In the emerging field of neuromorphic engineering, the neuromorphic computer is soon to be industrialized given an interesting Georgia Tech roadmap for developping neuromimetic devices.
Neuromimetic devices are analog information treatment, memorization and communication systems repeating the human cognition : a neuromorphic computer, having special electrical power requirements, is made of an integrated circuit with biological based neurons structures made of silicon and a FPAA for Field Programmable Analog Array, that permits to use the computing device.
Like field programmable gate arrays (FPGA), whiches are digital integrated circuits that are nearly ubiquitous in modern computing, the FPAA can be reconfigured by reprogramming it ("field-programmable."). There are scalable and we can forsee large-scale neuromorphic systems based on neuro mimetics, openning new frontiers in robotics and computing. To make cognitive applications easier to build and to help create an ecosystem of application developers, the team has created composable, scalable software cores called corelets.
Stanford scientists (including Krishna Shenoy, an electrical engineering professor at Stanford and Boahen's neighbor at the interdisciplinary Bio-X center), have recently developed microchips 9,000 times faster than a normal PC: Neurogrid, a circuit board consists of 16 custom-designed "Neurocore" chips, which can simulate 1 million neurons and billions of synaptic connections.
When we look the U.S. BRAIN Project – (Brain Research through Advancing Innovative Neurotechnologies), two projects comparable to Neurogrid are particularly interesting :
* IBM's SyNAPSE Project (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) has developped a chip code-named Golden Gate, to emulate neurons establishing communication with synaptic connections on the fly. At present a Golden Gate chip consists of 256 digital neurons.
* Heidelberg University's BrainScales project is developing analog neuromimetic mini network chips called HICANN (High Input Count Analog Neural Network). At present, the HICANN system can emulate 512 neurons
