Qualcomm, other vendors look to autos, drones, and other applications.
Mobile processors, also known as application processors, are well-known as the engines that run smartphones, tablet computers, and other wireless devices. But these chips increasingly are finding their way into autonomous vehicles, the Internet of Things, unmanned aerial vehicles, virtual reality, and other applications far beyond phone calls and text messages. Moreover, they are gaining in complexity as they are adapted for other markets.
This shift was evident at this month’s Consumer Electronics Show. Little attention was paid to smartphones and their apps, even though this is still one of the largest markets for complex SoCs. The real buzz was artificial intelligence, big data, machine learning, and self-driving cars.
As the market for smartphones and tablets flatten, processor vendors and their many suppliers in design software and services are branching out. The industry transition to 10nm chips, as exemplified by Qualcomm’s new Snapdragon 835 processor, is pointing to a future with more sophisticated components to deal with images, 4K videos, and voice commands, in addition to processing data.
Strategy Analytics estimates that Qualcomm commanded 39% of the global smartphone AP market during the first six months of 2016. MediaTek held second place, with 23%, and Apple took third, with 15% market share. The smartphone AP market reached sales of $10 billion in the first half of last year, a 3% gain from the same period in 2015, according to the market research firm.
“Qualcomm maintained its smartphone AP market share leadership despite MediaTek’s strong inroads in 1H 2016,” said Strategy Analytics’ Sravan Kundojjala in a statement. “After a less successful 2015, Qualcomm continued to recover its flagship market share with the newly launched Snapdragon 820, which featured in several flagships, including the Samsung Galaxy S7 Edge, LG G5, Xiaomi Mi 5, OnePlus 3 and others. Qualcomm’s mid-range Snapdragon 652, 650 and 400 series of chips also gained good traction during 1H 2016 and helped Qualcomm to maintain volume.”
That won’t be enough to sustain the kind of growth rates that companies like Qualcomm and Samsung have seen in the past, though, which explains why they are both getting into connected cars in big ways. Qualcomm’s proposed acquisition of NXP Semiconductors and Samsung’s purchase of Harman International Industries are evidence of this new way of thinking. This analysis, and this one, detail how Qualcomm and Samsung are diversifying their chip efforts into cars, drones, and other products.
NXP brought its new i.MX M8 Series application processors for IoT devices to CES, unveiling four models for audio, vision, and voice processing. Each IC has up to four 1.5-gigahertz ARM Cortex-A53 and ARM Cortex-M4F cores, along with full 4K UltraHD resolution and HDR video quality, and up to 20 audio channels and DSD512 audio.
Steve Roddy, senior group director, Tensilica marketing, at Cadence Design Systems, sees “an explosion of heterogeneous processing elements” in mobile processors, as “the number and complexity of digital signal processors really take off.” Mobile processors now are “always on, always listening,” he observes.
The Google Translate service currently runs on convolutional neural networks, Roddy notes. The upgrading of the language translation service now depends upon “how many compute islands are available,” he says. This kind of data processing is “more than megahertz,” he comments.
While processors are getting brawnier, their power efficiency is advancing by 20x, according to the Cadence executive. “We still need general-purpose CPUs,” he adds. At the same time, specialization for processors is requiring “more and more complex use cases,” Roddy says.
Today’s mobile processors are doing a lot more than taking selfies, though. Application processor companies are building on their research and development investments for handsets by “repurposing chipset devices” for other applications, notably automotive/infotainment electronics, drones, IoT, and security cameras, Roddy says.
Blurring the lines
Mobile processors historically have fallen into two main buckets, those with a modem and those without. But that distinction is becoming fuzzier.
“The line is blurring because cellular-connected tablets are gaining popularity due to interest from U.S. telecom carriers in pushing those products,” said Tom Wong, Cadence’s director of marketing for design IP. “As a result, tablets will bifurcate to Wi-Fi-enabled tablets and cellular-connected tablets. The tablet market will be clearly competitive, and many vendors are fighting hard for volume. White labels are getting squeezed out by sub-$100 tablets from leading providers when they clear out older models. A number of high-end smartphone providers have sufficient unit volume to justify using custom-designed AP SoCs (previously in the 28nm process) and moving to the most advanced process technology available today (16/14nm today, moving to 10nm and 7nm). Other high-end smartphone providers with lower unit volume will rely on commercial off-the-shelf offerings from leading mobile SoC providers.”
Regarding the industry transition to 10nm processes and the effect on mobile processor design, Wong adds, “The industry is driving toward more advanced geometries for power, performance and area advantages, namely lower power, smaller die size, and higher clock speed. 10nm was the first available, but it may prove to be like 20nm where the first went there, but those who followed just went directly to the next geometry. Whereas 10nm seems to be a transition like 20nm, the likely nodes will be 16 and 7. Then you have the marketing of what is really 10 and 7. Are they really that different, or is it just naming?”
While the numbers are fuzzy, the rising challenges in designing these chips certainly is not.
“At 10nm and 7nm, routability dictates profitability,” said Tobias Bjerregaard, founder and CEO of Teklatech. “Wires don’t scale. The only way around this is to improve the power integrity, which frees up resources for routing. One of the big problems chipmakers are facing is that there is not an economic benefit from scaling after 28nm. It takes a huge investment to work smart, and a high volume before you see an economic benefit. It’s only the really big guys with big volumes who will harvest the benefits of scaling after that.”
