Brain-mimicking microchips pave way for better robots
Scientists in the US have designed a circuit board modelled on the human brain they claim is 9,000 times faster than a typical PC.
The researchers at Stanford University say this “Neurogrid” could pave the way for greater advances in robotics: for example, chips as fast and efficient as the human brain that could operate prosthetic limbs as well as a real brain controls real limbs.
The new circuit board is also 40,000 times more energy-efficient than a PC, said researcher Kwabena Boahen, who has published an article on the work in the journal Proceedings of the IEEE. ‘From a pure energy perspective, the brain is hard to match,’ he said in a statement.
The tablet computer-sized Neurogrid consists of 16 custom-designed “Neurocore” chips, which together can simulate 1 million neurons and billions of synaptic connections.
By enabling certain synapses to share hardware circuits, Neurogrid can simulate orders of magnitude more neurons and synapses than other brain mimics on the power it takes to run a tablet computer.
The next step will be creating software to enable engineers and computer scientists with no knowledge of neuroscience to solve problems using Neurogrid.
‘Right now, you have to know how the brain works to program one of these,’ said Boahen. ‘We want to create a neurocompiler so that you would not need to know anything about synapses and neurons to able to use one of these.’
The researchers also hope to reduce the $40,000 (£24,000) cost of Neurogrid to $400 (£240) by replacing the chips made using 15-year-old fabrication technologies with those created from modern manufacturing processes at large volumes
However, there is still some way to go to reproduce the scale and energy-efficiency of the brain. ‘The human brain, with 80,000 times more neurons than Neurogrid, consumes only three times as much power,’ said Boahen.
‘Achieving this level of energy efficiency while offering greater configurability and scale is the ultimate challenge neuromorphic engineers face.’
Similar research efforts include IBM’s SyNAPSE Project (Systems of Neuromorphic Adaptive Plastic Scalable Electronics), which aims to redesign chips to emulate the ability of neurons to make a great many synaptic connections, and Heidelberg University’s BrainScales project to develop analogue chips that mimic the behaviours of neurons and synapses.
Each of these research teams has made different technical choices, such as whether to dedicate each hardware circuit to modelling a single neural element (e.g. a single synapse) or several (e.g.,by activating the hardware circuit twice to model the effect of two active synapses). These choices have resulted in different trade-offs in terms of capability and performance.