Physix FrontierPhysix Frontier · News Briefing Card (IT Home · Aug 14, 2026)
SiEngine Mass-Produces 7nm Auto AI Chip at Sub-$150 Cost
KEY FACTS
- On August 12, SiEngine Technology announced mass production of its 7nm automotive-grade AI accelerator chip, Tiangong 100.
- It uses a standalone plug-in architecture, allowing compute upgrades without rebuilding the vehicle's electrical/electronic architecture.
- It delivers 96 TOPS INT8 compute and 102 GB/s of dedicated memory bandwidth.
- It supports local execution of 3B-7B large models, with hardware cost kept in the thousand-yuan range.
- It has passed AEC-Q100 Grade 3 and ISO 26262 ASIL-B safety certifications.
KEY DATA
7nmProcess node
96 TOPSINT8 compute
102 GB/sMemory bandwidth
Thousand-yuan rangeHardware cost
PHYSIX OBSERVATION
The Tiangong 100 breaks the high barrier of traditional main-controller replacement with a plug-in add-on approach, letting existing vehicle models access on-device large models at low cost. This decoupled design not only avoids the risk of E/E architecture reconstruction, but also turns AI compute into a standardized accessory, signaling that in-vehicle intelligence is moving from closed integration toward modular ubiquity and offering smaller automakers a pragmatic path to quickly catch up on AI cockpits.
Source: IT Home report