SiEngine announced on August 12, 2026, that its self-developed 7nm automotive-grade AI acceleration chip "TianGong 100" (NNA100) has entered full mass production. The company has simultaneously begun batch deliveries to OEMs, Tier 1 suppliers, and ecosystem partners. The chip delivers 96 TOPS of INT8 computing power and 102GB/s of dedicated LPDDR5 memory bandwidth.
Mass Production Milestone: From Lab to Front-Line Deployment
On August 12, 2026, chip design company SiEngine officially announced that its self-developed automotive-grade 7nm AI acceleration chip "TianGong 100" (model NNA100) has entered full-scale mass production. Concurrently, the company began batch deliveries to original equipment manufacturers (OEMs), Tier 1 suppliers, and ecosystem partners. This milestone marks a critical step forward for China's automotive AI chip industry in the edge-computing domain.
The TianGong 100 is positioned as a plug-and-play, independent automotive-grade accelerator for large models, targeting edge-side AI applications. Its core mission is to address the current bottleneck in expanding edge AI computing power and provide a complete domestic computing foundation for the rapid deployment of in-vehicle local large models.
Technical Deep Dive: The Plug-and-Play Architecture Innovation
The TianGong 100's greatest technical breakthrough lies in its independent add-on automotive AI accelerator architecture. According to SiEngine, this is reportedly the only specialized AI accelerator in China that requires no restructuring of the vehicle's electronic/electrical (E/E) architecture and is compatible with cockpit main controllers across all brands. This capability allows both new and existing models to upgrade their AI computing power with minimal deployment — no modification to original infotainment hardware or software is required — significantly shortening delivery cycles.
Key performance specifications include:
- INT8 Computing Power: 96 TOPS
- Dedicated LPDDR5 Memory Bandwidth: 102 GB/s
- Supported Model Scale: 3B to 7B parameter large models, or multiple 4B models running simultaneously
- Design Standard: Automotive-grade, meeting functional safety requirements
The 102GB/s dedicated memory bandwidth ensures AI inference resources are isolated, avoiding contention with the vehicle's baseline computing systems. This design philosophy delivers AI inference performance without compromising the stability of existing vehicle systems.
Market Positioning and Industry Significance
The competition for in-vehicle intelligence is rapidly evolving from cockpit interaction to AI large-model empowerment. However, edge-side AI compute expansion has long been an industry pain point — automotive-grade chip development cycles are long, validation thresholds are high, and upgrading compute power in existing models often requires a complete E/E architecture overhaul at prohibitive cost.
The TianGong 100's "add-on" approach offers a new solution:
- Lower Upgrade Barrier: Existing models can access AI large-model capabilities without altering the original vehicle architecture
- Shorter Time-to-Market: Bypasses the complex process of vehicle E/E architecture restructuring
- Broad Compatibility: Adapts to all brands' cockpit main controllers, expanding the addressable vehicle pool
- Compute Isolation: Dedicated memory bandwidth prevents resource contention with in-vehicle baseline systems
| Parameter | TianGong 100 (NNA100) | Typical Cockpit SoC Integrated NPU |
|---|---|---|
| Process Node | 7nm | 5-7nm |
| INT8 Compute | 96 TOPS | Usually <30 TOPS |
| Dedicated Memory BW | 102 GB/s | Shared system bandwidth |
| Architecture Changes | None required | Requires E/E restructuring |
| Cockpit Brand Compatibility | All brands | Limited to SoC ecosystem |
Supply Chain and Commercial Outlook
SiEngine's batch deliveries to OEMs and Tier 1 suppliers indicate that the TianGong 100 has passed rigorous automotive-grade validation processes. For automakers seeking rapid deployment of in-vehicle AI large models, this "add-on" accelerator offers a lower-cost, shorter-cycle technical path.
From an industry-trend perspective, 2026 is the pivotal year when in-vehicle large models transition from proof-of-concept to scaled deployment. The mass production and delivery of the TianGong 100 are expected to accelerate the普及 (popularization) of edge AI computing power in China and globally. For more technical details and market updates, visit EX1000.COM.
