Broadcom Adds AI Features To Its Latest Network Chips To Data Centers

JAKARTA - Technology conglomerate Broadcom added an artificial intelligence feature to the latest version of one of its flagship network chips. This was conveyed by the company on Thursday, November 30, with the aim of helping move information in data centers more efficiently.

Allocating part of silicon in Broadcom's Trident network processor for artificial intelligence increases chip efficiency and its ability to handle other tasks such as network security. The new chip includes typical performance improvements such as lower power consumption and larger network bandwidth.

One of the biggest difficulties when building a large AI cluster is moving data in it.

"Chip, called Trident 5-X12, can help reduce some of the network traffic jams," said Robin Grindley, an executive at Broadcom's Core Switching Group, in an interview. There is certain network computing tasks, such as AI, which can only be improved by adding additional capabilities to the chip, as the software is too slow.

"That's what neural networks do - look at all the packages, all the traffic patterns, so try to identify these things that usually standard approaches won't be able to capture," Grindley said.

Artificial intelligence parts of these chips can be activated once customers build AI models based on traffic at their data centers. Data center operators can train models to detect unwanted types of traffic, such as non-available service attacks or network traffic jams. Once built, the shape of the model is put into the chip and can help steer traffic.

The company's engineers decided to add artificial intelligence features about two years ago after the company introduced programming capabilities to the previous version of the chip.

This latest version of Trident is built with a 5 nanometer manufacturing technology and is currently being delivered to "qualified" customers.