Neuromorphic technology based on new physics

A novel signal processing platform has been developed at the University of Oxford that takes on the characteristics of a neuron without suffering from the power dissipation problems of conventional electronic circuits. The device uses interactions of nonlinear acoustic waves in thin films of phase changing liquid crystals for neuromorphic computation. The platform can be used as part of a larger signal processing system or can be stacked seamlessly to build more complex signalling processing systems. It can also have other applications such as a physical or biomolecular sensor.

The semiconductor technology that fuelled the digital revolution has reached its limits. A further reduction in transistor size is being limited by quantum uncertainties, and next-generation supercomputers are struggling to withstand high temperatures that result from tremendous heat dissipation in these systems.

Apart from reaching their theoretical limits, non-biodegradable silicon-based technologies are not sustainable. Incremental solutions in the field are proving to be insufficient, demanding radical innovations. In dealing with these challenges, scientists have been taking inspiration from the extraordinary energy efficiency of the human brain, which has led to many innovations in computational hardware, known as neuromorphic computing.

Now inspired by the thermodynamics and material physics of neurons, a novel platform has been invented at the University of Oxford that closely mimics the computational properties and energy consumption of real neurons. The platform is based on the recently discovered phenomenon of nonlinear acoustic waves in liquid crystal thin films of lipids.

The nonlinear excitation and collision properties of these waves in a phase change material allow neuron-like computational capabilities in a substantially elastic and energy efficient system.
The salient features of the platform that set it apart from the competition are (a) mimics axonal computation including collisions, (b) computes in-material using analog non-linear spikes with efficiency close to real neurons, (c) operates at room temperature, and (d) uses biodegradable material.

The modular design of the signal processing unit will allow seamless integration for massively parallel computation and arbitrary network topology. Finally, the neuron mimicking property of the platform can also be used for sensing or developing new assays for neuro-pharmaceutical and toxins.

Request more information
about this technology

Ready to get in touch?

Contact Us
© Oxford University Innovation