This is very clearly the domain of the largest chipmakers, and expanding into new applications is the only way to remain out in front with rising costs of designing complex chips.
Bigger, faster, more complex
With these advanced SoCs, bigger, faster, and more heterogeneous are competitive advantages. Multi-core chips are almost required, and the vast majority have between four and eight cores.
“The approximately 10mm x 10mm die size still seems to be the sweet spot for optimal size and yield,” Wong says. “Everyone is trying to pack as many features and functionality into a 10×10 die size. But the real story is in the transition from LPDDR3 to LPDDR4 and eventually LPDDR4x memory. Lower I/O voltage, higher speed (3200) and higher density are driving this trend.”
Brian Jeff, director of product marketing for ARM’s CPU Group, also sees the preference for eight cores in mobile processors, citing marketing desirability in some regions. Eight, for example, is considered an auspicious number in China and other Asian countries.
“Pipelines are going to continue to scale,” Jeff says. ARM’s big.LITTLE architecture often means four big cores and four little cores in one chip, providing operational and energy efficiency, he adds. “CPUs themselves are growing more robust. The power budget is still a big strength (particularly in handset designs). You really have to manage power in a smartphone.”
Stefan Rosinger, a senior product manager at ARM, sees a lot of innovation in process technologies. “Mobile processors scale very well across multiple markets,” he notes, including Chromebooks, clamshell devices, and in-vehicle infotainment for cars.
Automotive manufacturers are pursuing processor stacks for advanced driver assistance systems, Jeff says, often mixing CPUs and graphics processing units in the mix. He sees “diversity in architectures” for autonomous driving. “We’re very active in that space,” the ARM executive adds.
When it comes to heterogeneous processing and CNNs, the big.LITTLE architecture lends itself to “optimizing system-level power,” Jeff says. Microcontrollers and other types of chips are being utilized in the mobile space. “You’ll always have a CPU,” he adds. “The question is what will be paired with the CPU. Logic is becoming smaller and smaller. RAMs don’t scale as well.”
When it comes to performance, power, and area, chipmakers want the highest performance for the best power, Rosinger says. “Die size is important for our partners,” he adds, and cost is important, too.
But the chips need to be secure, too. “The CPU is one piece of the puzzle,” says Jeff. “We’re seeing a lot more attention paid to that. Cybersecurity is a cat-and-mouse game. We’ll see an increase in that.”
Arvind Narayanan, product marketing architect at Mentor Graphics, looks to the next wave in mobile processors, with significant growth in AI, automotive-class chips, and IoT. Image sensors are key to self-driving cars.
“There’s so much demand in that area,” he says. “There’s going to be a lot more chips in these cars. That will help us in EDA.”
While most chipmakers assumed that auto makers would rely on older technologies, that is turning out not to be the case. As autonomous driving races well ahead of predicted rollouts, car companies are looking to the most advanced processor available because they basically are building supercomputers on wheels with radar, LIDAR, and high-speed wireless.
And then there are other markets, such as 3D, graphics, and virtual/augmented reality to consider.
“The transition from 28nm to 20nm was a big jump for us,” says Narayanan. “We now have to support different rules with multiple metal layers on a chip, Going further to 14nm/16nm is still a lot of work. Triple patterning has emerged. It’s another big jump. 7nm work is well along the way. Initial runs have gone through. Customers are mostly focused on mobile devices and autos. These are the dominant areas for growth. Device scaling has a lot more complexity for mobile processors now.”
But it’s also not slowing down. The change from one process node to another seems to have boiled down to 12 months now, a more compressed cycle time than earlier years, he notes.
Nor is it confined to traditional markets. Navraj Nandra, senior director, marketing, for DesignWare Analog and Mixed-Signal IP at Synopsys, characterizes the application processor market as an evolving push/pull process between mobile processor suppliers and their customers.
“Chip vendors are moving into different markets as companies try to increase their market share,” Nandra says. “The emerging markets include automotive, big-data analytics, and drones. But the pull is coming from companies and startups that have funding, and many of those companies are based in China. The push is from big companies. The pull is from new entrants. And the largest silicon foundries are all pursuing production of 7nm chips, which are being scoped for auto applications, driven by ADAS.”
For mobile processors, there is also a push into industrial automation, machine-to-machine communications, and wearable electronics. When the PC era ended, the main focus was power—how much computing could be done on a single battery charge. It has now shifted to PPA, he says. And there it is likely to stay for the foreseeable future. “I don’t see any letup,” Nandra says.
Despite these new applications, though, there are lingering questions about whether these complex chips are the right solution for some tasks for which they are being employed.
“The market attention on IoT came largely from application processors moving into new markets, and we’re more than two years into that,” said Drew Wingard, chief technology officer at Sonics. “For most things, these chips are overkill. It’s attractive from a design cost perspective, because if you take a chip you designed for a mobile phone and use that design in a smart watch, then the design cost is zero. But the battery life is bad. I want to be able to work all day, get on a plane and land in another country, and still have your watch work. I still can’t do that today.”
Executive Insight: Simon Segars
ARM’s CEO digs into the future of mobile and the IoT, why ecosystems are so important, and what’s changing in the data center.
The Zen Of Processor Design
AMD’s CTO talks about how to achieve more performance per watt and how chip architectures are changing.
CPU, GPU, Or FPGA?
Need a low-power device design? What type of processor should you choose?
Sensors Enable ADAS
Advanced driver assistance requires a careful and complex balance of hardware, software and security.