ARM on Tuesday announced new solutions based on Project Trillium, its suite of technologies comprised of scalable processors, software, and architecture, centered around machine learning (ML), object detection (OD), and neural networking (NN) that's said to be "the industry's most scalable, versatile" machine learning solution. Project Trillium will start out with ARM ML processors scaled to provide the most benefit to mobile devices. These will be used in tandem with ARM's more traditional SoCs, though they'll provide a significant boost in performance. ARM says the platform can offer up to 4.6 trillion operations per second and provide a rise of two to four times the effective throughput through intelligent data management. Moreover, the platforms are highly efficient, offering three trillion operations per second or better per watt, in addition to boasting improvements in terms of thermal performance.
Another piece of the Project Trillium puzzle comes in the form of ARM's new OD processors which the company says offer real-time detection of people and objects with "virtually unlimited objects per frame" in Full HD at 60 frames per second. If the platform is as scalable as the company claims and is sold at a reasonable price, it should result in some interested real-world use cases beyond the relatively expensive object detection processing available today. At the very least, it should cement the company as a real competitor in that market. Meanwhile, the new NN software is meant for use along with the ARM Compute Library and CMSIS-NN. That means it works well with the chipset maker's Cortex CPUs and Mali GPUs, as well as the OD and ML processors. In fact, it is intended to act as a bridge between ARM's components and the commonly used TensorFlow, Caffe, and Android NN neural networks. Finally, as mentioned above, the new tech will be scalable in both directions, which means it will see use beyond mobile in both smaller and much larger applications.
As of this writing, there's no official market branding for the Project Trillium tech that's still referred to by its internal codename and no specific timeframe with regard to when it will become available or how much it could cost. Bearing that in mind, limited availability of the technology for early preview access will be offered in April and the company says the tech will become more widely available toward the middle of the year.