Xiaomi officially launched its first high-bandwidth AI acceleration chip designed specifically for edge-side large models - Xuanjie O100. The chip uses the industry's first 6nm wafer-level vertical stacking (Wafer on Wafer) advanced packaging and introduces Hybrid Bonding hybrid bonding technology, achieving a world-leading 1.4 micrometer bonding pitch.
In terms of performance, Xuanjie O100 offers an ultra-high bandwidth of 1.22 TB/s, which is 16 times higher than traditional flagship smartphones, significantly enhancing the data throughput and inference capabilities of edge-side large models. Official data shows that its edge-side inference speed can reach up to 330 TPS, further improving the real-time response capability of AI models on terminal devices.
As edge-side AI evolves from lightweight applications to large model inference, high bandwidth and advanced packaging have become important technical directions for mobile AI chips. Xuanjie O100 improves data transmission efficiency through wafer-level vertical stacking and hybrid bonding, indicating that the competition in smartphone AI computing power is shifting from simply pursuing computing performance to a more comprehensive optimization of storage bandwidth, packaging technology, and edge-side large models.