Rajiv V. Joshi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Rajiv V. Joshi has a total of 165 co-authors. Here are the top co-authors by number of shared publications:
- Rouwaida Kanj
Electrical and Computer Engineering Department, Maroun Semaan Faculty of Engineering and Architecture, American University of Beirut, Beirut, Lebanon
9 shared publications - Said Hamdioui
Delft University of Technology
6 shared publications - Anteneh Gebregiorgis
Department of Quantum and Computer Engineering, Delft University of Technology, Delft, CD, The Netherlands
4 shared publications - Rajendra Bishnoi
Computer Engineering Lab, Delft University of Technology, Delft, The Netherlands
4 shared publications - Sumit Diware
Computer Engineering Laboratory, Delft University of Technology, Delft, The Netherlands
4 shared publications - Abhairaj Singh
IBM Research Europe, Rüschlikon, Switzerland
3 shared publications
Latest Publications
- 30 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 16 (2) , 327-337
ShiftMUX: Reconfigurable 4:1 MUX-Based In-Memory Shifter for Multiplication-Less Deep Neural Network Accelerations
This paper presents ShiftMUX, a reconfigurable 4:1 multiplexer (MUX)-based in-memory computing (IMC) architecture for multiplier-less deep neural network (DNN) acceleration. The proposed design completes its operation in only three cycles, compared with at least ten cycles in prior approaches. Beyond basic logic, ShiftMUX supports complex functions such as full adder, full subtractor, and comparator, eliminating the need for separate mapping and control peripherals. When cascaded in a crossbar, multiple MUX units realize N-bit shift operations including logical, arithmetic, circular, and universal shifts enabling efficient power-of-two multiplications for quantized DNNs. Circuit-level simulations in 45nm technology show write energy of 32 pJ, read delay of 9.14 ps, and a total energy consumption of 240 pJ for a 4-bit multiplier, outperforming state-of-the-art designs. A Vision Transformer model exhibits only a +2.66% accuracy deviation with shift-based multiplication, confirming the suitability of ShiftMUX for modern AI models. Moreover, the architecture reduces computation overhead by more than 90%, requiring only N +log2(N)+2 cycles, and achieves up to 9× speedup in multiplication. ShiftMUX thus provides a scalable, reconfigurable, and energy-efficient solution for low-latency DNN acceleration in edge and high-performance computing systems.
- 30 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 16 (2) , 327-337
ShiftMUX: Reconfigurable 4:1 MUX-Based In-Memory Shifter for Multiplication-Less Deep Neural Network Accelerations
This paper presents ShiftMUX, a reconfigurable 4:1 multiplexer (MUX)-based in-memory computing (IMC) architecture for multiplier-less deep neural network (DNN) acceleration. The proposed design completes its operation in only three cycles, compared with at least ten cycles in prior approaches. Beyond basic logic, ShiftMUX supports complex functions such as full adder, full subtractor, and comparator, eliminating the need for separate mapping and control peripherals. When cascaded in a crossbar, multiple MUX units realize N-bit shift operations including logical, arithmetic, circular, and universal shifts enabling efficient power-of-two multiplications for quantized DNNs. Circuit-level simulations in 45nm technology show write energy of 32 pJ, read delay of 9.14 ps, and a total energy consumption of 240 pJ for a 4-bit multiplier, outperforming state-of-the-art designs. A Vision Transformer model exhibits only a +2.66% accuracy deviation with shift-based multiplication, confirming the suitability of ShiftMUX for modern AI models. Moreover, the architecture reduces computation overhead by more than 90%, requiring only N +log2(N)+2 cycles, and achieves up to 9× speedup in multiplication. ShiftMUX thus provides a scalable, reconfigurable, and energy-efficient solution for low-latency DNN acceleration in edge and high-performance computing systems.
