We’re talking about CMOS-based MEMS sensors that are 1,000 times smaller than their traditional silicon-based MEMS counterparts, offering greater bandwidth and higher sensitivity, while also being more cost-effective and reliable. Seriously—what’s not to love?

I’ve just been chatting with Dr. Josep Montanyà, who, along with Dr. Marc Llamas, founded Nanusens. Headquartered in Edinburgh, UK, with a Research and Development office near Barcelona, Spain, Nanusens has been focused on finding a solution to address the limitations associated with traditional MEMS technologies. Spoiler: they’ve succeeded!

Just to set the scene, let’s kick off by reminding ourselves that from the 1950s through the 1990s, accelerometers were anything but tiny, cheap, or ubiquitous. These were not the casual, cheap-and-cheerful chips we sprinkle into consumer gadgets today; they were serious pieces of hardware, reserved for applications such as missiles, aircraft, rocket engines, and seismic labs—places where budgets were substantial, stakes were high, and engineers had forearms like blacksmiths.

In those days of yore, if you wanted to measure acceleration, you reached for a device that was essentially a miniature scientific instrument, not a component. A “navigation-grade” unit might include a pendulum suspended on jewel-like flexures, optical or capacitive pickoffs to sense its tiniest twitch, and a servo loop that used electromagnetic coils to yank the mass back to center. The entire arrangement resided inside a hermetically sealed metal can, approximately the size and weight of a small brick, supported by an entourage of analog electronics that generated more heat than a 1960s toaster.

Were they accurate? Absolutely… after a day of warm-up and a team of technicians had lavished them with esoteric calibration ceremonies. But they were also temperamental, expensive, and decidedly high-maintenance. Their outputs would fluctuate if the temperature changed or if an operator sneezed nearby. And heaven help you if you dropped one (suffice it to say that you wouldn’t expect to receive a Christmas bonus in the foreseeable future).

Even the simpler piezoelectric models used for vibration testing weren’t exactly pocket-friendly. They lived in chunky metal packages, required specialized amplifiers with input impedances approaching infinity, and only measured AC acceleration anyway—DC or slow drift was right out. So, OK for rocket engines, not so great for your pocket calculator.

Fast-forward to today, and the contrast borders on the absurd. The MEMS accelerometer in your phone—made from delicately etched silicon combs thinner than a human hair—does everything those hulking 20th-century beasts did, only faster, better, cheaper, smaller, and infinitely more reliable. What once filled a shoebox now fits into a space smaller than a breadcrumb and sips microwatts of power. Your smartwatch contains more inertial sensing capability than an entire Apollo-era inertial navigation platform, and it doesn’t even break a sweat.

Now it’s time to strap yourself in and hold onto your hat because—as I essentially said while introducing this column—the folks at Nanusens have come up with a way to fabricate CMOS-based MEMS sensors. These bodaciously small beauties are roughly three orders of magnitude smaller than their traditional silicon-based counterparts, delivering performance that borders on the ridiculous—higher sensitivity, broader bandwidth, lower power consumption, and improved reliability.

I don’t know about you, but I always enjoy learning about the origin story behind a company. In this case, it’s a doozy. Josep’s background is in telecommunications engineering. After completing his degree, he spent some time working in industry before returning to the university in 2000 to pursue his PhD. It was here that he was first introduced to the concept of MEMS technology.

While at university, one of Josep’s professors set a challenge for the students. He said that if someone could make an RF switch that actuated at low voltage, that would be something the industry would be very interested in. (FYI, making low-voltage RF switches is notoriously difficult and surprisingly hard to do.)

Josep accepted the challenge and focused his PhD on designing a low-voltage actuated RF switch. He managed to get some funding from the “Three F’s” (friends, family, and fools) and, after two years, he had a prototype working.

Then, with the help of external advisors, Josep traveled to the US, where he visited several semiconductor companies to showcase his work. As he told me: “It was at this point I discovered that my professor knew a lot about technology but had no idea about the semiconductor market because nobody was looking for the product I’d designed.”

I’m afraid I couldn’t help but laugh. Josep continued to say, “All was not lost because I discovered that what the industry really wanted was sensor integration with CMOS.”

There is much more to the story; suffice it to say that it culminated with the founding of Nanusens in 2014. Now, before we dive into Nanusens’s new approach, let’s quickly remind ourselves how traditional MEMS sensors are implemented, using an accelerometer as an example.

At a conceptual level, your bog-standard MEMS accelerometer is a remarkably simple contraption. Imagine a teensy cantilever—essentially a microscopic diving board—anchored at one end and free to bend at the other. Under acceleration, that free end deflects ever so slightly. That deflection changes the gap between the end of the cantilever and a fixed electrode beneath it. And since that gap forms part of a capacitor, the accelerometer measures acceleration by tracking the corresponding change in capacitance.

All this is easy to say but devilishly hard to do. These capacitance changes live in the attofarad realm (that’s 10^-18 farads). Even the slightest noise, drift, parasitic effect, or mischievous thought can drown out the signal.

Physically, a typical commercial MEMS accelerometer might measure approximately 740 μm x 510 μm, which is close enough for me to round up to one square millimeter without losing sleep. Inside that tiny footprint, the gap between the moving and fixed electrodes is usually around one micron. That gap defines a fundamental trade-off between bandwidth and sensitivity. You can have sensitivity or bandwidth, but not both, which is why traditional MEMS accelerometers always involve a compromise.

