Rambus’ Tim Messegee has penned an article for Semiconductor Engineering that takes an in-depth look at the importance of ensuring security at the intersection of AI and 5G. As Messegee notes, 5G represents nothing less than a revolution in mobile technology with performance that is poised to rival that of wireline networks.
“Relative to its 4G predecessor, 5G promises 10X the data rate, 100X the efficiency, and 1000X the capacity, at 1/100th the latency,” he elaborates. “With 1Gbps speed at 1ms latency, 5G makes it possible to offer a host of real-time applications and services.”
Real-time is critical, says Messegee, because the rise of artificial intelligence (AI) runs in parallel to the roll out of 5G.
“As AI increasingly moves into controlling devices in the physical world, from delivery drones to autonomous vehicles, the high-speed, Ultra-reliable Low Latency Communication (uRLLC) links that 5G provides become a critical enabler,” he explains. “Most of 5G’s ‘users’ will be the things of the Internet of Things (IoT). Human users will benefit both from the improved high-data rate mobile experience of 5G, and from the AI-enabled and 5G-connected devices that will make our world smarter, safer, and more convenient.”
The vast number of 5G-connected IoT devices, Messegee notes, will generate a torrent of data.
“In a true virtuous cycle, 5G networks will make possible the collection of this enormous quantity of data,” he adds. “AI training requires vast amounts of data, and AI will be the only practical means of managing all this data.”
In this way, says Messegee, the success of 5G and AI are inextricably tied. While independently they create enormous value, together they create exponentially more. And with significantly increasing value, the imperative to protect said value rises commensurately.
“By their very nature, 5G networks will have increased attack surface that adversaries will try to exploit. For instance, to meet its [low] latency targets, 5G architecture pushes more computing to the edge of the network,” he writes. “For AI, this will enable both inference and even training at the edge. This distributes valuable AI algorithms (more opportunities for attack) and takes them out of the hardened data center environment (a lower barrier for attack).”
According to Messegee, this is precisely why it is so critical to safeguard the data carried by 5G networks. For example, with AI-powered devices flying the skies, driving the roads, and protecting neighborhoods, an attack which compromises the data coursing to and from these devices can threaten privacy, property, and personal safety.
“Thirty years of the web have made it abundantly clear that software-level security alone is not up to the task of protecting the real-time, always-on world of 5G and AI,” he explains. “The whack-a-mole game of patching software vulnerabilities is far too risky given the stakes. Protecting 5G networks, and the AI-enabled IoT devices that depend on them, will require security anchored in hardware.”
More specifically, says Messegee, secure processing cores embedded in the chips at the heart of 5G and AI devices can enable a system-level security architecture that can protect the entire network.
“Provisioned at time of manufacture, these trusted devices can attest to the validity of electronic systems and the data they process and communicate,” he states. “Hardened against direct and side-channel attacks, they extend protection to the edge and to end-point devices. Intelligent and flexible, they can be managed in the field to adapt to an evolving threat landscape.”
As Messegee emphasizes, there are incredible synergies to be realized when 5G meets AI.
“It is imperative that security anchored in hardware is part of the fundamental design philosophy of 5G and AI systems given the great value at issue,” he concludes.