"We will bring to reality autonomous driving in future vehicles with our AI chips specialized for automobiles."

Jaehong Park, CEO of BOS Semiconductors, expressed this ambition ahead of the upcoming release of the company's first automotive semiconductor NPU accelerator, named 'N1', scheduled for December 2024.

BOS, a fabless company focusing on automotive semiconductors, was established in May 2022 and is now in its third year. Despite its short history, it is gaining attention in the global automotive market due to its strong technical capabilities. BOS's huge growth potential was recognized by the market as it secured investment from Hyundai Motor Company early on. As of June this year, the company has accumulated investments totaling 25 billion KRW.

In addition, the company has entered into a co-development collaboration last October with Tenstorrent, a Canadian AI company led by CEO Jim Keller, a legendary figure in the semiconductor industry. Under the collaboration, Tenstorrent will supply its NPU(Neural Processing Unit) IP to BOS for its integration into the N1 SoC(System-on-Chip).

CEO Park is a veteran in the field of system semiconductors with approximately 23 years of experience at Samsung Electronics, where he led the foundry business and Exynos SoC design. The majority of BOS's semiconductor design team, which consists of a total of 130 people, are experts with over 20 years of experience and hold advanced degrees, bolstering the company's confidence in its technical prowess.

First Automotive NPU 'N1' to Launch in December, Mass Production in 2026... Supports LLM AI Models

In May, BOS completed the tape-out process for the N1 automotive NPU accelerator using Samsung Electronics' 5nm processand plans to release its first samples in December. Tape-out signifies that the semiconductor product design is complete and ready for mass production. The N1 chip is set to receive ASIL-B (Automotive Safety Integrity Level) certification and is expected to enter mass production in September 2026.

CEO Park stated, "When the N1 chip samples are available at the end of this year, we will conduct PoC(Proof of Concept) verifications with our customers. The N1 chip can support AI models, including LLMs(Large Language Models), in addition to autonomous driving functions."

The N1 chip boasts an AI computation power of up to 400 TOPS(trillion operations per second) and a power efficiency of 12 TOPS/watts. It also supports PCIe(PCI Express) 5th generation and CXL(Compute Express Link) interfaces and utilizes LPDDR5/5X memory.

He further explained, "Vehicles that will be releaed in the market within the next 1-2 years will have Level 3 autonomous driving function and support AI with LLM capabilities. Currently, most vehicles support voice recognition, but it lacks accuracy and has limited interface capabilities. However, LLMs can significantly improve the interface between the vehicle and the driver, which is highly anticipated in the industry. On the other hand, the development of Level 4 and higher autonomous driving technology is progressing more slowly, and it will take more time to achieve fully autonomous driving."

Collaboration with Tenstorrent... Strength in AI Model Flexibility

The integration of Tenstorrent's NPU IP into the N1 chip is expected to greatly enhance BOS's global competitiveness.

CEO Park commented, "Tenstorrent's NPU IP excels in computational performance and low power consumption due to its use of RISC-V, and its greatest strength lies in the flexibility of product applications and AI model support. Supporting various AI models in the market, including autonomous driving algorithms and LLMs, Tenstorrent's AI compiler and software stack allow customers to easily implement AI functions when using our N1 semiconductor."

He added, "In addition to Tenstorrent's software support, BOS's own NPU software team will provide additional software solutions for our customers' AI models, enabling customized AI NPU operations."

German Subsidiary to be Established by Year-End... Initiating 'A1 Development' with 4nm Process in 2025

In line with the launch of N1 at the end of the year, BOS plans to establish a German subsidiary in the third quarter of this year to commence full-scale sales and business development activities in the global market, with additional branches to be established in major countries next year.

The company also announced plans to start developing the ADAS(Advanced Driver Assistance Systems) integrated application processor named "A1", which is its next-generation product, in 2025.

CEO Park revealed, "We will begin development of A1, based on Samsung Foundry's 4nm process, next year with sample releases scheduled for the first quarter of 2026. Unlike N1, A1 will be a one-chip solution with integrated CPU and camera processing engines, enabling autonomous driving capabilities with a single chip."

BOS has plans to expand its development workforce from 130(90 located in Korea, 40 in Vietnam) to 180 by the end of the year to support the development of these new chips.

CEO Park concluded, "Team BOS is a collection of veterans in system semiconductor design and software development with over 20 years of experience from global semiconductor companies including Samsung Electronics. We provide an environment where engineers can quickly grow their technical skills based on this wealth of experience and competitive edge. We ask for your continued interest in our growth and success."