脑机接口大盘点：从科幻到现实，谁在引领这场“读心术”革命?

Imagine being completely paralyzed from the neck down, able to see and think, but unable to communicate your thoughts.

That was Noland's situation.

In 2016, an accident left him paralyzed from the shoulders down.

But now, he's transformed.

He can play Civilization VI, chess , and even operate a computer faster than most people using their hands.

He doesn't use his hands, feet, or even look at the keyboard; he simply thinks, and the computer moves.

This isn't science fiction; it's a real event happening in 2024.

Noland Arbaugh's life has undergone such a dramatic change because last year he received a brain-computer interface chip implanted by Neuralink, Elon Musk's brain-computer interface company . By 2025, brain-computer interfaces— a technology that sounds like something out of The Matrix— are moving from the lab to reality.

Just recently, Neuralink announced that it has implanted brain-computer interface devices in 13 patients, just like Noland. Arbaugh

can type, browse the web, and play games using his thoughts.

Although Neuralink is the most well-known company in the field of brain-computer interfaces , the field is becoming increasingly competitive, with a number of veterans and newcomers catching up.

Projects using non-invasive and minimally invasive approaches are also quietly emerging, including many big names such as OpenAI CEO Sam Altman who have entered the fray.

It is certain that the global brain-computer interface market is experiencing explosive growth , and according to reports, the market size in the United States alone may reach $400 billion.

And this may only be the beginning.

In this video, we will talk about where this technology, which may change the future of humanity, is headed.

A year and a half ago, I visited the Harvard University brain-computer interface lab project in Boston . What new technological advancements have been made since then

. What new technological advancements have been made since then ? (Sorry, I broke the tab,

? (Sorry, I broke the tab, which further demonstrates the flexibility of our materials.)

Who is leading the way? Who may overtake us?

When will this technology be available to ordinary people?

First, let's talk about the principles of brain-computer interfaces.

Before discussing the major players today, we need to understand what brain-computer interfaces actually are.

The full English name for brain-computer interface is Brain-Computer Interface . Brain-computer interface

Interface . Brain-computer interface (BCI) is simply a direct communication channel established between the human brain and external devices.

It bypasses our traditional neuromuscular sensory system , allowing the brain to directly "communicate" with machines. For example, when we use a computer, we

with machines. For example, when we use a computer, we need to use our fingers to type on the keyboard and move the mouse to operate it.

But BCI technology allows you to skip this intermediate step and control the computer directly by "thinking."

This is not mind reading , but rather capturing the signals emitted by the brain and then using algorithms to translate these signals into machine-understandable instructions.

How does it do this?

Imagine that your brain has 86 billion neurons that are "speaking" every moment, that is, communicating through electrical signals.

So, what you can see in this video and understand these concepts is essentially your neurons firing. The working principle of brain- computer interfaces (BCIs) is actually quite simple.

First, signals are acquired by recording neuronal activity through electrodes or ultrasound, much like installing a high-precision listener in a chat group of billions of neurons in the brain.

Second, signals are decoded using AI algorithms to translate them and understand what the brain wants to do.

For example, when you want to move your finger , specific neurons in the motor cortex will fire according to a specific pattern.

The AI ​​learns to recognize this pattern and knows what you want to do.

Third, commands are output by sending the decoded commands to external devices such as computer cursors, robotic arms, wheelchairs , and even future humanoid robots.

Fourth, feedback loops are established . The most advanced BCIs can even work in reverse;

. The most advanced BCIs can even work in reverse; after the device performs an action, it sends feedback signals back to the brain, forming an interactive closed-loop system.

For example, if a BCI-controlled robotic arm picks up a cup, the brain can feel the touch and weight, thus forming a complete closed loop.

So, we now know "what" BCIs can "do," but "how" is a big question because it involves implanting chips in the most important and vulnerable part of the human brain.

The most crucial aspect is its safety.

Brain-computer interfaces have also developed into three major technical routes, making different trade-offs between safety and performance.

The first type is non-invasive.

The advantage is that it is the safest , but the disadvantage is that the signal is the weakest . These devices are like a "mind-reading cap"

. These devices are like a "mind-reading cap" that can be used by wearing on the head. The working principle is that electrodes placed on the scalp detect the weak electrical signals generated by brain activity.

The advantages are that it is completely non-invasive, does not require surgery , is convenient to use, and is relatively inexpensive, with consumer-grade devices costing only a few hundred to a few thousand dollars.

The disadvantages are also obvious: the signal is very weak , like listening to music through a thick wall.

