How To Test IoT Devices

Existing approaches may work initially, but new technology and methods will be necessary as the IoT gains steam.

popularity

At a recent event, test experts said the IC industry needs a new paradigm in testing chips for the Internet of Things (IoT).

The message was fairly simple to interpret. Existing automatic test equipment (ATE) is well suited to test today’s digital, analog, and mixed-signal chips, though it may be ill-equipped or too expensive to test IoT-based devices.

But what isn’t so easy to grasp is also clear: What is the IoT in the first place? And what chips are involved in the IoT? In basic terms, the IoT involves the seamless communication between various devices in a network. An IoT system may include three basic components — a microcontroller, an RF chip, and a sensor/MEMS device.

The next questions are also apparent. So if the IoT takes off, can existing ATE handle and test IoT-based devices in a cost-effective manner? Or does it really require a new class of testers? As it turns out, there are several ways to test IoT-enabled devices. It may require a combination of both traditional, and non-conventional, hardware approaches. And as with today’s chips, the goal with IoT devices is to keep test times and test costs in check.

“Devices driving the volume in the IoT, such as wearables, smart mobile devices and home automation, all have two things in common — wireless and sensors. These IoT chip characteristics are creating a serious challenge for the conventional ATE paradigm,” said Luke Schreier, senior manager of automated test marketing at National Instruments. “The chips that power these devices are usually low power with smaller digital pin counts, often with serial protocols, which can challenge the cost structure of traditional ATE. And then, to test the wireless and sensors, you need expensive RF and analog upgrade options as well as elaborate handling techniques.”

What is IoT?
To be sure, the IoT is still a nebulous topic. And it’s still unclear how all the pieces of technology will work in the real world. “The IoT is the network of dedicated physical objects that contain embedded technology to sense or interact with their internal state or external environment,” said Dean Freeman, an analyst at Gartner. “The IoT comprises an ecosystem that includes things, communication, applications and data analysis. Automotive is a big component of IoT, home will play a large role, as will office buildings.”

In automotive, for example, IoT could play a role in the predictive maintenance area, which involves the use of sensors throughout the engine, according to Gartner. LED lighting is another IoT application, in which “smart LEDs” are connected to a network and can sense the environment, according to the firm. And in the home, smart TVs and set-top boxes could be classified as IoT systems.

In total, Gartner forecasts that 4.9 billion connected things will be in use in 2015, up 30% from 2014, and will reach 25 billion by 2020. The processing, sensing and communications semiconductor device portion of the IoT market is projected to grow 36.2% in 2015, compared with the overall IC market growth of 5.7%, according to Gartner.

Testing IoT devices
To test IoT chips and sensors, there are several approaches. But first, chipmakers must get their arms around a simple question—What type of chips need to be tested? “For IoT, the primary piece is the attachment of an RF device, or RF capability, into a microcontroller,” said Jason Zee, manager of the consumer business unit at Teradyne.

The microcontroller could range from a 4- to 32-bit device. Each microcontroller is integrated with three main components—a digital core, analog-to-digital convertors and flash memory. “So with the microcontrollers, you start attaching RF protocols into them. That could be 802.11 or ZigBee. You could also have a low energy version of Bluetooth,” Zee said.

All told, the MCU and RF are still discrete devices. In the test flow, the chips go through the usual wafer sort step. In some cases, the separate MCU and RF chip may require two separate test platforms and programs. One platform tests the MCU, while the other handles the RF. “Starting at probe, you do two insertions. At that particular point, it’s not an integrated single die,” he said.

Then, many chipmakers are integrating the MCU and RF die into a system-in-package (SiP). This SiP undergoes a final test step. “Historically, you had to do two touchdowns on the final test, probably on two different platforms,” he said.

The problem is fairly apparent. The test flow requires too many platforms and insertions, thereby driving up test times and costs. To help solve the problem, Teradyne will offer an RF instrument option to its J750, a low-cost tester. “Now, you have a single flow on a single platform,” Zee said. “You can test both insertions (in wafer sort) on the J750. And you basically have one touchdown at final test. So, you don’t have to maintain multiple platforms and test programs.”

Test times and test costs
“At final test, the RF portion of test probably adds 5% to 10%,” he said. “But test costs actually go down. You are removing one insertion at final test. You can also combine a single touchdown with a higher site count.”

