Over a decade ago I was very interested in the concept of neuro search – searching for objects or concepts by thinking about them and even published a paper and got a patent granted before starting a company. However, in 2010 we failed to raise funding for it, and when Elon Musk got into this space, I was incredibly excited. When Elon picks up a concept, he usually generates a lot of hype and generates a lot of skepticism. But then in the spirit of Thomas Edison, he delivers, beats his own forecasts, and keeps the public continuously engaged, excited, and educated. So when Elon went on stage with the first surgical robot and the prototype in pigs (imagine how hard is it to get the ethics approval for a show like this), it was very tempting to check his academic publications on the subject to see if there are any. It is a common practice for life science companies to publish a few research papers in peer-reviewed journals demonstrating the proof of concept and to provide evidence, and confidence to the fellow scientists, investors, and the potential recruits. A quick PubMed query resulted in four papers, three of which were on COVID-19 (one high-profile, and two pre-prints) and one – The Neuralink Paper in the Journal of Medical Internet Research. I also checked the Altmetric score of the paper and was surprised to see that only ten news outlets mentioned the paper. So let’s take a closer look…

Neuralink is a brain-machine interface company co-founded by SpaceX and Tesla founder Elon Musk along with a team of experts in areas such as neuroscience, biochemistry and robotics. Other founding members of Neuralink include Paul Merolla, Vanessa Tolosa, Max Hodak, Dongjin Seo, Timothy Hanson, Philip Sabes, and Tim Gardner.

When the company was founded, however, these people wanted to create and develop cutting-edge brain computer interfaces (BCI), or micron sized devices that will help people with certain brain injuries like stroke and cancer.

Since Neuralink is a private company, we don’t know how much Musk invested in it or what its market capitalization is. However, according to financial news provider Millennial Money, Musk invested a total of $100 million into the company with additional $58 million in investments coming from private equity firms.

The startup, which was registered in 2016, is advancing the field of neuroscience and neuroengineering. But Neuralink is not really that revolutionary. Other companies have tried similar things, mostly with non-invasive interfaces. Companies working to create implants and non-invasive devices include Kernel, Synchron, Neurable, CereGate and Mindmaze. Likewise, Neuralink is not disrupting any industries – at least in the short term.

On July 16, 2019, a white paper detailing the purpose of the company and its project was released. According to this paper, Neuralink aims to create the future of brain interfaces. This includes building devices that will help people with paralysis and inventing new technologies that will expand our abilities.

Theoretically, these devices could improve memory and allow direct interfacing with computers. The devices would be completely integrated with the human brain and would give the brain the ability to connect with the cloud as well as with other brains – wirelessly.

The concept is described by Musk as the Fitbit for the brain. It will not work for mind uploading but will help advance this area of research. It is also very commercially viable as it will help understand many diseases and may be used for clinical research and formulating disease hypotheses.

In this white paper, Musk describes Neuralink’s first step towards a scalable brain-machine interface (BMI) system. Musk also outlines the end goal of the company, which is to restore functions using BMI’s. The system has three main components: ultra-fine polymer probes, a neurosurgical robot, and custom high-density electronics. When combined, this system serves as a research platform and prototype toward a fully implantable BMI.

For the first component, Neuralink developed and built arrays of small and flexible electrode threads, with as many as 3,072 electrodes per array distributed across 96 threads. Musk says that one of the goals of this approach is to maintain a small thread cross-sectional area to minimalize tissue displacement in the brain. To achieve this, the company designed and manufactured more than 20 thread and electrode types in these arrays. These fabricated threads range from 5 µm to 50 µm in width and incorporate recording sites of several geometries. Thread thickness is 4 µm to 6 µm, including up to three layers of insulation and two layers of conductor. Typical thread length is approximately 20 mm.

In order to keep all of this inside of a small package, the company developed a novel alignment and flip-chip bonding process. The electrode array is packaged into a coin-sized implantable device which contains custom chips for on-board amplification and digitization. The package for 3,072 channels takes less than 23×18.5×2 mm3.

Because of the flexible and thin fiber of these threads, Neuralink designed a neurosurgical robot capable of inserting six threads per minute. Each of these threads can be inserted into the brain individually with micron precision. The robot’s insertion head is mounted on a globally accurate, 400x400x150 mm travel, 10-µm three-axis stage and holds a needle-pincher assembly. Moreover, the robot has a feature that allows it to insert up to six threads, or 192 electrodes, per minute. The robot also enables precise targeting of anatomically defined brain structures by registering insertion sites to a common coordinate frame with landmarks on the skull.