The accuracy is low, it can only perform some simple controls , and it is easily affected by hair, sweat , and external electromagnetic fields.

Currently, many brain-computer interface products on the market also adopt non-invasive solutions.

Its simplicity and ease of use make it suitable for consumer-grade scenarios , but its effectiveness has also been questioned by some professionals .

We must respect the physical facts.

The brain is like this: each neuron operates in a frequency bandwidth of about 300 to 3000 Hz.

More importantly, in the 3000 Hz range, there is the neuron's action potential.

The skull and the membrane on the surface of the brain are very good low-pass sensors.

A low-pass filter blocks all signals above 40 Hz.

If you're using a non-invasive method, physically speaking, you won't get the signal from a single neuron; you'll get an average result.

Our skull is a perfect insulator , so most of the electrical signals developing in the brain are blocked by it.

To put it simply, the brain is like a symphony playing inside; all your thoughts are like a magnificent symphony, and our skull is the concert hall.

Placing electrodes outside the skull to non-invasively measure the brain's electrical signals is like listening to a symphony outside a concert hall.

Regardless of the specifics...

No matter how wonderful a symphony is , or how advanced the microphones and recording equipment you place outside the concert hall are , the signal you ultimately hear through the hall is a very weak and mixed signal.

This is one of the problems that non-invasive brain-computer interfaces currently face : they cannot obtain high-precision and high-bandwidth signals.

Then there is a second, intermediate approach called "semi-invasive."

called "semi-invasive." Semi-invasive is an "intermediate approach" that requires opening the skull, but the electrodes are placed only on the surface of the brain or on the epidural space, without penetrating brain tissue or implanted through blood vessels.

The advantage is that the signal quality is better than non-invasive , and the risk is slightly lower than that of fully invasive.

The disadvantage is that the number of channels is relatively small and the performance is not as good as that of fully invasive.

But this approach is a bit awkward.

Our guest told us that the highest risk is actually craniotomy, but after opening the skull, they don't collect data deeply.

It's like buying a ticket to a concert, entering the concert hall, but sitting in the last row.

Invasive brain-computer interfaces are divided into two main categories : one is called Surface Brain-machine Interface, and the other is called Depth Brain-machine Interface.

After removing the skull and exposing the brain, you can choose to attach electrodes to the surface of the brain to measure electrical signals , or you can choose to insert electrodes into the brain to measure signals.

The advantage of attaching them to the surface of the brain is to ensure the integrity of the brain structure , but the disadvantage is that the brain's electrical signals are actually in the depth.

So you still have the same problem as listening to music in a concert hall.

Now you are in the concert hall , but you are sitting in the last row. Which will give you a better experience : sitting in the last row with the best microphone or sitting in the first row with a worse microphone ?

This is a point of discussion.

Deep electrodes are essentially electrodes inserted directly next to each neuron that generates an electrical signal, obtaining the most firsthand, accurate, and high-throughput signal.

This method aims to reconstruct the brain's thoughts most completely.

These are two main schools of thought within invasive brain-computer interfaces.

Therefore, after craniotomy, how to collect data and how deep to insert the electrodes are key points that the industry is actively exploring and seeking solutions to.

Now let's look at the third school , invasive brain-computer interfaces.

As the name suggests, these invasive devices directly penetrate the cerebral cortex and neurons for "zero-distance contact."

They use tiny electrode needles inserted directly into brain tissue to record the activity of individual neurons .

The advantages are high signal strength, like listening to high-definition stereo , and extremely high precision , enabling complex control and large bandwidth for transmitting more information.

The disadvantages are the need for craniotomy, high risk, the possibility of long-term electrode implantation causing rejection or even infection, and the potential for electrode damage .

Another issue with time degradation is the implanted material.

The brain, especially a living brain , is a very soft tissue , like tofu .

However, all metal or silicon-based probes are extremely hard , like a steel knife.

So, when you insert this into the brain , which is constantly active, the electrodes, like a steel knife, will cut into the brain at a microscale, causing not only long-term mechanical damage but also electrode drift within the brain . This drift results in the inability to measure,

. This drift results in the inability to measure, even if you can detect neuronal signals, you cannot stably measure signals from the same neuron.

Furthermore, it causes a large number of immune rejection reactions in the brain.

Over time, this immune rejection at the implantation site leads to neuronal apoptosis and the proliferation of numerous immune cells , ultimately reducing the initial single-cell detection rate.

Unit action potentials (SAPs) gradually become undetectable.

For patients with implanted deep brain electrodes called Deep Brain Stimulators (DBS), the stimulation location needs to be changed every few months or even a year to effectively treat the disease, as it can cause significant damage to deep brain tissue.

