Sanghamitra Roy
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Sanghamitra Roy has a total of 37 co-authors. Here are the top co-authors by number of shared publications:
- Koushik Chakraborty
Utah State University
28 shared publications - Prabal Basu
Cadence Design Systems
9 shared publications - Pramesh Pandey
Bridge Lab, Electrical and Computer Engineering, Utah State University, Logan, UT 84321, USA
6 shared publications - Aatreyi Bal
USU BRIDGE Laboratory, Electrical and Computer Engineering Department, Utah State University, Logan, UT 84322 USA
5 shared publications - Arnd Hartmanns
University of Twente
5 shared publications - Noel Daniel Gundi
Electrical and Computer Engineering, Utah State University, College of Engineering, Logan, United States
5 shared publications
Latest Publications
- Open Access
- 17 August 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Design Automation of Electronic Systems
Dynamic Slack-Aware Clocking for Near-Threshold Tensor Processing Units (TPUs)
Operating Tensor Processing Units (TPUs) in the near-threshold computing (NTC) region significantly reduces energy consumption but introduces high delay sensitivity to process variation and data activity. Conventional designs typically rely on a conservative, fixed global clock to ensure safety, which leaves large portions of timing margin unexploited as most operations finish well before the clock edge. We propose Dynamic Slack-Aware Clocking (DSAC), a proactive framework that replaces worst-case timing with operation-specific adjustments. DSAC employs lightweight Hamming-Distance, Most-Significant-Bit, and Hybrid predictors to estimate the delay sensitivity of individual multiply-accumulate (MAC) operations and classify them into three timing tiers. These tiers are enforced locally via dummy-hold cycles under a fixed global reference clock, enabling fine-grained timing adaptation without global clock retuning or frequency scaling. A closed-loop feedback controller monitors timing violations and updates tier thresholds at runtime to maintain resilience. Experiments on quantized DNN benchmarks demonstrate that the MSB predictor maintains high inference accuracy, with an average loss of only 1% even at aggressive performance points. Furthermore, DSAC achieves up to 1.55 Ć better energy efficiency at 2.15 Ć frequency scaling compared to a baseline TPU, while incurring an area overhead as low as 13%.
- 17 August 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Design Automation of Electronic Systems
Open AccessDynamic Slack-Aware Clocking for Near-Threshold Tensor Processing Units (TPUs)
Operating Tensor Processing Units (TPUs) in the near-threshold computing (NTC) region significantly reduces energy consumption but introduces high delay sensitivity to process variation and data activity. Conventional designs typically rely on a conservative, fixed global clock to ensure safety, which leaves large portions of timing margin unexploited as most operations finish well before the clock edge. We propose Dynamic Slack-Aware Clocking (DSAC), a proactive framework that replaces worst-case timing with operation-specific adjustments. DSAC employs lightweight Hamming-Distance, Most-Significant-Bit, and Hybrid predictors to estimate the delay sensitivity of individual multiply-accumulate (MAC) operations and classify them into three timing tiers. These tiers are enforced locally via dummy-hold cycles under a fixed global reference clock, enabling fine-grained timing adaptation without global clock retuning or frequency scaling. A closed-loop feedback controller monitors timing violations and updates tier thresholds at runtime to maintain resilience. Experiments on quantized DNN benchmarks demonstrate that the MSB predictor maintains high inference accuracy, with an average loss of only 1% even at aggressive performance points. Furthermore, DSAC achieves up to 1.55 Ć better energy efficiency at 2.15 Ć frequency scaling compared to a baseline TPU, while incurring an area overhead as low as 13%.
