Brain-like chips (well, synapse-like chips) are – I can surprisingly say – nothing new. They’ve been around for about a decade (since at least 2007) and have been in use doing some amazing things with signal processing, image and speech recognition and other pattern recognition tasks for a long time.
But they’ve never really been massively power efficient. With the spinnaker chip bringing some incredible improvements over previous designs using traditional hardware, things improved, but just a few days ago, scientists took a massive leap forwards in neural-type architectures.
Using nanowire, korean scientists have brought down the power-usage of artificial neurons to about an eighth of that used by ‘real’ neurons.
As always, this technology is cutting edge and will require a lot of work before it is cost-effective to use, and even more work before it becomes useful in a world which currently doesn’t use neural nets for everyday tasks, but if you’ve ever been of the opinion that Moore’s Law is dead, that we’ve hit some sort of mythical final plateau of computing, you’d better think again.
And for those of us who can foresee a time when brain uploading might indeed be something less science fiction and more science fact, these sorts of technologies are a tantalizing first fumbling tippy-toe-like quantum step towards it.