ARM to Launch New AI Chip for Small Devices

ARM, a semiconductor technology firm owned by SoftBank Group, unveiled a chip technology aimed at putting artificial intelligence functions on tiny devices such as sensors designed to detect patterns in human speech or other streams of data.

ARM, a semiconductor technology firm owned by SoftBank Group, unveiled a chip technology aimed at putting artificial intelligence functions on tiny devices such as sensors designed to detect patterns in human speech or other streams of data.

ARM, which provides chip technology to mobile phone semiconductor suppliers like Qualcomm and end device makers like Apple, has been diversifying its customer base in recent years to markets such as self-driving cars.

It has also expanded to target areas like the Internet of things – in which many everyday devices such as traffic lights or agricultural irrigation systems will be fitted with Internet-connected sensors and automated.

ARM on Monday released the new Cortex M55 processor paired with what it calls its Ethos-U55 “neural processing unit.”

Chips with the technology will hit the market in 2021 and aim to carry out the special kinds of math needed by artificial intelligence (AI) software that can detect vibrations or pick out spoken keywords from a user.

The chips are designed to function with very low amounts of electricity. That allows devices such as sensors to last for years at a time on a small battery and to only connect to the Internet when needed.

Minimizing Internet connections can help protect privacy by processing data locally and sending only what is needed to remote servers while discarding the rest, ARM executives said.

Many fields such as health care will require data to be processed locally, with little or none of the results sent back to remote servers, Dipti Vachani, senior vice president and general manager of ARM’s automotive and Internet of things line of business, told Reuters.

“You may not want that data to move around,” Vachani said.

The new technology “is going to give you low-power processing, to process data where it is best suited, where you want to keep that data.”