On April 12th, BOS Semiconductors and HodooAI Labs, a company specializing in AI platform for unsupervised training, announced their collaboration for AI software optimization for application in AI semiconductor NPU (Neural Processing Unit). The kickoff meeting was held at BOS Semiconductors' headquarters at Pangyo, attended by BOS Semiconductors CEO Jaehong Park and HodooAI Labs CEO Jungwoo Lee.

Under the agreed partnership, both parties will co-develop and commercialize software lightweighting and optimization technology that will be applied to the AI NPU Accelerator System-on-Chip (SoC) currently being developed by BOS Semiconductors for autonomous driving cars. HodooAI Labs plans to leverage the current project as the basis for further developing a software ecosystem for future edge AI NPU applications.

HodooAI Lab has been spun off from Seoul National University research lab (led by Professor Jungwoo Lee) as a startup specializing in AI. HodooAI has commercialized an automatic model exploration solution called Autotrainer that targets usage in AI semiconductors. This solution was released at the "Vision 2021," the world's largest machine vision exhibition. It garnered significant attention in the AI industry for being the world's first "automatic learning software" with automatic neural architecture search (NAS) functionality, improving model optimization and lightweighting by 15% compared to existing technologies. They plan to expand this solution into lightweighting and optimization technology for AI semiconductors.

CEO Jaehong Park stated, "HodooAI Labs has outstanding competitiveness in the AI software field, and we expect to secure the competitiveness of AI software in our current AI SoC development through collaboration with them."

The AI accelerator semiconductor currently being developed by BOS semiconducotrs facilitates neural network model data processing used in various perception and decision-making tasks of autonomous driving. This automotive SoC product is being developed by BOS Semiconductors which has strengths in HW architecture design and core SW-related technologies.

Meanwhile, HodooAI Labs recently commercialized AI auto-training solutions for various industries, including bio-pharmaceuticals for the first time in Korea. Targeting the education sector, they have also finalized the supply of auto-training software with automatic model exploration capability, which can reduce the size of AI models by 1/20, to Seoul National University through consultations with the Seoul National University Information Technology Center. The solution is currently being used for research and education in various departments at Seoul National University.