- Open Access
- 9 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 11, 148-156
Understanding Reliability Trade-Offs in 1T-nC and 2T-nC FeRAM Designs
Ferroelectric random-access memory (FeRAM) is a promising candidate for energy-efficient non-volatile memory, particularly for logic-in-memory and compute-in-memory applications. Among the available cell architectures, 1T–nC and 2T–nC FeRAMs each offer distinct trade-offs in density, scalability, and reliability. In this work, we present a comparative study of these two architectures under both dimensional scaling (XY/Z shrinkage) and vertical integration (increasing stacked capacitors per cell). Using TCAD and circuit-level simulations, we analyze how scaling impacts ferroelectric capacitance, parasitic coupling, and floating-node dynamics, which together dictate sense margin and read stability. A key mitigation strategy—floating unselected capacitors—is applied to both architectures, effectively decoupling the sense margin from the number of stacked capacitors and enabling tractable analysis across scaling regimes. Results show that 1T–nC suffers more from charge sharing with the bitline, while 2T–nC benefits from transistor isolation and stronger low-voltage sensing at the cost of increased area. By systematically evaluating these behaviors across scaling directions, this work establishes the reliability trade-offs of 1T–nC and 2T–nC cells and provides design guidelines for high-density, vertically integrated FeRAM systems.
- 9 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 11, 148-156
Open AccessUnderstanding Reliability Trade-Offs in 1T-nC and 2T-nC FeRAM Designs
Ferroelectric random-access memory (FeRAM) is a promising candidate for energy-efficient non-volatile memory, particularly for logic-in-memory and compute-in-memory applications. Among the available cell architectures, 1T–nC and 2T–nC FeRAMs each offer distinct trade-offs in density, scalability, and reliability. In this work, we present a comparative study of these two architectures under both dimensional scaling (XY/Z shrinkage) and vertical integration (increasing stacked capacitors per cell). Using TCAD and circuit-level simulations, we analyze how scaling impacts ferroelectric capacitance, parasitic coupling, and floating-node dynamics, which together dictate sense margin and read stability. A key mitigation strategy—floating unselected capacitors—is applied to both architectures, effectively decoupling the sense margin from the number of stacked capacitors and enabling tractable analysis across scaling regimes. Results show that 1T–nC suffers more from charge sharing with the bitline, while 2T–nC benefits from transistor isolation and stronger low-voltage sensing at the cost of increased area. By systematically evaluating these behaviors across scaling directions, this work establishes the reliability trade-offs of 1T–nC and 2T–nC cells and provides design guidelines for high-density, vertically integrated FeRAM systems.
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 25 (3) , 66-79
7th IBM IEEE CAS/EDS AI Compute Symposium (AICS’24) (Global Industry-University Brain Storming Kumbh Mela) [CASS Conference Highlights]
Provides society information that may include news, reviews or technical notes that should be of interest to practitioners and researchers.
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 25 (3) , 66-79
7th IBM IEEE CAS/EDS AI Compute Symposium (AICS’24) (Global Industry-University Brain Storming Kumbh Mela) [CASS Conference Highlights]
Provides society information that may include news, reviews or technical notes that should be of interest to practitioners and researchers.
- 16 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Artificial Intelligence
- Vol. 2 (3) , 211-221
DREAM-CIM: A Digital SRAM-Based CIM Accelerator for Energy- and Area-Efficient Edge AI
With the rise of energy-constrained smart edge applications, there is a pressing need for energy-efficient computing engines that process generated data locally, at least for small and medium-sized applications. To address this issue, this paper proposes DREAM-CIM, a digital SRAM-based computationin-memory (CIM) accelerator. It targets an energy- and area-efficient implementation of the multiply-and-accumulate (MAC) operation, which is the core operation of neural networks. The accelerator is based on a multi-sub-array macro to increase parallelism, integrates multiplication operations within the memory cells such that they are executed while reading the cells, makes use of pipelining to further optimize the throughput of the MAC operations, and gets rid of the expensive adder-tree structures commonly used in State-of-The-Art (SOTA) digital CIM solutions by replacing them with a custom accumulation circuit to reduce power and area. The SPICE simulation results of the DREAM-CIM accelerator show an energy efficiency of 5097 TOPS/W (normalized to a 1-bit × 1-bit MAC operation) and an area efficiency of 3854 TOPS/mm2 using 22 nm technology node. The obtained circuit-level results were fed into a python-based system-level simulator to benchmark the system architecture using two applications, i.e., image classification (using MNIST and CIFAR-10 dataset on LeNet5 and Resnet-20 models) and object detection (using COCO dataset on the YoloV6 model). The system-level results show that DREAM-CIM can achieve an energy efficiency of 0.1mJ, 0.2mJ, and 11.02mJ per inference for the MNIST, YOLOv6, and CIFAR-10 datasets, respectively, while maintaining SOTA accuracy.
