOS-compatible doped hafnium oxide with the capability of large-scale CMOS integration in the advanced logic nodes were proposed. Considering the high scalability and CMOS-compatibility of the FeFETs, in this work, for the first time, we propose a LiM implementation utilizing a single ferroelectric fully-depleted-silicon-on-insulator (Fe-FDSOI) FET exploiting the unique drain erase phenomenon. In our proposed LiM implementation, inputs are applied at the gate and drain terminals using a novel input-to-voltage mapping scheme, and output is obtained as the current flowing through the Fe-FDSOI FET. We utilize an experimentally calibrated compact model of the ferroelectric capacitor connected to the baseline industry standard BSIM-IMG compact model for the FDSOI transistor for proof of concept demonstration. We also perform a comprehensive analysis of the performance metrics of the proposed LiM implementation. Our results indicate that we can realize at least 10 Boolean logic gates with high energy and area-efficiency utilizing the proposed scheme.
- 10 April 2025
- Published by American Physical Society (APS) in Physical Review Applied
- Vol. 23 (4)
Harnessing Gaussianlike transfer characteristics for ultraefficient computation in monolayer two-dimensional devices
- 10 April 2025
- Published by American Physical Society (APS) in Physical Review Applied
- Vol. 23 (4)
Harnessing Gaussianlike transfer characteristics for ultraefficient computation in monolayer two-dimensional devices
- 5 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (4) , 1547-1559
Ferroelectric FET-Based Bayesian Inference Engine for Disease Diagnosis
Probabilistic/stochastic computations form the backbone of autonomous systems and classifiers. Recently, biomedical applications of probabilistic computing such as Bayesian networks for disease diagnosis, DNA sequencing, etc. have attracted significant attention owing to their high energy-efficiency. Bayesian inference is widely used for decision making based on independent (often conflicting) sources of information/evidence. A cascaded chain or tree structure of asynchronous circuit elements known as Muller C-elements can effectively implement Bayesian inference. Such circuits utilize stochastic bit streams to encode input probabilities which enhances their robustness and fault-tolerance. However, the CMOS implementations of Muller C-element are bulky and energy hungry which restricts their widespread application in resource constrained IoT and mobile devices such as UAVs, robots, space rovers, etc. In this work, for the first time, we propose a compact and energy-efficient implementation of Muller C-element utilizing a single Ferroelectric FET and use it for cancer diagnosis task by performing Bayesian inference with high accuracy on Wisconsin data set. The proposed implementation exploits the unique drain-erase, program inhibit and drain-erase inhibit characteristics of FeFETs to yield the output as the polarization-state of the ferroelectric layer. Our extensive investigation utilizing an in-house developed experimentally calibrated compact model of FeFET reveals that the proposed C-element consumes (worst-case) energy of 4.1 fJ and an area $0.07~\mu m^{2}$ and outperforms the prior implementations in terms of energy-efficiency and footprint while exhibiting a comparable delay. We also propose a novel read circuitry for realising a Bayesian inference engine by cascading a network of proposed FeFET-based C-elements for practical applications. Furthermore, for the first time, we analyze the impact of cross-correlation between the stochastic input bit streams on the accuracy of the C-element based Bayesian inference implementation.
- 5 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (4) , 1547-1559
Ferroelectric FET-Based Bayesian Inference Engine for Disease Diagnosis
Probabilistic/stochastic computations form the backbone of autonomous systems and classifiers. Recently, biomedical applications of probabilistic computing such as Bayesian networks for disease diagnosis, DNA sequencing, etc. have attracted significant attention owing to their high energy-efficiency. Bayesian inference is widely used for decision making based on independent (often conflicting) sources of information/evidence. A cascaded chain or tree structure of asynchronous circuit elements known as Muller C-elements can effectively implement Bayesian inference. Such circuits utilize stochastic bit streams to encode input probabilities which enhances their robustness and fault-tolerance. However, the CMOS implementations of Muller C-element are bulky and energy hungry which restricts their widespread application in resource constrained IoT and mobile devices such as UAVs, robots, space rovers, etc. In this work, for the first time, we propose a compact and energy-efficient implementation of Muller C-element utilizing a single Ferroelectric FET and use it for cancer diagnosis task by performing Bayesian inference with high accuracy on Wisconsin data set. The proposed implementation exploits the unique drain-erase, program inhibit and drain-erase inhibit characteristics of FeFETs to yield the output as the polarization-state of the ferroelectric layer. Our extensive investigation utilizing an in-house developed experimentally calibrated compact model of FeFET reveals that the proposed C-element consumes (worst-case) energy of 4.1 fJ and an area $0.07~\mu m^{2}$ and outperforms the prior implementations in terms of energy-efficiency and footprint while exhibiting a comparable delay. We also propose a novel read circuitry for realising a Bayesian inference engine by cascading a network of proposed FeFET-based C-elements for practical applications. Furthermore, for the first time, we analyze the impact of cross-correlation between the stochastic input bit streams on the accuracy of the C-element based Bayesian inference implementation.
