IoT Myth Busting

The (IoT) means many things to a large number of people, but one thing is clear—every discussion involving the IoT invariably includes some rather dramatic growth predictions for how many connected devices will be sold and who will be the primary beneficiaries.

While that data helps spice up speeches, and typically gets people to read and quote analyst and financial reports, some of the underlying assumptions are wrong. And as the IoT begins to take shape across a number of markets, more companies are beginning to figure out what’s real and what isn’t.

Joe Costello, CEO at Enlighted, talked about the early days of the IoT in a speech at the recent Design Automation Conference. “There has been a tremendous amount of hype about the Internet of Things, and most of it was BS,” he said.

Several fallacies persist:

• Edge devices will be so cost-sensitive that it will be difficult for anyone but systems companies to profit from their development;
• All chips used in the IoT, and particularly for edge nodes, will be similar and do roughly the same thing;
• Much of the profit will be derived from cloud and services providers, which will reap the lion’s share of applications and services revenue.

What’s becoming clear is that as more devices are connected across a variety of sometimes related, sometimes unrelated markets, none of these assumptions is completely correct—and often they are completely wrong.

Fig. 1: IoT Reference Model. Mac Devine, IBM – IoT Slam 2015

Standard platforms or custom silicon?
While there is an emphasis on cost-cutting for edge-node devices, this is true for any semiconductor-based device in any market. Should edge devices be built from standard platforms or custom silicon? Arguments can be made that both solutions reduce cost, but there is almost universal agreement on this—and history to guide us.

“Microcontroller companies have been threading this needle for years,” explained Drew Wingard, CTO at Sonics. “They work hard to reduce the cost of design so they can deliver a wide range of products. They are not always different chip designs. Sometimes it is the same chip in different packages or with features disabled. They work to ensure a high degree of compatibility across the product line. That has been shown to enable the usage of a common set of software so you don’t have to start from scratch when you pick up a different part. For IoT edge devices, we should be looking at how to enhance microcontroller design methods to attack the cost issues. And we do have a need to run networking software stacks, which does imply a larger amount of software than was traditionally provided by the MCU vendor.”

The mobile industry took a different path. “In the early days of the mobile phone industry, TI had OMAP, and they believed that if you used that then everything would be simple,” said Anush Mohandass, vice president of business development for NetSpeed Systems. “What happened was they did follow it for a while to get the initial set of phones to market. Then they realized that to capture any market share they need differentiation and they took that in different directions. Some did it in software and others in hardware, and the ones who were really successful did it both areas.”

Both solutions have a role to play. “There will be space for both platform and custom solutions within the IoT space,” said Oliver King, CTO for Moortec. “Certainly there will be applications that will require a focused optimized chip, while the barrier to entry with a platform will bring new players to the market.”

For many companies, it will be a process. “With the IoT gold rush, platform providers are selling shovels and hardware to all those innovators looking to mine this new market,” asserted Graham Bell, vice president of marketing for Uniquify. “Platforms will allow for proof-of-concept solutions that can attract further investment. The next generation of a company’s IoT solution will use a more customized approach to lower cost, improve performance, add unique features, and provide a higher barrier to competitors copying the product’s success.”

Cost has both an NRE and a production component, which can achieve economies of scale. The optimization of this balance is no different than for any other device. An advantage they do have is that volumes are likely to be large.

“They have to be inexpensive since they will be produced in large volumes,” said Jeff Miller, product marketing manager at Mentor, a Siemens Business. “Many of them need to operate for years at a time on a single battery, so they will have to be extremely low power. For many use cases, physical size will be a constraint. These systems will have widely varying levels of local data processing horsepower. All of these requirements are driving higher levels of integration and customization.”

Latest technology or older nodes?
The need for increased levels of integration has been the driver for adoption of new nodes, but IoT edge devices may never need that magnitude of integration. “Older processes are being used because they are cheaper,” said , general manager of the IP Division of Silvaco. “We see significant revamps of design kits for the older processes to incorporate improvements that have been developed for the new processes. This effectively gives a second life to these less-expensive processes. In fact, it’s been widely reported that there is a lack of capacity in these older fabs are a result of this movement.”

That requires updates to the available IP as well. “We have been working closely with foundries to provide IoT specific memory and logic libraries,” said Ron Lowman, strategic marketing manager for IoT at Synopsys. “They have to be concerned with multiple power islands, , and this takes more effort.”

Distributed vs. centralized intelligence
One area in which there is little agreement is the amount of intelligence that should be put into the edge nodes. “A recent study showed that a system design’s best power efficiency and performance comes from centralized processing and no processing at the edge, where each element has its own software stack,” said Uniquify’s Bell. “By eliminating software stacks at the edge, and pulling all the input directly to the central processor, a more efficient closed system is possible. This approach is more application-specific and therefore more customized, for greater NRE, but smaller and cheaper silicon. Those companies that have the resources to do this—a small number I think—will reap the benefits.”

Not all accept this. “Edge devices do not need to be dumb, and they are not just sensors,” said Lowman. “These things are often quite complex. Even low cost does not mean that processing does not happen locally, and there are some very efficient ways to do that. Getting the right data to the right place at the right time, that is what is critical for creating value.”

There are multiple reasons to put more intelligence into edge nodes. “Edge devices have to get smarter and while never as smart as the datacenter, they will get smarter in a number of ways,” said Mohandass. “For example, they will do local processing of information and sending compressed data to the core.”

