Google Research, the tech giant's advanced research organization, has unveiled a next-generation "Edge AI" semiconductor that significantly reduces power consumption.

During the annual "Research@" event held at its headquarters in Mountain View, California, on the 14th (local time), Google Research previewed a low-power AI chip from the "Astra" series, developed in partnership with semiconductor company Synaptics.

Developed for Edge AI, the chip incorporates "M2," the successor to the Coral Neural Processing Unit (NPU), an AI computing processor first introduced by Google last year.

Edge AI is a technology that performs AI computing directly on wearable smart devices such as wristwatches and glasses, rather than sending data to cloud servers.

This approach reduces the time required for data transmission and responses, while also lowering concerns over leaks of personal and sensitive data.

However, because it must run on small devices and is primarily installed in wearables, it requires high power efficiency and minimal heat generation.

Avinatan Hassidim, Vice President of Google Research, explained that current edge chips are slow and consume too much energy, making an ambient AI experience impossible. He added that the newly unveiled Coral NPU M2 version uses 40% less energy than existing products.

Consequently, the chip is expected to be used for real-time processing of data such as voice, video, motion, and context detection in wearable devices.

Rahul Patel, CEO of Synaptics, emphasized that the chip will not remain in the lab but will also be embedded in Google devices.

However, he did not specify which devices it will be applied to or when it will be released.

Synaptics plans to reveal the detailed specifications of the chip and conduct a live demonstration at its company event on the 20th.

With the development of this chip, following its Tensor Processing Unit (TPU) chips for large-scale data centers, Google is expanding its ecosystem in the Edge AI sector and positioning itself to capture leadership in the on-device AI market.

In addition, Google Research showcased a wide range of research achievements at the event, including "Google Earth AI" for predicting global environmental conditions and public health, "Deep Review" for verifying the logic and factual accuracy of technical documents, and "TimesFM," a time-series data prediction model.
 
(Photo: Yonhap News)