Musk's genius brain-computer interface plan

Two years ago, on July 17, 2019, Musk and his neurotechnology company Neuralink announced that they had implanted a "thread" probe with a diameter of several microns into the brain of a laboratory mouse using a neurosurgery robot, which communicated with external devices through a customized chip. The project was based on ASIC, op amp, ADC, FPGA and other technical means, and has successfully achieved multi-channel synchronous reading, amplification and transmission of brain signals.

Previous studies have been able to achieve neural prosthesis control of computer mice, robotic arms, etc., but the number of electrodes used in these studies did not exceed 256. These electrodes are limited in number and are only placed in the cerebral cortex, making it difficult to record the activity information of tens of thousands of neurons.

In addition, many electrodes implanted in the brain are made of rigid materials, such as metals or semiconductors, while Neuralink reported that it uses flexible polymer probes, which help alleviate immune responses and reduce brain tissue damage.

In order to make it easier to implant flexible probes into the brain, the research team also developed a complete surgical robot that can implant six "wire" probes per minute, each with 32 electrodes (electrode contacts), which means that approximately 192 electrodes can be implanted per minute.

Gold electrode materials use simple structure, small energy gap, high conductivity polymers PEDOT and IrOx (iridium oxide) to reduce electrophysiological impedance and increase the surface effective charge carrying capacity. PEOT has low impedance, but its long-term stability and biocompatibility are not as good as IrOx.

The main substrate and medium of these linear probes are polyimide with gold thin film lines on them. The research team designed a thin film array, each of which contains a "linear" probe area and a "sensor" sensor area. The "line" is used to collect raw signals, and the "sensor" area integrates signal amplification and ADC modules.

The research team used a wafer-level processing technology, with 10 thin-film devices on a wafer, and each thin-film device has 3072 electrodes. These "wires" end with (16 × 50) square micron rings to accommodate needle threading. Each "wire" in the figure below has 32 electrodes.

This high-density channel weak signal acquisition ( 3072 channels @ <10µVRMS ) requires the integration of amplifiers and digitization modules into the thin film array, otherwise the requirements for connectors and cables will be very strict.

The system designed by the research team uses 12 ASICs, each of which includes 256 independent programmable amplifiers, each of which is called an analog pixel, integrating on-chip ADC and interface control circuits. The amplifiers (analog pixels) can be configured with a high degree of freedom, and the gain can be calibrated and the filter properties can be modified. The ADC sampling rate is 19.3Khz@10bit, the power consumption of each pixel is 5.2uW, and the power consumption of the ASIC is about 6mW.

The system made up of these ASICs also includes an FPGA, real-time temperature, acceleration and magnetic field detection sensors, and a USB C interface for wired data transmission. The entire system is packaged in a titanium metal casing and coated with Parylene C to provide a moisture barrier and extend service life.

The Ethernet-connected base station converts the data streams from these systems into 10G UDP packets, allowing downstream users to process the data in a variety of ways, such as real-time visualization of the data, storage, and other functions.

However, these "wire" probes are soft and thin. In order to facilitate installation and reduce bleeding and immune response, the research team also designed and produced a special surgical robot. The needle on the robot is driven by a linear motor, with an adjustable insertion speed and can be quickly pulled out, allowing the needle to be quickly separated from the wire probe. The pincher clamp is a 50um tungsten wire that is bent at the tip and can move horizontally and rotationally to support and guide the threading.

The inserter head has 6 independent light modules, each capable of independent illumination with 405 nm, 525 nm and 650 nm or white light. 405 nm illumination excites fluorescence from the polyimide substrate of the "thread" to locate the (16 × 50) square micron thread ring on the "thread". Illumination with 525 nm light allows precise estimation of the position of the cortical surface.

The robot registers the insertion site to a common coordinate system on the skull, which, when combined with depth tracking, allows precise positioning of brain structural information. The system can pre-select all insertion locations, enabling planning and optimization of the insertion path to minimize entanglement and strain on the wires and avoid vascular bleeding as much as possible. The robot has an automatic insertion function, inserting 6 "wires" (192 electrodes) per second, with an insertion success rate of 87.1 ± 12.6%, and is compatible with sterile covers and can clean the needles ultrasonically.

At the same time, human intervention can also be used to fine-tune the insertion accuracy to avoid micro-bleeding. After fine-tuning, the insertion speed is slightly reduced to about 29.6 "threads" per minute. The following figure shows the process of the implant robot inserting the "thread" into the "agar jelly".

To allow the mice to move freely, the research team used commutation cables and an online detection algorithm to process digital signals in real time to identify spike movements.

This device is the first unit that Musk has used to realize his dream of brain-computer interface. It can be expanded to human experiments in the future to realize the reading of brain signals and even two-way communication. The 21st century is the century of brain science. What breakthroughs will brain science bring that will shock the world in the near future? Let us wait and see.