Research & Development

We grow based on verifiable science, not just ideas. As a Deep Tech company, CREAIM continuously produces proprietary research, experiments, and internal verification reports (Whitepapers) on core technologies, with some research preparing for global top conference submission.

Technical Whitepaper · Preparing for DAC 2025 Submission

Autonomous Silicon Design: Closing the Loop with LLM and PPA Feedback

Proposes an autonomous design-verification feedback loop structure where AI receives natural language requirements, generates RTL, and relearns from PPA (Power, Performance, Area) simulation feedback.

Key Contributions
  • •LLM-based RTL Generation + PPA Closed-loop Optimization
  • •90% Design Time Reduction
  • •Design Quality Increase (Auto Deduplication & Bug Reduction)
View Abstract (PDF)
Internal Research Report · Preparing for NeurIPS Workshop Submission

Truth Kernel: Real-time Verification Layer for Hallucination Suppression

Proposes a new architecture suppressing hallucination rate below 0.1% by verifying factuality, logic, and evidence at the kernel layer when AI generates answers.

Key Contributions
  • •Verification Kernel Layer operating above LLM
  • •Logic & Fact-based Filtering Engine
  • •Applicable in Medical, Legal, Finance fields
View Abstract (PDF)
Technical Report · Under Internal Peer Review

Cognitive DNA: Structured Long-term Memory for AI Systems

Describes an AI-dedicated long-term memory system storing relationships, structures, and contexts beyond vector DBs.

Key Contributions
  • •Definition of 'Memory Unit (DNA Token)'
  • •6x Context Retention Improvement
  • •Multi-session Memory Stability
View Abstract (PDF)
Internal Whitepaper · Preparing for Computer Vision Journal Submission

Visual DNA: Consistency Preservation in Generative AI

Defines Visual DNA (Style & Structure Vector) to solve character and style consistency collapse in generative models, introducing consistency control technology based on it.

Key Contributions
  • •Drastic Improvement in Character, Background, Costume Style Retention
  • •Suitable for Multi-cut & Multi-scene Content Production
  • •Core Foundation for Webtoon, Animation, Game Production Automation
View Abstract (PDF)

Research Partners

MIT CSAIL
Stanford HAI
KAIST AI
SNU