Jaeyong Chung

Prof. Jaeyong Chung


Research Focus & Metrics

Academic Career Length

18Years (2009-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
12 May 2022 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Verified
Journal Article
Closed access
Kashif InayatKashif Inayat
Kashif Inayat
JCJaeyong Chung
Jaeyong Chung
Deep learning applications have become ubiquitous in today’s era and it has led to vast development in machine learning (ML) accelerators. Systolic arrays have been a primary part of ML accelerator architecture. To fully leverage the systolic arrays, it is required to explore the computer arithmetic data-path components and their tradeoffs in accelerators. We present a novel factored systolic array (FSA) architecture, in which the carry propagation adder (CPA) and carry-save adder (CSA) perform hybrid accumulation on least significant bit (LSB) bits and most significant bits (MSB) bits, respectively, inside each processing element. In addition, a small CPA to complete accumulation for MSB bits along with rounding logic for each column of the array is placed, which not only reduces the area, delay, and power but also balances the combinational and sequential area tradeoffs. We demonstrate the hybrid accumulator with partial CPA factoring in “Gemmini,” an open-source practical systolic array accelerator and factoring technique does not change the functionality of the base design. We implemented three baselines, original Gemmini and two variants of it, and show that the proposed approach leads to overall significant reduction in area within the range 12.8% – 50.2% and in power within the range 18.6% – 41% with improved or similar delay in comparison to the baselines.

12 July 2012 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
Verified
Journal Article
Closed access
JCJaeyong Chung
Jaeyong Chung
JXJinjun Xiong
Jinjun Xiong
VZVladimir Zolotov
Vladimir Zolotov
JAJacob A. Abraham
Jacob A. Abraham
In the face of large-scale process variations, statistical timing methodology has advanced significantly over the last few years, and statistical path selection takes advantage of it in at-speed testing. In deterministic path selection, the separation of path selection and test generation is known to require time consuming iteration between the two processes. This paper shows that in statistical path selection, this is not only the case, but also the quality of results can be severely degraded even after the iteration. To deal with this issue, we consider testability in the first place by integrating a satisfiability (SAT) solver, and this necessitates a new statistical path selection method. We integrate the SAT solver in a novel way that leverages the conflict analysis of modern SAT solvers, which provides more than 4X speedup without special optimizations of the SAT solver for this particular application. Our proposed method is based on a generalized path criticality metric whose properties allow efficient pruning. Our experimental results show that the proposed method achieves 47% better quality of results on average, and up to 361X speedup compared to statistical path selection followed by test generation.

22 March 2021 in IEEE Access
Verified
Journal Article
Open access
PPPhuoc Pham
Phuoc Pham
JAJacob A. Abraham
Jacob A. Abraham
JCJaeyong Chung
Jaeyong Chung
Quantizing weights and activations of deep neural networks is essential for deploying them in resource-constrained devices, or cloud platforms for at-scale services. While binarization is a special case of quantization, this extreme case often leads to several training difficulties, and necessitates specialized models and training methods. As a result, recent quantization methods do not provide binarization, thus losing the most resource-efficient option, and quantized and binarized networks have been distinct research areas. We examine binarization difficulties in a quantization framework and find that all we need to enable the binary training are a symmetric quantizer, good initialization, and careful hyperparameter selection. These techniques also lead to substantial improvements in multi-bit quantization. We demonstrate our unified quantization framework, denoted as UniQ, on the ImageNet dataset with various architectures such as ResNet-18,-34 and MobileNetV2. For multi-bit quantization, UniQ outperforms existing methods to achieve the state-of-the-art accuracy. In binarization, the achieved accuracy is comparable to existing state-of-the-art methods even without modifying the original architectures.

20 February 2020 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Verified
Journal Article
Closed access
IUInayat Ullah
Inayat Ullah
JYJoon-Sung Yang
Joon-Sung Yang
JCJaeyong Chung
Jaeyong Chung
Static random access memory (SRAM)-based ternary content-addressable memory (TCAM) on field-programmable gate arrays (FPGAs) is used for packet classification in software-defined networking (SDN) and OpenFlow applications. SRAMs implementing TCAM contents constitute the major part of a TCAM design on FPGAs, which are vulnerable to soft errors. The protection of SRAM-based TCAMs against soft errors is challenging without compromising critical path delay and maintaining a high search performance. This brief presents a low-cost and low-response-time technique for the protection of SRAM-based TCAMs. This technique uses simple, single-bit parity for fault detection which has a minimal critical path overhead. This technique exploits the binary-encoded TCAM table maintained in SRAM-based TCAMs for update purposes to implement a low-response-time error-correction mechanism at low cost. The error-correction process is carried out in the background, allowing lookup operations to be performed simultaneously, thus maintaining a high search performance. The proposed technique provides protection against soft errors with a response time of 293 ns, whereas maintaining a search rate of 222 million searches per second on a 1024 x 40 size TCAM on Artix-7 FPGA.

1 January 2019 in IEEE Access
Verified
Journal Article
Open access
WKWoochul Kang
Woochul Kang
JCJaeyong Chung
Jaeyong Chung
Deep learning is a state-of-the-art approach that provides highly accurate inference for many cyber-physical systems (CPS) such as autonomous cars and robots. Deep learning inference often needs to be performed locally on mobile and embedded devices, rather than in the cloud, to address concerns such as latency, power consumption, and limited bandwidth. However, existing approaches have focused on delivering ‘best-effort’ performance to resource-constrained mobile embedded devices, resulting in unpredictable performance under highly variable environments of CPS. In this paper, we propose a novel deep learning inference runtime, called DeepRT, that supports multiple QoS objectives simultaneously against unpredictable workloads. In DeepRT, the Multiple Inputs/Multiple Outputs (MIMO) modeling and control methodology is proposed as a primary tool to support multiple QoS goals including the inference latency and power consumption. DeepRT’s MIMO controller coordinates multiple computing resources, such as CPUs and GPUs, by capturing their close interactions and effects on multiple QoS objectives. We demonstrate the viability of DeepRT’s QoS management architecture by implementing a prototype of DeepRT. The evaluation results demonstrate that, compared to baseline approaches, DeepRT can support the desired inference latency as well as power consumption for various deep learning models in a highly robust manner.

18 July 2018 in Real-Time Systems
Verified
Journal Article
Closed access
PDF file
WKWoochul Kang
Woochul Kang
JCJaeyong Chung
Jaeyong Chung