Dmitry Petrov Back

IBM Chip Processes Data Similar to the Way Your Brain Does

IBM has just presented the computer chip with the fundamentally new architecture that is really different from Von Neuman arctitecture in all the ways. It’s still built with transistors but at the end it’s a hardware made neural network, where these transistors work as perceptrons or synapses.

On the tech demo it was demonstrated that such a chip can do image recognition from a live video stream with a very little energy consumed and at the same time the decent notebook could only deal with 10 times slow video stream and consumed 10,000 more power.

Since it’s a neural network it requires very specific programming. It has not distinct processor and memory, because such network does both. And to program it correctly it’s required to program how parts blocks of the chip should be interconnected and how much influence they will have on each other. And then one should run the code in the chip simulator and only after that the code could be run on chip.

So the chip is really great in terms of future potential but provides a very big challenge in terms of programming because it’s so different from anything else.