Computers

Million-core neuromorphic supercomputer could simulate an entire mouse brain

Million-core neuromorphic supercomputer could simulate an entire mouse brain
A newly built supercomputer is able to simulate up to a billion neurons in real time, enough to emulate a whole mouse brain and simulate sections of the human brain for pharmaceutical testing
A newly built supercomputer is able to simulate up to a billion neurons in real time, enough to emulate a whole mouse brain and simulate sections of the human brain for pharmaceutical testing
View 3 Images
A newly built supercomputer is able to simulate up to a billion neurons in real time, enough to emulate a whole mouse brain and simulate sections of the human brain for pharmaceutical testing
1/3
A newly built supercomputer is able to simulate up to a billion neurons in real time, enough to emulate a whole mouse brain and simulate sections of the human brain for pharmaceutical testing
The supercomputer features one million processing cores
2/3
The supercomputer features one million processing cores
A detail of the interconnection network between the processing cores
3/3
A detail of the interconnection network between the processing cores
View gallery - 3 images

After12 years of work, researchers at the University of Manchester in England have completed construction of a "SpiNNaker" (Spiking Neural Network Architecture) supercomputer. It can simulate the internal workings of up to a billion neurons through a whopping one million processing units.

The human brain contains approximately 100 billion neurons, exchanging signals through hundreds of trillions of synapses. While these numbers are imposing, a digital brain simulation needs far more than raw processing power: rather, what's needed is a radical rethinking of the standard computer architecture on which most computers are built.

"Neurons in the brain typically have several thousand inputs; some up to quarter of a million," Prof. Stephen Furber, who conceived and led the SpiNNaker project, told us. "So the issue is communication, not computation. High-performance computers are good at sending large chunks of data from one place to another very fast, but what neural modeling requires is sending very small chunks of data (representing a single spike) from one place to many others, which is quite a different communication model."

The researchers tackled this problem by devising a massively parallel architecture where each of the million cores is able to send tiny "packets" of information (up to just 72 bits in size) that are routed to their destinations by an internal communication network.

A detail of the interconnection network between the processing cores
A detail of the interconnection network between the processing cores

With this architecture, the supercomputer should be easily capable of simulating the 100 million neurons inside a mouse's brain. Even an ad-hoc design, however, isn't nearly enough on its own: to build a proper brain model, you'll also need to get the wiring right.

"To build a mouse brain model we need, in principle, to know every neuron and its connections to every other neuron in the brain," Furber told New Atlas. "In practice this an infeasible amount of data to collect, so we have to settle for statistical distributions of neurons types and statistical connectivity data, so that we can construct a statistically representative brain model.

"Such models do now exist, though they are very rough cut in places – they have been compared to the first attempts to draw a map of the globe, which had highly variable accuracy and missed out Australia altogether as it hadn't been discovered then."

Though one-to-one neuronal mapping may not happen anytime soon, even a somewhat rough lay of the land could provide interesting results. For instance, researchers could build a computer model of the visual cortex of a mouse, "show" it an image that would be translated into a stream of spikes down the optic nerve, and learn much about how such a signal is processed by the cortex, even using the output to control the movement of a virtual mouse or a physical robot.

Furber tells us that the system also has the potential to uncover more about how high-level functions such as learning work inside the brain.

"We already support a fair amount of work on learning processes at the synaptic level, including dopamine reinforced plasticity which is a biologically-plausible form of reinforcement learning. But though putting such local plasticity rules together into a high-level brain-like learning system is possible on SpiNNaker, it is stretching our understanding to generate such a system that we can then claim 'is how the brain learns.'"

The team has already used the system to simulate a region of the brain called the Basal Ganglia, an area affected in Parkinson's disease. Indeed, there is potential for this technology to provide advancements in the medical field, particularly with regard to pharmaceutical testing, though the researchers believe the impact of his research on real patients could take decades to materialize.

Furber and his colleagues are now working on a second-generation machine,"SpiNNaker2," which uses upgraded silicon technology to deliver10 times the functional density and energy efficiency. This would enable, among other things, the creation of a whole insect brain model in a system that could fit on top of a drone.

Prof. Forber provides more details on the SpiNNaker project in the video below.

SpiNNaker: 1 million core neuromorphic platform

Source:University of Manchester

View gallery - 3 images
1 comment
1 comment
usugo
so, it is an approximation, of an approximation, of an approximation, ...