Paul van Gerven
28 April

Currently, MEMS testing basically comes in two flavors: highly automated but tailored to a very specific device that’s manufactured in high volumes, and, essentially, manual inspection for everything else. A consortium of ‘neighboring’ companies and research institutes aims to bridge that gap by developing an all-electric, universal MEMS testing solution suitable for all volumes.

Imagine having to discard a product at the final stages of its production process because it turns out a chip is malfunctioning. This doesn’t happen very often in the electronics industry because chips are rigorously tested before they’re mounted. But not all companies, particularly manufacturers of specialized MEMS chips, have the luxury of low-cost and highly automated testing procedures.

Because of their mechanical nature, MEMS are more complex to test than electrical ICs. Large MEMS manufacturers like Analog Devices and Bosch can afford to develop highly specialized testing equipment, but for companies with lower volumes, sometimes as low as a few thousand devices per year, that’s never going to be cost-effective. Nor are there any specific ‘MEMS testing houses’ to outsource to. These companies have to resort to labor-intensive ‘manual’ testing, which often isn’t enough to prevent that a percentage of the final assembled product won’t work.

Three companies and two research institutes, all located in the east of the Netherlands, have partnered up to solve this problem. Co-financed by the European investment fund OP-Oost, the Meteoriet consortium is looking to develop a universal MEMS testing technology. Make no mistake, though: large manufacturers stand to benefit, too, as their testing time per unit can be reduced. All in all, the new testing methods will remove a major bottleneck for the adoption and application of MEMS technology.

Great value

Most chips only need to have electrical signals run through them to find out whether they work as intended. MEMS, on the other hand, require another type of physical input for testing, such as a temperature increase, a movement or a changing magnetic field. A MEMS airbag sensor chip, for example, is tested by subjecting it to a massive deceleration. And a MEMS microphone device is exposed to a range of sound frequencies. Each type of device requiring a different physical stimulus makes MEMS testing hard to standardize.

But perhaps it doesn’t need to be. The Meteoriet partnership is convinced that it’s possible to test MEMS chips the same way regular ICs are tested: by electrical means only. “The movement in a MEMS chip is associated with electrical signals, which can be measured and used to obtain information about the displacement that has occurred. It’s also possible to induce the desired movement electrically. Taking together, you can test a MEMS chip electrically,” explains Paul van Ulsen, CEO of test technology company Salland Engineering, which spearheads Meteoriet.

An all-electrical MEMS test could look like this: an electrical stimulus is applied to induce a displacement of a particular mechanical part of the chip. Next, an appropriate electrical measurement is performed to verify whether the displacement did indeed take place. Other functionalities, such as temperature sensors or heaters, can be tested in a similar fashion.

The advantages of such a relatively simple procedure are myriad. It can be performed on the wafer, allowing for the elimination of malfunctioning devices at the early stages of the production process, reducing waste and saving costs. Wafer-level testing intrinsically also speeds up testing because many chips can be tested at once, and can also be automated quite easily.

Salland Engineering
A Salland Enginneering 8-channel PXI instrument to measure extremely low capacitance, which is required for electrical MEMS testing. Credit: Salland Engineering

“Electrical testing of standard electrical parameters is already part of the MEMS testing toolbox, but a generic, universally applicable test solution for the physical parameters doesn’t yet exist,” says Van Ulsen. Clearly, it would be very lucrative for his company if the Meteoriet partners succeed in developing one, especially given the rosy growth prospects of the MEMS market. “MEMS devices are often sensors, and sensors are an integral part of the IoT and Industry 4.0. Thus, speeding up and simplifying MEMS testing would help drive maturation of these technologies by reducing production cost.”

For Salland, the deliverable of the Meteoriet project is a testing solution, along with the associated IP. The Zwolle-based company will deploy that in a variety of ways. “We’ll sell stand-alone test instruments, but we’ll also offer add-ons for existing equipment, so they can be made suitable for MEMS testing. Lots of existing IC testing equipment can be upgraded to be able to handle MEMS testing, which we believe is a great value proposition. On top of that, we can start offering test services for small and medium-sized companies that don’t have the critical mass to do testing in-house.”

Machine learning

One of the project’s main challenges is to figure out which stimuli and measurements are effective for testing different MEMS chips and their components. This is why the University of Twente (UT) and the Saxion University of Applied Sciences are involved in Meteoriet. These institutes are tasked with developing models that relate the non-electric physical quantities (movement, temperature, gas flow, and so on) to electrical characteristics to test the MEMS.

