Our brains are not only wondrous works of nature; they are extremely efficient. Computing power has increased rapidly even as the size of electronic devices has decreased, but that all pales in comparison with the speed and efficiency of the brain. For a frame of reference, a computer simulation of the functions of a mouse's brain runs slower than the actual mouse's brain by a factor of 9,000 - and consumes 40,000 times more energy to boot.
Researchers at Stanford University have been considering computer modeling of the brain from a power consumption standpoint, and their work has resulted in a circuit board that is specifically modeled to mimic the human brain and neural connections. This work, and similar efforts on computer simulation of the brain, was discussed in an article in the recent Proceedings of the IEEE.
Neurogrid is the name of this new board, with sixteen so-called "Neurocore" chips that were custom-designed and manufactured by the research team. Neurogrid can simulate one million neurons, translating into billions of connections. With the reduction of power consumption in mind, the design specifically allowed sharing of some circuits by synaptic connections.
The resulting board has considerably more simulation power than other brain-mimicking programs - by orders of magnitude - and only consumes the power of a current tablet computer. It still has a significant gap to close to match the human brain, which has 80,000 times more neurons to work with than Neurogrid and consumes only three times its power.
Neurogrid represents another advance toward human style robotics, and because of its low power usage and high speed, it is being considered in other areas of research. Colleagues from Stanford are considering the Neurocore style of chip for incorporation into prosthetic devices to assist the disabled.
Even so, Neurocore has limitations. Cost is one of the first issues, as the prototype Neurogrid board cost approximately $40,000. However, the research team used fabrication techniques that are several generations old to make the chips, and it is entirely possible that newer and more efficient manufacturing methods could reduce the cost by a factor of a hundred.
Also, at this point the Neurocore doesn't have any sort of user interface or compiling software that a computer scientist could use without knowledge of the brain's functions. The research team has already begun work on this software.
With improved software and lower costs, the Neurogrid could become a standard for modeling and control of computer-operated biological systems. Having said that, there are several other entries in the field of computer brain simulation, including the SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) project by IBM, the BrainScales project from Heidelberg University, the EU's Human Brain Project, and the BRAIN (Brain Research through Advancing Innovative Neurotechnologies) Project.
All of these projects are mentioned in the IEEE paper along with Neurogrid, and they are all taking slightly different angles on neural research. Together, they are making significant advances in understanding of the brain, robotics, and enhancement of biological systems. Expect a series of breakthroughs in the field over the next few decades.