Automotive AI semiconductors take an average of five years from order to mass production. Because they are developed to prepare for the future, close collaboration with global automakers who will use them is crucial for success.

On the 10th, the Artificial Intelligence Semiconductor Forum held its 12th Artificial Intelligence Semiconductor Forum Breakfast Lecture under the theme, "The Key to Dominating the Automotive AI Market - Semiconductor Competitiveness: What's Important?"

Won-Joo Park, CSO of Boss Semiconductor, who was in charge of the presentation that day, said that the automobile industry moves based on the market five years in the future, and that autonomous driving and in-vehicle infotainment are at the center of AI innovation.

Park Won-joo, CSO, said that with the rapid evolution of autonomous driving and vehicle infotainment technologies, automobiles are no longer simply a means of transportation but are transforming into “moving living and working spaces,” and in particular, he cited the development of AI semiconductors as a key driving force accelerating this change.

In particular, automotive semiconductors take an average of five years from order to mass production. It takes two years of design and three years of in-vehicle testing before it can be released.

Park Won-joo, CSO, said, “The chip we are developing now is targeting the market in the early 2030s,” and added, “Close collaboration with global automakers will determine success or failure.”

Boss Semiconductor develops both application-specific integrated circuits (ASICs) and general-purpose AI chips, and has a design and software support system tailored to customer needs.

The goal is to simultaneously implement AI model optimization and high-performance computing, especially in the fields of autonomous driving and infotainment.

According to the announcement, future vehicles will communicate with drivers to plan their journeys, and passengers will become "digital capsules" that enjoy music, videos, and work.

To achieve this, edge computing performance capable of processing large-scale AI calculations in real time inside the vehicle is essential.

“Processing solely through the cloud will result in delays and cost issues,” said Park Won-joo, CSO. “High-performance semiconductors in vehicles are essential.”

Bose Semiconductor's flagship chipset, 'Eagle N', provides 250 TOPS of computational performance and reduces memory bottlenecks by arranging cores dedicated to data movement.

Additionally, it uses a 'safety island' structure that meets safety standards to maintain core functions even in emergency situations.

The chip design adopted a modular (UCIe) approach, which increased production yield and reduced costs compared to large single chips.

This allows the same architecture to be applied from low-end to premium, reducing the maintenance burden on automakers.

Boss Semiconductor produces various products including CNN, transformer, and LLM. Lightweight AI models to fit the vehicle environment.

In particular, it supports video-language combined models (VLM) and autonomous driving integrated models, processing multi-camera, radar, and lidar data in real time.

“Model optimization and hardware design must be intertwined to achieve both performance and power efficiency,” said Park Won-joo, CSO.

At the end of his presentation, he pointed out the vulnerability of the domestic system semiconductor industry. While non-memory semiconductors account for 75% of the global semiconductor market, Korea's share is only around 3%.

Park Won-joo, CSO, suggested, “National support is needed not only for AI semiconductors but also for system semiconductors in general,” and “Domestic automakers and large IT companies should cooperate with fabless companies from the beginning and provide incentives when adopting domestic chips.”