“Our models are mathematical representations of devices that, given the input, reveal what the measured output says about the physical quantity that sits in between. From these models, Salland can derive the requirements for its testing instruments,” explains Dennis Alveringh, assistant professor at UT’s Integrated Devices and Systems group and part-time research scientist at Salland Engineering.

UT MEMS chip
An early version of the MEMS chip Meteoriet will be using for calibration. Credit: University of Twente

“The model of a die can be viewed as a black box that takes the mechanics or other quantities like temperature changes out of the equations. The input and output are electrical, yet they yield information about something non-electrical happening inside the device,” adds Aleksandar Andreski, associate professor in nanophysics at Saxion.

The goal of the project, however, isn’t just to find out whether a MEMS chip (or a MEMS component) works or doesn’t work. Alveringh: “We also want to know how well it works. Because of random variations in the production process, some chips may have an above-average sensitivity in a certain measuring range, for example. Others may have an excellent response time. We can find out which test results are associated with such characteristics. This will allow manufacturers to select chips and match them with different product lines that are based on the same chip. This so-called product binning is a common practice in the semiconductor industry. And, finally, we might even be able to predict what the right calibration will be once the final product has been assembled.”

In addition, Saxion will be validating the results. “At the end of the day, Salland needs to prove its claims about purely electrical MEMS testing to customers. At Saxion, we’ll collect the extensive data set to do so. Of course, as research institutes, we want to publish our research results, for which proof is required,” says Andreski.

Finally, the consortium wants to know more about the nature of the defects and other anomalies encountered, and what effect these have on the test results. This is where a fourth consortium partner comes in, Enschede-based Maser Engineering, a failure analysis specialist (among other things). “There will be defects that won’t show up with electrical testing, and we’d like to know what type of defects these are. Maser has the equipment and expertise to paint a detailed picture. We’re also working on complementary automated testing methods to identify and characterize these defects. We’re developing machine learning software that can do just that from pictures of the wafer. Optical inspection already is more or less standard in MEMS testing, and we can easily upgrade Salland’s testing equipment to handle both electrical and optical testing – adding even more value,” Andreski continues.

Dear to my heart

Now, there’s just one piece of the puzzle missing: MEMS chips to test. This need will be partially satisfied by UT, which will design a state-of-the-art MEMS chip, taking into account requirements for testability. In parallel, a company itching for more effective testing procedures will do the same: mass flow sensor manufacturer Bronkhorst High-Tech from Ruurlo.

“We have MEMS-based products on the market for about 15 years now. Currently, about 5-10 percent of our offerings are MEMS-based. Recently, we’ve identified some very interesting new opportunities in the gas flow sensor markets that we’d like to capitalize on. However, scaling up our MEMS activities will require faster and more effective testing methods,” says Joost Lötters, science officer at Bronkhorst and professor in microfluidic handling systems at UT.

Bronkhorst IQFlow
The IQ+Flow is one of Bronkhorst’s product series that’s already based on MEMS technology. Credit: Bronkhorst High-Tech

Bronkhorst manufactures its own MEMS chip in the cleanroom of the MESA+ Institute for Nanotechnology, a shared production facility for micro and nanotechnology located on the UT campus. After production, these chips are subjected to a number of manually performed tests, among which an optical inspection and an electrical test of certain components, but that’s not enough to ensure the product works as intended. During final testing and calibration of the assembled module, some chips turn out to be malfunctioning after all.

At Bronkhorst’s current needs of several thousands of MEMS devices per year, this situation is manageable. Expecting volumes to grow to tens of thousands of chips per year or more, however, the company’s labor-intensive testing procedures clearly won’t cut it anymore. “These volumes are still not high enough for foundries. We really need access to comprehensive, automated testing,” states Lötters. That’s why his company gladly participates in Meteoriet to help Salland develop the testing solutions. Evidently, the chip that’s being designed as a part of the project will be used in next-generation mass flow meters.

And so everything neatly falls into place within Meteoriet. Amazingly, all partners are located no more than 50 kilometers from one another, as the crow flies. “There’s a lot of knowledge and expertise in this part of the country, but we find it difficult to come together. I get it, there are always plenty of other priorities, right? Still, I think that cooperation between Dutch tech companies is very important to be able to compete in the global marketplace,” says Van Ulsen.

“The Meteoriet project is particularly dear to my heart because it involves automation. Thanks to automation, costs are lowered, meaning manufacturing operations – and the associated jobs – that have been outsourced to Asia or Eastern Europe could return to the Netherlands,” Van Ulsen concludes. “That’s why I hope Meteoriet will inspire companies across the country to put in the effort and start working together more often.”