- 19 February 2026
- Published byĀ Institute of Electrical and Electronics Engineers (IEEE)Ā inĀ IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol.Ā 34 Ā (4) , 1206-1216
LOgIQ: Log-Domain Optimization for Hardware-Efficient Inference and Quantization of Transformers
Quantization enables efficient large language model (LLM) inference by reducing memory and computation cost, but most existing methods sacrifice hardware simplicity through floating-point fallback, dequantization logic, or metadata routing. These overheads disrupt datapath regularity and limit gains in timing and area, especially in multiply-accumulate (MAC)-heavy transformer layers. We propose LogiMAC, a hardware-aligned quantization scheme that encodes weights in the log domain using shift-based arithmetic. LogiMAC embeds all decoding logic within the MAC unit, eliminates runtime scaling, and supports sublevel refinement without multipliers or look-up tables. The resulting datapath supports both shift directions with optional gating for unused paths. We evaluate LogiMAC across seven modern LLMs (ranging from 2B to 8B parameters) and compare it to strong baselines including SmoothQuant, GOBO, and SpinQuant. Our results show that while functional accuracy remains competitive, LogiMAC reduces area by $1.8\times $ and dynamic power by $5.9\times $ (in power-gated mode) compared to standard integer baselines. Timing analysis reveals over 50% of LogiMAC paths complete within just 12% of the normalized delay seen in other schemes. Validated over 34 million generated tokens, our findings demonstrate that accurate low-bit quantization can coexist with extreme hardware efficiency.
- 19 February 2026
- Published byĀ Institute of Electrical and Electronics Engineers (IEEE)Ā inĀ IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol.Ā 34 Ā (4) , 1206-1216
LOgIQ: Log-Domain Optimization for Hardware-Efficient Inference and Quantization of Transformers
Quantization enables efficient large language model (LLM) inference by reducing memory and computation cost, but most existing methods sacrifice hardware simplicity through floating-point fallback, dequantization logic, or metadata routing. These overheads disrupt datapath regularity and limit gains in timing and area, especially in multiply-accumulate (MAC)-heavy transformer layers. We propose LogiMAC, a hardware-aligned quantization scheme that encodes weights in the log domain using shift-based arithmetic. LogiMAC embeds all decoding logic within the MAC unit, eliminates runtime scaling, and supports sublevel refinement without multipliers or look-up tables. The resulting datapath supports both shift directions with optional gating for unused paths. We evaluate LogiMAC across seven modern LLMs (ranging from 2B to 8B parameters) and compare it to strong baselines including SmoothQuant, GOBO, and SpinQuant. Our results show that while functional accuracy remains competitive, LogiMAC reduces area by $1.8\times $ and dynamic power by $5.9\times $ (in power-gated mode) compared to standard integer baselines. Timing analysis reveals over 50% of LogiMAC paths complete within just 12% of the normalized delay seen in other schemes. Validated over 34 million generated tokens, our findings demonstrate that accurate low-bit quantization can coexist with extreme hardware efficiency.
- 13 January 2026
- Published byĀ Springer Nature
- p. 383-407
Probabilistic Verification for Modular Network-on-Chip Systems
- 13 January 2026
- Published byĀ Springer Nature
- p. 383-407
Probabilistic Verification for Modular Network-on-Chip Systems
- Open Access
- 13 January 2026
- Published byĀ MDPI AGĀ inĀ Journal of Low Power Electronics and Applications
- Vol.Ā 16 Ā (1) , 3
Exploring Runtime Sparsification of YOLO Model Weights During Inference
In the pursuit of real-time object detection with constrained computational resources, the optimization of neural network architectures is paramount. We introduce novel sparsity induction methods within the YOLOv4-Tiny framework to significantly improve computational efficiency while maintaining high accuracy in pedestrian detection. We present three sparsification approaches: Homogeneous, Progressive, and Layer-Adaptive, each methodically reducing the modelās complexity without compromising its detection capability. Additionally, we refine the modelās output with a memory-efficient sliding window approach and a Bounding Box Sorting Algorithm, ensuring precise Intersection over Union (IoU) calculations. Our results demonstrate a substantial reduction in computational load by zeroing out over 50% of the weights with only a minimal 6% loss in IoU and 0.6% loss in F1-Score.