To avoid repeatedly cutting the implanted electrodes and chips in the soft, delicate human brain , the materials must be soft.

Professor Liu Jia and his team made a breakthrough , co-founding Axoft with Dr. Ye Tianyang.

They believe that by making hard materials extremely thin , they can become soft, just as thick iron blocks can be bent into tin foil and thick plastic sheets can be made into plastic wrap.

This discovery immediately inspired the entire industry, with companies like Neuralink adopting similar approaches because thinner wires are more flexible and less likely to damage brain tissue . However, a new problem arises:

. However, a new problem arises: electrodes that are too thin are prone to breakage.

If it breaks , it might be impossible to remove , and materials that are too thin cannot accommodate too many electronic components.

To increase the number of channels , more electrodes need to be inserted.

This led to Neuralink's complex "sewing machine" surgical robot, which we will discuss in detail later.

So, Liu Jia's team took another step forward.

Instead of thinning hard materials, they decided to create a new material that is inherently soft yet tough.

They used a special polymer material that is elastic like rubber , resistant to liquid corrosion , can support more channels , and can be repeatedly inserted and removed without breaking.

In fact, if you actually saw that material, you would hardly be able to see it with the naked eye. So even if it breaks after being implanted in the brain... While

the surface appearance isn't necessarily a problem , it's completely unacceptable clinically.

Firstly, and secondly, the overall device lacks scalability.

I might only have 10 sensors per device, but I need to implant 1000 or 10000 electrodes, or 100 or 1000 brain probes.

This is why Neuralink places such importance on its surgical robot— because it requires implanting a vast number of brain probes to achieve the desired number of channels.

The reason is the same: if you use a very rigid material to make it very thin and then create a brain-computer interface , the number of channels on each brain probe is very limited . If you want to expand the number of channels

. If you want to expand the number of channels ... Integrating more materials and devices

... Integrating more materials and devices into a single brain probe causes it to harden, and once it hardens, all the problems reappear.

Therefore, truly solving this problem requires developing truly flexible electronic materials.

Professor Liu Jia told us that he spent a lot of time researching this issue at Stanford and they actually found a special photoresist.

You can imagine our material as a kind of rubber; when you stretch it , it's elastic , like a chewy candy . Yes, our brain is also an elastic material.

. Yes, our brain is also an elastic material.

Our brain doesn't expand when we breathe, so tearing it further highlights the softness of our material.

Because our material is flexible and elastic , and its surface is very smooth , you won't feel any cutting sensation when it scrapes your hand.

After these comparisons, we believe our material is more suitable for brain-computer interfaces than the polycyanide material commonly used in the market . It is precisely because of this that we used this material as a basis

. It is precisely because of this that we used this material as a basis to create the softest brain-computer interface currently on the market.

It is very soft and still performs like a normal electronic material. Why can't

traditional very soft materials be used for photolithography and brain-computer interfaces?

Because when a material is very soft...

You can imagine why it's so soft . It's actually called a glass transition temperature, which is below room temperature.

. It's actually called a glass transition temperature, which is below room temperature.

You can imagine that at room temperature , it's a cross-linked liquid , or rather, a macroporous cross-linked material. It's easily

cross-linked material. It's easily swollen by the organic or aqueous solvents used in multilayer photolithography , preventing the achievement of highly precise photolithography.

Furthermore, the salts and ions in our body fluids easily corrode it, gradually causing the electronic devices to fail.

What we ultimately achieved was finding a chemical solution to this problem.

We needed an elastomer or a soft material that could be used for photolithography , overcome all aqueous or oily solvents , and prevent ion corrosion.

Chemically speaking, only perfluorinated elastomers can achieve this.

Therefore, we developed a series of perfluorinated elastomer photoresists , solving the problem.

After solving the material problem, new challenges arose.

Signal volume exploded; the previous few hundred channels became millions of data points—a terrifying amount . Algorithms and chips had to be redesigned,

. Algorithms and chips had to be redesigned, and this is where AI comes in.

Yes, AI is completely changing the game in this field.

If electrodes and chips are the "body" of brain-computer interfaces, then AI... This is its "soul."

then AI... This is its "soul."

Without AI, there would be no miracles like "typing with thoughts" or "speaking with brainwaves" that you see today . Imagine scientific experiments ten years ago: researchers could only listen to a maximum of twenty neurons simultaneously— a pitifully small number of brain signals.

They could slowly analyze and manually label them, like a craftsman polishing a piece of jewelry.

But now, tens of thousands of neurons are firing signals simultaneously.

