Eras of human history are named for the materials that compose the tools of the time, i.e., the stone age, the bronze age and the iron age. In many ways, the early 21st Century could be called “the ARM age,” given that so many products in everyday life—from smartphones to tablets, to the internet-of-things devices—are built using ARM Ltd.’s processors. Now, in a landmark development that could make artificial intelligence as ubiquitous as the ARM processor itself, ARM has announced new products that have built in hardware designed to accelerate AI tasks like machine learning.
ARM announced the Cortex-A75 and A55, the company’s first processors to use DynamIQ technology. DynamIQ allows companies building chips using ARM technology to employ flexible combinations of processor cores, allowing mixtures of computing capabilities that are optimized for specific tasks, such as running AI algorithms. DynamIQ also soups up AI performance using a redesigned memory subsystem and modified caches work, yielding a doubling of memory streaming performance relative to preceding processors in the ARM line.
The combination of DynamIQ and other enhancements—such as new processor instructions, microarchitectural improvements, and software optimizations—are expected to produce dramatic improvements in performance. ARM predicts DynamIQ and the other improvements will yield a fiftyfold increase in AI performance on its processors during the next three to five years.
Microchips using these processor designs are expected to start shipping in early 2018.
With the two processors, ARM is addressing a broad gamut of applications. The Cortex-A75 is designed for higher-performance uses, including mobile phones, laptop computers, network infrastructure gear, automotive designs, and servers. The Cortex-A55 is a more power-stingy device, suited for applications ranging from edge devices to the cloud, according to ARM.
All these applications could benefit from the addition of machine learning and AI capabilities, from smart surveillance cameras, to autonomous cars that become better drivers over time. ARM’s highest profile market—mobile devices—also is adopting various AI features, from object-recognition-enabled digital assistants to facial recognition systems for user authentication.
Given the prevalence of ARM, this means AI and machine learning could soon see much more widespread adoption. ARM estimates that more than 86 billion devices using its processor technology have shipped cumulatively as of last year.
Moreover, ARM is not the only company adding AI enhancements to its central processing unit (CPU) technology. Intel this year will test first silicon on its Lake Crest coprocessor, which is designed to accelerate the performance of neural networks. The company predicts it will be able to boost the performance of neural networks by a factor of 100 by the end of the decade.
The integration of AI into CPUs will allow machine learning and other artificial intelligence technologies to spread to wherever these microchips are used. Because of this, future historians may look back at this time as the start of the “AI age.”
Stephan Cunningham is vice president, product management at Veritone. Working in concert with core internal teams including industry-specific general managers and engineering as well as directly with clients and prospects, he leads the disciplines and business processes which govern the Veritone Platform.