BCI Bin is not designed to be an all-in-one learning platform but instead focuses on guiding you to find the best BCI resources available.

Each unit consists of a resource index on the left and blobs of information on the right that either is direct material from the resources or contains summaries/information about it.

While most websites want to keep you on their page for as long as possible, BCI Bin's goal is for you to frequently leave the site to directly explore the resources that sound the most interesting to you.

While the different units of BCI Bin are numbered, they don't have to be learned in any order, so feel free to hop around.

Lastly, BCI Bin is always looking to improve, so if you have any suggestions while exploring, feel free to message me on X.

BCIs can be broken down into two main categories: sensing and stimulating.

The majority of research in the field of BCIs is focused on sensing BCIs, which are used to read brain activity. These readings can then be processed and used as input for computers or robotic limbs.

On the other hand, stimulating BCIs are used to send signals to the brain. These devices aim to help people who have lost the ability to move, see, or have other neurological disorders that require stimulation. Since most of these devices are still in the early stages of R&D, BCI Bin will primarily focus on sensing BCIs.

For more information on both categories of BCIs, BrainMind provides a visual map that explores their current landscape.

There are two types of sensing BCIs which you'll explore in this unit Invasive and Non-Invasive. Both have their own unique set of tradeoffs.

To understand these tradeoffs, there are a few sets of criteria we'll need to become familiar with. These criteria are: scale, temporal resolution, spatial resolution, and invasiveness.

This section of an online lecture from UC Berkeley provides a high-level overview of the tradeoffs between different types of BCIs.

Invasiveness is the broadest category used to differentiate between different types of BCIs. It refers to whether the BCI device penetrates the brain and by how much.

From exclusively a data collection standpoint, the most invasive BCIs are advantageous. However, more invasive BCIs come with cons such as higher costs, risks of complications during surgery, and risks of the body rejecting the device post-surgery.

The first table from this article from Front Syst Neurosci provides a more in-depth look at the data superiority that invasive BCIs provide. The following subsections in this unit will focus on this comparison as well.

Resolution is a measure of how detailed the data collected by a BCI is. It can be broken down into two types: temporal and spatial resolution.

Temporal resolution is a measure of how long it takes to collect and make use of data from a system. High temporal resolution means the data is usable faster.

Spatial resolution is a measure of how closely data comes to telling you how individual neurons are firing. High spatial resolution means the data is sampled from small areas with high specificity.

Lastly, scale is a measure of how many neurons can be simultaneously recorded by a BCI. Higher scale means more neurons can be recorded at once.

If any of these criteria are unclear, don't worry. The next sections will provide context through examples of each of them.

In his article on Neuralink, author Tim Urban mentions that "[T]he long-term goal is to have all three of your cakes and eat them all. But for now, it's always a question of which one (or two) of these criteria are you willing to completely fail?"

Until a significant breakthrough occurs, tradeoffs will have to be made between scale, resolution, and invasiveness when choosing a BCI type to work with.

The first, and most widely used and researched BCI type is the non-invasive BCI. These BCIs use sensors on or near the scalp to measure the electrical potentials or magnetic fields that are produced by the brain.

This article by NeuroTechEDU has plenty of great visuals and is a well-rounded starting point for learning about all BCI types, with a good emphasis on non-invasive BCIs.

Scale: High | Temporal Resolution: High | Spatial Resolution: Very Low

EEG (Electroencephalogram) devices are the most common non-invasive BCI type by a wide margin. They measure the electrical activity of the brain using electrodes placed on the surface of the scalp.

The underlying cause of the brain's electrical activity is the firing of neurons, a topic which will be discussed in more detail in the next unit. But for now, all you need to know is that when large groups of neurons fire together, they provide enough signal to measure from outside the skull.

The BCI Guys provide a comprehensive introduction to neurotechnology course that covers EEGs as well as nearly every other BCI type mentioned in this unit.

While EEGs are a great entry point into the world of BCIs due to their low cost, portability, and high temporal resolution, (ability to get signals from the brain in real-time), they have the lowest spatial resolution of any BCI type.

This means that while EEGs can provide a lot of data quickly, the data is not very specific, as each electrode measures the average charges of millions to billions of neurons.

On top of that, EEG spatial resolution also suffers from the interference that the scalp, skull, and other intermediary layers of tissue produce.

Scale: High | Temporal Resolution: Medium | Spatial Resolution: Low

From the outside, an fNIRS (functional Near-Infrared Spectroscopy) device may look very similar to an EEG device, however, the way they read brain activity is quite different.

An fNIRS device uses infrared light to measure the amount of oxygenated and deoxygenated blood in areas of the brain. This is done by emitting light into the brain, which passes through the skull and tissue, and measuring the light that is reflected back.