Another ATE vendor, Advantest, is moving in a similar direction with its test platforms. But in IoT applications, chipmakers will not only test the MCU and RF chip, but they must also deal with a separate sensor/MEMS component. At times, the three devices may reside in the same SiP, which presents more challenges on the test floor. “As an example, an inertial sensor integrated together with the MCU and Bluetooth may go through three insertions with different ATE configurations,” said Dave Armstrong, director of business development at Advantest. “As the volumes increase, the multiple insertion approach crashes against the quality and cost constraints.”

Testing MCUs and RF chips are well understood and conducted with traditional ATE. But testing a sensor/MEMS device requires a different strategy. “What’s typically needed to test these parts is ATE plus some sensory stimulus in a calibrated setting,” Armstrong said. “The sensory stimulus, of course, depends on what type of sensor is being tested and how the test is being done.”

There are several ways to test a MEMS device. Both Advantest and Teradyne can test MEMS-based devices using existing ATE. The ATE may also require specialized circuitry on the device-under-test (DUT) board. It also requires another major element. “In semiconductor devices, you think of digital or analog,” Teradyne’s Zee said. “When it comes to MEMS, you inject a third dimension. It could be vibration. It could be humidity.”

For example, to test an accelerometer, the part must not only undergo an electrical test, but also a vibration test. In this case, an OEM or device maker would develop a customized test cell, which would literally make the tester vibrate. “To test a MEMS part electrically, there is nothing complicated about it. What’s complicated is the stress or stimulus you put into it,” Zee said.

Screen Shot 2014-12-10 at 10.18.31 AM

So today, it appears that existing ATE can test MCUs, RF and MEMS. But over time, the ATE industry may need to devise a new class of low-cost, highly parallel testers to address the IoT. In concept, this could involve a “universal pin” architecture, which would support mixed-signal, RF and digital. “In order to support this, creative solutions will be needed,” Advantest’s Armstrong said. “A preferable approach would be a ‘Swiss Army knife’ approach, using the similar type of ATE resource for all the device pins.”

Another viable way to test MEMS/sensors is to use specialized, MEMS-based handlers/test systems, which are different than traditional ATE and IC handlers. A conventional IC handler automatically picks up a chip and puts it in a test socket in the tester. Then, the part either passes or fails in the ATE.

In general, MEMS-based handlers/test systems provide thermal conditioning, mechanical stimulus and electrical signal path testing for accelerometers, gyroscopes and other MEMS devices. “The MEMS test business is really a handling and stimulus business,” said Richard Chrusciel, vice president at FocusTest, a company that specializes in the design of MEMS-based test systems. “That complicates the living daylights out of this whole thing.”

In general, a MEMS-based test system may include a socket, which could provide some form of an electrical test. “It also has to provide some form of mechanical stimulus. For example, in the case of an accelerometer or gyroscope, I flip the device or rotate the device to all of the axis. If it’s a high-g device, it’s typically done with vibration stimulus,” Chrusciel said. “That socket also needs to maintain the electrical contact. And this whole thing has to be under temperature. That’s what makes this real difficult.”

The challenges are widely known in the industry. “The test times are very long,” he said. “I could spend 10 to 12 seconds mechanically positioning the device in order to electrically test a parameter.”

Cost is also an issue. In 2010, an ASIC and accelerometer, which were housed in the same package, sold for about $2. A staggering 30% to 40% of the $2 figure was spent to test that MEMS device. “Now, in the last three or four years, the ASPs of most of the MEMS-based commercial sensors have dropped by 50% to 60%,” he said. “That means the device I sold for $2 is now $1 or less. My volumes have gone up by at least twofold or more, but my test cost has stayed the same. That’s where this industry was, and still is, in a lot of cases. The test cost is killing these guys.”

What’s the solution? The answer: Testing MEMS in parallel. “You need to test tons of them in parallel,” he said. “You have to test a bunch of these guys or you can’t make money.”

In fact, the MEMS test/handler players, including Multitest, Spea and Tesec, have separately rolled out systems with highly parallel test features. (FocusTest designed a MEMS tester for Tesec. Tesec makes and sells the system in the market.)

What’s next?
Clearly, the MEMS test/handler players are moving towards traditional ATE. And the traditional ATE players are attempting to address MEMS. In fact, Xcerra, formerly called LTX-Credence, acquired Multitest in late 2013, propelling the ATE company into the MEMS test business. And if the IoT does indeed take off, Advantest and Teradyne may need to respond, and perhaps consider a similar route.