Building an insect brain on a chip
Think like a
bee
Honeybees have no problem flying through a field of flowers. They won’t bump into anything and they’re even quick enough to dodge a hand that wants to slap them away, all the while looking for their beehive. ‘It’s impressive that a honeybee’s brain can perform all these tasks at the same time’, says Elisabetta Chicca, newly appointed professor in bio-inspired circuits and systems. ‘They require complex calculations to be performed extremely fast.’
Many neuroscientists study real brains to try to understand how they work, while others do simulations on computers. But Chicca has chosen another route: she wants to create a brain on a silicon chip.
A natural brain consists of an intricate network of neurons and synapses, where one neuron can communicate with another because they are connected through the synapses. The processes of millions of neurons passing information are extremely complicated, so doing a simulation won’t really help her understand them, she feels. ‘The famous physicist and Nobel Prize winner Richard Feynman once said: “What I cannot create, I do not understand.” My research is radically founded on that principle.’
A piece of the brain
She picks up an old circuit board the size of an iPhone from the small museum of hardware in her office. ‘We use more modern versions of these electric circuits to mimic the workings of neurons and synapses in the brain’, she explains.
What I cannot create, I do not understand
Chicca isn’t trying to imitate a full-size brain just yet, but rather complex neural networks which emulate a piece of the brain. Specifically of the honeybee brain, because so much is already known about it.
Currently, she’s focusing on the system that ensures the bee avoids obstacles while flying. ‘The honeybee needs to make the decision where to fly to in a very short time’, she says. ‘Otherwise, it would have a big problem.’
Building on previous research, her PhD student Thorben Schoepe is currently mimicking that process on a neuromorphic processor. This device resembles the neural network and aims to solve the computations with minimum energy and maximum speed. But that’s easier said than done.
Time-consuming
The brain can compute extremely quickly because neurons and synapses behave in such a complex way. But therein lies the problem: computers can only perform a calculation one loop at a time, which means replicating this behaviour is very time-consuming. ‘But on the physical model that we’re making, there are multiple devices that perform the calculations in parallel’, Chicca explains. These devices are interconnected, resembling the neural network in natural brains.
A honeybee needs to decide where to fly in a very short time
Of course, you could achieve the same by using a fast computer, but the size and power consumption could never compare to those of a honeybee brain. And then they wouldn’t have been able to build a tiny robot.
The bee-bot they’ve engineered can drive through all kinds of complex environments while sensing where obstacles are, moving to avoid colliding with them. ‘We had an idea of how these calculations worked in the insect brain’, says Chicca. ‘Now I’m very excited that we were able to implement this in a physical system and that we can test it in the real world.’
Physical constraints
But why is it necessary to make this artificial insect brain on a silicon chip?
Building a physical representation of a neural network has its advantages, Chicca explains. You’re working with physical constraints, just like the ones the brain has. ‘For example: our brain needs to function even while our body temperature is rising. It doesn’t suddenly stop working when it’s hot.’
Our devices will solve very specific problems
One of Chicca’s challenges is to make her circuit boards robust to ‘noise’, like changing temperatures, and compatible with physical constraints. A simulation doesn’t have to deal with varying parameters like that. ‘I’m not saying that the environment in the circuit boards that we use is exactly the same as in the brain, but similar principles are conserved.’
Research centre CogniGron is a great fit for her, she says. It aims to develop ‘smart’ materials that can be used in computer systems inspired by the brain. ‘I was watching the creation of this new multidisciplinary centre and it was very interesting to me. I didn’t have the opportunity at my previous university in Bielefeld to collaborate with materials scientists who are strongly focused on the development of cognitive systems.’
Practical applications
Chicca’s main drive is to understand biology and work together with neuroscientists to find out how a biological brain works. However, when she discovers a new principle and sees an interesting application for it, she won’t shy away from that. ‘Because we only look at specific parts of the brain, our devices will solve very specific problems’, she says. A device that knows how to avoid an obstacle like a honeybee is not very useful on its own. ‘That’s why what we build has to be embedded into more complex systems. The advantage is to enable autonomous behaviours in these systems.’
A small robot that can avoid obstacles autonomously can be very useful in rescue missions, for example, when it’s looking for survivors under debris after an earthquake. Being able to swerve around obstructions is ‘only one feature of the drone, but it is a crucial one’, says Chicca.
So… when will she be done? When will she have built that ‘real’ artificial brain? She smiles. ‘It’s not likely going to be done anytime soon. For sure not before I retire.’ That doesn’t hold her back, though. ‘I never felt that I needed motivation to go on, because I love what I do. I couldn’t imagine myself doing something else.’