- 13 January 2026
- Published byĀ MDPI AGĀ inĀ Journal of Low Power Electronics and Applications
- Vol.Ā 16 Ā (1) , 3
Open AccessExploring Runtime Sparsification of YOLO Model Weights During Inference
In the pursuit of real-time object detection with constrained computational resources, the optimization of neural network architectures is paramount. We introduce novel sparsity induction methods within the YOLOv4-Tiny framework to significantly improve computational efficiency while maintaining high accuracy in pedestrian detection. We present three sparsification approaches: Homogeneous, Progressive, and Layer-Adaptive, each methodically reducing the modelās complexity without compromising its detection capability. Additionally, we refine the modelās output with a memory-efficient sliding window approach and a Bounding Box Sorting Algorithm, ensuring precise Intersection over Union (IoU) calculations. Our results demonstrate a substantial reduction in computational load by zeroing out over 50% of the weights with only a minimal 6% loss in IoU and 0.6% loss in F1-Score.
- Open Access
Probabilistic Verification for Modular Network-on-Chip Systems (extended version)
Quantitative verification can provide deep insights into reliable Network-On-Chip (NoC) designs. It is critical to understanding and mitigating operational issues caused by power supply noise (PSN) early in the design process: fluctuations in network traffic in modern NoC designs cause dramatic variations in power delivery across the network, leading to unreliability and errors in data transfers. Further complicating these challenges, NoC designs vary widely in size, usage, and implementation. This case study paper presents a principled, systematic, and modular NoC modeling approach using the Modest language that closely reflects the standard hierarchical design approach in digital systems. Using the Modest Toolset, functional and quantitative correctness was established for several NoC models, all of which were instantiated from a generic modular router model. Specifically, this work verifies the functional correctness of a generic router, inter-router communication, and the entire NoC. Statistical model checking was used to verify PSN-related properties for NoCs of size up to 8x8.
Open AccessProbabilistic Verification for Modular Network-on-Chip Systems (extended version)
Quantitative verification can provide deep insights into reliable Network-On-Chip (NoC) designs. It is critical to understanding and mitigating operational issues caused by power supply noise (PSN) early in the design process: fluctuations in network traffic in modern NoC designs cause dramatic variations in power delivery across the network, leading to unreliability and errors in data transfers. Further complicating these challenges, NoC designs vary widely in size, usage, and implementation. This case study paper presents a principled, systematic, and modular NoC modeling approach using the Modest language that closely reflects the standard hierarchical design approach in digital systems. Using the Modest Toolset, functional and quantitative correctness was established for several NoC models, all of which were instantiated from a generic modular router model. Specifically, this work verifies the functional correctness of a generic router, inter-router communication, and the entire NoC. Statistical model checking was used to verify PSN-related properties for NoCs of size up to 8x8.
- Open Access
- 10 January 2025
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Design Automation of Electronic Systems
- Vol.Ā 30 Ā (2) , 1-25
STRIVE: Empowering a Low Power Tensor Processing Unit with Fault Detection and Error Resilience
Rapid growth in Deep Neural Network (DNN) workloads has increased the energy footprint of the Artificial Intelligence (AI) computing realm. For optimum energy efficiency, we propose operating a DNN hardware in the Low-Power Computing (LPC) region. However, operating at LPC causes increased delay sensitivity to Process Variation (PV). Delay faults are an intriguing consequence of PV. In this article, we demonstrate the vulnerability of DNNs to delay variations, substantially lowering the prediction accuracy. To overcome delay faults, we present STRIVEāa post-fabrication fault detection and reactive error reduction technique. We also introduce a time-borrow correction technique to ensure error-free DNN computation.
- 10 January 2025
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Design Automation of Electronic Systems
- Vol.Ā 30 Ā (2) , 1-25
Open AccessSTRIVE: Empowering a Low Power Tensor Processing Unit with Fault Detection and Error Resilience
Rapid growth in Deep Neural Network (DNN) workloads has increased the energy footprint of the Artificial Intelligence (AI) computing realm. For optimum energy efficiency, we propose operating a DNN hardware in the Low-Power Computing (LPC) region. However, operating at LPC causes increased delay sensitivity to Process Variation (PV). Delay faults are an intriguing consequence of PV. In this article, we demonstrate the vulnerability of DNNs to delay variations, substantially lowering the prediction accuracy. To overcome delay faults, we present STRIVEāa post-fabrication fault detection and reactive error reduction technique. We also introduce a time-borrow correction technique to ensure error-free DNN computation.
