Experts at the table, part 1: Rising costs, massive complexity, and challenges in testing and simulating these devices will require some fundamental shifts in the communications market.
Semiconductor Engineering sat down to talk about challenges and progress in 5G with Yorgos Koutsoyannopoulos, president and CEO of Helic; Mike Fitton, senior director of strategic planning and business development at Achronix; Sarah Yost, senior product marketing manager at National Instruments; and Arvind Vel, director of product management at ANSYS. What follows are excerpts of that conversation.
SE: What do you see as the big issues in moving to 5G?
Fitton: It’s a fascinating time for 5G, but if you ask 10 different people about this, you get 11 or 12 different answers about what 5G is going to be. It’s different from a geographic point of view. There are different things coming in Release 16 from Europe, China and North America. And then there are so many different waves of 5G, starting with sub-6 GHz first, and then millimeter wave after that. Our view is this is going to be a much longer deployment than we saw with 4G. There’s going to be a reasonable amount of uncertainty, lots of different applications coming through, and a requirement for continued investment and innovation. The question of where the tail is going to be for 5G, and how we mitigate cost. In Korea, there is a joint venture of three service providers and one ISP. They are developing one 5G network among the four of them to reduce their cost. There are going to be conflicting challenges of lots of changes, but a key one is trying to optimize everything and get it to pay off. What we see, from our point of view, is the need for flexibility and reprogrammability to adapt to these new standards coming in. We’re also seeing, with the really high bandwidths, a challenge that translates to custom products like ASICs and ASSPs. Tying all of those things together, we need a way to embrace changes that occur with Ultra Reliable Low-Latency Connectivity or mMTC (massive Machine Type Communication) or new frequency bands. So there will be a long deployment of sub-6GHz, and them millimeter wave.
Yost: No one has a good pricing model figured out yet. I’m not going to spend any more for my next cell phone than what I’ve already spent. The new iPhone is more than $1,000. People don’t want to spend $2,000 just to have a millimeter-wave chipset on it. But there’s also question of how the infrastructure providers make money. The current model is that these companies make so many dollars per bit. But if you’re looking at the IIoTand IoT, you’re using tiny amounts of data that go out to hundreds of devices. It’s not clear how that will work. If you think about the connected car, who’s going to pay for that? For the connected car to be successful, we’re going to have to do a pretty significant reboot to our infrastructure. Consumers don’t want to pay for hundreds of new macro cells to be put up, and neither do telecomms. So is it the car manufacturers? How are we going to change the monetization around that? We have challenges in taking some of these new technologies to market and figuring out how to get that cost model down. We’ve figured out how to make millimeter wave products with large numbers of antennas. This has existed for a long time in aero/defense.
SE: And they charge a lot of money for that.
Yost: Yes. So first we need to take this technology and get it commercialized to the point where we can manufacture it and make the physical chip. And then you have to look at how we’re going to test it. So if you look at the way we’re testing things today, if you think about going from 20MHz up to 800MHz, and then you take that chunk and add in the fact that you’re going to have an antenna array, and you test one configuration per element, you’re looking at a 2,500X increase in the time it would take to test that device. And then on the millimeter wave side, you also have to consider the battery power of the devices, because the arrays take up massive amounts of power to get a signal out that’s strong enough. There are a lot of technology challenges and a lot of business challenges, and then there is the idea of taking an initial technology and bringing it to broad commercialization.
Koutsoyannopoulos: You would be extremely surprised today if someone put out a cell phone with a price tag of $2,000. But you would be equally surprised if we had this discussion 10 years ago and someone said they were going to charge $1,000 for a cell phone. But today millions and millions of people are paying that price for something that 10 years ago would have been unbelievable. I wouldn’t bet for or against the $2,000. Seven years from now it might be reasonable to pay $3,000. It’s so complex to figure out how this dynamic will evolve on the consumer side, and we as engineers know how to design and test the product, and we know how to design to test the product. We want to be given a spec. But we cannot imagine how things will evolve. With 5G in particular, currently we have no idea why we will need 1 gigabit per second or 20 gigabits per second. If you give me 20 gigabits per second today, what additional value will I have? We have to look at the application and how that will evolve. If you look at the landscape from the semiconductor ecosystem point of view, we provide the solution to make this high bandwidth available. It may involve a chip, a system, an FPGA core that will enable the modem you’re testing to reach 20 gigabits. But somehow we may have to be involved in shaping the landscape for the end user. The problem from the design point of view is that we have many more radios on the same chip and broader bandwidth, and we have complexities we cannot see right now at the mobile system level. From the transceiver perspective, there are 20-plus radios on the same chip, maybe more, and from a design perspective this is beyond challenging. For the EDA community, this is an opportunity. We are not totally there with a solution yet. But to integrate that many radios on a single die, or in a single package, and to provide that level of bandwidth in a cell phone of the size we see today is not an easy exercise. This is where we see most of the design problems.
Fitton: From the form factor point of view, everyone has become used to the size of today’s cell phone. But it’s also about the other stuff, as well—the IoT and the Ultra-Reliable Low Latency Communication—in which case the integration challenges will be completely different.
Koutsoyannopoulos: It’s hard to imagine beyond the cell phone as a device. The current screens of six-plus inches are too large for me. But I do believe the form factor will be redefined. We still think of it as a talking device, or as a talking-plus-Internet browsing device. It’s not going to be like that.
Fitton: I agree. My favorite example of this is SMS (short message service). It was never intended as an application. It was an engineer at Vodafone who sent his boss a seasons greeting message using something that was just meant for mechanistic purposes. A few years later, it was a huge revenue stream for operators, and then it was replaced again. If you think about that progression, it’s startlingly fast. We’re not very good at predicting what the killer app will be.
Vel: What’s clear is that people have not been satiated with enough data yet. Having access to large amounts of data—not just on your cell phone—and having the bandwidth to transmit that data in real-time is going to be important. The mobile industry is obviously one of the drivers of that need for data, but there are others, as well. The automotive industry is very quickly catching up. With autonomous driving and ADAS, there are going to be huge volumes of data that have to be processed on board, but also transmitted and received in real-time. Along with that comes the infrastructure that is built to handle these autonomous vehicles. And along with that infrastructure comes commercial and industrial infrastructure, which connects to the cloud. So you have all these areas where communication, and the bottleneck of communication, is going to be the driver of 5G technology. It’s not just the end device that’s going to be affected by the problems of 5G. It’s the complete end-to-end market, from an IoT device or an autonomous vehicle or cell phone, to the front-end network, the back-haul network, and then to the cloud. The entire infrastructure will need to be overhauled for 5G. Different pieces of these networks have different challenges. If you’re looking at the challenges of designing the next generation of antennas for your cell phone, those are completely different from a base station, where you will need repeaters every 200 or 300 meters. Who’s going to fund that? Who’s going to build those? Who’s going to test those? It’s an opportunity for semiconductor and device manufacturers, as well as EDA providers to provide real-time solutions of ASICs and platforms. If you you have a 64- or 128-array antenna, just understanding how the beams are going to get formed, you cannot test it anymore. You cannot build prototypes and test them. You have to simulate them even before you take the first step in building them.
Fitton: I was looking recently at applying machine learning to beam-forming. How do you test that? That’s a really interesting example about how 5G enables machine learning, but also needs machine learning. If you think about getting the data from the edge.
Yost: If you’re looking at the beam combinations of 64-array antennas, there are millions of them. You need something to smartly figure out how to smartly track beams.
Fitton: And if you take into consideration how difficult it is to test that, if you have some kind of recursive neural network in between, no one really understands how something gets optimized in the middle of that. – Semiconductor Engineering