IBM announced that it has built two prototype chips that process data more analogous to the way humans digest information than the chips that now power PCs and supercomputers. The chips represent a significant milestone in a six-year-long project that has involved 100 researchers, a number of academic institutions and funding from the government’s Defense Advanced Research Projects Agency, or DARPA.
The system is capable of “rewiring” its connections as it encounters new information, similar to the way biological synapses work. Researchers believe that by replicating that feature, the technology could start to learn. Cognitive computers may eventually be used for understanding human behaviour as well as environmental monitoring.
The SyNAPSE system uses two prototype “neurosynaptic computing chips”. Both have 256 computational cores, which the scientists described as the electronic equivalent of neurons. One chip has 262,144 programmable synapses, while the other contains 65,536 learning synapses.
In humans and animals, synaptic connections between brain cells physically connect themselves depending on our experience of the world. The process of learning is essentially the forming and strengthening of connections. We know a machine cannot solder and de-solder its electrical tracks. However, it can simulate such a system by “turning up the volume” on important input signals, and paying less attention to others. Instead of stronger and weaker links, such a system would simply remember how much “attention” to pay to each signal and alter that depending on new experiences. A learning algorithm instructs the chip where to focus its attention.