How Testing MEMS, Sensors Is Different

These devices require more than an electrical input and output.


When it comes to testing microelectromechanical system devices and sensors, sometimes you have to shake and bake.

MEMS and sensors are physically different from standard ICs. They require a specific type of stimulus to get the required testing results. Most chips only need to have an electrical charge run through them to gauge their pass/fail status – electrical input, electrical output — so it’s all digital, all the time.

MEMS and sensors, however, are more on the analog side of the fence. Most vendors of automatic test equipment are easily disposed to testing of digital or mixed-signal chips. When it comes to MEMS devices or sensors, specialized handling equipment is often required.

Temperature sensors, for example, may be subject to heating in testing. MEMS gyroscopes must be spun around in a gimbal controlled by stepper motors in a special module.

“Testing MEMS and sensors is significantly more complex than testing ICs,” says Frank Shemansky, chief technology officer of the MEMS & Sensors Industry Group. “Unlike ICs, MEMS and sensors transduce a physical input to an electrical output. Thus, they require a physical input for testing, such as motion, sound, or chemical, which complicates testing. Because each sensor type requires different inputs, standardization is very difficult. In addition, the sensors often have to be in a system to function fully, which adds yet another layer of intricacy.”

Advanced packaging technology is a given for many MEMS and sensor chips. “We typically see the same challenges in testing as with conventionally packaged MEMS and sensors with the additional complexity of essentially testing a subsystem when employing 2.5 and 3D packaging techniques,” Shemansky says. “Assuring known good die or individual components meeting required specifications in a complex package is particularly challenging and impacts yield.”

The challenge is that what starts out as a known good die may not end up as a known good die, and being able to detect that isn’t so easy.

“A lot of these devices are very sensitive, so how you test often depends on the application,” explains Joey Tun, principal market development manager at National Instruments. “With automotive, this is even harder because they do see fairly extreme temperatures, so testing needs to be more intensive.”

Testing needs to be done before MEMS devices are assembled, and they have to be tested after assembly. “If you’ve got an SiP module with a MEMS chip, you have to worry about signal conditioning and local computing, which is basically noise,” Tun says. “The addition of big data capabilities offers some interesting insights here. There are companies working on the ability to trace problems down to the exact wafer and where the problems came from. If you can pinpoint what went wrong at what point, that’s very good.”

What’s different?
One big difference with the MEMS and sensors testing, compared with other chips, is there is a stimulus in addition to the electrical input and output.

“A stimulus can be any type of physical, inertial, pressure, magnetic, optic, sound, gas, light, whatever — any stimulus that needs to be measured is applied to a device and that is also measured,” says Ram Praturu, director of test product technology marketing at STATS ChipPAC. “From a tester component standpoint, for a standard test, you have a tester and you have a handler. Whereas, for MEMS test, you have the same tester, you have the handler, but also you have a stimulus component, a module, that you need to apply this external stimulus to the device.”

So for a MEMS microphone, the stimulus is sound. “You apply sound to it, and apply it in such a way that it receives it, so you need to be physically able to apply it. You cannot use the same type of handler that’s available in the market because a standard IC test handler may not have the option to apply the external stimulus. Depending on the device and the package, you may have to get a different handler. The same thing is true if it needs to be probed. Again, a stimulus needs to be applied on whatever test is required at the probe stage.”

He notes that what typically happens in the MEMS and sensors market is a control IC is integrated into a chip, or an ASIC is added that controls the MEMS device, receives the data from the sensor, analyzes it, and passes it along in digital format. “They used to put it on a side by side, like a module type, but now they’re integrating it into multiple die—one atop of another, like flip chip. There are through-silicon vias, which are being used in the industry. For accelerometers, you can put a controller and the sensor itself, and then you can put a TSV to connect between these two devices or to the substrate. You can call that a 3D-IC, because of the TSV that is being used.”

For automotive applications, the test cost for MEMS and sensors is a little higher than a standard IC, according to Praturu. “The key thing is for MEMS and sensors, all of our testing, giving a stimulus into the device, and know-how to test it, whether it be a shake, rattle, and roll, or put a pressure in, is a little bit on the analog side. So it’s a little more complex than a standard digital and mixed-signal testing. It’s an art, at the end of the day.”

Chips that move
Gerard John, senior director of advanced test at Amkor Technology, characterizes a MEMS device as “something that moves.” That’s one way they differ from standard ICs.

