For the past two years, quantum sensing technology specialist Richard Norte (Department of Precision Microsystems Engineering) has been working intensively with machine learning expert Miguel Bessa (Department of Materials Science & Engineering) - a collaboration that has led to a real breakthrough in science: “We’ve been able to experimentally demonstrate the world’s best vibration isolation at the micro-scale which has exciting implications for next-generation technologies like ultra-sensitive sensing, non-GPS navigation and quantum Internet.”

Mechanical resonators

Working together as part of the Faculty of Mechanical, Maritime and Materials Engineering’s (3mE) Cohesion Projects initiative, Norte and Bessa set out to produce new types of sensors that could measure weak forces like gravity and dark matter. These forces are notoriously hard to measure because the sensors - mechanical resonators - need to be very well isolated from outside vibrations. “The only way to isolate resonators from surrounding thermal ‘noise’ right now is to put them in very large, extremely expensive refrigerators, each costing about half a million Euros, so if you want to have a whole network of resonators, each with their own ‘fridge, that becomes very expensive. But if one can attain high levels of vibration isolation on a microchip working at room temperature, then one can really start to unlock a lot of groundbreaking technologies.”

We’ve basically made the most vibration-isolated object ever made and we did it with a design that was really simple.

Inspired by nature’s spider webs and guided by machine learning, Richard Norte (left) and Miguel Bessa (right) demonstrate a new type of sensor in the lab. [Photography: Frank Auperlé]

Machine learning

At a certain point Norte felt that his background in nano- and microchip technology could only get him so far: “If I look at all the research that I’ve done in the past, it’s all been me sitting there going “Well I think this would work - let’s go and try it and painstakingly testing all these different designs… But then I realised maybe machine learning could help me over this experimental loop that I constantly find myself in – that maybe machine learning could be used to augment our creative possibilities.”

So Norte sent an email to Bessa: “I’m more of a computational person. I’m into machine learning, Artificial Intelligence, and I try to apply AI to designing new materials or new structures either at the very small or the very large scale.”

Bessa describes Norte as the ideal partner because “he can develop nanoscale devices that very few people can and that allows the AI to be very creative.” This means that instead of the researchers having ideas and then controlling everything in the process, the researchers try to delegate at least part of the tasks to machine-learning algorithms and optimisation algorithms. “So when Richard reached out and told me about these applications – and I have to be honest I knew very little about photonics and phononics – he was very patient with me. Eventually, I realised that these structures are governed by conventional mechanical principles but due to their size being so extreme they start behaving more quantum mechanically. Once Richard and I found a common language to communicate without the jargon of the field, we were ready to tackle the intended challenge: can machine learning and optimisation techniques really help in this context?”

Spider webs

However, they still needed a good initial concept design. Then after months of hard work learning something about each other’s fields of expertise, inspiration hit - “partly as a result of staring into dusty corners during corona lockdown,” jokes Norte. Spider webs! “We realised that nature has been evolving vibrational sensors over millions of years. And if you look at a spider web, it’s designed to sense very small vibrations while also blocking out other things like the wind from the trees or other environmental noise.”

So the team asked a specific machine learning algorithm called Bayesian Optimisation to come up with a really good design based on the basic geometry of a spider web, and this threw up something totally unexpected: “We saw that there was some kind of torsional vibration in some of the outer threads of the web construction,” says Bessa, “and that was actually not something we’d thought about. So this was really an outcome of the exploration by the algorithm.”

Artist impression of an artificial spider web probed with laser light” Artist: Felix Vu [Image credit: Optics lab TU Delft]

World record

With this new insight, the team’s post-doctoral researcher, Dongil Shin, was able to come up with simulations that mimic the experiments very closely. “And excitingly, when we did the simulations, the algorithm said “you can break the world record on one of the properties, a so-called Quality Factor, integral to this particular design,” says Bessa. “According to the algorithm, our simulation had a Quality Factor of 2 billion - much larger than anything we could find in the literature and with a design that was far simpler than anything proposed previously.”


But would the results found in the simulations hold up in practical experiments? This challenge was taken on by PhD student Andrea Cupertino, who built the web-inspired resonator using ultra-thin ceramics from materials often used on microchips, and then tested the new design in the lab. “It was far from easy,” says Bessa, “and it took a while to make sure that all the experimental conditions were similar to the simulation but eventually it was validated, it was real! And so the AI really did help.”

“And the cool thing about this design is that it’s able to produce great isolation with a minimalist geometry,” adds Norte, “so now we can make these devices really easily, very cheaply and you don’t need fridges because it can all be done at room temperature.”

The spiderweb sensors vibrate in a vacuum chamber to avoid contact with air and probed with laser light using optical fibers. [Photography: Frank Auperlé]

It was truly the symbiosis between these two different worlds that produced something really quite nice.

Cohesion Project

Both Bessa and Norte are convinced that this breakthrough was the result of collaboration between scientists from very different disciplines: “Sometimes when you’re not in the field you’re like OK let’s do that and put this together and you’re done but it took us two years - really two years  - to achieve our goal even using a fairly simple algorithm. The experimental technique was also very well known by Richard so it was truly the symbiosis between these two different worlds that produced something really quite nice.”

“And I gained a new intuition out of this,” adds Norte “a new way of looking at things and that’s much simpler than I could have ever imagined. We thought we were going to come up with some super-complicated design and it ended up being this incredibly simple idea and that I think is the power of AI, the machine learning algorithms.”