Arm announced 'Project Trillium'.

This is a family of Arm IPs, including new processors with exceptional scalability, delivering enhanced machine learning (ML) and neural network (NN) capabilities. While the current technologies are focused on the mobile market, they will power a new class of ML-enabled devices with advanced compute capabilities, including cutting-edge object recognition capabilities.

“As AI accelerates into edge devices, it becomes increasingly challenging to deliver sufficient compute performance while maintaining power efficiency. To address this need, Arm is announcing Project Trillium, our new ML platform,” said Rene Haas, president of the IP Products Group at Arm. “New devices will require the high-performance ML and AI capabilities that these latest processors deliver, and our partners will be able to leverage the flexibility and scalability of the Arm platform to expand the range of implementations possible across a wide range of devices.”

Current ML technologies tend to focus on specific device types or specific requirements. Arm’s Project Trillium changes this by providing ultimate scalability. While the initial launch of Arm ML products focuses on mobile processors, future products will deliver relevant performance for sensors, smart speakers, home entertainment, and beyond.

Arm’s new ML and object recognition processors deliver massive efficiency gains over existing CPUs, GPUs, and accelerators, while also outperforming traditional DSPs. The Arm ML processor was designed from the ground up for ML. It is based on the highly scalable Arm ML architecture, maximizing performance and efficiency for ML applications:

Arm OD processors are specifically designed to efficiently identify people and other objects, with an almost infinite number of objects per frame. OD processors offer even greater performance, providing a high-performance and power-efficient people identification and recognition solution. This enables low-power, high-resolution, detailed face recognition in real time on smart devices.