GF spoke 5G and the world listened at Mobile World Congress 2018
The Mobile World Congress in Barcelona, Spain is the wireless industry’s leading annual event, and this year’s edition in late February was buzzing with talk of 5G wireless technology. GLOBALFOUNDRIES seized the moment on the show’s very first morning with a special program on 5G’s evolving uses and technology requirements.
First, Gregg Bartlett and Dr. Bami Bastani, Sr. Vice Presidents of GF’s CMOS and RF business units, respectively, outlined 5G-related semiconductor challenges and opportunities for an audience of device developers, networking specialists and high-performance computing architects. 5G will impact all these areas because it enables smarter devices to feed through higher-bandwidth connections to ever-more-powerful data centers.
Then, a panel discussion moderated by Mike Cadigan, GF’s Sr. Vice President of Global Sales and Business Development, and head of GF’s ASIC business unit, took place. The panel was made up of invited experts from Nokia Mobile Networks, Mobile Experts LLC and TU-Dresden.
They gave insights into why 5G networks aren’t likely to roll out on a nationwide scale, why a one-millisecond network latency is “magical,” how working directly with a foundry can support more holistic solutions, and many other important considerations.
5G Computing Demands Optimized Silicon
Bartlett said 5G will drive profound changes in the computing requirements for devices and data centers, because the complexity and volume of network traffic are growing exponentially as the result of more users, more transactions per user and richer content per transaction.
“Data center applications will require very fast processors and near-100% uptime, while edge-connected devices will require chips with extremely low-power/low-leakage performance, and with embedded memory for storage and RF for wireless connectivity,” he said.
Both applications also will make use of artificial intelligence (AI) functionality but they will do so differently, he said. Data centers will use AI to learn, anticipate and direct the behavior of devices and networks, while edge-connected devices such as automotive cameras will use it locally for real-time processing and inference. 5G bandwidth is essential to support all these uses.
Bartlett said many companies will find it difficult to take advantage of 5G opportunities because of the significant investments required in design tools, EDA, intellectual property (IP) development and verification. “Many new, innovative companies can’t absorb these development costs, and they need technology solutions offering both competitive advantage and cost reductions going forward,” he said.
He explained how GF’s dual-technology roadmap offers this flexibility, with advanced FinFET CMOS technology for high-performance computing, and FD-SOI technology for wireless and battery-powered applications, both of which can be integrated with best-in-class RF functionality. Application-specific ICs, or ASICs, are another path forward to 5G, and GF offers the leading ASIC IP portfolio and more than 1,000 experienced engineers.
While many customers are clamoring for such wide-ranging, flexible foundry solutions, not all foundries are able to respond. “We have a growing portfolio of what I call ‘revolutionary’ customers, who are using new silicon as a wedge to break or change their industry’s traditional competitive framework,” he said. “They are demanding easier access to silicon and we have aligned ourselves accordingly to provide the optimized solutions they need.”
5G Connectivity Brings More Complexity
On the connectivity side, Bami Bastani said 5G will be rolled out in stages, leveraging the existing 4G/LTE backbone. First there will be enhancements to the existing system, then an initial rollout of sub-6GHz bands with massive MIMO architectures for high-rate transmission, and then a second rollout to expand network capacity and drive even higher data transmission rates by leveraging mmWave bands.
“This all means a more complex radio is required, one that works not only with new network protocols but also with legacy protocols and bands,” he said. “Thus, front-end modules (FEMs) must evolve in many ways as the transition from 4G to 5G takes place.”
Bastani said GF’s rich RF portfolio of silicon-on-insulator (SOI) and silicon germanium (SiGe) technology platforms creates differentiation for customers, as these optimized solutions can address specific customer performance, complexity and cost demands. He gave two examples.
For 5G basestations, control of the antenna arrays will require much more complex signal processing circuitry. “This process is called beamforming, and it can be done with analog, digital or hybrid circuitry depending on the size of the array. How the system is partitioned drives the choice of technology, and GF has a rich set of offerings to address all requirements,” he said.