- Open Access
- 10 April 2024
- Published byĀ MDPI AGĀ inĀ Electronics
- Vol.Ā 13 Ā (8) , 1431
Voltage Scaled Low Power DNN Accelerator Design on Reconfigurable Platform
The exponential emergence of Field-Programmable Gate Arrays (FPGAs) has accelerated research on hardware implementation of Deep Neural Networks (DNNs). Among all DNN processors, domain-specific architectures such as Google’s Tensor Processor Unit (TPU) have outperformed conventional GPUs (Graphics Processing Units) and CPUs (Central Processing Units). However, implementing low-power TPUs in reconfigurable hardware remains a challenge in this field. Voltage scaling, a popular approach for energy savings, can be challenging in FPGAs, as it may lead to timing failures if not implemented appropriately. This work presents an ultra-low-power FPGA implementation of a TPU for edge applications. We divide the systolic array of a TPU into different FPGA partitions based on the minimum slack value of different design paths of Multiplier Accumulators (MACs). Each partition uses different near-threshold (NTC) biasing voltages to run its FPGA cores. The biasing voltage for each partition is roughly calculated by the proposed static schemes. However, further calibration of biasing voltage is performed by the proposed runtime scheme. To overcome the timing failure caused by NTC, the MACs with higher minimum slack are placed in lower-voltage partitions, while the MACs with lower minimum slack paths are placed in higher-voltage partitions. The proposed architecture is implemented in a commercial platform, namely with Xilinx FPGA and academic platform with 22 nm, 45 nm and 130 nm FPGAs. Any timing error caused by NTC can be caught by the Razor flipflop used in each MAC. The proposed voltage-scaled, partitioned systolic array can save 3.1% to 11.6% of dynamic power in and tools, respectively, depending on the FPGA technology, partition size, number of partitions and biasing voltages. The normalized performance and accuracy of benchmark models running on our low-power TPU are very competitive compared to existing literature.- 10 April 2024
- Published byĀ MDPI AGĀ inĀ Electronics
- Vol.Ā 13 Ā (8) , 1431
Open AccessVoltage Scaled Low Power DNN Accelerator Design on Reconfigurable Platform
The exponential emergence of Field-Programmable Gate Arrays (FPGAs) has accelerated research on hardware implementation of Deep Neural Networks (DNNs). Among all DNN processors, domain-specific architectures such as Google’s Tensor Processor Unit (TPU) have outperformed conventional GPUs (Graphics Processing Units) and CPUs (Central Processing Units). However, implementing low-power TPUs in reconfigurable hardware remains a challenge in this field. Voltage scaling, a popular approach for energy savings, can be challenging in FPGAs, as it may lead to timing failures if not implemented appropriately. This work presents an ultra-low-power FPGA implementation of a TPU for edge applications. We divide the systolic array of a TPU into different FPGA partitions based on the minimum slack value of different design paths of Multiplier Accumulators (MACs). Each partition uses different near-threshold (NTC) biasing voltages to run its FPGA cores. The biasing voltage for each partition is roughly calculated by the proposed static schemes. However, further calibration of biasing voltage is performed by the proposed runtime scheme. To overcome the timing failure caused by NTC, the MACs with higher minimum slack are placed in lower-voltage partitions, while the MACs with lower minimum slack paths are placed in higher-voltage partitions. The proposed architecture is implemented in a commercial platform, namely with Xilinx FPGA and academic platform with 22 nm, 45 nm and 130 nm FPGAs. Any timing error caused by NTC can be caught by the Razor flipflop used in each MAC. The proposed voltage-scaled, partitioned systolic array can save 3.1% to 11.6% of dynamic power in and tools, respectively, depending on the FPGA technology, partition size, number of partitions and biasing voltages. The normalized performance and accuracy of benchmark models running on our low-power TPU are very competitive compared to existing literature. - Open Access
- 9 January 2024
- Published byĀ MDPI AGĀ inĀ Journal of Low Power Electronics and Applications
- Vol.Ā 14 Ā (1) , 4
Understanding Timing Error Characteristics from Overclocked Systolic MultiplyāAccumulate Arrays in FPGAs
Artificial Intelligence (AI) hardware accelerators have seen tremendous developments in recent years due to the rapid growth of AI in multiple fields. Many such accelerators comprise a Systolic MultiplyāAccumulate Array (SMA) as its computational brain. In this paper, we investigate the faulty output characterization of an SMA in a real silicon FPGA board. Experiments were run on a single Zybo Z7-20 board to control for process variation at nominal voltage and in small batches to control for temperature. The FPGA is rated up to 800 MHz in the data sheet due to the max frequency of the PLL, but the design is written using Verilog for the FPGA and C++ for the processor and synthesized with a chosen constraint of a 125 MHz clock. We then operate the system at a frequency range of 125 MHz to 450 MHz for the FPGA and the nominal 667 MHz for the processor core to produce timing errors in the FPGA without affecting the processor. Our extensive experimental platform with a hardwareāsoftware ecosystem provides a methodological pathway that reveals fascinating characteristics of SMA behavior under an overclocked environment. While one may intuitively expect that timing errors resulting from overclocked hardware may produce a wide variation in output values, our post-silicon evaluation reveals a lack of variation in erroneous output values. We found an intriguing pattern where error output values are stable for a given input across a range of operating frequencies far exceeding the rated frequency of the FPGA.