Neuralink can monitor thousands of channels, which is already the most in academia.

Each channel outputs tens of thousands of data points per second, and the entire system is a gigabyte-level "data tsunami" per second . At this scale, it

. At this scale, it is simply impossible for humans to read the signals.

It's like asking you to count ten thousand grains of sand in one second.

At this point, AI is no longer just "helping out," but the key to whether the brain-computer interface can function.

But the role of AI goes far beyond helping to process data.

The ultimate task of the brain-computer interface is actually an extremely difficult "translation" problem: translating the brain's electrical signals into actions or language that humans can understand.

Machines can translate Chinese into English and look up words in a dictionary and read grammar .

But the input of a brain-computer interface is thousands of neurons firing chaotically, while the output may be a robotic hand lifting a cup or a complete sentence.

This is like translating an "alien language" that no one can understand.

But AI... Deep learning, in particular , excels at this kind of task.

In 2015, a team from UC Berkeley and UC San Francisco achieved a breakthrough : a woman who had lost her speech for 18 years due to a stroke regained her ability to speak via a brain-computer interface.

Scientists implanted 256 electrodes in the language region of her brain.

AI translated the electrical signals in her brain into the smallest unit of speech, phonemes, in real time with an accuracy rate exceeding 90%.

Combined with large language models and N-Gram models , the system understood context and automatically corrected errors.

The final step involved training a speech clone version using her past recordings, allowing her to read aloud what she wanted to say in her own voice.

The result?

An overall accuracy rate of 97%. With a delay of less than a second, she can chat with her family at home without needing an engineer to debug it.

This sounds like something out of a science fiction movie , but AI has truly made it a reality.

More importantly, I think everyone is increasingly realizing , especially in this AI era , that brain-computer interfaces are inseparable from AI because there are two aspects to this.

First , the new brain-computer interface technology has caused a dramatic increase in neural bandwidth—the measurement of neurons —requiring a large number of AI algorithms to decode it.

These AI algorithms are now ready.

Second, understanding the brain itself is a crucial aspect for further improving AI in the future.

Therefore, you can see that everyone working on AI-related projects will inevitably... The idea

of ​​moving into the field of brain-computer interfaces is more about technological layout and competition for future technological dominance.

If "reading" brain signals is inseparable from AI, then "writing" brain signals , or neuromodulation, requires even more AI assistance.

Neuromodulation , simply put, involves using electrical stimulation to influence brain activity.

The brain is like a vast symphony orchestra, with thousands of neurons each playing a complex symphony.

Can you imagine using a sledgehammer to "conduct" a symphony?

This is the problem with traditional technology— too crude and cumbersome.

Now, with the number of electrodes skyrocketing to thousands of channels, precise neuromodulation has finally become possible.

However, to make thousands of electrodes coordinate like musicians, a sufficiently intelligent "conductor" —AI—is needed. Agent

—AI—is needed. Agent AI doesn't need to fully understand why the brain works the way it does.

It relies on reinforcement learning to continuously try different stimuli, observe the brain's response , and then adjust its strategy based on the results. It can

adjust the frequency, intensity, duration, and even which electrodes should work simultaneously within milliseconds, thus customizing the most suitable stimulation plan for each individual.

This is somewhat similar to AlphaGo playing Go.

The brain is a black box, and the AI ​​is also a black box , but the two black boxes kept playing against each other and eventually figured out the optimal solution.

So why is AI so crucial for brain-computer interfaces?

To summarize, there are four points: First, the data volume is exploding.

In the past, we only looked at a few dozen neurons; now we have to process tens of thousands simultaneously , which is impossible for humans . Second, the real-time requirements are high.

. Second, the real-time requirements are high.

The latency from "thinking" to "doing" must be less than 100 milliseconds , otherwise the experience will be unnatural.

Third, there are significant individual differences.

Everyone's brain signals are different, and AI can automatically learn and adapt.

Fourth, the brain is a dynamic system ; its signals change over time.

AI can continuously learn and correct itself to keep performance consistently online.

In short, AI is no longer just an "assistant" to brain-computer interfaces, but the "core brain" that determines whether they can truly be implemented.

Without AI, this dream of "human-machine symbiosis" can only remain in the laboratory.

Having discussed AI... Let's take a look at which companies are currently leading the brain-computer interface (BCI) field.

First and foremost is Neuralink, founded by Elon Musk in 2016. This company

has truly brought BCI into the public eye, becoming a "celebrity" in the field.

Neuralink's system consists of three parts: the N1 implantable chip, the Thread flexible electrodes , and the R1 surgical robot.

