Taesik Na
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
Taesik Na has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Jong Hwan Ko
Sungkyunkwan University
7 shared publications Saibal Mukhopadhyay
Department of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, 30332 USA
6 shared publications- Mohammad Faisal Amir
School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA
3 shared publications - Saibal Mukhopadhyay
Department of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA
3 shared publications - Duckhwan Kim
School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA
2 shared publications - Priyabrata Saha
School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA
2 shared publications
Latest Publications
- 18 February 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 56 (8) , 2402-2415
A Universal Modular Hybrid LDO With Fast Load Transient Response and Programmable PSRR in 14-nm CMOS Featuring Dynamic Clamp Strength Tuning
This article presents a universal modular hybrid low-dropout regulator (MHLDO) to provide any desired combination of the power supply rejection ratio (PSRR) and power conversion efficiency (PCE) with in-compliance output ripple, load transient response, and operating range while minimizing losses and decoupling capacitor. The hybrid architecture eliminates the need for the fine quantization of the digital LDO power gates and prevents any associated limit cycle oscillation while keeping the overheads low. It is configurable at design-time and robustly self-adjusts across different operating points via a scalable architecture. The modular topology overcomes significant challenges of developing a large variety of analog and digital LDOs in order to meet the varying PSRR and power budget requirements for systems on a chip (SoCs) in scaled CMOS. A nonlinear control (NLC) feature provides an energy-efficient way to respond to fast load transients, while the dynamic clamp strength tuning (DCST) technique prevents unnecessary oscillations stemming from the input parasitic inductances and improves stability while lowering switching losses. The designed MHLDO provides a programmable PSRR capability of up to -42 dB with a quiescent current of less than 27.3 μA as ALDO and a 133-mV worst droop against a >1-A/ns fast di/dt load change as DLDO. A new figure of merit (FoM) with improved accuracy demonstrates performance of 83 fs.
- 18 February 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 56 (8) , 2402-2415
A Universal Modular Hybrid LDO With Fast Load Transient Response and Programmable PSRR in 14-nm CMOS Featuring Dynamic Clamp Strength Tuning
This article presents a universal modular hybrid low-dropout regulator (MHLDO) to provide any desired combination of the power supply rejection ratio (PSRR) and power conversion efficiency (PCE) with in-compliance output ripple, load transient response, and operating range while minimizing losses and decoupling capacitor. The hybrid architecture eliminates the need for the fine quantization of the digital LDO power gates and prevents any associated limit cycle oscillation while keeping the overheads low. It is configurable at design-time and robustly self-adjusts across different operating points via a scalable architecture. The modular topology overcomes significant challenges of developing a large variety of analog and digital LDOs in order to meet the varying PSRR and power budget requirements for systems on a chip (SoCs) in scaled CMOS. A nonlinear control (NLC) feature provides an energy-efficient way to respond to fast load transients, while the dynamic clamp strength tuning (DCST) technique prevents unnecessary oscillations stemming from the input parasitic inductances and improves stability while lowering switching losses. The designed MHLDO provides a programmable PSRR capability of up to -42 dB with a quiescent current of less than 27.3 μA as ALDO and a 133-mV worst droop against a >1-A/ns fast di/dt load change as DLDO. A new figure of merit (FoM) with improved accuracy demonstrates performance of 83 fs.
- 19 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 10 (4) , 444-457
Cross-Layer Noise Analysis in Smart Digital Pixel Sensors With Integrated Deep Neural Network
Digital pixel based image sensors with embedded deep neural network (DNN) allow many mission critical surveillance applications. However, image noise caused by variations and non-idealities in the sensor aggravates the quality of image and further degrades the performance of a DNN. We propose a digital pixel-DNN cross-layer simulation methodology for accurate training and evaluation of a DNN under image noise induced from sensors. In particular, this paper focuses on the image noise derived from device mismatches in digital pixel circuits with 3D integrated and pixel-parallel readout circuits, and studies the effect of the resulting image noise on the accuracy of a DNN. The simulation results show that the device mismatch in the digital pixel creates distinct noise structure on output image and should be accurately considered while training a DNN. We also present design space explorations using our cross-layer simulation.
- 19 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 10 (4) , 444-457
Cross-Layer Noise Analysis in Smart Digital Pixel Sensors With Integrated Deep Neural Network
Digital pixel based image sensors with embedded deep neural network (DNN) allow many mission critical surveillance applications. However, image noise caused by variations and non-idealities in the sensor aggravates the quality of image and further degrades the performance of a DNN. We propose a digital pixel-DNN cross-layer simulation methodology for accurate training and evaluation of a DNN under image noise induced from sensors. In particular, this paper focuses on the image noise derived from device mismatches in digital pixel circuits with 3D integrated and pixel-parallel readout circuits, and studies the effect of the resulting image noise on the accuracy of a DNN. The simulation results show that the device mismatch in the digital pixel creates distinct noise structure on output image and should be accurately considered while training a DNN. We also present design space explorations using our cross-layer simulation.
