AI Briefing
KO

Rebellions Accelerates Development Tasks Including Physical Design by Building AWS-Based EDA Environment

·2026.09.18 13:46

Key point

Rebellions adopted AWS cloud infrastructure to significantly reduce execution times for AI NPU design verification and Physical Design tasks, thereby increasing development speed.

1 / 4

Details

AI inference-focused fabless startup Rebellions has built an elastic EDA environment based on AWS to accelerate the development of its next-generation NPU. The existing on-premises environment served as a bottleneck delaying development schedules, requiring a minimum of 20 weeks for new server procurement and 6 to 8 months for installation.

Performance Improvement Effects

By combining AWS's latest instances (x8aedz, c8g) with Amazon FSx for OpenZFS storage, Rebellions achieved groundbreaking performance improvements in key stages of design verification.

  • File Access Performance: Sequential read speeds improved by approximately 18.3 times, and 4KB random access latency decreased to approximately 1/30 of the original level.
  • Physical Design: Execution time was reduced from a total of 266 hours to 129 hours, a reduction of approximately 2.1 times.
  • DFT and Logic Simulation: Design for Test (DFT) execution time became approximately 2.5 times faster, and Logic Simulation became 2.3 to 2.6 times faster during concurrent tasks.

Operational Efficiency and Security

Using AWS ParallelCluster, Rebellions dynamically creates EC2 nodes to match queued jobs and terminates them when idle, minimizing resource waste. Additionally, design files are securely managed without exposure to the internet through VPC private subnets and Site-to-Site VPN, while activities and logs are audited using CloudTrail and CloudWatch. Through this, Rebellions plans to reduce the burden of large initial investments such as storage and strengthen its competitiveness in global AI inference solutions.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.