First, the N1 implantable chip , about the size of the brain and roughly the size of a coin, is implanted under the skull.

The chip integrates 1024 signal acquisition channels and can record the activity of hundreds or even thousands of neurons simultaneously, making it one of the implantable devices with the most channels currently available.

Wireless transmission and inductive charging are the neural center of the system.

Neuralink can record signals from 1000-2000 neurons, which I think is acceptable.

However, in terms of quantity, 2000 neurons are negligible compared to 86 billion neurons . Therefore, I say it's a local signal decoding.

. Therefore, I say it's a local signal decoding.

From a spatial perspective, I can also give you some data: the area where Musk's Neuralink electrodes are inserted on the surface of the brain region is 1.3/1000 of the total brain surface area. 999/1000 of the entire brain surface area has not yet been recorded.

But in terms of timing, the electrical signals are quite good because now the electrodes are in direct contact with neurons.

The electrodes can record the neuronal firing at contact time, within 10 microseconds. It can

collect many spikes (neuronal firings) per second and remember the neuronal firing signals.

You can understand electrical brain... The key feature of the machine interface is that the timing of the electrical signal is very close to real-time, but its spatial coverage is only 1.3/1000.

This is only the surface and does not include the interior of the brain.

There are many functional areas inside the brain, which it can never penetrate.

It can only insert 3 millimeters into the surface.

The brain is 8 centimeters thick.

If you want to insert it into the deep brain, it needs to be 8 centimeters.

Secondly, the Thread flexible electrode is one of Neuralink's core innovations.

Traditional electrodes are rigid.

Using traditional electrodes is like inserting a steel needle into tofu.

Over time, it will cause damage.

Neuralink's Thread electrode is soft and flexible.

It "breathes" with the brain.

The wire is only 5 micrometers wide and can conform to the tiny movements of brain tissue.

Each wire has 32 recording points.

The R1 robot will precisely implant 64 Threads several millimeters deep into the cortex to keep them stable in the micro-movements of brain tissue.

Finally, there is the R1 surgical robot, which is probably the coolest part of the whole system.

Musk calls it a "sewing machine".

Because of the extremely thin wires and the dense network of blood vessels in the brain, Neuralink developed the R1 robot to perform the implantation of this "sewing machine-like" device.

It can implant approximately six wires per minute, automatically identifying and avoiding blood vessels to complete the surgery in just a few hours.

This combination makes Neuralink one of the most densely packed and wirelessly integrated invasive brain-computer interfaces currently available . In 2024, Neuralink completed its first human implantation

currently available . In 2024, Neuralink completed its first human implantation in Noland, the patient we mentioned at the beginning. Arbaugh's

surgery was a success.

A few weeks later, Noland was able to control a computer cursor with his thoughts , even setting a world record with an information transmission rate of 8 bits per second.

This was the fastest brain-computer interface at the time, as everyone has witnessed.

By 2025, Neuralink had completed 13 human implantations.

A recently released 20-minute video of a complete surgery by Neuralink shows how they implanted the chip into the human brain.

This release contains a wealth of information, so let's see how it was done.

The core of the entire surgery is a sophisticated surgical robot.

The doctor first makes a hole the size of a coin in the patient's skull.

Then, this robot has a needle thinner than a red blood cell.

It picks up more than 100 flexible electrode wires, thinner than a human hair, and "sews" them one by one into the cerebral cortex.

This sounds simple , but it is actually extremely difficult.

The human brain is as soft as tofu, as we just mentioned , and it is densely packed with blood vessels.

These 128 needles must not pierce any blood vessels.

And what is most impressive is the increase in speed . The previous version of the robot took 17 seconds to insert a needle

. The previous version of the robot took 17 seconds to insert a needle , but now it only takes 1.5 seconds.

This means the entire surgery time is significantly shortened, and the patient's risk is reduced accordingly.

To achieve this, Neuralink used six microscopes and OCT scanning technology to form a super-precise "eye." This system can track the movement of brain tissue at the millisecond level, predict the path in advance, and accurately avoid all blood vessels.

Currently, 13 people have had this device implanted, and the usage time has exceeded 15,000 hours.

In addition to controlling electronic signals by thought, the latest demonstration also showed that a robotic arm can be controlled by thought.

The current chip is only 4 millimeters deep, and it has already achieved these functions.

Neuralink's next goal is to implant it more than 50 millimeters deep, so as to reach deeper areas of the brain and achieve more complex functions.

Currently, more than 10,000 people are waiting in line for this surgery.