- 14 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 9 (3) , 498-508
CAMEL: An Adaptive Camera With Embedded Machine Learning-Based Sensor Parameter Control
Today’s cameras are designed to approximate what they observe in a manner that preserves entropy. However, time critical autonomous applications such as autonomous driving, surveillance and defense systems require task critical information at the highest quality. With rapid advances in frame rates and resolutions, observing scenes at the highest quality raises concerns for the transmission bandwidth. In this paper, we introduce a new paradigm of smart camera that captures only taskcritical information at the highest quality. Embedded deep neural network (DNN) algorithms within the camera enhance quality of information through real-time control of sensor parameters. We show the hardware feasibility of this camera by demonstrating a 3D-stacked architecture with a Digital Pixel Sensor (DPS). We demonstrate a number of high-level vision applications that benefit through this task-guided control including object detection, object tracking and activity recognition. Finally, we present the unique challenges faced created as a result of feedback and show how software/hardware innovations can be used to mitigate them.
- 14 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 9 (3) , 498-508
CAMEL: An Adaptive Camera With Embedded Machine Learning-Based Sensor Parameter Control
Today’s cameras are designed to approximate what they observe in a manner that preserves entropy. However, time critical autonomous applications such as autonomous driving, surveillance and defense systems require task critical information at the highest quality. With rapid advances in frame rates and resolutions, observing scenes at the highest quality raises concerns for the transmission bandwidth. In this paper, we introduce a new paradigm of smart camera that captures only taskcritical information at the highest quality. Embedded deep neural network (DNN) algorithms within the camera enhance quality of information through real-time control of sensor parameters. We show the hardware feasibility of this camera by demonstrating a 3D-stacked architecture with a Digital Pixel Sensor (DPS). We demonstrate a number of high-level vision applications that benefit through this task-guided control including object detection, object tracking and activity recognition. Finally, we present the unique challenges faced created as a result of feedback and show how software/hardware innovations can be used to mitigate them.
- 3 July 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 26 (12) , 2781-2794
ReRAM-Based Processing-in-Memory Architecture for Recurrent Neural Network Acceleration
We present a recurrent neural network (RNN) accelerator design with resistive random-access memory (ReRAM)-based processing-in-memory (PIM) architecture. Distinguished from prior ReRAM-based convolutional neural network accelerators, we redesign the system to make it suitable for RNN acceleration. We measure the system throughput and energy efficiency with the detailed circuit and device characterization. Reprogrammability is enabled with our design, and an RNN friendly pipeline is employed to increase the system throughput. We observe that on average the proposed system achieves 79x improvement of computing efficiency compared with graphics processing unit baseline. Our simulation also indicates that to maintain high accuracy and computing efficiency, the read noise standard deviation should be less than 0.2, the device resistance should be at least 1 MΩ, and the device writes latency should be minimized.
- 3 July 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 26 (12) , 2781-2794
ReRAM-Based Processing-in-Memory Architecture for Recurrent Neural Network Acceleration
We present a recurrent neural network (RNN) accelerator design with resistive random-access memory (ReRAM)-based processing-in-memory (PIM) architecture. Distinguished from prior ReRAM-based convolutional neural network accelerators, we redesign the system to make it suitable for RNN acceleration. We measure the system throughput and energy efficiency with the detailed circuit and device characterization. Reprogrammability is enabled with our design, and an RNN friendly pipeline is employed to increase the system throughput. We observe that on average the proposed system achieves 79x improvement of computing efficiency compared with graphics processing unit baseline. Our simulation also indicates that to maintain high accuracy and computing efficiency, the read noise standard deviation should be less than 0.2, the device resistance should be at least 1 MΩ, and the device writes latency should be minimized.
- 2 May 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 8 (3) , 591-602
An Energy-Quality Scalable Wireless Image Sensor Node for Object-Based Video Surveillance
Wireless image sensor nodes are required deliver better visual information of the region-of-interest (ROI) under tight energy constraints. The energy-quality scalability of the sensor node can be improved by incorporating ROI-based image processing. This paper presents an energy-quality scalable wireless image sensor node using ROI-based processing for object-based surveillance. After detecting the ROI using a low-power noise-robust method, frame images are encoded by a low-complexity ROI coding scheme with better rate-quality performance. The ROI-based processing parameters are controlled by a simple on-line rate controller to minimize the buffer requirement. We integrate the proposed approach into the wireless image sensor node, and analyze the performance assuming a base station with human operators and deep neural networks. The results show that the proposed approach provides higher quality under the available energy budget. Better energy-quality scalability yields higher energy-accuracy performance for the intelligent surveillance system with deep neural networks.
