Increasingly, artificial intelligence has been transferred to edge devices with higher computing powers. ARM, the UK silicon IP manufacturer, has unveiled several solutions in this regard.
Artificial intelligence has become one of the most discussed topics these days, with both security and non-security applications. Increasingly, that intelligence has been transferred to edge devices with higher computing powers.
ARM, the UK silicon IP manufacturer, has unveiled several solutions in this regard.
ARM officials discussed this topic during a technology symposium held in Taipei in early November, part of its technology roadshows that took place in several other countries in Asia, including China, Japan and Korea.
AI has already found its way into various security and non-security applications, including digital assistant, augmented reality, fingerprint identification, facial recognition, predictive text and speech recognition. In terms of security, deep learning-based video analytics can help identify problems or irregularities with further precision and accuracy. When applied to post-event investigation, for example, the system can locate and pinpoint problems or special situations in a matter of seconds from piles upon piles of recording.
According to John Ronco, VP and GM of Embedded Line of Business at ARM, accelerating AI adoption begins at the edge.
“Today a lot of AI takes place in the cloud. But for AI to be truly successful, more and more AI computing will need to move to the edge. There are several reasons for this, including bandwidth, efficiency and cost,” Ronco said. “If you are processing lots of videos it’s just not practical to send the video streams to the backend. Being able to do more processing locally on the edge near the user is something that’s essential.”
ARM solutions
According to Ronco, ARM has unveiled several silicon intellectual property (IP) products, or the building blocks of systems on chips, in this regard.
With more computing taking place on devices, rather than in the cloud, that data is processed faster and more securely. For this ARM has unveiled DynamIQ, a platform on which different processors are built. It allows dedicated, advanced processing instructions for AI, so more information stays on the device, giving users more privacy and security. Additionally, DynamIQ can integrate with accelerators through a new dedicated port that delivers a significantly quicker response time, enhancing the user experience.
In terms of processors, ARM has announced the Coretex-A75 CPU with applications in smartphones, intelligent home devices like DTV, servers, and telematics. According to Ronco, the CPU enables over 20 percent more integer core performance and high-performance NEON and FPU engines, providing a significant performance boost compared to the Cortex-A72 and Cortex-A73 processors. The CPU also provides additional performance for advanced workloads, like machine learning.
The Mali-G72 GPU, meanwhile, ensures arithmetic optimizations tailored to complex machine learning and high-fidelity mobile gaming use cases, providing 25 percent higher energy efficiency, 20 percent better performance density and 40 percent greater overall performance than devices based on previous generation Bifrost GPU, Ronco said.
Importance of cybersecurity
While devices are becoming smarter than ever, they are at the same time subject to greater vulnerabilities especially cyberattacks. In October last year, DDoS attacks were launched against Dyn, an Internet performance and management company based in New Hampshire. The result was a shutdown of service across various famous sites including Amazon, the Financial Times and Netflix. It was later found out that poorly protected networked devices, including cameras and NVRs, were used as robotic attackers after being affected by a malware called Mirai.
“We are in danger of creating an insecurity of things if we are not careful, and we are in danger of eroding trust from the customers. It’s certainly an area we believe we need to evolve more in,” Ronco said.
ARM has also built cybersecurity into its solutions. These include ARM’s platform security architecture as well as the Cryptoisland-300 and SDC-600 IPs.