The manufacturing cost of the needle clamp has been reduced by 95%.

The previously expensive experimental components can now be mass-produced.

The robot can also be adapted to 99% of the global population.

All of these are preparing for large-scale application. Neuralink is

not just developing experimental medical devices ; they are building a system that can be scaled up and replicated, from surgical robots and implanted chips to neural signal decoding algorithms. Musk 's dream remains grand: he says that any device that can be controlled by a computer or mobile phone will one day be controlled by Neuralink.

This is a profound breakthrough that will fundamentally change the meaning of "being human."

Of course, this technology still faces many challenges, including ethical issues, privacy concerns , and long-term security.

However, from a technological perspective, brain-computer interfaces are moving from science fiction to reality, and Neuralink's progress may be a key step towards commercialization in this field .

In its future plans, Neuralink hopes to move from "restoring function" to "enhancing function."

The company is developing a "speech decoding" project , attempting to implant electrodes in the speech cortex to directly convert "what you want to say" into text or speech.

Another research project called Blindsight has received FDA Breakthrough Device designation. The certification

goal is to restore partial visual perception in blind people by stimulating the visual cortex . Musk even envisions

. Musk even envisions a future where this vision might perceive infrared or ultraviolet wavelengths.

More ambitiously, Musk's vision for the future is that by 2028, the number of electrodes in the implanted device is expected to exceed 25,000 , enabling interaction with deeper brain regions and further exploring integration with AI systems. The longer-term goal is to allow paralyzed patients to control Tesla's humanoid robot Optimus through a brain-computer interface , thus regaining mobility in the physical world.

This scenario sounds familiar; isn't that what Avatar was like?

Because with their technology, they double the number of channels every two years.

This is a concept we discussed in the industry when working on electrical brain-computer interfaces, known as Moore's Law —basically, the number of electrodes we can insert into the brain doubles every 18 months.

This is achievable , but it depends on whether it's one device or several implanted; it might be a combination . Currently, electrodes can only be inserted 3-5 millimeters into the brain.

. Currently, electrodes can only be inserted 3-5 millimeters into the brain.

They are indeed planning to work on the next generation of electrodes.

Electrodes that can be inserted 6-7 centimeters deep into the brain nuclei can be doubled in a year and a half.

I think doubling in a year and a half is a good time because the requirements for doubling are very high.

The chip needs to be expanded, its communication capabilities, its computing power, and its size reduced, but heat dissipation also needs to be controlled.

It's like a bucket; every weak link needs to be doubled for the whole to double.

Speaking of which, I think doubling in a year and a half is already very good.

Although these ideas are still in the research and development stage , Neuralink has already received strong support from the capital market . In June 2025,

. In June 2025, the company completed a $650 million funding round with a valuation of approximately $9 billion.

However, Neuralink also faces challenges , including the first patient experiencing 85% electrode signal withdrawal.

Although officially attributed to surgical issues , This also exposed potential problems with the technology's stability.

Secondly, the FDA approval process was extremely difficult ; it was rejected at least six times and investigated for ethical issues related to animal testing.

There were also controversies surrounding its technical approach.

It uses a "flexible" solution that thins rigid materials instead of truly soft materials, resulting in issues such as material fragility and difficulty in removal.

However regardless Neuralink as the most noteworthy company in this field , has recently made very rapid progress.

Another company worth noting in this industry is Synchron.

If Musk's Neuralink is a hardcore player in "skull surgery and electrode insertion," Synchron takes a completely different path.

It enters the brain through the blood vessels without surgery, without robots , and without a long recovery period.

This company was founded in 2012 by Tom... Oxley

by Tom... Oxley

, a physician specializing in cerebrovascular surgery, founded a company whose core philosophy is to "unlock the brain's natural highways," specifically the vascular system.

Why?

Because Synchron chose a completely different technological approach: implanting the device into a blood vessel without opening the skull.

Synchron's device, called Stentrode, resembles a heart stent.

Doctors only need to make a small incision in the jugular vein to "send" it along the blood vessel to the brain, eventually placing it on the blood vessel wall near the motor cortex.

The device then unfolds and stabilizes like a stent.

The entire process takes only about two hours, and the patient can be discharged the next day.

Stentrode has 16 electrodes attached to the blood vessel wall , which is almost flush with the brain surface.

Therefore, the electrodes can record the electrical signals of nearby neurons.

These signals are then wirelessly transmitted to a computer via a small device in the chest. In

2020, Synchron completed the world's first human implantation surgery, four years earlier than Neuralink.

Several ALS patients used it to send emails, browse the web , and even typed the first tweet in human history using their brain: "Hello World."

