There will be billions of edge devices and cloud servers, but the real money is in optimizing the pieces and making them all work together.
Nobody knows how many tens of billions of semiconductors will be used in the IoE, but it’s a sure bet it won’t be a few chips replicated billions of times.
Most IoE devices will need to be customized for specific applications. Many will need to be highly reliable for many years. And all of them will need to be secure and power-efficient. Yet they also will need to connect to heterogeneous networks, be rolled out quickly in small batches, and for many markets they will need to be as inexpensive as mass-produced ASICs. In effect, they will require the benefits of economies of scale and advanced engineering, but only be produced in small batches with low average selling prices.
These are seemingly contradictory goals when looked at from the traditional semiconductor design-through-manufacturing standpoint. In fact, that was the IC industry’s initial conclusion about the IoE’s profit potential. But more recently a different view has started to emerge, namely that there are huge profits to be made if the industry can shift to new ways of doing things. And as more companies begin adjusting their sights and redirecting their resources, a race has begun to find profitable opportunities that crisscross vertical markets, as well as unique segments within those markets, and even across industries that previously had nothing in common.
As Subramani Kengeri, vice president of global design solutions at GlobalFoundries, observed: “Five years from now, everything will have a radio. Whether that will be very advanced logic or planar FDX or 2.5D remains to be seen. But all of them will need to be done at a very low cost.”
For most of the semiconductor ecosystem, figuring out a way to survive in the IoE Age is critical. The number of companies following Moore’s Law is shrinking at each new node, and the demand for chips made at the latest process node is consolidating into a handful of large systems makers. That has set off a series of rather dramatic shifts in recent months, including a long list of strategic acquisitions and spinoffs, new technology approaches for companies—there are many projects under development that have not yet been made public—and a rethinking of core competencies that continue to raise eyebrows.
Today, for example, Mentor Graphics will enter the business of selling intelligent IoT gateways, which it intends to offer for everything from edge devices to industrial automation to the Cloud. Two weeks ago, ARM bought Carbon Design Systems so it can offer virtual prototypes to speed time to market. And in August, memory IP developer Rambus announced it would begin selling controller hardware for DDR4.
This follows some other significant shifts in the past year, including Marvell‘s push into a 2.5D modular architecture so chips can be quickly assembled for the IoE, Synopsys‘ continuing commitment to software development (which included more acquisitions), Cadence‘s deepening focus on embedded vision, and the licensing of PRPL’s security outside of just the MIPS architecture. And then, of course, there are some very high-profile acquisitions including NXP‘s pending acquisition of Freescale, Avago’s purchase of Broadcom, not to mention broadening shifts into hybrid emulation and verification and more choices from foundries and OSATs.
Viewed individually, these moves tend to get lost in a dynamic $344.5 billion global industry, which is the Semiconductor Industry Association’s estimate for 2015. But looked at together, they create a picture of an industry in the midst of a huge transition—one that collectively accounts for tens of billions of dollars of new investments with no end in sight.
The big issue for many companies is figuring out where to play in the IoE, because it encompasses every device that can be connected to the Internet, from a remote sensor to a server in a datacenter. Achieving economies of scale requires finding a common thread that makes a business model scalable across multiple markets, including some that were never connected in the past.
Mentor Graphics’ push into IoE gateways is one such example. The challenge was to make its design flexible enough so it could be produced in small volumes, or even used as a reference platform, according to Warren Kurisu, Mentor’s director of product management for runtime solutions.
“We’ve productized it, but you also can create a reference design or lease a gateway that will be managed for updates and monitoring,” said Kurisu. “And with a customized interface, you can do a design in eight weeks. We’ve also added security for data at rest, work and in transit using ARM’s TrustZone.”
Mentor’s entry into the integrated device market with a standalone product is certainly a big shift, and it pits the company against potential competitors in different markets that it has never faced before. But what’s really worth noting is the business model that enables this shift, which includes variable volume, customizable hardware, software and security, and different options for buying, leasing or licensing. The strategy, according to Kurisu, is to be able to tap into a broad swath of markets, ranging from industrial to automotive to consumer.
In August, Rambus likewise expanded its business model to include hardware. The market focus for Rambus is different, because the opportunity is in DDR4 controllers primarily for server memory. But the two efforts share a common thread of finding new ways to leverage internal expertise across markets that weren’t obvious a year or two ago. Ely Tsern, vice president of the memory products group at Rambus, said it was “getting tougher to find expertise in interface design.” As with Mentor’s push into hardware, Rambus’ is highly scalable as the amount of data being processed in the Cloud continues to increase.