It also depends upon the application. “The trend and key differentiation is to include some form of embedded intelligence in these devices,” adds Saraj Mudigonda, senior business development manager for Imagination Technologies. “There is also a combination of IoT edge sensor and cloud, especially with voice controlled devices where limited vocabulary and Natural Language Processing (NLP) is processed in edge device and uses the cloud for a more comprehensive dictionary. There is more scope for differentiation in a smart, high-performance processor for applications, such as connected network cameras with analytics such as face detection, person detection and people count.”

Put a different way, the companies that were out in front on the IoT hype are not necessarily going to win the market with words.

“Cisco and other companies that were initially promoting IoT fed the concept of simple edge nodes,” said Lowman. “They saw all data going back to the cloud, and that is not how the IoT provides enough value. They have been working in the cloud and they want to justify the fact that everything goes back to them. There is a level of magnitude difference in where they make money: in services. There is a cost of transmission and associated power consumption. If you do processing locally, and we have seen this is some new applications, in vision and voice devices, when done locally you can do some tremendous things.”

Total cost of ownership
Total cost of ownership may be a more appropriate metric for edge nodes, particularly those that are battery powered. Consider the cost to change a battery for each device in the field. It is likely that cost will swamp the initial cost of the silicon meaning that extra NRA used to lower total energy may quickly pay for itself.

“Battery and power consumption is a great example and then next is security,” said Lowman. “For security, the overall architecture for the end product far outweighs the cost of adding a little piece of security hardware where we may be talking fractions of a penny to the die.”

There is growing consensus around that. “Security is the heart of the new IoT platforms and part of their architecture,” adds Bell. “Rolling your own solution might be tempting, but getting it robust enough so that partners and users trust it will be a daunting task.”

But how do you get these devices down to the minimum power levels? “Power, by my observation, is a black art,” said Grant Pierce, Sonics’ CEO. “It is usually specialized engineers who don’t socialize with the guys who design the rest of the chip. They are called in late, not given a lot of time to do things that are optimal, and they are super pressured to get the chip to market. The CPU guys are getting good at it, but the rest of the market, and those likely to originate IoT devices, have no facilities in their design to address power. It will be expensive for people to create an expertise in order to attack the power problem.”

That indicates that IP and EDA have to change to target these new designers.

Business impact
IoT will breed a new kind of designer. “The tall thin engineer is not a specialist in one area but spans many as a generalist,” said Laurie Balch, a chief analyst with Gary Smith EDA. “This is being driven by IoT. We need to develop tools for their needs which are different from the mentality necessary for specialists. Verticals are providing more influence than in the past. IoT is having a market impact. We never looked at tailoring tools to verticals in the past and that is changing.”

, managing partner at Lanza techVentures, is in total agreement. “The design automation segment of the industry needs to be alert and in front of the pack. IoT design is unique because it is not driven by the next node, and we’ll need to understand the driving factors to be able to support the evolving needs of IoT. And now that we know the next segment of design challenges may not be just the increase in complexity of SoC, we need to become strategically alert. It may be the optimization of silicon-on-package or IP-on-substrate driven by power or the ability to make memories connect in a more effective way, or FPGAs become the foundation of inexpensive innovative ideas or…who knows what?”

IP needs to adapt as well. “Knowledge can be packaged in IP solutions that can put given to tall thin engineers to go build cost effective solutions,” said Pierce. “What is different for IoT guys is that the skillset they need in their tool bag will expand. The future of IP is bright because of this.”

But change is hard. “I don’t see a lot of evidence that the EDA market is adapting,” said Savage. “IP is usually paid for on project-based budgeting along with the design/verification efforts. EDA has avoided being part of that and is usually sold at the enterprise level, where design groups pay a corporate tax for those tools. If EDA were licensed like IP, as part of the project budget, there would be more incentives on the EDA side to do things that could reduce the overall schedule, which reduces the project cost. This would open a new selling proposition for EDA, but at the same time upset the apple cart that has been serving them for 30 years. You just have to remember that the objective of an EDA company is to design a product that engineers spend more time in, rather than less.”

There are some unique challenges inherent to creating these devices. “The high levels of integration require that the engineering teams be able to design, implement and verify systems consisting often of a combination of , analog, digital, and RF disciplines,” points out Mentor’s Miller. “EDA will have to adapt to this by providing tools to these new design teams that allow them to easily move between disciplines, to leverage advanced packaging and fabrication techniques, and to simulate complete systems at various levels of abstraction to capture interactions between electro-mechanical, photonic, analog, digital and RF subsystems.”

Conclusion
Cheap is a relative term that can be applied to anything. All devices strive to provide the cheapest implementation that does the job effectively and that has to take all of the requirements into consideration. Data has value, information has even more value, and being able to make that actionable can create a lot of money and that must be shared between all of the providers of the system.

At the same time, the industry can do a lot to help drive down the cost of creating and maintaining these devices. NRE, fabrication cost and total cost of ownership are important considerations and may require a different focus from the IP and EDA companies.

Related Stories
When Digital, Physical Worlds Merge
What the IoT will look like in the next five years, and what problems need to be solved to get there.
Custom Hardware Thriving
Predictions about software-driven design with commoditized IoT hardware were wrong.
IoT Designs Evolving
Unique approaches to IoT hardware begin to take shape; companies adapt their strategies.
IoT Startups Rake In Cash
Funding is free-flowing for the field, but hurdles persist.
New Memories And Architectures Ahead
So far there is not widespread adoption, but many see change as inevitable.