This is where things get a bit messy. Unlike solid-state integrated circuits—the vast majority of which share the same basic process called CMOS—MEMS has no manufacturing standard. None. Zero. Zip. Every MEMS manufacturer has its own proprietary process flow. Worse still, each MEMS manufacturer often has different processes for each of its products. An accelerometer might use one sacrificial-layer approach, a gyroscope another, and a pressure sensor may employ a completely different technique. This results in no standardization, no economies of scale, huge development costs, long lead times, limited flexibility… and that’s the good news. In other words, it works, but it’s clunky.

Additionally, although MEMS structures are typically made from silicon, most companies build two separate dies—one for the mechanical sensor and one for the electronics—and package them together. That creates parasitic picofarad-level capacitances on the signals between the two dies. One picofarad might not sound like much, but when you’re trying to detect attofarad-level changes, it’s like trying to hear a pin drop inside a jet engine.

This mishmash of processes might have been tolerable when MEMS were niche devices, but it is no longer so today, when billions of sensors are deployed every year into virtually every product category imaginable. This is where Nanusens is set to upend the entire MEMS industry. Instead of designing a sensor and then inventing a specialized MEMS process to manufacture it, Nanusens starts with the most mundane, predictable, and high-volume semiconductor process on Earth: standard CMOS. Their trick is breathtakingly elegant:

• Run a normal CMOS process. No custom materials. No special layers. No weird doping. Just standard wafers flowing through TSMC, GlobalFoundries, etc.

• At the very end, apply a single post-processing step using vapor-phase hydrofluoric acid (HF). This 30-minute etch selectively removes some of the silicon dioxide in the metal interconnect stack. This creates empty cavities, freeing the top-layer metal features to move like tiny cantilevers. In other words, MEMS structures are carved out of the same metal layers that are used for wiring.

This sounds simple—and conceptually it is—but pulling it off required over a decade of R&D and more than 50 patents. Those metal layers were never intended to act as mechanical devices, so Nanusens had to invent entirely new design methodologies to overcome thin-film stress, layer rigidity, geometric limits, and all sorts of devilish physical quirks.

Traditional silicon-based MEMS sensor (left, Source: Yole) vs Nanusens equivalent (right, Source: Nanusens)

One way to look at this is that traditional MEMS companies design the sensor first and then conjure up a bespoke process to manufacture it. Nanusens has flipped that model on its head: they’ve defined a single CMOS-friendly process first, leaving them free to develop an entire zoo of sensors that can be created using that same flow.

Happily, the payoff is proving to be worth the effort. Since the structures are defined in the metal stack, the gaps between capacitor plates can be hundreds of nanometers, even tens of nanometers, depending on the CMOS node. This means:

• Huge sensitivity
• Huge bandwidth
• No need for the traditional trade-off
• Higher yield
• Higher reliability
• Dramatically lower parasitics

And the electronics live on the same die as the sensor, thereby eliminating any awful picofarad-scale parasitic paths. Instead of fighting one pF of unwanted capacitance, the MEMS connects directly to the ASIC’s input, whose capacitance might be as little as 15 femtofarads. Going from attofarad variations riding on 1pF of parasitic capacitance to attofarads riding on just 15fF isn’t an improvement—it’s a revolution.

Even better, since every sensor is built using the same standard CMOS process, you can mix-and-match as many sensor types as you like—accelerometers, gyros, magnetometers, pressure sensors, tunable capacitors, whatever—and fabricate them all on the same die, right alongside their associated electronics. In other words, what traditionally required multiple MEMS dies (each with its own quirky manufacturing flow) can now be integrated monolithically, using a single, uniform, high-volume CMOS process.

Nanusens’s immediate plan is to bring its own products to market—beginning with a tunable capacitor and a voice accelerometer. But that’s just the opening act. Their long-term ambition is to have their MEMS-in-CMOS magic baked directly into the PDKs of mainstream CMOS processes and transition to becoming an IP vendor. At that point, any silicon designer could drop accelerometers, gyros, tunable caps, microphones, or whatever else they fancy right into their chip layout, just as easily as adding another op-amp or PLL.

The semiconductor industry in 2025 is estimated to be a $700 billion colossus driven by the relentless scaling, standardization, and volume economics of CMOS manufacturing. By comparison, the entire MEMS industry—encompassing every accelerometer, gyroscope, microphone, pressure sensor, and RF MEMS switch—amounts to only about $15 billion. That’s not small change, but it’s barely a rounding error in the larger semiconductor world.

Having said this, the demand for sensors is exploding. They’re showing up everywhere: smartphones, laptops, wearables, AR/VR headsets, industrial IoT nodes, automobiles, drones, smart homes, robotics, medical devices, and a growing flood of ambient-AI systems. Every year we embed more sensors into more products at more price points—and we expect them to be smaller, better, cheaper, more reliable, and available in unimaginable volumes.

This is precisely why Nanusens’s CMOS-native MEMS approach feels like a genuine inflection point. By moving MEMS into the same manufacturing ecosystem that builds the rest of the world’s silicon, they’re aligning themselves with the largest, fastest, most scalable industrial machine humanity has ever created. Sensors that used to require bespoke processes, dedicated fabs, multi-die packaging, and painful trade-offs could soon be dropped into any CMOS design as easily as adding another standard cell.

It’s common for me to finish a column like this with a polite “call to action” along the lines of “Feel free to contact the folks at Nanusens to learn more and tell them Max says Hi.” In this case, however, I’m tempted to address the big MEMS players—Bosch, STMicroelectronics, Broadcom, Infineon, TDK, Analog Devices, and the rest—and say: “Keep your eyes peeled. If one of your competitors snaps up Nanusens, you may wake up one morning to discover that they’re not just eating your lunch—they’ve taken over the kitchen and changed the locks!” I’m just sayin’.