The requirements are different for small mobile devices. “You’re now dealing with smaller arrays which require higher power per element to achieve the same radiated power. The good news is we can now do much of the beamforming digitally, thereby leveraging the scaling of advanced nodes like 22FDX to achieve low power and cost for these applications,” he said.
Industry Experts Outline the 5G Future
The discussion then shifted to a panel of experts including Joe Madden, principal analyst at Mobile Experts, Professor Frank Fitzek, head of the Deutsche Telekom Chair of Communication Networks at TU Dresden, and Michael Reiha, head of RF IC R&D at Nokia Mobile Networks.
Joe Madden started the panel dialogue, commenting that 5G networks will roll out differently than previous networking technologies. These were characterized by rapid surges of deployment because they enabled existing, widely used applications such as email to go wireless. By contrast, he said, 5G primarily benefits network operators and as-yet non-existent markets.
“From a network operator’s viewpoint the real advantage of 5G is cost. Today, it costs about $1.50 to deliver 1 GB of data over an LTE network, but with mmWave 5G it might be 5 cents or less,” he said, which implies there will be islands of deployment initially, such as in urban centers with dense network traffic or where it’s specifically needed for certain IoT applications.
Moving on to the topic of 5G standards, Cadigan asked Prof. Fitzek to describe the ways in which they are evolving, and how it relates to foundry technology. “Transporting more data isn’t really the issue, it’s all about latency. In that regard, why do we keep arguing that a 1-ms latency requirement is so magical? Well, it has to do with the physics of feedback loops,” said Prof. Fitzek. (Latency is the inherent delay in the network.)
He gave the example of a 50 Hz power plant feeding electricity into a smart power grid. A mere 10-ms of latency in the grid would result in such large phase shifts in the generator’s electrical output that it could be damaged, he said, whereas 1-ms of latency would be adequate.
“Many people think that if you put the wrong figure for latency into the standard, you can just fix it later. But it will be hard to fix, and to get the full value of 5G networks it must be there from the start.” This doesn’t pose a problem for semiconductor technologists, he said, because they are already very familiar with feedback loops.
The need for low latency is a major reason why Nokia designed its recently introduced 5G Reefshark chipsets itself instead of working with a fabless semiconductor company, according to Reiha. Cadigan asked him what that might mean for future foundry relationships.
Reiha said that to achieve such low latencies one needs to look at 5G requirements holistically, with a vision of the future that semiconductor solutions are flexible enough to support. “Nokia Bell Labs literally wrote the book on massive MIMO, and this enables us to understand the system-based challenges. We also understand the importance of seamless integration of semiconductor functions,” he said.
“What we expect from our foundries is an honest dialog and open access to IP to maintain our quality standards. We need quality IP because we can’t do everything, we’re not experts in all domains,” he said.
Cadigan went on and asked the panelists for their perspective on the approach GLOBALFOUNDRIES is taking in the 5G space. Madden said that GF’s ability to integrate various technologies is very important. “As we go to massive MIMO arrays, there is pressure to reduce the size of radio arrays as well as receivers. There can’t be large transmission lines, and multichip modules where everything is tightly integrated are essential,” he said. Cadigan noted the advanced packaging technology which came to GF from IBM.
Reiha said GF has the best-in-class RF capability, and that from Nokia’s perspective the continuation of ongoing device model improvements for RF is key. “This is especially needed for thermal device models and also for technologies such as SOI to enable more of a seamless mixed-signal simulation environment that would let us build many more sensors and put more control on our RF die, which would really let us focus on having an AI footprint at the antenna interface,” he said.
Prof. Fitzek talked about the importance of software and the openness of GF’s technology. “Because at this point you can’t really foresee what users will do, and machine learning will have its own purposes, your software APIs will only become more important in the future.”