Putting a brain on a chip
By NTU College of Engineering Posted 4yr(s) ago Reading Time: A few minutes
Inspired by the human brain, researchers at EEE have developed a smart chip that could imbue portable and wearable devices with artificial intelligence at minimal energy use and cost.
The researchers tested the small and light- weight prototype, that can be mounted on headgear, and were able to decode a monkey’s brain waves and predict how it wanted to move its fingers with 99.3% accuracy. This could pave the way for creating implants or portable devices to help paralysed people regain use of their limbs.
The researchers exploited manufacturing imperfections to make their smart chip smaller, lighter and less power-hungry than counterparts on the market. Electronic devices consist of transistors that amplify electric currents or act as switches to start and stop various processes. While transistors are supposed to have specified sizes and properties, they invariably deviate to some degree because of design differences and manufacturing inconsistencies.
The NTU chip takes advantage of these unintended variations in the transistors’ threshold voltage, which determines their ability to produce current for a given input voltage. It uses mathematical formulas that multiply input signals such as brain waves, temperatures, pressure, sounds and images using the array of transistors with varying threshold voltages. This fleshes out similarities and differences among the signals. The chip can then better compare each signal to specified patterns.
Prof. Basu of EEE said the the chip’s low power usage and compact size allows it to be embedded in sensors so that they can sort through raw data and transmit only refined or specified information. He added that wearable medical devices that monitor physiological parameters such as heartbeats, blood oxygen levels and brain activity are becoming more common, but that most of them simply collect the data and display it in visual form with a smart-phone.
The team has received funding from the Singapore – MIT Alliance for Research and Technology to improve its prototype, including its ability to recognise images.
“Our prototype can give meaning to the data by matching it with knowledge obtained from patient databases, so it can give useful, refined information to users instead of raw data.”