Jinwook Oh

ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware Training

Janghwan Lee, Sihwa Lee, Jinseok Kim, Yongjik Kim et autres

Large Reasoning Models (LRMs) achieve strong problem-solving through long chain-of-thought, but their deployment is constrained by the high cost of full-precision inference and growing KV cache footprints. Microscaled FP4 formats enable efficient FP4 deployment; however, fully quantizing weights, activations, and KV caches …

ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware Training

Janghwan Lee, S Lee, Jinseok Kim, Yong Wook Kim et autres

Large Reasoning Models (LRMs) achieve strong problem-solving through long chain-of-thought, but their deployment is constrained by the high cost of full-precision inference and growing KV cache footprints. Microscaled FP4 formats enable efficient FP4 deployment; however, fully quantizing weights, activations, and KV caches …

A Quad-Chiplet AI SoC with Full-Chip Scalable Mesh Over 16Gb/s UCIe-Advanced Die-to-Die Interface for Large-Scale AI Inferencing

Chang-Hyo Yu, Jaewan Bae, Jinseok Kim, Hongyun Kim et autres

A 4nm-based quad-chiplet with an advanced packaged LLM accelerator achieving 56.8TPS on LLaMA v3.3 70B with single-batch 2k/2k input/output sequences. The architecture combines chiplet-based design, low-latency die-to-die interfaces, unified mixed-precision compute, holistic synchronization, and HBM3E with advanced power schemes to sustain bandwidth, …

kr (code pays fourni par la source)

IGZO/PbS QD/Ga 2 O 3 -Based Optical Synapse Transistors With High PPF for NIR Detection

Yong Jun Jeong, He Young Kang, Jinwook Oh, Gwang‐Bok Kim et autres

In this paper, we present a high-PPF optical synapse transistor with an IGZO/PbS QD/Ga2O3structure for NIR sensing applications. The fabricated device exhibited superior NIR detection capabilities, with responsivities of 214.5 ± 4.1, 202.4 ± 5.6, and 188.5 ± 5.8 A/W and high …

kr (code pays fourni par la source)

CTDM: Resource-Efficient FPGA-Accelerated Simulation of Large-Scale NPU Designs

Hyunje Jo, Han-Sok Suh, Hyungseok Heo, Jinseok Kim et autres

This paper proposes a novel approach to accelerate large Neural Processing Unit (NPU) simulations on FPGA through Chain-based Time-Division Multiplexing (CTDM) and its automatic compiler. CTDM replaces repeated logic patterns with a single logic pattern and register chains, which can take advantage …

kr, us, ch (code pays fourni par la source)

AMXFP4: Taming Activation Outliers with Asymmetric Microscaling Floating-Point for 4-bit LLM Inference

Janghwan Lee, Jiwoong Park, Jin-Seok Kim, Jungju Oh et autres

As large language models (LLMs) grow in parameter size and context length, computation precision has been reduced from 16-bit to 4bit to improve inference efficiency.However, this reduction causes accuracy degradation due to activation outliers.Rotation-based INT4 methods address this via matrix calibration, but …

kr, gb (code pays fourni par la source)

AMXFP4: Taming Activation Outliers with Asymmetric Microscaling Floating-Point for 4-bit LLM Inference

Janghwan Lee, Jiwoong Park, Jin-Seok Kim, Jungju Oh et autres

As large language models (LLMs) grow in parameter size and context length, computation precision has been reduced from 16-bit to 4-bit to improve inference efficiency. However, this reduction causes accuracy degradation due to activation outliers. Rotation-based INT4 methods address this via matrix …

Improvement in Performance and Stability of PbS QD/IGZO Phototransistors Through the Introduction of Ga2O3 Film for Broadband Sensor Applications

Yong Jun Jeong, Gwang‐Bok Kim, Min Jae Kim, Jinwook Oh et autres

The development of broadband photosensors has become crucial in various fields. Indium–gallium–zinc oxide (IGZO, In:Ga:Zn = 1:1:1) phototransistors with PbS quantum dots (QDs) have shown promising features for such sensors, such as reasonable mobility, low leakage current, good photosensitivity, and low-cost fabrication. …

kr (code pays fourni par la source)

2.4 ATOMUS: A 5nm 32TFLOPS/128TOPS ML System-on-Chip for Latency Critical Applications

Chang-Hyo Yu, Hyoeun Kim, Sungho Shin, Kyeongryeol Bong et autres

The growing computational demands of AI inference have led to widespread use of hardware accelerators for different platforms, spanning from edge to the datacenter/cloud. Certain AI application areas, such as in high-frequency trading (HFT) [1–2], have a hard inference latency deadline for …

LightTrader: A Standalone High-Frequency Trading System with Deep Learning Inference Accelerators and Proactive Scheduler

Sungyeob Yoo, Hyunsung Kim, Jinseok Kim, Sunghyun Park et autres

Recent research shows that artificial intelligence (AI) algorithms can dramatically improve the profitability of high-frequency trading (HFT) with accurate market prediction, overcoming the limitation of conventional latency-oriented approaches. However, it is challenging to integrate the computationally intensive AI algorithm into the existing …

kr, ca, gb (code pays fourni par la source)

LightTrader : World’s first AI-enabled High-Frequency Trading Solution with 16 TFLOPS / 64 TOPS Deep Learning Inference Accelerators

Hyunsung Kim, Sungyeob Yoo, Jaewan Bae, Kyeongryeol Bong et autres

We present the world’s first AI-enabled high-frequency trading (HFT) system, LightTrader , which integrates the custom AI accelerators and the FPGA-based conventional HFT pipeline for the low-latency-high-throughput trading solutions with a reduced query miss rate. For better utilization, adaptive job scheduling methods …

gb, ca (code pays fourni par la source)

A 7-nm Four-Core Mixed-Precision AI Chip With 26.2-TFLOPS Hybrid-FP8 Training, 104.9-TOPS INT4 Inference, and Workload-Aware Throttling

Sae Kyu Lee, Ankur Agrawal, Joel A. Silberman, Matthew M. Ziegler et autres

Reduced precision computation is a key enabling factor for energy-efficient acceleration of deep learning (DL) applications. This article presents a 7-nm four-core mixed-precision artificial intelligence (AI) chip that supports four compute precisions—FP16, Hybrid-FP8 (HFP8), INT4, and INT2—to support diverse application demands for …

us, de, kr (code pays fourni par la source)