The Huashan series is a high-compute, automotive-grade autonomous driving chip platform developed by Black Sesame Technologies, engineered to cater to comprehensive intelligent driving applications ranging from Level 2+ driver assistance systems to Level 4 Robotaxis. Anchored by proprietary core IP (NPU and ISP) and scalable multi-chip architecture, the series stands as a formidable contender within China's domestic autonomous driving semiconductor landscape.
The Huashan series currently comprises two core product generations: the A1000 family and the A2000 family:
Huashan A1000 Family (16nm process):
A1000: Peak compute performance of 70 TOPS, power consumption below 8W, with dual-chip cascading delivering up to 140 TOPS, tailored for Level 3/Level 4 autonomous driving applications
A1000L: Lower-compute variant tailored for entry-level ADAS and Level 2.5 driver assistance systems
In series production since 2022, the line-up is deployed across production models including the Geely Galaxy E8, Lynk & Co 07/08 EM-P, Dongfeng eπ007, BYD, and FAW Hongqi, representing one of China’s longest-running and most widely adopted automotive AI chips
Huashan A2000 Family (7nm process):
The A2000 family is Black Sesame Technologies' high-compute platform engineered for next-generation AI models. It successfully completed tape-out in 2025, received global commercial clearance in January 2026, and is scheduled for series production and vehicle integration in the second half of 2026.
Built on a shared technical foundation, the family offers four distinct compute tiers:
| Model | Compute Power | Target Applications |
|---|---|---|
| A2000N | 200 TOPS | Cockpit AI Box, lightweight ADAS |
| A2000L | 400 TOPS | Cost-effective Urban NOA (Navigate on Autopilot) |
| A2000U | 700 TOPS | Full-scenario generalised intelligent driving (Level 3-ready) |
| A2000X | 1000 TOPS | Level 3/Level 4 autonomous driving, Robotaxi |
The Huashan series' proprietary dual-core IP forms its primary technological moat:
Jiushao NPU Architecture:
Engineered specifically for large AI models, featuring native hardware acceleration for Transformer/VLA (Vision-Language-Action) architectures and world models
Full-pipeline support for INT4/INT8/FP8/FP16/FP32 mixed-precision computing, enabling FP16 models to execute natively without prior quantisation
Utilises near-memory computing architecture with dedicated high-speed on-chip cache boasting 8TB/s bandwidth and hundreds of megabytes in capacity, drastically cutting data transfer latency and power draw
Dedicated hardware acceleration for Transformer non-compute operators (such as reshape and transpose)
Xingmou ISP:
Features 4-exposure processing, 150dB high dynamic range (HDR), and 3D noise reduction (3DNR), ensuring reliable imaging under challenging conditions such as backlit tunnel exits, pitch-dark night driving, and glare from wet weather
Enables direct RAW data pass-through to the NPU, preserving the maximum dynamic range of the raw sensor feed
Functional Safety: The A2000 family incorporates a "3L" integrated functional safety architecture and has attained ISO 26262 ASIL-D certification
The Huashan series has made substantial headway in terms of mass production and commercial deployment:
A1000: Already featured across numerous production models from carmakers including Geely, BYD, Dongfeng, FAW, and JAC
A2000: Has secured multiple mass-production design wins with leading OEMs, with initial vehicle rollouts anticipated in the second half of 2026; the FAD Tianyan L3 Platform unveiled at the 2026 Beijing Auto Show harnesses dual A2000U chips, pre-engineered for China's mandatory national Level 3 autonomous driving standards taking effect in July 2027
Delivering 58–70 TOPS on a 16nm node, the Huashan A1000 outperformed the contemporary NVIDIA Xavier (32 TOPS), whilst the flagship A2000 variant with 1,000 TOPS establishes itself firmly within the industry's top echelon. Through a multi-tiered portfolio comprising "Huashan (high-compute automated driving) + Wudang (cross-domain cockpit-driving integration) + Yizhi (low-power edge AI)", Black Sesame establishes a complete scalable compute spectrum from 0.5 TOPS to 1,000 TOPS. This covers the entire gamut from entry-level ADAS to full Level 4 driverless mobility, positioning the brand distinctively against rivals such as Horizon Robotics' Journey series and Huawei's Ascend line-up.