- Open Access
- 13 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of the Electron Devices Society
- Vol. 13, 822-830
Compact XOR/XNOR-Based Adders and BNNs Utilizing Drain-Erase Scheme in Ferroelectric FETs
Compact and energy-efficient computing avenues such as in-memory computing and processing-in-memory (PIM) are being actively explored to address the limitations of the sparse von-Neumann computing systems. The recent advancements in the field of emerging non-volatile memories (e-NVMs), such as FeFETs, RRAMs, MRAMs, etc., have propelled the development of the PIM technique where the logic operations are performed in situ (where the operands are stored) to reduce the energy draining data movement. Considering the promising potential of the doped-hafnium oxide (HfO2) based FeFETs, such as CMOS compatibility, high scalability, high integration density, and field-driven programming capability, in this work, for the first time, we propose a novel input-to-voltage mapping scheme and exploit drain-erase phenomenon to realize compact and energy-efficient majority logic gate using a single Fe-FDSOI FET, XOR and XNOR logic gates using two Fe-FDSOI FETs. Furthermore, utilizing the proposed FeFET-based XOR and XNOR logic design, we demonstrate a compact implementation of a half adder (using 3 FeFETs) and full adder (utilizing only 9 FETs) which outperforms the CMOS and prior eNVM-based implementations in terms of area and energy. Moreover, we also propose a modified XNOR-cell design utilizing 4 FeFETs for performing bitwise count operations in binary neural networks.
- 13 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of the Electron Devices Society
- Vol. 13, 822-830
Open AccessCompact XOR/XNOR-Based Adders and BNNs Utilizing Drain-Erase Scheme in Ferroelectric FETs
Compact and energy-efficient computing avenues such as in-memory computing and processing-in-memory (PIM) are being actively explored to address the limitations of the sparse von-Neumann computing systems. The recent advancements in the field of emerging non-volatile memories (e-NVMs), such as FeFETs, RRAMs, MRAMs, etc., have propelled the development of the PIM technique where the logic operations are performed in situ (where the operands are stored) to reduce the energy draining data movement. Considering the promising potential of the doped-hafnium oxide (HfO2) based FeFETs, such as CMOS compatibility, high scalability, high integration density, and field-driven programming capability, in this work, for the first time, we propose a novel input-to-voltage mapping scheme and exploit drain-erase phenomenon to realize compact and energy-efficient majority logic gate using a single Fe-FDSOI FET, XOR and XNOR logic gates using two Fe-FDSOI FETs. Furthermore, utilizing the proposed FeFET-based XOR and XNOR logic design, we demonstrate a compact implementation of a half adder (using 3 FeFETs) and full adder (utilizing only 9 FETs) which outperforms the CMOS and prior eNVM-based implementations in terms of area and energy. Moreover, we also propose a modified XNOR-cell design utilizing 4 FeFETs for performing bitwise count operations in binary neural networks.
- Open Access
- 1 November 2024
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 136 (17)
Harnessing room-temperature ferroelectricity in metal oxide monolayers for advanced logic devices
Two-dimensional ferroelectric materials are beneficial for power-efficient memory devices and transistor applications. Here, we predict out-of-plane ferroelectricity in a new family of buckled metal oxide (MO; M: Ge, Sn, Pb) monolayers with significant spontaneous polarization. Additionally, these monolayers have a narrow valence band, which is energetically separated from the rest of the low-lying valence bands. Such a unique band structure limits the long thermal tail of the hot carriers, mitigating subthreshold thermionic leakage and allowing field-effect transistors (FETs) to function beyond the bounds imposed on conventional FETs by thermodynamics. Our quantum transport simulations reveal that the FETs based on these MO monolayers exhibit a large ON/OFF ratio with an average subthreshold swing of less than 60 mV/decade at room temperature, even for short gate lengths. Our work motivates further exploration of the MO monolayers for developing advanced, high-performance memory and logic devices.