- 16 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Artificial Intelligence
- Vol. 2 (3) , 211-221
DREAM-CIM: A Digital SRAM-Based CIM Accelerator for Energy- and Area-Efficient Edge AI
With the rise of energy-constrained smart edge applications, there is a pressing need for energy-efficient computing engines that process generated data locally, at least for small and medium-sized applications. To address this issue, this paper proposes DREAM-CIM, a digital SRAM-based computationin-memory (CIM) accelerator. It targets an energy- and area-efficient implementation of the multiply-and-accumulate (MAC) operation, which is the core operation of neural networks. The accelerator is based on a multi-sub-array macro to increase parallelism, integrates multiplication operations within the memory cells such that they are executed while reading the cells, makes use of pipelining to further optimize the throughput of the MAC operations, and gets rid of the expensive adder-tree structures commonly used in State-of-The-Art (SOTA) digital CIM solutions by replacing them with a custom accumulation circuit to reduce power and area. The SPICE simulation results of the DREAM-CIM accelerator show an energy efficiency of 5097 TOPS/W (normalized to a 1-bit × 1-bit MAC operation) and an area efficiency of 3854 TOPS/mm2 using 22 nm technology node. The obtained circuit-level results were fed into a python-based system-level simulator to benchmark the system architecture using two applications, i.e., image classification (using MNIST and CIFAR-10 dataset on LeNet5 and Resnet-20 models) and object detection (using COCO dataset on the YoloV6 model). The system-level results show that DREAM-CIM can achieve an energy efficiency of 0.1mJ, 0.2mJ, and 11.02mJ per inference for the MNIST, YOLOv6, and CIFAR-10 datasets, respectively, while maintaining SOTA accuracy.
- 7 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 72 (6) , 3327-3334
Demonstration of Vertical 2T-nC FeRAM Hybrid Cell and Its Scalability for High-Density 3-D Ferroelectric Capacitor Memory
In this work, we present a comprehensive experimental and modeling study on the scaling of vertical 2T-nC ferroelectric random access memory (FeRAM) hybrid cells, comprising n metal–ferroelectric–metal (MFM) capacitors, to demonstrate a high-performance and high-density 3-D capacitor memory. Our contributions include: 1) successful process integration of vertical 2T-3C FeRAM cells by stacking MFM structures on top of Si CMOS transistors; 2) experimental validation of memory cell functionality, confirming the feasibility of the vertical 2T-nC FeRAM architecture; 3) an analysis of scaling effects on parasitic capacitance in densely integrated 3-D arrays, using 3-D technology computer-aided design (TCAD) simulations; 4) exploration of aggressive stacking of write bitlines (WBLs) to enhance memory density, where ferroelectric linear capacitance (CFE) enables self-boosted inhibition under the VW/2 scheme, but renders the VW/3 scheme ineffective due to intolerable write disturbances; and 5) assessment of horizontal scaling, revealing significant increases in read disturbances caused by interplane capacitance between adjacent WBLs (CZ). This work represents an early exploration into the potential of 2T-nC FeRAM as a scalable and efficient 3-D memory solution.
