Brain-machine interfaces (BMI) show promise to restore lost sensory or motor function. Similarly, they offer a potential treatment for a variety of neurological disorders. According to the researchers, BMI's have not been utilized by the populus. This is due to limitations in their physical design and ability to transmit information. This paper details Neuralinks first attempt at scalable high-bandwidth BMI. The researchers have built an array that uses small threads with at most 3072 electrodes per array across 96 threads. Each thread is inserted individually with submicron precision to avoid vasculature and enable targeting of regions in the brain. The electrode array is packaged into a small implantable device. The package is 23x18.5x2 mm3 and utilizes a single USB type C cable to provide full-bandwidth data streaming. The array is also capable of simultaneous recording and analysis of all 3072 electrodes. This system also achieved a spiking yield of up to 70%% in long-term implanted electrodes. This approach to a BMI is unprecedented in its packaging density and scalability for clinical applications.
Brain-machine interfaces show promise in helping people with a variety of clinical disorders. BMI’s with 256 electrodes have been used to allow control of a computer cursor, robotic limbs, and speech synthesizers. These demonstrate a proof of concept, but they are limited by an inability to record a large amount of neural input simultaneously. Noninvasive electrodes placed on the outside of the skull provide distorted results. They are only capable of averaging the activity of millions of neurons which makes them incredibly non-specific. Similarly, invasive electrodes placed on the skull can only measure the activity of thousands of neurons. They are also incapable of reading deeper brain activity.
Microelectrodes are the standard technology for recording action potentials from neurons. No large-scale systems exist for implantation. This requires a system with high biocompatibility, safety, and longevity. Such a system would also need to be compact and low power to allow wireless operation. Current systems use rigid metals or semiconductors. Such a design permits deeper penetration into the brain, but can accelerate an immune response. It also limits its potential applications as certain portions of the brain can not be accessed due to vasculature.
Here, Neuralink uses an alternative approach. Flexible polymer probes have 32 electrodes per probe. Their greater flexibility and smaller size offer great biocompatibility. However, the increased flexibility creates issues with their insertion. The researchers used a robot to insert a large number of probes efficiently and independently across varying brain regions. Neuralinks system increased the amount of channels by a factor of 10. This system has three main portions: 1) ultra-fine polymer probes, 2) a neurosurgical robot, 3) and a custom high density electronic array. This system is designed for long-term implantation. It also permits full broadband streaming of the neuronal information. A proprietary neuronal spike-detection software was developed for high accuracy, low latency detection.
A custom process was devised to fabricate such delicate threads. The developed microfabrication process allows for easy and fast production of these threads. Each of the 96 threads contain 32 independent electrodes. The researchers have developed over 20 different types of threads each with three layers of insulation and two layers of conductors.
The insertion head contains six light modules each capable of illuminations with 405nm, 525nm, 650nm, or white light. The 405nm module allows the robot to visualize for alignment. The 525nm light in conjunction with stereoscopic cameras and specialized software allow for estimation of the surface of the brain surface. This complex setup allows for the robot to implant electrodes around previously identified microvasculature. This system has demonstrated an 87.1%% insertion success rate over 19 surgeries. Small manual adjustments were made. This resulted in an unprecedented insertion rate of 29.6 electrodes per minute.
The custom chip produced by Neuralink provides unparalleled signal amplification and digitization within an impressively small package. Their system rejects background noise, and digitizes the amplified signals to be streamed out of the unit for real time-processing. This system uses minimal power and takes up a fraction of the size of other systems. The researchers currently have two configurations. System A has better overall performance measures, but less channels for communication. System B has more channels but slightly worse performance. The data is then streamed to a computer for live data analysis.
System A and B were implanted into Long-Evans rats. System A was able to record 1344 of 1536 channels simultaneously. The exact channels can be changed according to what is desired. System B was able to record from all 3072 channels simultaneously. The use of the researchers custom online spike-detection software was instrumental in decreasing their systems energy usage, latency, size, and accuracy.
The systems developed in this experiment serve two main purposes. First, it is a research platform for use in rodents. Secondly it serves as a prototype of a human version used in clinical applications. The system is easily implantable and allows for rapid refinement. The next step is to modulate neuronal activity. This could potentially give the sense of touch back or allow movement of paralyzed individuals. The researchers system was designed to allow electrical stimulation through every channel, but it was not demonstrated in this paper. The research system offers distinct advantages over previous approaches. The size and composition of the probes is unparalleled in size and biocompatibility. Similarly, the ability to choose where the probes are placed allows greater flexibility and avoidance of vasculature. Finally, the system allows very high channel counts with a low power consumption and small packaging. In theory multiple systems could be readily implanted onto a human to interface with more neurons. Various technological improvements must be made before practical. This offers an unparalleled scalable prototype from which to move forward. This system shows the potential to offer paralyzed patients the ability to control a digital mouse or keyboard. Additionally, it could feasibly be used to restore motor function.