IoT Designs Evolving

Unique approaches to IoT hardware begin to take shape; companies adapt their strategies.


IoT hardware is beginning to take shape across a variety of vertical markets, and devices are looking far different from the initial concepts. They’re smarter, more targeted, and in most cases custom-built for specific applications.

The concept of connected things is hardly a new one. Students at Carnegie Mellon University added sensors into a vending machine in the early 1980s to remotely monitor the temperature of soda cans and send that information over Arpanet. That was followed by more sophisticated pricing of drinks, depending upon the outside temperature.

When Cisco started publicizing the vast potential for an Internet of Things several years ago, the idea of connected things had grown both in scope and market opportunity. Cisco’s view was a world filled with relatively simple sensors connected through gateways to the cloud, spawning thousands of entries for wearable computing at CES in 2015 that had little impact on the market because of a lack of must-have applications and poor battery life. Initial reactions ranged from ridicule to skepticism, based on projections of tens to hundreds of billions of devices.

Much has changed in the past year, though. Consolidation at the leading edge of design, fueled by cheap capital and flattening growth projections for mobile phones, has forced companies that initially scoffed at the IoT to look seriously at new markets opportunities. Even Intel, which bet a fortune on mobile phone technology before ultimately pulling the plug, is now pitching IoT as a primary growth engine, both for edge devices and cloud servers.

But unlike computers or phones, IoT devices are unique, and they’re becoming even more narrowly defined as end markets come into focus. That means many more highly specific designs that will need to be manufactured in much smaller batches.

“It’s a lot of custom engagements,” said Amy Wong director of product marketing for Marvell’s IoT Business Unit. “If you look at media over wireless, particularly voice-based systems, there is definitely going to be evolution there—how to tie it to cloud services and interesting new ways to uses those services. There will be different implementations. We’re also going to see a drive to make processing data more intelligent. So if you have media sensor data processing, you want to get the right data. Data is bits, and you need to find a way to use it and figure out what to use.”

Wong said that requires better matching of technology to the task, rather than limiting the task to the available technology. In the wearable market, Marvell has swapped from a four-core CPU, for example, to a two-core CPU based on TSMC’s new 28nm HPC process. The result is better graphics for less power. “We can run the chip at 0.65 volts with minimal leakage and isolate blocks into different power domains, keeping enough state retention to get back to that state on wakeup. It’s half the power of the predecessor design.”

System approaches
28nm is the bleeding edge of IoT designs. Even 40nm is considered advanced for many IoT applications. Whether these devices migrate to finFET-based processes anytime soon is unknown, but complexity and cost may make it unsuitable for many IoT markets, where time-to-market and low cost are critical factors.

Still, that doesn’t mean these devices are simple. Far from it, in fact, but the real complexity is less about the process and more about the systems these chips will be included in or connected to. Those systems are increasingly as unique as the market segments they serve.

“The most challenging technology issues are still in the mobile space at the leading process nodes,” said Qi Wang, vice president and chief of staff at Cadence. “But with IoT, so many things change. You’ve got packaging changes, customer demand changes, cost issues. You need to be careful about the package, board, and scheduling for integration of software and firmware.”

Wang noted that the communications infrastructure will need to change to 5G to really take full advantage of the IoT’s potential, but that will introduce a whole new set of rules and challenges. “Right now, communication is through Bluetooth or WiFi, and that will not fly if it’s truly an IoT network. You will need 5G for that. But what also will change is the software layer below, because software will be related to a particular market. Right now you have emulation for hardware/software system verification. But automotive, for example, might be different. You will need special technologies for reliability and safety, and they may not be fully addressed by emulation. If you talk to system integrators and OEMs, what they most care about is how to integrate with their software. We will need to invest in these areas.”

In fact, the whole ecosystem needs to approach this from a systems perspective rather than as discrete parts and steps. That includes everything from the package to security for each market slice, and this is where chipmakers are running into problems.

“If you look at microcontrollers, the mbed OS provides secure software for the IoT,” said Chrisopher Seidl, technical marketing manager at ARM. “But it’s not only the chip that needs to be secure. It’s also the partition between the system and the edge nodes. That communication needs to be secure, as well.”

There is no simple solution, either, because use models can vary significantly. In a smart phone, with billions of units, there is room for generalization about use models and enough field testing to eke out solutions to common problems. In IoT devices, that’s no longer possible. Markets are smaller, and problems may not show up for years—particularly if the same technology is used across different markets. For example, Seidl said that industrial machinery in Germany frequently utilizes a multi-touch graphical user interface. But that same technology also could find its way into automobiles, where there are different sets of requirements, different standards, and far different use cases.

“What used to be done in communications over a PCB using shared memory is now done on the same chip communicating directly over an internal bus structure,” he said. “So how good is the support for writing software? If you can’t program these devices, they’re useless.”

Finding new opportunities
That approach does open up a slew of new options for companies to get to market quickly, though, and it opens the door for tools companies that can find enough commonality in processes to begin automating them. But at the same time, everything has to be reconsidered in context, from tooling to methodologies to non-standard ways of putting things together. Even PCBs are undergoing change for IoT applications.

“Making PCBs is generally a subtractive process,” said Hemant Shah, product management group director at Cadence. “But a new methodology is to build them up. To make that work you need ways to reduce the number of layers, and to fit everything you have to reduce the feature sizes so you can add more features. This plays into the package side, where you have multiple dies on a package. You can stack them up and get to market first, then consider putting them on the same IC to reduce cost. That kind of approach also allows the company building a special-purpose chip to get chips from different sources.”

What’s happening in the EDA market is a reflection of these changes. Mentor Graphics, Synopsys and Cadence are all making significant changes in how they approach markets, from acquisitions to how they bundle together their capabilities. In the past year Mentor, for example, has begun selling secure IoT gateways and, most recently, equipment to test the reliability of chips in hybrid and electric vehicles, which may be one of the most advanced IoT devices on the market today.

“What we’ve done is straddle the mechanical and electrical world,” said John Parry, electronics industry manager for Mentor’s Mechanical Analysis Division. “In a hot climate, there is less thermal headroom to cool it. Under-the-hood electronics struggle to stay in the acceptable range. And with LED headlights, this is particularly important because you need a way to test for fogging. Unlike regular headlights, LEDs don’t put out enough heat to eliminate the fogging.

This is hardly business as usual in the EDA market, though. And as more markets begin to merge, others from all sides will likely begin dipping their toes into uncharted regions.

Related Stories
IC Industry Waking Up To Security
More companies recognize cybersecurity needs to be built-in from the beginning.
The Mightier Microcontroller
MCUs will play an increasingly critical role in cars and IoT devices, but they’re also becoming much more difficult to design.
Grappling With IoT Security
Updating connected devices creates a whole new challenge as threats continue to evolve.