- 7 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 72 (6) , 3327-3334
Demonstration of Vertical 2T-nC FeRAM Hybrid Cell and Its Scalability for High-Density 3-D Ferroelectric Capacitor Memory
In this work, we present a comprehensive experimental and modeling study on the scaling of vertical 2T-nC ferroelectric random access memory (FeRAM) hybrid cells, comprising n metal–ferroelectric–metal (MFM) capacitors, to demonstrate a high-performance and high-density 3-D capacitor memory. Our contributions include: 1) successful process integration of vertical 2T-3C FeRAM cells by stacking MFM structures on top of Si CMOS transistors; 2) experimental validation of memory cell functionality, confirming the feasibility of the vertical 2T-nC FeRAM architecture; 3) an analysis of scaling effects on parasitic capacitance in densely integrated 3-D arrays, using 3-D technology computer-aided design (TCAD) simulations; 4) exploration of aggressive stacking of write bitlines (WBLs) to enhance memory density, where ferroelectric linear capacitance (CFE) enables self-boosted inhibition under the VW/2 scheme, but renders the VW/3 scheme ineffective due to intolerable write disturbances; and 5) assessment of horizontal scaling, revealing significant increases in read disturbances caused by interplane capacitance between adjacent WBLs (CZ). This work represents an early exploration into the potential of 2T-nC FeRAM as a scalable and efficient 3-D memory solution.
- Open Access
- 13 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Solid-State Circuits Society
- Vol. 5, 60-74
SRAM and Mixed-Signal Logic With Noise Immunity in 3-nm Nano-Sheet Technology
A modular 4.26Mb SRAM based on a 82Kb/block structure with mixed signal logic is fabricated, characterized, and demonstrated with full functionality in a 3nm nanosheet (NS) technology. Designed macros utilize new circuits for supply boosting, read, and write assist techniques. The proposed circuits are evaluated extensively and compared to prior techniques. Statistical simulations are used to predict the benefits of these circuits in the context of dual supply use. Through programmable local clock and wordline (WL) pulsewidths, SRAM cell margins and speeds are demonstrated through hardware measurement. Stability assists as well as dual supply techniques are used to demonstrate how noise can be suppressed during traditional memory operations (single WL on), as well as to support mixed-signal logic block operation (multiple WLs on). Functionality is shown down to a cell supply of 0.45V with an estimated margin/speed of 6 GHz for SRAM cells (High Density-0.026μm2, and High Current -0.032μ m2).
- 13 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Solid-State Circuits Society
- Vol. 5, 60-74
Open AccessSRAM and Mixed-Signal Logic With Noise Immunity in 3-nm Nano-Sheet Technology
A modular 4.26Mb SRAM based on a 82Kb/block structure with mixed signal logic is fabricated, characterized, and demonstrated with full functionality in a 3nm nanosheet (NS) technology. Designed macros utilize new circuits for supply boosting, read, and write assist techniques. The proposed circuits are evaluated extensively and compared to prior techniques. Statistical simulations are used to predict the benefits of these circuits in the context of dual supply use. Through programmable local clock and wordline (WL) pulsewidths, SRAM cell margins and speeds are demonstrated through hardware measurement. Stability assists as well as dual supply techniques are used to demonstrate how noise can be suppressed during traditional memory operations (single WL on), as well as to support mixed-signal logic block operation (multiple WLs on). Functionality is shown down to a cell supply of 0.45V with an estimated margin/speed of 6 GHz for SRAM cells (High Density-0.026μm2, and High Current -0.032μ m2).
- 19 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 24 (3) , 49-53
6th IBM IEEE CAS/EDS AI Compute Symposium (AICS’23) [CASS Conference Highlights]
The 6th IBM IEEE CAS/EDS AI Compute Symposium was held hybrid at the T. J. Watson Research Center on 28 November 2023. The event was extremely successful and well attended by over 2000 folks from all over the world (in-person and virtual). The symposium featured 8 distinguished speakers (7 from industry and 1 from academia), over 30 student in-person posters, best poster awards, and a panel discussion. The registration list spanned citizens of 53 countries. The theme of the symposium, “From Chips to Chiplets,” turned out to be an opportune and important topic for the current semiconductor industry direction. The symposium served as an educational as well as a brainstorming session for industry/academia/students across the world. The symposium covered a range of topics from emerging device technology, innovative circuits, chip and chiplet architecture, advanced packaging technologies, such as 2D to 3D packaging elements, and how these topics drive the rapid growth of AI and generative AI. Dr. Rajiv Joshi, General Chair and IEEE Life Fellow opened the symposium with welcoming remarks along with the goals and accomplishments of this symposium under the auspices of CAS and IBM.