“When it comes to testing MEMS, what’s different compared with standard product in a sense is that you have a transducer that you have to activate,” John said. “The transducer takes the analog signal, converts it to digital, and then you have an A-to-D converter. You have an amplifier. That part of it is pretty much standard to regular electronic devices. The transducer is the big change that goes from MEMS to other devices. In order to activate the transducer, you need to have a stimulus. The stimulus box makes MEMS testing different from other electronic testing. The stimulus could be an audio signal, it could a motion, like you have for the IMUs, it could be something that spins in a certain direction, it could be a gas, it could be some pressure, it could be temperature, it could be humidity. You can have different kinds of stimulus. Within the MEMS test process, you have these different kinds of devices that you have to test, and you have to have different stimuluses for each of them.”

Adds Adrian Arcedera, Amkor’s vice president of MEMS and sensors: “The easiest way to go look at it is you have to move the sensor. There is that third element – of not just an electrical input, but actually a mechanical input, whether it be extending from motion to light, to everything, depending on what type of sensor you’re testing. There’s that element of the environment coming into the sensor.”

MEMS sensors, like most sensors, are an interface into the physical/analog world. On the input side, that can be analog or digital. The input is run through a transducer to convert into an electrical signal, which can then be processed.

The inertial measurement units going into smartphones are typically all in one package, combining several sensors, according to Arcedera. With MEMS packaging, “you’re always combining a MEMS and an ASIC,” he says. “Every device needs to be tested. That’s the key, especially for MEMS and sensor testing. The calibration happens during testing. That’s another unique thing about MEMS and sensor testing.”

The process is similar to binning and trimming in semiconductor testing.

Evolution of MEMS test
Rob O’Reilly, senior member of the technical staff at Analog Devices, has a long-term perspective on MEMS, having worked on the devices at ADI since 1993. O’Reilly notes that MEMS and sensor testing is not an in-socket test, as in standard semiconductor testing. He was involved in developing a shaker stimulus chip test handler back 24 years ago, as well as in testing a dual-axis accelerometer, which was “orders of magnitude more difficult than doing a standard test configuration,” he recalls. “One man’s shaking was another man’s noise,” O’Reilly says.

Back in the 1990s, ADI’s MEMS were all going into automotive applications. Technology eventually moved on to tilt sensors for mobile devices. MEMS microphones are “their own animal,” he comments. “The more test you can design out at the device level, you’ll win.”

ATE costs have risen over time, making the goal of “no test” more appealing to MEMS manufacturers, he adds. Reducing test times has been a constant objective.

Qualcomm and Intel have been leaders on setting sensor standards, along with the MEMS & Sensors Industry Group, with the work being passed up to the IEEE, O’Reilly says. The aim is make testing less expensive, but not cheap. Having standardized packaging, where possible, is a big benefit.

“We work really rapidly at the beginning of the MEMS process to use standard package technology where we can,” says O’Reilly. “In the consumer realm, we’re in all standard configurations. At some test houses, they don’t know that it’s a MEMS device. In the automotive world, we’ve done the same thing. Maybe we’re in older packages, the usual surface-mount components, generally with leads for all that automotive stuff, but a lot of it’s lead-less now and looks just like anything else. From a handling perspective, in terms of the technologies we used to move the assembly and test, there’s a lot more standard packaging coming through now than there was. What you had was a proliferation of handling equipment. If you went to a standard subcon, and they were testing two or maybe three vendors’ MEMS devices, they could have 15 different platforms out there. Because the accelerometers were in different packages than the gyros, which were in different packages than the IMUs, they needed technology for all of those.”

Much has changed since then. There is more merging of technology being driven by the OEMs.

“They’re starting to dictate – obviously, smaller is better – package widths and heights,” he says. “The semiconductor industry has been working toward trying to get down to sub-1mm, and there’s also a push obviously in wafer-scale, where there is no package. There is a bigger challenge there. We ship a ton of wafer-scale stuff today. That solves a lot of packaging issues. It also pushes a lot of the testing back on the probe systems.”

Related Stories
MEMS: Improving Cost And Yield
Second in a series: New packaging options could help boost profitability, but testing and thermal issues remain problematic.
Rethinking The Sensor
As data gathering becomes more pervasive, what else can be done with this technology?
Flexible Sensors Begin Ramping
Technology opens up new possibilities for the Internet of Everything.
The Trouble With MEMS
Severe price erosion is putting this whole sector under pressure at a time when demand is growing.

  • Great article and lots of things to think about. As for us, with controlled microline diamond indenter, we were able to de-lid / de-cap the MEMS package with no damage (no mechanical impact) and at room temperature for Failure Analysis purposes

  • fouffylafouffe

    Hi Jeff,
    Very interesting article.
    A great addition would have been to get the inputs from the people who make the handlers, stimuli or complete test cells for MEMS. There are only very few on the market, like Xcerra…