Arm debuts first-gen machine learning processor at Hot Chips

Arm debuts first-gen machine learning processor at Hot Chips

NEWSBYTE Semiconductor specialist Arm debuted its first-generation machine learning processor this week at the annual Hot Chips conference in Cupertino, California.

The Trillium architecture brings together functionality, map compression, programmability, and low-power operations in an open source technology which, some analysts suggest, could cause problems for makers of devices that major in any one of these individual areas.

Speaking at Hot Chips, Ian Bratt, director of technology at the British-founded, Cambridge-based company (now majority owned by Japan’s SoftBank), said that the aim was to create a generalist high-spec chip that would be equally at home in a data centre or a connected device, such as an autonomous vehicle.

“We could see how GPUs, CPUs, and DSPs were being used for machine learning but we started to see how we could leverage technology in each of these distinctly,” he said.

“We could leverage technology in CPUs around control and programmability and from GPUs in terms of data compression, data movement and arithmetic density, for instance, all with the efficiency uplift of a DSP and with an open source software stack.”

The processor brings together 16 SRAM slices, MAC and DMA engines, and Arm’s own Cortex technology to control the chip, along with static scheduling – which removes the need for cacheing – efficient processing, bandwidth reduction mechanisms, low energy usage, and inbuilt flexibility.

Internet of Business says

Licensees of Arm’s technology or architecture, which include Apple, NVIDIA, AMD, Huawei, Infineon, Intel, Texas Instruments, Broadcom, Qualcomm, and Microsoft – stand to benefit from a generalist chip design that brings together the best of several worlds.

IP for the chip is set to be released to partners later this year.