Building a computer like a human brain: a technological revolution
For decades computers have been growing exponentially in computational power. However, the current technology is nearing a threshold: computing simply requires too much power.
In March 2017, the local TV station Mid-West reported that, once completed, Apple’s new data centre in Viborg, Denmark, will use so much power that it corresponds to three per cent of the entire electricity consumption in the country.
Three per cent. One single data centre. That is a lot of power to run servers and keep them cool. In fact, power consumption is becoming a rapidly narrowing bottleneck in developing new hardware: too much power is wasted on cooling. Why? Because the technology behind modern computers generates heat.
Imagine for a moment that computers worked like the human brain. No excess heat there. If our brains were to heat up when thinking, we would have a huge problem – instead our minds are capable of monumental computational tasks without even breaking a sweat.
Building a computer just like the human brain is actually what a team of engineers are working on with a new project at the Department of Engineering, Aarhus University. The project is entitled PHOTON-NeuroCom, and it focuses on mimicking the brain, which works in an analog way – not digital.
"When it comes to image and video processing, the human brain is far superior to any supercomputer of today in terms of energy efficiency. Once I have met a person, for instance, I can easily identify the same person in a crowd without too much energy consumption. The person has been imprinted on my memory. For a computer, the same task is not that easy. A computer has to process and categorise every single feature on every single face in the crowd in order to identify the object. The computer gets inputs, converts them to 1s and 0s, does the signal processing and then the detection, which may turn out to be negative in many cases. A lot of work and energy for nothing," says Associate Professor Farshad Moradi, Department of Engineering, Aarhus University.
There is a huge gap between the energy consumption in the brain and the energy consumption of a single computer. Since a very large part of the Internet is concerned with image and video processing, an extreme amount of energy could be saved with more ‘brain-like’ computers.
"If you tackle the problem of the large amount of energy consumption involved in image and video processing by using new hardware to fill the gap between the power density of the brain and the existing processors, you are probably in good shape to develop a next-generation, power-efficient and brain-like computer," says Associate Professor Moradi.
"The current von Neumann computers will not work in the future because the power consumption required to deal with the exploding big data industry will be too great. In this project, we have started a journey to construct a cognitive computing system as a proof of concept, using special very low-power nanoscale devices called spintronic nano-oscillators, which will function as neurons in the brain-like computers of the future," he continues.
About the project
Partners: Aarhus University & CEA/SpinTEC
Budget: DKK 1.58 million