Recently we reviewed the interesting work of FACETS, a large European collaboration developing hardware-level designs for computer circuits that mimic the architecture of our brain. Another group here in the United States at Stanford University is taking an alternate hard-wiring approach to designing a brain in silico.
They hope to create a computer that works nearly as powerfully as the human brain–and be “affordable” at the same time. In addition, they also anticipate that not only will their work take a step forward to a deeper understanding of human brain function, but it will also provide the computational power to help other neuroscientists better analyze and simulate neural activity to advance their own research.
The research team, lead by Kwabena Boahen, is developing a neuromorphic chip: a computer that is not based on the classic transistor developed in 1947, but instead is composed of individual mini-circuits designed like a human neuron, developed some 250,000 or more years ago. More specifically, the ion-flow regulated in the neuron’s membrane is replicated by electron flow in the silicon device. And with quite a bit of clever foresight, the interconnections between the “silicon neurons” are not permanently hardwired on the circuit. Instead, each silicon neuron is identified by a memory address, like in a typical RAM chip, and their electrical activity is referenced by the controlling software. This allows for the same chip to be soft-wired to model the interconnectivity of any sort of neural network that is desired to be used for a particular computational application.
Read more about the specific details of how the neuromorphic chips are designed, fabricated, and tested at the Brains in Silicon group’s website. [ VISIT ]
K Boahen, “Neuromorphic Microchips,” Scientific American, vol 292, no 5, pp 56-63, May 2005. [ READ (pdf) ]