This works due to two factors:

This video from Artinis Medical Systems provides a great introduction and technical deep dive into how fNIRS works, along with a comparison of it against other BCI types.

fNIRS is less susceptible to noise than EEG, giving it a slightly higher spatial resolution, but it still has a long way to go before it can compete with the spatial resolution of invasive BCIs.

Additionally, fNIRS has a lower temporal resolution than EEG, as the changes in blood oxygenation are more delayed when compared to the electrical signals that EEG measures.

While the majority of technical content on BCI Bin will focus on EEGs due to their accessibility (i.e. $), UCLA provides a free 8-part course that covers fNIRS in depth in case you're interested in exploring the technology further.

Scale: High | Temporal Resolution: Low | Spatial Resolution: Medium

The remaining non-invasive BCI types to be discussed are various imaging technologies such as fMRI (functional Magnetic Resonance Imaging), PET (Positron Emission Tomography), and MEG (Magnetoencephalography).

fMRI is an imaging technique that, similar to fNIRS, relies on the fact that cerebral blood flow and neuron activity are coupled. When an area of the brain is in use, blood flow to that region also increases.

It works due to hemoglobin, the oxygen-carrying protein in the blood, having different magnetic properties depending on whether it's oxygenated. This difference allows an MRI machine, essentially a huge magnet combined with a scanner, to detect which brain areas are active.

PET is another imaging technique that measures blood flow in the brain, but instead of using magnetic properties, it uses radioactive tracer materials ("radiotracers") that are injected into the bloodstream before the scan and reach the brain.

These radiotracers attach to glucose molecules, which are the brain's main source of energy. As different areas of the brain use more glucose, the radiotracers will accumulate in those areas, allowing the PET scanner to detect which areas are active.

MEG is an imaging technique that measures the magnetic fields produced by the brain's electrical activity. It uses a massive helmet-like device that contains sensors that can detect these magnetic fields.

If this sounds similar to EEG, that's because it is. The main difference is that MEG offers a much better spatial resolution due to magnetic fields' ability to pass through the scalp and skull, whereas electrical fields used by EEG are volume conducted through these layers, which introduces noise.

While the underlying technologies behind these devices are different, they all share similar tradeoffs:

Due to their high cost, size, and complexity, these devices are only found in research facilities or hospitals. This makes them much less accessible to those who aren't in academia when compared with the other non-invasive BCI types, therefore BCI Bin will focus less on them.

The second, more specialized BCI type is the invasive BCI. The most defining feature of invasive BCIs is that they require surgery to implant the device into or onto the brain.

As previously mentioned, invasive BCIs currently provide the highest quality data of any BCI type, however, the cost of this data is a host of cons that make them less practical outside of specialized applications and research.

As we begin exploring the wide variety of invasive BCIs that exist, it's important to note that these devices represent the cutting edge of neurotechnology. As such, the majority of the resources in this unit will be at a higher difficulty level than the previous unit.

That being said, don't let the advanced nature of these resources intimidate you. Even if you just skim through and get a high-level understanding, being exposed to these cutting-edge topics can spark curiosity and open up new areas of interest.

To start, this article from the MIT Media Lab provides a good overview of the current state of invasive BCIs, of which, most have an in-depth article in the following subsections.

Scale: High | Temporal Resolution: High | Spatial Resolution: Low

ECoG (Electrocorticography) is a type of invasive BCI that sits between the non-invasive and fully invasive BCIs. It involves placing a blanket of electrodes on the surface of the brain, without penetrating the brain tissue.

In this unit, ECoG will be considered an invasive BCI, as it requires surgery to implant the electrodes under the skull. However, due to the relative simplicity and lower risk of the surgery, (deploying the electrode blanket through a small hole in the skull), it is often referred to as a semi-invasive BCI.

That being said, ECoG is a great middle ground between non-invasive and invasive BCIs, as it provides a good balance between the quality of data from invasive BCIs while still being relatively safe due to the minimally invasive surgery required.

The core concept behind ECoG is the same as EEG, using an array of electrodes to measure the electrical activity of the brain, with each electrode measuring the average activity across millions of neurons near the surface of the brain.

While having these electrodes on the surface of the brain significantly reduces noise introduced by the skull and scalp, therefore producing cleaner data, ECoG still has a relatively low spatial resolution due to the large area that each electrode covers.

Today, ECoG is rarely used outside of research and specialized medical applications. Due to this, BCI Bin will not focus on it. However, if you're interested in learning more, this academic article from NeuroImage provides a great overview of the current state of ECoG research.

Scale: Low | Temporal & Spatial Resolution: High

Shallow implants are an invasive BCI that are placed just below the surface of the brain. They are used to record the electrical activity at a higher resolution than ECoG but at a lower scale.

This paper from Suzhou Institute of Nano-Tech and Nano-Bionics provides an excellent comparison of a variety of both these shallow implants, as well as some depth implants which will be discussed in the next subsection.

