Liquid Nanofluidic Synapse Enables In-Memory Computing
Researchers at EPFL's School of Engineering have fabricated a novel nanofluidic device for memory that allows for the first time the connection of two 'artificial synapses'. The device paves the way for liquid hardware designs inspired by the brain. The work is published in the latest Nature Electronics.
Unlike the brain, which computes directly on stored data, computers shuttle data back and forth between memory units and the central processing unit. This inefficient separation – the von Neumann bottleneck – is increasingly driving up computers' energy costs.
Since the 1970s, researchers have been working on memristors – electronic devices that can compute and store data, like synapses. But EPFL's Laboratory of Nanoscale Biology has built a functional nanofluidic memristor that relies on ions, not electrons. This brings the device closer to the brain's energy-efficient approach to information processing.
Whereas electronic memristors rely on the movement of electrons and holes to process digital information, the new memristor exploits the flow of different species of ions. The device is created on a silicon nitride membrane by patterning a nanopore. Layers of palladium and graphene are added to create an ionic nanofluidic channel. When a current is applied across the chip, ions flow through the channel and accumulate at the pore, forming a bubble between the chip surface and the graphene. As the graphene layer is pushed upwards by the bubble, the device becomes more conductive, switching its memory state to 'on'. Applying a negative voltage brings the layers back into contact, resetting the memory to 'off'.
The researchers patterned what they call a highly asymmetric channel – analogous to ion channels in the brain that undergo structural changes within synapses – with the shape of the channel along the central pore directing the flow of ions. The team successfully connected two highly asymmetric channels to an electrode, creating a logic circuit based on the flow of ions.
By simply immersing the device in an electrolyte aqueous solution containing ions, the memory of the device can be tuned – how it switches between on and off and how much memory it stores – by changing the ions, such as potassium, sodium and calcium. This is a unique feature. The researchers also show how the device can be integrated with fluidics to create fully liquid circuits. This could not only allow for built-in cooling but also pave the way for the development of biocompatible devices, which could have applications in brain-computer interfaces and neuromedicine.
