Xiaomi has unveiled details of its custom semiconductor roadmap, introducing two specialized AI processors alongside its next-generation mobile system-on-chip: the 6-nanometer Xring O100 near-memory AI accelerator and the 3-nanometer Xring D100 autonomous driving chip. Both processors are manufactured by TSMC and have completed hardware validation ahead of planned commercial rollouts.
The announcements follow a reported investment of more than 21 billion yuan ($3.1 billion) by Xiaomi into in-house silicon design, backed by an engineering team of nearly 3,000 specialists.
Near-Memory Architecture on the Xring O100
The Xring O100 is engineered as a dedicated coprocessor to accelerate large language models, including Xiaomi's proprietary MiMo model family, across smartphones, electric vehicles, and robotics platforms.
To address memory bandwidth bottlenecks common in local LLM inference, the 6nm processor utilizes a 3D near-memory packaging architecture. By vertically stacking silicon dies and memory layers connected through high-density vertical interconnect channels, the O100 achieves near-memory bandwidth of up to 1.22 terabytes per second (TB/s). This configuration is designed to reduce data transfer latency between compute units and memory during autoregressive token generation. Commercial integration of the O100 is scheduled for 2027.
3nm Xring D100 for Autonomous Driving and 200B Models
The Xring D100 represents Xiaomi's first custom 3nm automotive processor. Built for intelligent driving systems, the chip features:
- A 20-core central processing unit (CPU)
- A 16-core neural processing unit (NPU)
- Support for up to 160 GB of unified system memory
The hardware architecture is designed to execute models up to 200 billion parameters locally within vehicles, eliminating cloud round-trip latency for perception, path planning, and multi-modal decision-making. Xiaomi currently relies on Nvidia Thor and Orin chips for its production electric vehicle fleet, and plans to introduce the D100 to commercial vehicles starting in 2027.
Desktop Demonstration on AI Cube Prototype
To demonstrate multi-chip coordination, Xiaomi showcased the silicon inside an engineering hardware terminal called the AI Cube Prototype. The desktop enclosure pairs the new Xring O3 flagship mobile processor (slated to launch in September with the Xiaomi 18 Fold) with the O100 and D100 chips.
In live tests, the integrated prototype executed 120-billion-parameter and 3-billion-parameter models concurrently on-device, supporting dynamic switching between fast lightweight routines and larger reasoning workflows.
