Specifications
Device Type
Chiplet-based automotive AI accelerator SoC
NPU Performance
Up to 250 TOPS (INT8)
Scalable Performance
Chiplet scaling to 2,000+ TOPS across multiple chiplets
NPU IP
Tenstorrent Tensix cores tailored for automotive applications
Tensix Core Composition
Array math unit for tensor operations, SIMD unit for vector operations, a network-on-chip, five 'baby RISC-V' control processors and up to 1.5 MB of SRAM
Host Interfaces
PCIe Gen 5 and UCIe high-speed interfaces
Integration Model
Add-on NPU and AI subsystem that connects to any existing ADAS or IVI processor
Functional Safety
ISO 26262 ASIL-B compliance, with ISO 26262 functional safety management certification held
Qualification
AEC-Q100 Grade 2
Security
Built-in hardware virtualisation and security engine
First Silicon
First engineering sample operated correctly across CNN vision, transformer LLMs and multimodal foundation models
Sibling Product
Eagle-A — a 150 TOPS all-in-one ADAS SoC
Product Family
Eagle chiplet SoC family for automotive and robotics
Company
BOS Semiconductors, a Korean automotive fabless company founded in 2022 and based in Seongnam
Workload Coverage
CNN vision, transformer LLMs and multimodal foundation models on first-pass silicon
Presented
Chiplet-based Eagle-N presented at Hot Chips 2026, in the automotive session alongside Waymo
Overview
Eagle-N is an automotive AI accelerator delivered as a chiplet rather than a monolithic SoC. It supplies up to 250 TOPS of INT8 NPU performance built on Tenstorrent's Tensix architecture — each Tensix core containing an array math unit for tensor operations, a SIMD unit for vector work, a network-on-chip for core-to-core and chip-to-chip data movement, five 'baby RISC-V' control processors and up to 1.5 MB of SRAM.
The design premise is that automakers and Tier-1s should not have to replace their existing ADAS or IVI domain computer to add AI capacity. Eagle-N attaches over PCIe Gen 5 or UCIe as an add-on NPU and AI subsystem, so it can be layered on top of a platform already in production. Chiplet scaling extends the architecture beyond 2,000 TOPS where higher autonomy levels demand it.
Automotive-grade qualification is central to the proposition: ISO 26262 ASIL-B functional safety compliance, AEC-Q100 Grade 2 qualification, and hardware virtualisation with a dedicated security engine. On its first engineering sample BOS demonstrated correct operation across CNN vision models, transformer LLMs and multimodal foundation models. The smaller Eagle-A is a 150 TOPS all-in-one ADAS SoC in the same family. QS Compute supplies edge AI accelerators, automotive-grade compute modules and embedded AI systems — request a quote.
Key Benefits
Additive upgrade path: bolt AI capacity onto existing ADAS/IVI processors over PCIe Gen 5 or UCIe. Scalable to 2,000+ TOPS: add chiplets as autonomy requirements rise. Automotive qualified: ASIL-B functional safety and AEC-Q100 Grade 2. Open NPU IP: RISC-V based Tensix cores rather than a closed accelerator block.
Applications
Advanced driver assistance systems (ADAS), automated driving domain controllers, in-vehicle infotainment AI, robotics and drones, industrial automation, and intelligent-space inference at the edge.
Request a Quote — BOS SEMICONDUCTORS EAGLE-N — 250 TOPS AUTOMOTIVE AI CHIPLET SOC
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →