Arm Ltd. today pulled back the curtains on its latest batch of chip designs, which includes two neural processing units optimized to run machine learning models and a graphics processing unit aimed at midrange phones.

NPUs have become a staple feature of premium handsets in recent years. There’s one in the A13 Bionic system-on-chip that drives Apple Inc.’s latest iPhones, while Google  LLC and Samsung Electronics Co. Ltd. ship AI-optimized silicon with own their flagship handsets. But cost constraints have so far prevented the technology from being widely adopted in devices on the lower rung of the price ladder, which is what Arm hopes to change.

The Ethos-N57, the first of the new NPUs, can perform up to two trillion operations per second at a 1-gigahertz clock rate. It’s designed for use in midrange phones and smart home hubs. The Ethos-N37, the other NPU, maxes out at 1 trillion operations per second and targets entry-level handsets as well as connected devices like security cameras.

The chips join the Ethos-N77, formally the Arm Machine Learning Processor, which the company introduced in May to meet the needs of premium mobile devices. It can perform about twice as many operations as the midrange Ethos-N57.

“Once the preserve of premium devices, immersive experiences such as AR, high-fidelity gaming and new AI-based use cases across mobile and home are now being demanded by the mainstream market,” Paul Williamson, the head of Arm’s client computing business, detailed in a blog post. He wrote that the vision behind the Ethos line is to “deliver premium AI experiences on our everyday devices.”

Joining the NPUs is the Mali-G57 graphics module. Similarly to the Ethos-N57, Arm designed it with the goal of bringing midrange devices closer to flagship phones in speed. The GPU provides 30% better performance than last year’s Mali-G52 for a “range” of apps, according to the company, and runs machine learning models up to 60% faster while consuming 70% of the power.

Capping off today’s announcements is the Mali-D37 DPU. A DPU, or display processing unit, is a tiny integrated circuit that offloads some graphics rendering tasks from a device’s GPU to make visuals smoother. The Mali-D37 is hailed as the smallest chip of its kind to date. Semiconductor manufacturers that license the design can implement it on a die with a surface area of just one square millimeter, according to Arm. 

“For the end-user, this means better visuals and performance on lower cost devices where area matters most, such as entry-level smartphones, tablets, and small display screens up to 2K” in resolution, Arm’s Williamson wrote.

“When Arm brings out products, that technology becomes available to hundreds of thousands of developers, signaling it is mainstream,” said Patrick Moorhead, president and principal analyst at Moor Insights & Strategy. “This is Arm’s first discrete AI accelerator, signaling that AI is mainstream and is unique enough of a workload that it requires special silicon versus leveraging CPUs and GPUs.”

Photo: Arm