Scientists from Stanford University have generated a model capable of simulating brain activity (Benjamin BV et al). It’s capable of simulating a million neurons with billions of synaptic connections. The amount of power required is similar to that necessary for running a tablet computer.
They’re calling the system Neurogrid. It’s composed of 16 Neurocore chips that are integrated in a circuit board that consumes three watts of electricital energy. Every Neurocore can support 65536 neurons meaning that around a million neurons can be stimulated simultaneously with billions of synaptic connections in real-time.
Neurogrid is a neuromorphic system that realizes the function of biological neural systems by emulating their structure.
Designers of such systems face three major design choices:
1) Whether to emulate the four neural elements—axonal arbor, synapse, dendritic tree, and soma—with dedicated or shared electronic circuits.
2) Whether to implement these electronic circuits using analog or digital means.
3) Whether to interconnect these arrays of silicon neurons with a mesh or tree network.
The authors of this work emulated all neural elements except the soma with shared electronic circuits; this choice maximized the number of synaptic connections. They realized all electronic circuits except those for axonal arbors using analog computation; this choice maximized energy efficiency. Finally, they interconnected neural arrays in a tree network in order to maximized throughput.
