We've long understood that brain-like computers represent a promising future—especially in this age of large-scale data processing. Researchers have now created a photonic synapse that may be the next step in making computer brains function more like our own.

Conventional computing has always revolved around separate processing units combined with circuitry to process and store data. However, that hasn’t been considered to be anywhere near the optimal usage of hardware. For decades now, scientists have been attempting to develop the idea of a computer that processes and stores information the same way that our brains do. This idea, called neuromorphic computing, has made some significant advances in the future of our computing power.

A research paper recently published in the journal Science Advances describes a new type of on-chip photonic synapse that may be able to realize the neuromorphic computing architecture.  


The Problem of the Von Neumann Architecture

Our traditional computing is based on something called the von Neumann architecture. This architecture involves separate distinct components that develop our stored-program computers. Currently, we have an input that runs through a processor consisting of a logic gate, a control unit that directs all other components, a memory that can be saved to external storage, and an output.


The Von Neumann architecture. Image courtesy of Chris-martin. [CC BY-SA 3.0]


With this architecture, fetching instructions and data operation cannot simultaneously occur. This has led to a situation in modern computing known as the von Neumann bottleneck. While there have been optimizations such as modern caching and low-latency command channels, the underlying bottleneck issue still exists.

A multinational research team believes they're a step closer to solving this issue. The team consists of members from the University of Exeter as well as colleagues from Münster and the University of Oxford. Together, they created a photonic chip that can replicate the architecture of the synapses in our brains, possibly eliminating the need for data to be sent back and forth between the processing unit and the memory. The team developed the device using a photonic integrated circuit approach, constructed with a series of phase change materials (PCM) commonly found in optical drives with silicon nitride waveguides.


How is this neuromorphic computing?

Our brains are made up of near 100 billion neurons, and each neuron is connected to potentially tens of thousands of others. What connects these neurons is a synapse, tiny structures that transfer signals between neurons. Our typical brain will have somewhere from 100 trillion to a quadrillion synapses. To put that into computing power, that would be a trillion-bit processor.

The concept of the photonic synapse was meant to mimic this same structure and can be seen here. The waveguide has a series of phase change materials seated on top of it to replicate a synapse, while the pre-neuron and post-neuron are the connects for the input and output.

A component called an optical circulator connects the post-neuron with the output as well as controls the optical pulses that change the synaptic weight. This allows optical transmission to be measured through the pre and post neurons with the energy level varying with the synaptic weight.


The symbol for an optical circulator. Image courtesy of Geek3 [CC BY 3.0]


The research is still a work in progress, but the researchers believe this is the right step in the direction of neuromorphic computing. Notably, these photonic synapses can operate at nearly 1000 times the speeds of our human brains. In the future, we may see computing on the level of singularity, using the tremendous advantages of photonic systems.

Professor Wolfram Pernice from Exeter, a co-author of the paper, explained “Since synapses outnumber neurons in the brain by around 10,000 to 1, any brain-like computer needs to be able to replicate some form of synaptic mimic. That is what we have done here.”


Featured image used courtesy of the University of Oxford.