Chip manufacturers and packaging houses have been preparing for this shift for the past several years. While they are still focused on solving incredibly complex and very public problems at the leading edge of Moore’s Law, they have been actively developing alternatives such as FD-SOI, 2.5D, fan-outs and low-power versions of processes at older nodes. There is a general recognition that with the IoE, most of the semiconductors being manufactured won’t use the latest processes because that takes too long, costs too much, and isn’t easily customizable.
What’s critical in all of these implementations is time to market and unique implementations. What works in one factory or car or baby monitor may not work in another, but that doesn’t mean the whole chip has to be designed and built from scratch. This is the idea behind Marvell’s Modular Chip architecture, a 2.5D implementation that relies on an internally developed interconnect to combine a number of modules into an SoC.
“Because it’s modular, it can be separated, contained, interconnected at will, and manufactured in different processes,” said Mordi Blaunstein, senior director of product marketing at Marvell. “When you stack this on an interposer, the sky is the limit for what you can do. Over time, packaging, yield and known good die costs will go down.”
What’s becoming obvious in the IoE world is there are lots of blind alleys, bets that in theory should work but don’t, and less obvious approaches that are viable, at least for a time. This is evident with wearable electronics, where the number of companies vying for a piece of the market far outnumbers even the potential market.
The same has been true in security, even though it seems like a necessary and obvious market opportunity. The general consensus is that all chips will need some level of security. The question is how that security is implemented and just how profitable security becomes, namely whether it is an add-on, or whether it is a prerequisite for licensing that technology.
ARM, Synopsys, Imagination Technologies and Andes Technology all have been working on security for their respective cores, and Rambus has been developing IP through its Cryptography Research Division to thwart side-channel attacks. Despite some mammoth security breaches, though, this market has been slower to materialize than expected, particularly as an independent revenue source. That could change, particularly with over-the-air updates.
One thing that could change that significantly is the need for over-the-air updates.
“Suppose you develop the most secure light bulb on the planet,” said Rob Aitken, an ARM fellow. “Now also suppose that bulb has a seven-year life. In the year 2020, the hacking techniques are going to be a lot better than they are today. Do you have to update it? The answer is yes. That’s why security is becoming very real very quickly.”
Software and IP
Another way to tackle the IoT market is with software and IP, and a lot of the attention in this space is focused on the Cloud, particularly for Big Data mining and management, and for utilities that can improve the quality and efficiency of software as well as the efficiency of the hardware beneath it.
Yet there are less-obvious options that cross traditional boundaries, like everything in the IoE world. Type one hypervisors, which run on bare metal, are an example. Mentor has developed one for the industrial and automotive market for security reasons. The prpl foundation has been working with one from Imagination Technologies, as well, to improve security with stronger spatial isolation across both ARM and MIPS platforms.
The key is understanding where the money will be, where a company’s expertise is, and matching those. Cadence Fellow Chris Rowen said a lot of time was spent on exactly this kind of market dissection and number crunching before committing to embedded vision as a company direction. “We need to drive economics and technology,” he said.
Synopsys made a similar decision on software design, starting with the purchase of Coverity in 2014, and following that with more security-related software purchases. Coverity tested source code for defects, basically doing for software what linting did for hardware IP. In an interview last spring, Synopsys chairman and co-CEO Aart de Geus called the move very strategic because it reaches into embedded software as well as applications that span many different platforms and markets.
It’s hard to make generalizations about the IoE because it is so vast and contains so many different pieces and market segments. But there also are layers of commonality between the billions of parts that will make up the IoE, such as the communications infrastructure, determining where processing makes sense (local vs. edge node vs. cloud), the need for security, and low power everywhere.
The challenge will be tackling these markets with strategies that are new and unique, and which can leverage economies of scale in ways that never existed when markets were separate, engineering teams had 18 to 24 months to design SoCs and ASICs, and they could be assured that just moving to the next process node would guarantee improvements in power, performance and cost—as well as a ready market for their products. But for those companies that get it right, the upside can be as lucrative, far more creative, with more possibilities for growth than have existed in the semiconductor industry in a very long time.
But there is one final caveat. Getting it right will require taking on new risk, and the semiconductor industry has been traveling more or less in a straight line for the past few decades in an effort to minimize risk. “If you look at DRAM, SRAM and even embedded flash, those have not changed in the last 25 to 30 years,” said Charlie Cheng, CEO of Kilopass. “Even there differentiation is being blocked by a fear of risk.”