It is important to note that although the entire insertion process can be fully automated, the surgeon retains full control and can make manual adjustments, if desired.

The third component of the device, custom high-density electronics, demonstrates how the device is monitored by a Neuralink application-specific integrated circuit (ASIC), which provides tracing of the electrophysiological data in real time. The ASIC consists of 256 programmable amplifiers, on-chip analog-to-digital converters, and peripheral control circuitry for the digitized outputs. Musk detailed that the gains and filter properties on the ASIC can be calibrated to account for variability in signal quality, driven by process variations and the electrophysiological environment. In other words, the ASIC forms the core of a modular recording platform. 

Musk also detailed Neuralink’s two platforms aimed at targeting the brain for neuroprosthetic applications. Neuralink developed two configurations called System A and System B, each differing according to variables. To test this system, Neuralink implanted both systems in male Long-Evan rats and recorded all electrophysiological recordings when the animals explored an arena consisting of a commutated cable that allowed unrestricted movements.

Resultantly, they found that System A, which includes a 1,536-channel recording system, had the ability to record 1,344 channels simultaneously, while System B, which consists of a 3,072-channel recording system, recorded from all channels simultaneously.

Musk detailed that Neralink found a permissive filter that allows an estimated false-positive rate of about 0.2 Hz. So, the company set a threshold of >0.35 Hz to quantify the number of electrodes that recorded spiking units.

Moreover, in this experiment, 40 out of 44, or about 90% of the attempted insertions were successful for a total of 1,280 implanted electrodes.

Musk’s approach has many advantages over previous attempts at similar experiments. Firstly, the small size and compactness of the thin-film probes are a better match for insertion in the brain. In addition, the option to choose where to insert the probes allows for custom-built geometrics, which in turn gives space to targeting specific brain regions while avoiding vasculature. Lastly, the design of the ASIC offer flexibility and supports high channel counts within size and power constraints.

The field of neurosurgery is facing new challenges; particularly with regards to a high-bandwidth device suitable for clinical application. However, as Musk’s claims in the paper, is it plausible that patients with spinal cord or certain brain injuries could control digital devices.

At the same time, there are many questions that arise with ideas like Neuralink, including its viability and efficacy in clinical practice. Questions about the way in which these devices could affect autonomy and free-will are also valid concerns. However, when did such concerns stop Elon?

Another challenge is that as of June 2021, most of the co-founders have left Neuralink, with Max Hodak becoming the latest co-founder to leave the company in May. It is worthwhile highlighting Max Hodak, as also a visionary serial entrepreneur. In the past he started Transcriptic, a cloud robotics company performing services for the scientific community. When he exited, the company merged with 3Scan to form Strateos Health. However, when did executive exodus ever stop Elon?

Elon not only managed to keep the company together but also advance it to the level where he could go on stage and do the famous demo of reading pig’s brain activity and presenting the surgical robot. And while the presentation is rather technical, just looking at Youtube statistics for the various cuts of his presentation, I can guess that over five million English speakers watched it. So it eclipsed the popularity of the original paper many times over. The following demo of the monkey playing “MindPong” generated even more interest and demonstrated the resilience of Neuralink to the changes in team composition.

There are many people who criticize Elon Musk for all kinds of sins ranging from growing his companies using government funds, being brutal to the employees, self-promotion, reproduction via IVF, failed relationships, et cetera. However, most of the fellow entrepreneurs deeply admire him as pretty much every company he touched without having prior experience in the industry, he turned into a success story. And there is every reason to believe that Neuralink will deliver great products for the people with diseases and also for us healthy (but aging) people.

My first impression after reading the paper is that while not necessarily new, it is a very important concept and we are very likely to see it implemented very soon in research studies of all kinds of neurological diseases where real-time continuous monitoring of brain activity is required. In many of these diseases we need this “Fitbit for the brain” to elucidate the mechanism of disease. Just by looking at this market, it is a slam-dunk from the commercial perspective. To tease the transhumanist crowd I would speculate that Neuralink will be sine qua non for the first brain transplantation and revival. But as for the digital afterlife, in my opinion, we are still at least a couple decades away from this concept. And it would likely require the technologies described in Neal Stephenson’s “Fall; or, Dodge in hell” to advance to make it possible. Plus, there is a high chance that we already live in a recursive simulation.