One of the most well-known shallow implants is Neuralink's wireless N1 device. Its BCI type is considered a flexible microwire array and its original white paper was published in 2019.

Since then, Neuralink has made significant progress in the development of the N1 device, however, minimal technical details have been released to the public.

Fortunately, there are many deep dives that individuals have made on the limited information (blogs and patents) that has been released. This one from Mikael Haji is a great starting point.

The core concept that Neuralink's microwire array leverages, as well as all other shallow implants, is called Local Field Potential (LFP).

LFP works by picking up the average of the electrical charges from all of the neurons within a small radius of the electrode. This allows for a much higher spatial resolution but at the cost of an extreme reduction in scale.

This video from Neuromatch Academy offers a quick technical introduction to how LFPs work.

While the flexible electrode array associated with Neurlink's N1 device is the most publicly well-known shallow implant, many other shallow implants are currently being researched. These include, but are not limited to, Utah Arrays and distributed free-floating implants.

The Utah Array, invented at the University of Utah in the 80s, consists of a grid of probes with rounded electrode tips, inserted into the surface of the brain.

Since the creation of the Utah Array, other types of probes have been developed that allow for more electrodes to be placed per individual probe.

The most known of these is the Michigan Probe. This paper from the University of Michigan, where the probe was invented, provides an analysis of the current state of both the Utah array and Michigan Probe.

The University of Chinese Academy of Sciences provides a paper that gives an in-depth view of the manufacturing process behind multi-electrode probes, such as the Michigan Probe, and a great analysis of the state of the aforementioned flexible electrode array.

The next shallow implant, flexible electrode arrays, are a very broad category of implants that are currently at the forefront of BCI research.

The core idea behind them is that brain tissue is very soft, and traditional electrode materials are many orders of magnitude stiffer than the brain. This mismatch in properties can easily cause brain damage over time.

To combat this, natural materials such as flexible polymers are used to create probes and electrodes that are inserted into the brain. Depending on the desired region of neural activity to read, these electrodes can be inserted into the brain at shallow or deep levels.

The final type of shallow implant that will be discussed in this unit is distributed free-floating implants.

These implants are a relatively new concept that involves placing a large number of wireless, isolated, flexible electrodes into the brain.

The electrodes are then able to float freely in the brain, allowing for a higher scale than traditional implants while still maintaining the high spatial resolution that invasive BCIs provide.

As mentioned previously, all these shallow implants rely on listening to LFP signals. This paper from MIT and Boston University provides an overview as to why these are the optimal signals to record.

Scale: Low | Temporal & Spatial Resolution: High

Very similar to shallow implants, depth implants are devices that are inserted into the brain, but instead of sitting near the surface, they are inserted into deeper layers of the brain.

They're implanted through the cortex and sit at different depths based on both the length of the probe and the region of the brain they're targeting, (e.g. hippocampus, thalamus, corpus callosum). The details of these areas will be covered in the next unit of BCI Bin.

Both Michigan Probes and flexible electrode arrays can take the form of shallow and depth implants. In this subsection, we'll look at some implants that are exclusively used for depth.

One of these implants is SEEG (Stereo-electroencephalography) probes, which are hollow cylindrical electrodes used to record the electrical activity of the brain in patients with epilepsy. This paper from the University of Freiburg provides an overview of how and why SEEG probes are used.

Similar to a Michigan Probe, the electrodes on the SEEG probe are placed along the length of the cylinder and record LFP signals.

This paper from Seoul National University provides an overview of the current state of microfabricated probes, including the cylindrical probes used for SEEG.

Another type of depth implant, which is currently highly experimental, is the syringe-injected mesh implant.

This implant is made up of a mesh of electrodes that are injected into deeper levels of the brain using a syringe. The mesh then unfolds and expands, allowing for a large number of electrodes to be placed in areas that are otherwise difficult to reach.

Due to the novelty of this technology, there are very few resources available on it. However, if you're interested in learning more about it, this paper from Harvard University provides an overview of the technology and its potential applications.

Scale: Low | Temporal Resolution: High | Spatial Resolution: Medium

One last invasive BCI that didn't fit into the previous categories is the endovascular implant.

Endovascular BCIs surgically enter the brain through its blood vessels and use a stent to press electrodes against the walls of the vessels to record activity.

As of today, the only endovascular BCI that has been implanted in humans is the Synchron Stentrode. This paper from the Journal of NeuroInterventional Surgery provides an overview of the device and its initial testing results.

Like Neuralink, Synchron is a for-profit organization, and as such, they have not released much technical information about the Stentrode. As more information is released on the Stentrode and other endovascular implants, this section will be updated.

That's all for the first unit of BCI Bin. If you have any questions or feedback, please let me know by sending me a message on X.