Synchron has also received "Breakthrough Device" designation from the US FDA , becoming the first brain-computer interface company approved to conduct long-term implantation trials in the US.

As of 2025, Synchron had completed implantation tests on multiple patients, raising approximately $345 million in funding.

Its investor lineup includes several billionaires, such as the Bezos family foundation and the Gates Foundation.

Interestingly, they have begun compatibility testing with Apple devices, allowing patients to operate iPads or Vision Pro with their thoughts . Although Synchron only has 16 channels

. Although Synchron only has 16 channels , fewer than Neuralink's thousands , its approach is more "realistic" and focused, aiming to enable paralyzed individuals to truly communicate, express themselves, and live their lives.

This company may well become the first to bring brain-computer interfaces into hospitals and into the lives of ordinary people.

The third company is Paradromics.

If Synchron pursues "safety and practicality," then Paradromics pursues "performance limits."

Founded in 2015 in Austin, Texas, the company was founded by Matt... Angle

, a neuroscience geek from Stanford, had a simple idea: to achieve truly seamless communication between the brain and computers , the data channels needed to be wider and faster.

So, they turned the incurable problem of the brain into a solvable technological problem , aiming to surpass all competitors in performance and create a "superhighway for the brain."

Paradromics' research device, Argo, is most notable for its ability to simultaneously record 65,536 channels of neural signals —64 times that of Neuralink . Argo uses a microwire electrode array,

. Argo uses a microwire electrode array, making it arguably the world's highest-density invasive BCI device, capable of simultaneously recording the activity of tens of thousands of neurons , essentially implanting tens of thousands of "listeners" inside the brain.

Why so many channels?

Because certain brain functions, especially language, involve the coordinated work of multiple brain regions.

Therefore, achieving fluent speech decoding requires recording the activity of a large number of neurons.

It's important to note that this isn't the chip currently implanted in the human body , but an experimental system used to demonstrate their ability to collect an astonishing amount of brain signals . Their clinically-oriented product, Connexus,

. Their clinically-oriented product, Connexus, aims to achieve "high-speed, high-density" data transmission, allowing machines to more accurately understand what we want to say or do.

Their concept is quite apt; past brain-computer interfaces were like "dial-up internet" —usable but slow.

Paradromics' goal is to achieve... "Fiber optic cable into the brain" allows signals to flow as smoothly as broadband, especially in complex brain activities such as language and thought.

The more channels and the faster the signal, the more accurate the AI ​​decoding. In

June 2025, Paradromics completed its first human trial at the University of Michigan, temporarily placing a Connexus device in the brain of an epilepsy surgery patient and successfully recording neural signals . The entire surgery took only twenty minutes.

. The entire surgery took only twenty minutes.

This was a short-term trial, and the device was subsequently removed , but it marked their entry into the clinical stage.

Paradromics has obtained "Breakthrough Medical Device" designation from the US FDA and has been included in the TAP Accelerator Program, a program that opens a "fast track" for innovative medical devices.

In terms of funding, the company completed a $33 million Series A funding round in 2023, bringing its total funding to nearly $100 million.

The company's core advantage lies in its ability to record more neurons , making it more suitable for complex tasks such as natural language processing, laying the foundation for future cognitive enhancement applications.

Another company worth mentioning is Blackrock.

If Neuralink is the "celebrity" of the brain-computer interface world, then Blackrock Neurotech is the industry's "veteran."

Incidentally, this has no connection to the financial company Blackrock.

Blackrock Neurotech was founded in 2008.

Its predecessor was Blackrock Microsystems, headquartered in Salt Lake City, Utah, holding a position equivalent to Intel in the field of brain-computer interfaces.

Why?

Because Blackrock's Utah Array electrode array is almost the "gold standard" for invasive brain-computer interface research.

Most brain-computer interface research worldwide uses Blackrock's devices and platforms. The Utah Array was invented in the 1990s by Professor Richard Normann of the University of Utah and received FDA approval in 2005 , becoming the first approved high-channel-count implantable electrode.

From 2004 to the present, almost all invasive BCI human studies have used it.

Those news reports of "paralyzed people controlling robotic arms with their thoughts, and aphasic people regaining their speech" likely all have the Utah Array behind them.

It has 96 silicon needle-like electrodes, with up to 1024 channels per system, capable of recording the peak discharge of a single neuron, resulting in an extremely high signal-to-noise ratio.

Long-term safety is Blackrock's greatest advantage, with the longest implantation time exceeding 8 years and currently in normal use, accumulating over 30,000 days of implantation time with zero FDA reports of serious adverse events.

