Seungjae Moon is Co-Founder and AI Hardware Architect at HyperAccel, a Seoul-based AI semiconductor startup pioneering the Latency Processing Unit (LPU) for large language model inference. His work sits at the frontier of generative AI infrastructure, building purpose-built silicon to replace expensive GPU-based inference stacks.

Seungjae earned his M.S. in Electrical Engineering from KAIST and is currently pursuing a Ph.D. in Electrical Engineering there, conducting research at the KAIST CastLab on hardware accelerators for LLM inference and dataflow optimization. Prior to co-founding HyperAccel, he was a researcher at the AI Semiconductor Systems Research Center and at CastLab, where he developed accelerators for convolutional neural networks and LLM inference.

Seungjae was awarded the DAC 2023 Engineering Best Presentation Award for the IP track at the 60th Design Automation Conference, recognizing his presentation of HyperAccel's LPU. He has co-authored multiple peer-reviewed papers published at top venues including MICRO, ISCA, and IEEE Micro.

At HyperAccel, Seungjae leads hardware architecture efforts, focusing on dataflow optimization and memory-efficient chip design that achieves around 90% memory bandwidth utilization using LPDDR instead of costly HBM. He has represented HyperAccel in high-profile engagements with teams from Meta and Aramco, and the company's Bertha 500 chip has taped out on Samsung 4nm process.

HyperAccel has raised over $37.7M in a Series A round led by Korea Investment Holdings, with participation from 11 investors including Mirae Asset and BonAngels, and has since secured a Series B round of 200 billion won led by Krafton. The company's LPU delivers approximately 5x more tokens per second per unit of compute compared to Nvidia Hopper-generation GPUs, and around 20x the throughput per dollar versus the H100.