The electronic brain gets closer
EDN staff - November 29, 2012
An IBM supercomputer has achieved a new milestone toward DARPA's vision - as called for in its Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) program - of developing electronic neuromorphic (brain-simulation) machine technology that scales to biological levels. At the recent Supercomputing 2012 conference, IBM announced it had simulated an unprecedented scale of 2.084 billion neurosynaptic cores containing 53x1010 neurons and 1.37x1014 synapses running only 1542× slower than real time.
The DARPA SyNAPSE program ultimately calls for building a cognitive computing architecture with 1010 neurons and 1014 synapses - a number inspired by the estimated number of synapses in the human brain. What IBM has achieved, using its Compass scalable simulator for the company's TrueNorth Cognitive Computing architecture, is not a biologically realistic simulation of the complete human brain but a simulation of "a novel modular, scalable, non-von Neumann, ultra-low power, cognitive computing architecture at the scale of DARPA SyNAPSE metric."
The computing resources required for the latest simulation involved 1,572,864 processor cores and 1.5 PB of main memory. The simulation was powered by the Lawrence Livermore National Lab Blue Gene/Q Sequoia supercomputer, which is currently ranked second fastest in the world (having been just displaced from the top spot by the Cray Titan).
Cognitive computing research promises systems that will be able to "learn through experiences, find correlations, create hypotheses, and remember - and learn from - the outcomes." Below is a short (5:15) video from IBM explaining cognitive computing:
For more on the latest milestone, see the original IBM Research Report ("1014" (PDF)) on the Supercomputing 2012 presentation. Also see Dharmendra S Modha's Cognitive Computing Blog. Modha is a manager and lead researcher of the Cognitive Computing group at IBM Almaden Research Center.