- 19 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 24 (3) , 49-53
6th IBM IEEE CAS/EDS AI Compute Symposium (AICS’23) [CASS Conference Highlights]
The 6th IBM IEEE CAS/EDS AI Compute Symposium was held hybrid at the T. J. Watson Research Center on 28 November 2023. The event was extremely successful and well attended by over 2000 folks from all over the world (in-person and virtual). The symposium featured 8 distinguished speakers (7 from industry and 1 from academia), over 30 student in-person posters, best poster awards, and a panel discussion. The registration list spanned citizens of 53 countries. The theme of the symposium, “From Chips to Chiplets,” turned out to be an opportune and important topic for the current semiconductor industry direction. The symposium served as an educational as well as a brainstorming session for industry/academia/students across the world. The symposium covered a range of topics from emerging device technology, innovative circuits, chip and chiplet architecture, advanced packaging technologies, such as 2D to 3D packaging elements, and how these topics drive the rapid growth of AI and generative AI. Dr. Rajiv Joshi, General Chair and IEEE Life Fellow opened the symposium with welcoming remarks along with the goals and accomplishments of this symposium under the auspices of CAS and IBM.
- 15 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 24 (2) , 7-15
Past, Present, and Future of CASS Educational Programs and Initiatives [Feature]
The Circuits and Systems Society (CASS) places a strong emphasis on advancing education within its domain. Various educational programs and resources have contributed to the professional development of CASS members for more than 20 years now. A major transformation within the Society was started in 2021 with a clear vision to become a premier educational resource within circuits and systems areas of interest. This article provides a comprehensive review of flagship educational programs and past initiatives. Insights from the lessons learned and future trends in the education ecosystem guided society’s strategies and frameworks for improving and developing new educational products and services. The benefits of learning with the CASS program are presented, highlighting the Society’s deep commitment to serving the community at large in educational pursuit for years to come.
- 15 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 24 (2) , 7-15
Past, Present, and Future of CASS Educational Programs and Initiatives [Feature]
The Circuits and Systems Society (CASS) places a strong emphasis on advancing education within its domain. Various educational programs and resources have contributed to the professional development of CASS members for more than 20 years now. A major transformation within the Society was started in 2021 with a clear vision to become a premier educational resource within circuits and systems areas of interest. This article provides a comprehensive review of flagship educational programs and past initiatives. Insights from the lessons learned and future trends in the education ecosystem guided society’s strategies and frameworks for improving and developing new educational products and services. The benefits of learning with the CASS program are presented, highlighting the Society’s deep commitment to serving the community at large in educational pursuit for years to come.
- 15 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 24 (2) , 34-46
Cryogenic CMOS Design for Qubit Control: Present Status, Challenges, and Future Directions [Feature]
This article will review recent progress in cryogenic CMOS designs for future scaled quantum computing applications. After introducing the scaling challenges associated with qubit control and readout electronics operating at room temperature, approaches taken to date to cryogenic control electronics design will be discussed, focusing on the most recent relevant publications. Elements of ultra-low power circuit and system design approaches for cryogenic controllers in scaled CMOS nodes (40nm to 14nm) will be reviewed, including a discussion of current state-of-the art cryogenic controller performance and power efficiency. Note that leading designs, when operated as transmon qubit state controllers, have achieved gate error rates in the range of 10-4 to 10-3 achieving spurious free dynamic range (SFDR) of ~40dB while consuming 4-23mW of power per qubit under active control, with power efficiency strongly driven by the complexity of the digital processor integrated in the controller design. These demonstrations, while significant, are just the first steps toward achieving the performance, efficiency, and scalability that will be required for future systems. This review article will discuss fundamental tradeoffs in CMOS cryogenic designs in order to address the needs of future scaled quantum computing systems.
