Google 推进 Coral NPU 端侧 AI 硬件，发布开放架构、编译工具并与 Synaptics 合作推出商用芯片

Google AI· @GoogleAI · X·2026-10-23 00:28· 55 分钟前AI 评分51

AI 导读

Google 宣布推进其端侧 AI 硬件蓝图 Coral NPU，发布完整的开放硬件蓝图与软件、实际芯片和硬件套件。Coral NPU 在数据所在位置单步完成计算，避免传统芯片因数据来回搬运消耗电量，让可穿戴设备、智能传感器等小型设备低功耗运行复杂 AI；Google 正与 Synaptics 等伙伴推出实体商用芯片。博客：https://goo.gle/4i1iUEZ

正文 · 原文

https://x.com/i/article/2113304858849587200

【X 长文】How we're bringing powerful AI to small devices without draining the battery.

We’re advancing Coral NPU, our blueprint for on-device AI hardware, even further with a new open architecture, compiler tools, and commercial chips.

Some of the most useful AI experiences happen right on the devices around us, from wearables that understand context in real time to smart sensors and robots. But running complex AI drains batteries because traditional chips spend a lot of energy moving data back and forth to compute. Last year, @GoogleResearch introduced Coral NPU to fix this.

Coral NPU now handles all that heavy lifting in a single step, right where the data lives, giving small devices massive brain power while using very little battery.

So, how did we do it?

Right now, making Edge AI work on smartwatches or home robots is a huge headache for developers because every device speaks a different, complicated language. To solve this, we created a blueprint, the Coral NPU open hardware standard. It’s an open design that tells chip makers exactly how to build this chip so it processes AI efficiently without the energy-draining data commute, and the new software acts as the master translator.

Today, we’re taking a major leap forward with these tools by releasing:

— The completed open hardware blueprint and software — Real-world chips and hardware kits

We’re teaming up with partners like @SynaCorp to roll out the actual physical chips, meaning the foundation for the next generation of smart, battery-friendly gadgets is just around the corner.

Read the blog to learn more: https://goo.gle/4i1iUEZ

来源：Google AI · x.com