- 2 May 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
- Vol. 8 (3) , 591-602
An Energy-Quality Scalable Wireless Image Sensor Node for Object-Based Video Surveillance
Wireless image sensor nodes are required deliver better visual information of the region-of-interest (ROI) under tight energy constraints. The energy-quality scalability of the sensor node can be improved by incorporating ROI-based image processing. This paper presents an energy-quality scalable wireless image sensor node using ROI-based processing for object-based surveillance. After detecting the ROI using a low-power noise-robust method, frame images are encoded by a low-complexity ROI coding scheme with better rate-quality performance. The ROI-based processing parameters are controlled by a simple on-line rate controller to minimize the buffer requirement. We integrate the proposed approach into the wireless image sensor node, and analyze the performance assuming a base station with human operators and deep neural networks. The results show that the proposed approach provides higher quality under the available energy budget. Better energy-quality scalability yields higher energy-accuracy performance for the intelligent surveillance system with deep neural networks.
- Open Access
- 21 March 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 18 (10) , 4187-4199
3-D Stacked Image Sensor With Deep Neural Network Computation
This paper investigates the power and performance trade-offs associated with integrating deep neural network (DNN) computation in an image sensor. The paper presents the design of Neurosensor-a CMOS image sensor with 3-D stacking of pixel array, read-out circuits, memory, and computing logic for DNN. The analysis shows integrating DNN reduces transmit latency (and energy), but at the expense of processing and memory access latency (and energy). Hence, given a specific DNN and transmission bandwidth, there exist an optimal number of layers that should be computed in the sensor to maximize energy-efficiency. In general, it is often more efficient to integrate memory within the sensor stack and/or implement only the feature extraction layers on the sensor, and optimized configurations can achieve up to 90× improvement in energy efficiency compared to the baseline. Further, coupled power, thermal, and noise simulation demonstrates that integrating DNN computation can increase pixel-array temperature resulting in higher noise, and hence, lower classification accuracy.
- 21 March 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 18 (10) , 4187-4199
Open Access3-D Stacked Image Sensor With Deep Neural Network Computation
This paper investigates the power and performance trade-offs associated with integrating deep neural network (DNN) computation in an image sensor. The paper presents the design of Neurosensor-a CMOS image sensor with 3-D stacking of pixel array, read-out circuits, memory, and computing logic for DNN. The analysis shows integrating DNN reduces transmit latency (and energy), but at the expense of processing and memory access latency (and energy). Hence, given a specific DNN and transmission bandwidth, there exist an optimal number of layers that should be computed in the sensor to maximize energy-efficiency. In general, it is often more efficient to integrate memory within the sensor stack and/or implement only the feature extraction layers on the sensor, and optimized configurations can achieve up to 90× improvement in energy efficiency compared to the baseline. Further, coupled power, thermal, and noise simulation demonstrates that integrating DNN computation can increase pixel-array temperature resulting in higher noise, and hence, lower classification accuracy.
- 2 February 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 38 (1) , 109-121
Design and Analysis of a Neural Network Inference Engine Based on Adaptive Weight Compression
Neural networks generally require significant memory capacity/bandwidth to store/access a large number of synaptic weights. This paper presents design of an energy-efficient neural network inference engine based on adaptive weight compression using a JPEG image encoding algorithm. To maximize compression ratio with minimum accuracy loss, the quality factor of the JPEG encoder is adaptively controlled depending on the accuracy impact of each block. With 1% accuracy loss, the proposed approach achieves 63.4× compression for multilayer perceptron (MLP) and 31.3× for LeNet-5 with the MNIST dataset, and 15.3× for AlexNet and 10.2× for ResNet-50 with ImageNet. The reduced memory requirement leads to higher throughput and lower energy for neural network inference (3× effective memory bandwidth and 22× lower system energy for MLP).
