BOS Semiconductor's Eagle-N Pairs Tenstorrent NPU with Samsung 5nm for Automotive AI
BOS Semiconductor, a South Korean venture founded in May 2022 by former Samsung EVP Jaehong Park, has built Eagle-N, an automotive AI chip on Samsung's SF5 process with 250 TOPS of dense compute.
1 source 5h ago
Reporting from 1 source: ASCII.jp.
Why it matters
BOS is staking its automotive SoC on another company's NPU IP rather than building its own, so Eagle-N's case for carmakers rests on how well the Tenstorrent Tensix block integrates with the Samsung 5nm host.
Key Facts
- BOS Semiconductor was founded in May 2022 by Jaehong Park, who spent 22 and a half years at Samsung Electronics and was EVP of its automotive and custom SoC businesses.
- Eagle-N is manufactured on Samsung's SF5 process and delivers 250 TOPS of dense compute.
- BOS Semiconductor announced a partnership with Tenstorrent for automotive chip development in October 2023.
- Eagle-N's NPU uses 12 Tenstorrent Tensix Neo cores and a quad Cortex-A53 host, with a Cortex-M7 security island.
- Eagle-N was demonstrated at CES 2025.
Its NPU uses Tenstorrent's Tensix Neo cores, under a partnership the two companies announced in October 2023. The SoC pairs a quad Cortex-A53 host with a Cortex-M7 security island and 12 LPDDR5x channels.
The chip's security block is a Cortex-M7 paired with what the die slide labels RT-640, which reads as Rambus's RS-640 hardware security module. Across the interconnect, Eagle-N offers general-purpose I/O alongside UCIe x32 and PCIe Gen5 x8. Memory is 12 LPDDR5x channels at 16 bits each, a 192-bit total width running at 8533 MHz, which works out to roughly 204 GB/s.
The NPU side carries 12 Tensix Neo cores with a dispatch engine and additional logic that the slide calls Trinity. Four NoC2AXI units are arranged to reach 256 bytes per cycle of I/O, since each unit handles 64 bytes per cycle. BOS demonstrated Eagle-N at CES 2025.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google