- 1 November 2024
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 136 (17)
Open AccessHarnessing room-temperature ferroelectricity in metal oxide monolayers for advanced logic devices
Two-dimensional ferroelectric materials are beneficial for power-efficient memory devices and transistor applications. Here, we predict out-of-plane ferroelectricity in a new family of buckled metal oxide (MO; M: Ge, Sn, Pb) monolayers with significant spontaneous polarization. Additionally, these monolayers have a narrow valence band, which is energetically separated from the rest of the low-lying valence bands. Such a unique band structure limits the long thermal tail of the hot carriers, mitigating subthreshold thermionic leakage and allowing field-effect transistors (FETs) to function beyond the bounds imposed on conventional FETs by thermodynamics. Our quantum transport simulations reveal that the FETs based on these MO monolayers exhibit a large ON/OFF ratio with an average subthreshold swing of less than 60 mV/decade at room temperature, even for short gate lengths. Our work motivates further exploration of the MO monolayers for developing advanced, high-performance memory and logic devices.
- Open Access
- 9 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Enhancing the Capabilities of Quantum Transport Simulations Utilizing Machine Learning Strategies
The ongoing pursuit of exploring novel materials for potential future device applications continues to strengthen the vital role of Technology Computer-Aided Design (TCAD) simulations within the device community. However, its computationally intensive and time-consuming nature necessitates novel methodologies to overcome the limitations. Machine learning (ML) is a potential remedy in our contemporary data-driven society. In this work, we present a unique approach based on Neural Networks (NNs) to generate an efficient potential profile guess. The predicted ML potential enhances the convergence of the coupled Schrodinger and Poisson equation and speeds up the simulation process, ~ 2.5x, by reducing the number of self-consistent iterations at each bias point. Moreover, our method demonstrates versatility by providing reasonable prediction capacity and accuracy for grid reduction, doping, and channel material variations.
- 9 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessEnhancing the Capabilities of Quantum Transport Simulations Utilizing Machine Learning Strategies
The ongoing pursuit of exploring novel materials for potential future device applications continues to strengthen the vital role of Technology Computer-Aided Design (TCAD) simulations within the device community. However, its computationally intensive and time-consuming nature necessitates novel methodologies to overcome the limitations. Machine learning (ML) is a potential remedy in our contemporary data-driven society. In this work, we present a unique approach based on Neural Networks (NNs) to generate an efficient potential profile guess. The predicted ML potential enhances the convergence of the coupled Schrodinger and Poisson equation and speeds up the simulation process, ~ 2.5x, by reducing the number of self-consistent iterations at each bias point. Moreover, our method demonstrates versatility by providing reasonable prediction capacity and accuracy for grid reduction, doping, and channel material variations.
- Open Access
- 19 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of the Electron Devices Society
- Vol. 12, 516-524
Efficient Implementation of Mahalanobis Distance on Ferroelectric FinFET Crossbar for Outlier Detection
The developments in the nascent field of artificial-intelligence-of-things (AIoT) relies heavily on the availability of high-quality multi-dimensional data. A huge amount of data is being collected in this era of big data, predominantly for AI/ML algorithms and emerging applications. Considering such voluminous quantities, the collected data may contain a substantial number of outliers which must be detected before utilizing them for data mining or computations. Therefore, outlier detection techniques such as Mahalanobis distance computation have gained significant popularity recently. Mahalanobis distance, the multivariate equivalent of the Euclidean distance, is used to detect the outliers in the correlated data accurately and finds widespread application in fault identification, data clustering, single-class classification, information security, data mining, etc. However, traditional CMOS-based approaches to compute Mahalanobis distance are bulky and consume a huge amount of energy. Therefore, there is an urgent need for a compact and energy-efficient implementation of an outlier detection technique which may be deployed on AIoT primitives, including wireless sensor nodes for in-situ outlier detection and generation of high-quality data. To this end, in this paper, for the first time, we have proposed an efficient Ferroelectric FinFET-based implementation for detecting outliers in correlated multivariate data using Mahalanobis distance. The proposed implementation utilizes two crossbar arrays of ferroelectric FinFETs to calculate the Mahalanobis distance and detect outliers in the popular Wisconsin breast cancer dataset using a novel inverter-based threshold circuit. Our implementation exhibits an accuracy of 94.1% which is comparable to the software implementations while consuming a significantly low energy (27.2 pJ).
