Researchers from the Federal Polytechnic School of Lausanne (EPFL) have developed a brain microchip that is smaller and more efficient than the Neuralink chip. The research that led to the new chip was published in the IEEE Journal of Solid-State Circuits.
This device, known as a miniaturized brain-computer interface (MiBMI), is a very small and consists of two thin chips with a total area of only 8 mm². In comparison, the Neuralink device is larger – approximately 23×8 mm.
MiBMI uses very little power, is minimally invasive and has a fully integrated system that processes data in real time. Unlike Neuralink, which requires placing 64 electrodes in the brain and processing the data through an app on an external device, MiBMI simply reads brain activity and converts it into text.
“MiBMI allows converting complex neural activity into readable text with high accuracy and low power consumption,” – says Mahsa Shoaran, head of EPFL's Integrated Neurotechnology Laboratory. “This achievement brings us closer to practical implantable solutions that can significantly improve the communication capabilities of people with severe motor impairments.”
The new chip monitors the electrical activity of the brain and, using data from previous studies, converts this activity into text. MiBMI is able to read the brain signals generated when imagining writing a letter and convert them into text.
Although the chip has not yet been tested in live conditions, it has been tested on real neural recordings and achieved 91% accuracy in converting neural activity into text.
One of the reasons for the success of the new chip lies in a new way of reading the language processes of the brain. EPFL researchers discovered specific neural markers that are activated when each letter is imagined. These markers became a short code for each letter, allowing MiBMI to process less data, which reduces power consumption and chip space.
MiBMI is now able to decode 31 characters, which is a record for similar systems. Researchers plan to increase this number to 100 characters.