In comparison, Neuralink's longest implantation time is less than 2 years , and it is currently widely used in clinical applications .

Blackrock's Utah... Array

has made breakthroughs in multiple fields, including motor function recovery, such as controlling a robotic arm to grasp objects , operating a wheelchair, and restoring basic movements like shaking hands . It can also allow paralyzed individuals to feel the touch of the robotic hand.

. It can also allow paralyzed individuals to feel the touch of the robotic hand.

In a 2017 Stanford University study, patients used Utah... Array

achieved a typing speed of 40 characters per minute, setting a world record for brain-computer interfaces at the time.

In November 2021, Blackrock's MoveAgain system received FDA Breakthrough Device designation.

The system consists of three parts: an electrode array implanted in the brain, an AI algorithm for decoding motor intentions , and a wireless transmission module.

The design goal is to allow paralyzed patients to control a mouse, keyboard, wheelchair, prosthesis, etc. with their thoughts . In 2022, Blackrock released its next-generation technology concept, Neuralace , a flexible electrode with an ultra-high channel count, aiming to achieve whole-brain data capture.

As an industry leader, Blackrock has over 2,000 research papers using Blackrock devices, more than 1,000 research institutions as customers, and has published 116 peer-reviewed journal articles.

This ecosystem advantage is difficult for Neuralink and Paradromics to replicate in the short term.

If the above companies are industry veterans who have been preparing for a long time , some rising startups have also chosen the non-invasive route.

Just this year (2025), OpenAI CEO Sam Altman founded a company called Merge Labs' brain-computer interface company is aiming to raise $250 million at a valuation of approximately $850 million.

It explores a solution combining gene therapy and ultrasound technology , modifying neurons to respond to ultrasound signals to achieve a less invasive interface, or one different from traditional electrode implantation.

Meanwhile, Nudge, a non-invasive brain interface company founded by Coinbase co-founder Fred Ehrsam, announced the completion of a $100 million Series A funding round.

Forest Neurotech, which uses ultrasound BCI projects, received $14 million in funding from former Google CEO Eric Schmidt . Its goal is to use ultrasound to

. Its goal is to use ultrasound to sense and modulate brain activity without craniotomy or implanted electrodes.

Additionally, there is SPIRE, a brain technology company pursuing a "no-surgical" approach.

Therapeutics has developed a device called Diadem, which currently resembles using focused ultrasound to deliver sound waves deep into the brain without craniotomy or electrode implantation.

Its target audience is patients with chronic pain or depression —those facing problems that "medication and treatment cannot solve."

So, to summarize, Neuralink is betting on human-computer interaction with Musk's ambition and resources ; Synchron is betting on safe use with its ingenious vascular implantation approach; Paradromics is betting on technological excellence with extreme performance; Blackrock is betting on steady progress with its 20 years of experience; and these Silicon Valley upstarts are betting on non-invasive methods, forging a different path.

Which one do you think will succeed the most?

We've now summarized the overview of the 2025 brain-computer interface race.

A high-stakes gamble on the future of humanity is unfolding in Silicon Valley and globally.

However, technological progress is always accompanied by ethical challenges, especially for brain-computer interfaces, as they directly touch the most core part of the human body—the brain.

Advances in brain-computer interfaces are forcing us to rethink the meaning of being human.

They allow thoughts to be read, emotions to be perceived, and behaviors to be controlled by external systems . This has blurred the lines between privacy, will, and identity

. This has blurred the lines between privacy, will, and identity like never before.

Who owns our brainwave data?

When machines can influence emotions or decisions, will our autonomous will remain intact?

Will those who can afford to enhance their brains gain an asymmetric advantage in the future?

Due to time constraints, we won't discuss these questions here , but if you're interested, we can continue our in-depth analysis of brain-computer interfaces.

Who will win may not be important, because the significance of brain-computer interfaces lies not only in controlling machines with thoughts , but also in reconnecting severed pathways, allowing the paralyzed to move again, the aphasic to express themselves again , and potentially helping the blind regain their sight.

For ordinary people, brain-computer interfaces also represent a new direction ; perhaps one day, thought will be digitally perceived and transmitted like touch , and the boundary between humans and machines is becoming more blurred than ever before.

Where will brain-computer interfaces lead us?

Perhaps we will see the answer soon.

That's all for this video.

Don't forget to follow our account and don't miss updates.

Your likes, comments, and shares are the best motivation for Silicon Valley 101 to create in-depth technology and business content.

I'm Chen Qian, see you in the next video! Bye!