Xiaomi has expanded its self-developed semiconductor efforts with three new Xring chips, including the Xring O100, a dedicated AI accelerator designed to handle more demanding artificial intelligence workloads directly on devices.
Founder, chairman and CEO Lei Jun announced the new chips during Xiaomi’s Autumn flagship product launch event.
Built for Faster On-Device AI
The Xring O100 delivers up to 1.22TB/s of memory bandwidth and is described by Xiaomi as the world’s first 6nm 3D wafer-level stacked advanced packaging solution.
It uses an advanced hybrid bonding process with an extremely small bonding pitch.
According to Xiaomi, the design provides 16 times the memory bandwidth of traditional mobile phones and enables on-device AI inference speeds of up to 330 tokens per second.
3D Stacking Reduces the Memory Bottleneck
The chip vertically stacks logic and high-speed DRAM dies using wafer-on-wafer technology and hybrid bonding.
This reduces the physical distance between processing components and memory, allowing data to move more quickly.
Xiaomi says the approach is designed to address a major limitation for running large language models on edge devices. Even when mobile processors have enough computing power, AI performance can still be restricted by slow data transfer between memory and processing units.
The O100’s high bandwidth and dense connections are intended to improve token-generation speeds for local AI models.
Multi-Core NPU and Dynamic Data Routing
The Xring O100 uses a multi-core NPU architecture optimized for different stages of AI inference.
It also supports dynamic data routing, allowing workloads to be handled more efficiently depending on the task.
Xiaomi has already demonstrated prototype devices using the chip, including tablet-style systems and its Xiaomi AI Cube mini-PC.
These prototypes combine the O100 with other Xring processors to support sustained local AI workloads.
Commercial Devices Expected in 2027
Xiaomi expects commercial applications for the Xring O100 to begin in 2027.
Potential uses include smartphones, tablets, robotics and other devices that require powerful local AI processing.
The O100 forms part of Xiaomi’s wider Xring chip strategy, which also includes flagship mobile processors and chips designed for vehicles.
Rather than focusing only on smaller manufacturing processes, Xiaomi is also investing in improved memory bandwidth and advanced packaging technologies.
The company says the goal is to make generative AI faster, more private and usable even without an internet connection.
The post Xiaomi Xring O100 Can Run AI at 330 Tokens Per Second appeared first on ProPakistani.