- 15 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Circuits and Systems Magazine
- Vol. 24 (2) , 34-46
Cryogenic CMOS Design for Qubit Control: Present Status, Challenges, and Future Directions [Feature]
This article will review recent progress in cryogenic CMOS designs for future scaled quantum computing applications. After introducing the scaling challenges associated with qubit control and readout electronics operating at room temperature, approaches taken to date to cryogenic control electronics design will be discussed, focusing on the most recent relevant publications. Elements of ultra-low power circuit and system design approaches for cryogenic controllers in scaled CMOS nodes (40nm to 14nm) will be reviewed, including a discussion of current state-of-the art cryogenic controller performance and power efficiency. Note that leading designs, when operated as transmon qubit state controllers, have achieved gate error rates in the range of 10-4 to 10-3 achieving spurious free dynamic range (SFDR) of ~40dB while consuming 4-23mW of power per qubit under active control, with power efficiency strongly driven by the complexity of the digital processor integrated in the controller design. These demonstrations, while significant, are just the first steps toward achieving the performance, efficiency, and scalability that will be required for future systems. This review article will discuss fundamental tradeoffs in CMOS cryogenic designs in order to address the needs of future scaled quantum computing systems.
- Open Access
- 29 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 47469-47482
Reliable and Energy-Efficient Diabetic Retinopathy Screening Using Memristor-Based Neural Networks
Diabetic retinopathy (DR) is a leading cause of permanent vision loss worldwide. It refers to irreversible retinal damage caused due to elevated glucose levels and blood pressure. Regular screening for DR can facilitate its early detection and timely treatment. Neural network-based DR classifiers can be leveraged to achieve such screening in a convenient and automated manner. However, these classifiers suffer from reliability issue where they exhibit strong performance during development but degraded performance after deployment. Moreover, they do not provide supplementary information about the prediction outcome, which severely limits their widespread adoption. Furthermore, energy-efficient deployment of these classifiers on edge devices remains unaddressed, which is crucial to enhance their global accessibility. In this paper, we present a reliable and energy-efficient hardware for DR detection, suitable for deployment on edge devices. We first develop a DR classification model using custom training data that incorporates diverse image quality and image sources along with improved class balance. This enables our model to effectively handle both on-field variations in retinal images and minority DR classes, enhancing its post-deployment reliability. We then propose a pseudo-binary classification scheme to further improve the model performance and provide supplementary information about the model prediction. Additionally, we present an energy-efficient hardware design for our model using memristor-based computation-in-memory, to facilitate its deployment on edge devices. Our proposed approach achieves reliable DR classification with three orders of magnitude reduction in energy consumption over state-of-the-art hardware platforms.
- 29 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 47469-47482
Open AccessReliable and Energy-Efficient Diabetic Retinopathy Screening Using Memristor-Based Neural Networks
Diabetic retinopathy (DR) is a leading cause of permanent vision loss worldwide. It refers to irreversible retinal damage caused due to elevated glucose levels and blood pressure. Regular screening for DR can facilitate its early detection and timely treatment. Neural network-based DR classifiers can be leveraged to achieve such screening in a convenient and automated manner. However, these classifiers suffer from reliability issue where they exhibit strong performance during development but degraded performance after deployment. Moreover, they do not provide supplementary information about the prediction outcome, which severely limits their widespread adoption. Furthermore, energy-efficient deployment of these classifiers on edge devices remains unaddressed, which is crucial to enhance their global accessibility. In this paper, we present a reliable and energy-efficient hardware for DR detection, suitable for deployment on edge devices. We first develop a DR classification model using custom training data that incorporates diverse image quality and image sources along with improved class balance. This enables our model to effectively handle both on-field variations in retinal images and minority DR classes, enhancing its post-deployment reliability. We then propose a pseudo-binary classification scheme to further improve the model performance and provide supplementary information about the model prediction. Additionally, we present an energy-efficient hardware design for our model using memristor-based computation-in-memory, to facilitate its deployment on edge devices. Our proposed approach achieves reliable DR classification with three orders of magnitude reduction in energy consumption over state-of-the-art hardware platforms.