- 9 January 2024
- Published byĀ MDPI AGĀ inĀ Journal of Low Power Electronics and Applications
- Vol.Ā 14 Ā (1) , 4
Open AccessUnderstanding Timing Error Characteristics from Overclocked Systolic MultiplyāAccumulate Arrays in FPGAs
Artificial Intelligence (AI) hardware accelerators have seen tremendous developments in recent years due to the rapid growth of AI in multiple fields. Many such accelerators comprise a Systolic MultiplyāAccumulate Array (SMA) as its computational brain. In this paper, we investigate the faulty output characterization of an SMA in a real silicon FPGA board. Experiments were run on a single Zybo Z7-20 board to control for process variation at nominal voltage and in small batches to control for temperature. The FPGA is rated up to 800 MHz in the data sheet due to the max frequency of the PLL, but the design is written using Verilog for the FPGA and C++ for the processor and synthesized with a chosen constraint of a 125 MHz clock. We then operate the system at a frequency range of 125 MHz to 450 MHz for the FPGA and the nominal 667 MHz for the processor core to produce timing errors in the FPGA without affecting the processor. Our extensive experimental platform with a hardwareāsoftware ecosystem provides a methodological pathway that reveals fascinating characteristics of SMA behavior under an overclocked environment. While one may intuitively expect that timing errors resulting from overclocked hardware may produce a wide variation in output values, our post-silicon evaluation reveals a lack of variation in erroneous output values. We found an intriguing pattern where error output values are stable for a given input across a range of operating frequencies far exceeding the rated frequency of the FPGA.
- Open Access
- 6 June 2022
- Published byĀ MDPI AGĀ inĀ Journal of Low Power Electronics and Applications
- Vol.Ā 12 Ā (2) , 32
Implementing a Timing Error-Resilient and Energy-Efficient Near-Threshold Hardware Accelerator for Deep Neural Network Inference
Increasing processing requirements in the Artificial Intelligence (AI) realm has led to the emergence of domain-specific architectures for Deep Neural Network (DNN) applications. Tensor Processing Unit (TPU), a DNN accelerator by Google, has emerged as a front runner outclassing its contemporaries, CPUs and GPUs, in performance by 15Ćā30Ć. TPUs have been deployed in Google data centers to cater to the performance demands. However, a TPUās performance enhancement is accompanied by a mammoth power consumption. In the pursuit of lowering the energy utilization, this paper proposes PREDITORāa low-power TPU operating in the Near-Threshold Computing (NTC) realm. PREDITOR uses mathematical analysis to mitigate the undetectable timing errors by boosting the voltage of the selective multiplier-and-accumulator units at specific intervals to enhance the performance of the NTC TPU, thereby ensuring a high inference accuracy at low voltage. PREDITOR offers up to 3Ćā5Ć improved performance in comparison to the leading-edge error mitigation schemes with a minor loss in accuracy.
