From Stephen Hawking to Matrix: Making Science Fiction Come True (3)

Matrix AI Network
6 min readJan 6


Invasive BCI: Reading Detailed Motor Signal Through Neural Chips

Since EEG signals are not precise or stable enough, why not insert a neural sensor into the brain to read and record the activities of each motor nerve cell, or insert a BCI that stays in the brain forever? In 1998, Philip Kennedy, founder of an American start-up named Neural Signals, tried implanting microelectrodes in the human brain to record the discharge of nerve cells in the motor area, hoping to help a disabled person who had completely lost his motor ability reconnect with the outside world. Since Philip couldn’t use as many electrodes as he wants, the experiment couldn’t achieve an interpreting accuracy sufficient for practical use. Despite this, his preliminary research and Dr. Gray Walter’s brave attempt in the 60s inspired a group of researchers in neuroelectro-physiological motor function, including John Donoghue of Brown University, Andrew Schwartz of the University of Pittsburgh, Miguel Nicolelis of Duke University and Jose Carmena of Berkeley University. They had been in the electrophysiological research of primate motor nerves for a long time and had a deep understanding of the mechanism of the monkey brain motor cortex nerve cells that control upper limb movement. As mentioned above, the movement of hands and feet is controlled by the contralateral sensorimotor band. But the next question is: how is information such as the direction, acceleration, and the force of hand and foot movement encoded through the nerve cell activities of the sensorimotor band?

Research on this question can be traced back to Charles Sherrington, the founder of modern neuroscience and a winner of the Nobel Prize in 1932. He discovered that muscles are controlled by the motor nerve cells in the spinal cord in a relationship more complicated than a simple one-to-one match. The idea behind ​​this population coding inspired future research on brain motor cortex nerve cell coding. In the early 1980s, Apostolos Georgopoulos and others discovered that the neurons of the motor cortex used “democratic voting” to collaboratively code the arms’ direction of movement. If we use a vector to represent the discharges of these nerve cells, the direction of the vector is the preferred direction of the cell, and the length of the vector is the number of discharges per second (discharge rate), then the sum of the recordable “discharge vector” of all the cells is the target direction of the arm.

The Birth of Invasive BCI

The first invasive BCI studies were conducted on trained monkeys. By 2002, the monkeys in Donoghue’s laboratory could use the nerve cell activities in the motor area of their brains to control the cursor on a computer screen. Then the monkeys in Schwartz’s laboratory could freely control objects in a three-dimensional space through a similar BCI. Now, monkeys in Nicolelis’ lab can even remotely control a robotic arm at MIT from Duke University. As electrode implantation and neural activity interpretation become more precise, the number of parameters that monkeys can control through BCI is increasing. By 2008, the monkeys in Schwartz’s laboratory could freely control up to ten parameters of a robotic arm’s movement and operate the robotic arm through BCI to fetch marshmallows for themselves. This shows the bright future for disabled people who need invasive BCI to control prostheses.

The development of invasive BCI from primates to humans should be credited to three brave disabled people. The first was Matthew Nagle, a patient with high paraplegia. In 2006, he agreed to implant a hundred-channel microelectrode array in the area of ​​his cerebral motor cortex responsible for hand control. This electrode array was connected to the BrainGate BCI system developed by Donoghue’s lab capable of using thoughts to control the cursor, open emails, or play ping-pong games. Unfortunately, Matthew died of an infection the following year. The second test subject was Cathy Hutchinson, who lost her motor ability due to a stroke. In 2012, with the help of BrainGate’s second-generation BCI system, she could use a robotic arm to bring coffee to her mouth and take a sip for the first time in 15 years. The third test subject was Ian Burkhart, a paraplegic patient due to a car accident. In 2016, with the help of second-gen BrainGate, his motor nerve signals were translated into electrical pulses to drive the muscle stimulation electrodes on his arms so he could move his arms again to grasp, rotate his wrists and even play a couple of strums on the guitar. Over the past 15 years, invasive BCI has completed its transition from monkey to human. In addition to neuroscientists and biomedical engineers, neurosurgeons have also played an important role in this. The technologies involved in these three clinical trials are very complicated, and without exception, it is the teamwork of researchers from all three fields that made it happen.

Challenges for Invasive BCI

Seeing these exciting research results, you may think invasive BCI is perfect and that clinical application is close at hand, but you will be disappointed. So far, 15 to 20 severely disabled people have participated in clinical trials of invasive BCI, and they all encountered a problem: the implanted electrodes gradually failed due to the fact glia cells will wrap around them so they can no longer record nerve cell discharges. The above clinical experiments all used Utah arrays. Although it is only 4x4 mm, the 100 silicon electrode microneedles on it have to penetrate the surface of the brain and insert themselves among nerve cells to record nerve discharge signals. It will inevitably cause inflammatory responses from glia cells. In the worst-case scenario, these electrodes will become ineffective after 2 to 3 months. At best, they can last for 2 to 3 years, but the signal quality will gradually decline, and so will the performance of BCI. If this problem can’t be solved, long-term implantation is impossible, and the cost of replacing electrodes every 2 to 3 years is too high. In addition, the wireless transmission of neural signals is also a problem. All of the above clinical experiments have installed wired plugs in the patient’s head to transmit nerve signals and power supply, which would greatly increases the risk of infection. To address these two problems, scientists have considered many solutions, such as reducing the reaction of glia cells by adding anti-inflammatory substances on the electrode surface. Wireless and low-power neural chips are also being developed. Some people have even proposed to spread nano-dust around nerve cells to form Neural Dust around the cells so that neural signals can be obtained through an ultrasonic power supply.

To solve these two problems, low-invasive BCI was proposed in 2013: the neural electrodes will only be embedded in the surface of the cerebral cortex but not penetrate the cortex. BCI can be established through the field potential of a few electrodes rather than neural discharge activities, which helps avoid inflammatory responses and glia cell overgrowth. Meanwhile, wireless data exchange and power supply are both possible, since the number of electrodes and the signal sampling rate are low. Such a low-invasive solution may capture less neural information than electrode arrays, but it’s reliable in the long run. Under this principle, we have designed a new BCI typing method, placing electrodes in the key regions of ​​the visual brain area that process moving objects, and using high-frequency field potential changes above 60 Hz in the area to distinguish the patient’s attention and focus. This way, we can achieve fast, accurate and long-term stable EEG typing.

Read article on website:



The Matrix AI Network was founded in 2016. In 2023, we enter Matrix 3.0 blending neuroscience with our previous work to realize the vision of the Matrix films.


Website | GitHub | Twitter | YouTube

Telegram (Official) | Telegram (Masternodes)

Owen Tao (CEO) | Steve Deng (Chief AI Scientist) | Eric Choy (CMTO)



Matrix AI Network

The Matrix AI Network was founded in 2017. In 2023, we enter Matrix 3.0 blending neuroscience with our previous work to realize the vision of the Matrix films.