Arm's Trillium: An IoT Project to Watch


By: R. Scott Raynovich

Processing power for artificial intelligence (AI) and machine learning (AI) is all the rage. You can see that in Nvidia's (NVDA) parabolic rise in stock price after it bet on AI technology with a specialized GPU. Now Softbank-owned Arm is rolling out machine-learning technology that will be embedded in its chip line targeting the Internet of Things (IoT).

As intelligent devices proliferate, they'll need more horsepower. Arm on Tuesday announced Project Trillium, a broad strategy for embedding machine learning code targeting neural network (NN) engines at the edge. Arm says this machine learning technology will be embedded in its Arm ML Processor, OD Processor, and Arm NN processors, powering a range of applications ranging from mobile camera processing to neural network frameworks such as TensorFlow and Caffee.

The chips are targeting the IoT edge and designed for running machine-learning and NN apps, not training them. Processors such as the Arm ML will be specifically designed for processing ML at the edge, offering performance of up to 4.6 teraops (TOPs). Arm will also offer a library of ML resources that can be used across Arm Cortex CPUs and Arm Mali GPUs. The software supports the Arm machine learning processor; for Cortex-A CPUs and Mali GPUs via the Compute Library; and for Cortex-M CPUs via CMSIS-NN. Arm is also offering an OD processor designed to detect images in real time with full HD at 60 frames per second. This could power a wide range of image-recognition apps.

Arm's moves are indicative of an Arms race (no pun intended) at the edge, as technology players look to upgrade chips, ML processing, and tools specifically designed for IoT devices that will be ingesting and processing large amounts of data. This is fueling excitement for a wide range of applications using augmented reality, such as using mobile devices for gaming or sensors in car for self-driving vehicles.

The craze for all things intelligent IoT has also fueled massive investment, with the expectation that the explosion of smart IoT devices will fuel entirely new chip markets. This is part of what led Softbank to lay down $31 billion to acquire Arm in 2016. Likewise, Qualcomm's merger with NXP Semiconductor, which has been inching along the regulatory approval process but is expected to close this year.