- 6 June 2022
- Published byĀ MDPI AGĀ inĀ Journal of Low Power Electronics and Applications
- Vol.Ā 12 Ā (2) , 32
Open AccessImplementing a Timing Error-Resilient and Energy-Efficient Near-Threshold Hardware Accelerator for Deep Neural Network Inference
Increasing processing requirements in the Artificial Intelligence (AI) realm has led to the emergence of domain-specific architectures for Deep Neural Network (DNN) applications. Tensor Processing Unit (TPU), a DNN accelerator by Google, has emerged as a front runner outclassing its contemporaries, CPUs and GPUs, in performance by 15Ćā30Ć. TPUs have been deployed in Google data centers to cater to the performance demands. However, a TPUās performance enhancement is accompanied by a mammoth power consumption. In the pursuit of lowering the energy utilization, this paper proposes PREDITORāa low-power TPU operating in the Near-Threshold Computing (NTC) realm. PREDITOR uses mathematical analysis to mitigate the undetectable timing errors by boosting the voltage of the selective multiplier-and-accumulator units at specific intervals to enhance the performance of the NTC TPU, thereby ensuring a high inference accuracy at low voltage. PREDITOR offers up to 3Ćā5Ć improved performance in comparison to the leading-edge error mitigation schemes with a minor loss in accuracy.
- 3 September 2021
- Published byĀ Institute of Electrical and Electronics Engineers (IEEE)Ā inĀ IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol.Ā 29 Ā (10) , 1790-1799
EFFORT: A Comprehensive Technique to Tackle Timing Violations and Improve Energy Efficiency of Near-Threshold Tensor Processing Units
Modern deep neural network (DNN) applications demand a remarkable processing throughput usually unmet by traditional Von Neumann architectures. Consequently, hardware accelerators, comprising a sea of multiplier-and-accumulate (MAC) units, have recently gained prominence in accelerating DNN inference engine. For example, tensor processing units (TPUs) account for a lionās share of Googleās datacenter inference operations. The proliferation of real-time DNN predictions is accompanied by a tremendous energy budget. In quest of trimming the energy footprint of DNN accelerators, we propose Energy eFFicient and errOr Resilient TPU (EFFORT)āan energy optimized, yet high-performance TPU architecture, operating at the near-threshold computing (NTC) region. EFFORT promotes a better-than-worst case design by operating the NTC TPU at a substantially high frequency while keeping the voltage at the NTC nominal value. In order to tackle the timing errors due to such aggressive operation, we employ an opportunistic error mitigation strategy. In addition, we implement an in situ clock gating architecture, drastically reducing the MACsā dynamic power consumption. Compared to a cutting-edge error mitigation technique for TPUs, EFFORT enables up to $2.5\times $ better performance at NTC with only 4% average accuracy drop across six out of eight DNN benchmarks.
- 3 September 2021
- Published byĀ Institute of Electrical and Electronics Engineers (IEEE)Ā inĀ IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol.Ā 29 Ā (10) , 1790-1799
EFFORT: A Comprehensive Technique to Tackle Timing Violations and Improve Energy Efficiency of Near-Threshold Tensor Processing Units
Modern deep neural network (DNN) applications demand a remarkable processing throughput usually unmet by traditional Von Neumann architectures. Consequently, hardware accelerators, comprising a sea of multiplier-and-accumulate (MAC) units, have recently gained prominence in accelerating DNN inference engine. For example, tensor processing units (TPUs) account for a lionās share of Googleās datacenter inference operations. The proliferation of real-time DNN predictions is accompanied by a tremendous energy budget. In quest of trimming the energy footprint of DNN accelerators, we propose Energy eFFicient and errOr Resilient TPU (EFFORT)āan energy optimized, yet high-performance TPU architecture, operating at the near-threshold computing (NTC) region. EFFORT promotes a better-than-worst case design by operating the NTC TPU at a substantially high frequency while keeping the voltage at the NTC nominal value. In order to tackle the timing errors due to such aggressive operation, we employ an opportunistic error mitigation strategy. In addition, we implement an in situ clock gating architecture, drastically reducing the MACsā dynamic power consumption. Compared to a cutting-edge error mitigation technique for TPUs, EFFORT enables up to $2.5\times $ better performance at NTC with only 4% average accuracy drop across six out of eight DNN benchmarks.