- 19 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of the Electron Devices Society
- Vol. 12, 516-524
Open AccessEfficient Implementation of Mahalanobis Distance on Ferroelectric FinFET Crossbar for Outlier Detection
The developments in the nascent field of artificial-intelligence-of-things (AIoT) relies heavily on the availability of high-quality multi-dimensional data. A huge amount of data is being collected in this era of big data, predominantly for AI/ML algorithms and emerging applications. Considering such voluminous quantities, the collected data may contain a substantial number of outliers which must be detected before utilizing them for data mining or computations. Therefore, outlier detection techniques such as Mahalanobis distance computation have gained significant popularity recently. Mahalanobis distance, the multivariate equivalent of the Euclidean distance, is used to detect the outliers in the correlated data accurately and finds widespread application in fault identification, data clustering, single-class classification, information security, data mining, etc. However, traditional CMOS-based approaches to compute Mahalanobis distance are bulky and consume a huge amount of energy. Therefore, there is an urgent need for a compact and energy-efficient implementation of an outlier detection technique which may be deployed on AIoT primitives, including wireless sensor nodes for in-situ outlier detection and generation of high-quality data. To this end, in this paper, for the first time, we have proposed an efficient Ferroelectric FinFET-based implementation for detecting outliers in correlated multivariate data using Mahalanobis distance. The proposed implementation utilizes two crossbar arrays of ferroelectric FinFETs to calculate the Mahalanobis distance and detect outliers in the popular Wisconsin breast cancer dataset using a novel inverter-based threshold circuit. Our implementation exhibits an accuracy of 94.1% which is comparable to the software implementations while consuming a significantly low energy (27.2 pJ).
- 3 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 71 (7) , 4381-4388
Utilizing Dual-Port FeFETs for Energy-Efficient Binary Neural Network Inference Accelerators
Neuromorphic and in-memory computing architectures using emerging nonvolatile memories (e-NVMs) have emerged as promising solutions for area-and energy-efficient deep neural network (DNN) accele-rators. However, the inherent nonideal behavior of e-NVMs such as limited tuning precision (for multibit synapses), nonlinearity, temporal (cycle-to-cycle), and spatial (device-to-device) variability significantly degrades the performance of DNN accelerators. Recently, binary neural networks (BNNs), with 1-bit weights and activations, have been shown to offer an alternative relaxed approach for training and inference with high accuracy. However, the limited endurance and stuck-at faults of e-NVMs such as resistive (R)RAMs, charge trap memory, and so on limit the efficient implementation of BNN accelerators. Considering the ultrahigh endurance, ultralow switching energy, and CMOS-compatibility of the ferroelectric (Fe)FETs, it becomes imperative to explore their potential for BNN accelerators. To this end, in this work, we present a novel approach for the implementation of BNN inference accelerators utilizing an array of dual-port ferroelectric FETs (FeFETs) as current sinks. The dual-port FeFETs not only decouple the read and write paths (leading to reduced read disturbances) but also exhibit high reliability and voltage compatibility with existing peripheral circuit design since the write voltages are low. Furthermore, we utilize a comparator column approach that requires only half area when compared to other differential weights-based BNN accelerators. Our comprehensive analysis utilizing an experimentally calibrated compact model for dual-port FeFETs indicates that the proposed vector-matrix-multiplication (VMM) implementation exhibits an energy efficiency of 789.89 TOPS/W with a throughput of 0.06 TeraOp/s while achieving an accuracy of 96.39% and 80.8% for image classification task on the MNIST and Fashion MNIST datasets after ex-situ training.
- 3 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 71 (7) , 4381-4388
Utilizing Dual-Port FeFETs for Energy-Efficient Binary Neural Network Inference Accelerators
Neuromorphic and in-memory computing architectures using emerging nonvolatile memories (e-NVMs) have emerged as promising solutions for area-and energy-efficient deep neural network (DNN) accele-rators. However, the inherent nonideal behavior of e-NVMs such as limited tuning precision (for multibit synapses), nonlinearity, temporal (cycle-to-cycle), and spatial (device-to-device) variability significantly degrades the performance of DNN accelerators. Recently, binary neural networks (BNNs), with 1-bit weights and activations, have been shown to offer an alternative relaxed approach for training and inference with high accuracy. However, the limited endurance and stuck-at faults of e-NVMs such as resistive (R)RAMs, charge trap memory, and so on limit the efficient implementation of BNN accelerators. Considering the ultrahigh endurance, ultralow switching energy, and CMOS-compatibility of the ferroelectric (Fe)FETs, it becomes imperative to explore their potential for BNN accelerators. To this end, in this work, we present a novel approach for the implementation of BNN inference accelerators utilizing an array of dual-port ferroelectric FETs (FeFETs) as current sinks. The dual-port FeFETs not only decouple the read and write paths (leading to reduced read disturbances) but also exhibit high reliability and voltage compatibility with existing peripheral circuit design since the write voltages are low. Furthermore, we utilize a comparator column approach that requires only half area when compared to other differential weights-based BNN accelerators. Our comprehensive analysis utilizing an experimentally calibrated compact model for dual-port FeFETs indicates that the proposed vector-matrix-multiplication (VMM) implementation exhibits an energy efficiency of 789.89 TOPS/W with a throughput of 0.06 TeraOp/s while achieving an accuracy of 96.39% and 80.8% for image classification task on the MNIST and Fashion MNIST datasets after ex-situ training.