- 2 February 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 38 (1) , 109-121
Design and Analysis of a Neural Network Inference Engine Based on Adaptive Weight Compression
Neural networks generally require significant memory capacity/bandwidth to store/access a large number of synaptic weights. This paper presents design of an energy-efficient neural network inference engine based on adaptive weight compression using a JPEG image encoding algorithm. To maximize compression ratio with minimum accuracy loss, the quality factor of the JPEG encoder is adaptively controlled depending on the accuracy impact of each block. With 1% accuracy loss, the proposed approach achieves 63.4× compression for multilayer perceptron (MLP) and 31.3× for LeNet-5 with the MNIST dataset, and 15.3× for AlexNet and 10.2× for ResNet-50 with ImageNet. The reduced memory requirement leads to higher throughput and lower energy for neural network inference (3× effective memory bandwidth and 22× lower system energy for MLP).
- Open Access
- 26 June 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 64 (9) , 2401-2413
Clock Data Compensation Aware Digital Circuits Design for Voltage Margin Reduction
Tolerating timing error due to power supply noise (PSN) in digital circuits can be done with adding voltage margins. Conservative addition of voltage margins leads wastes of power reducing the battery life in Internet of Things (IoT) devices. This paper aims to provide guidelines to avoid over-design due to PSN especially for the low-cost IoT devices. To this end, we first present an accurate time-domain behavioral model of timing slack variation due to PSN accounting for the clock-data compensation. The accuracy of the model is verified against SPICE for complex designs, including AES engine and LEON3 processor. To prove the effectiveness of our model for reducing voltage margin, we utilize our model in standard VLSI design flow for various examples, such as timing slack versus noise frequency analysis, determining optimal value of an on-die capacitor, analyzing the effects of time borrowing technique, and PVT variation simulations. The analysis shows that the model helps reduce pessimism in estimated timing slack.
- 26 June 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 64 (9) , 2401-2413
Open AccessClock Data Compensation Aware Digital Circuits Design for Voltage Margin Reduction
Tolerating timing error due to power supply noise (PSN) in digital circuits can be done with adding voltage margins. Conservative addition of voltage margins leads wastes of power reducing the battery life in Internet of Things (IoT) devices. This paper aims to provide guidelines to avoid over-design due to PSN especially for the low-cost IoT devices. To this end, we first present an accurate time-domain behavioral model of timing slack variation due to PSN accounting for the clock-data compensation. The accuracy of the model is verified against SPICE for complex designs, including AES engine and LEON3 processor. To prove the effectiveness of our model for reducing voltage margin, we utilize our model in standard VLSI design flow for various examples, such as timing slack versus noise frequency analysis, determining optimal value of an on-die capacitor, analyzing the effects of time borrowing technique, and PVT variation simulations. The analysis shows that the model helps reduce pessimism in estimated timing slack.
- 7 June 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 64 (9) , 2295-2307
A Single-Chip Image Sensor Node With Energy Harvesting From a CMOS Pixel Array
This paper presents a single-chip image sensor node with energy harvesting from the pixel array. The design includes a 128 × 96 pixel array that can be reconfigured to form an on-chip photovoltaic cell to harvest energy. An onchip power management unit harvests energy from the pixel array, and delivers multiple regulated output voltage domains to power the sensor, image processor, and memory. The image processor is a low-overhead moving object detection unit to reduce the volume of transmitted data. The proposed sensor node is implemented on a single die in 130-nm technology. The pixel array demonstrates the peak harvested power of 2.1μW. The power dissipation of sensor is reduced by utilizing low-power circuit techniques, including block-level pipelining, power gating, and low-voltage memory. The system can be self-powered to process a frame at every 7 s. We discuss design approaches for improving the self-power performance. The noise characteristic of the reconfigurable sensor is analyzed, and the need for noise-robust moving object detection is evaluated for better image quality and improved self-power performance.
- 7 June 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 64 (9) , 2295-2307
A Single-Chip Image Sensor Node With Energy Harvesting From a CMOS Pixel Array
This paper presents a single-chip image sensor node with energy harvesting from the pixel array. The design includes a 128 × 96 pixel array that can be reconfigured to form an on-chip photovoltaic cell to harvest energy. An onchip power management unit harvests energy from the pixel array, and delivers multiple regulated output voltage domains to power the sensor, image processor, and memory. The image processor is a low-overhead moving object detection unit to reduce the volume of transmitted data. The proposed sensor node is implemented on a single die in 130-nm technology. The pixel array demonstrates the peak harvested power of 2.1μW. The power dissipation of sensor is reduced by utilizing low-power circuit techniques, including block-level pipelining, power gating, and low-voltage memory. The system can be self-powered to process a frame at every 7 s. We discuss design approaches for improving the self-power performance. The noise characteristic of the reconfigurable sensor is analyzed, and the need for noise-robust moving object detection is evaluated for better image quality and improved self-power performance.