Brainlet

A brainlet is a hardware neural network compatible with the human brain. Brainlets are the standard memory and processing unit used in modern Midorian technology.

Brainlet types
Brainlet manufacturing can be of three types:
 * 1) Biochemical (wet) brainlets. These brainlets are based on artificial neurons created with patented processes. Their optoelectronic transceivers allow them to work using light pulses, and operate at a fraction of light speed, but are twice as fast as carbon-based photonic brainlets. Advantage over other brainlet types are their biocompatible energy requirements and a simulated one-write-per-read capability due to their nucleic photonic cache. Disadvantages include the need for a biochemical feeding system, required rest conditions to perform neural connections (the other half of the write cycle), and failure due to write-exhaustion in certain conditions. The android behavioral module uses this kind of biochemical brainlets. Due to the non-zero probability of brainlet exhaustion caused by high-frequency consecutive write cycles, it is possible for androids to become berserk when the behavioral module becomes overloaded. Most biochemical brainlets include a self-diagnosis system that regulates write speed and slows down processing when oxygen and glucose intake is limited. But brainlet failure is non-catastrophic most of the time, meaning that with certain rest and repairing intake, a brainlet can be restored to its former operative state.
 * 2) Rare-Earth photonic (solid-state) brainlets. These brainlets operate at near light-speed (regulated through a software-configurable photonic clock), but write cycles are (as usual) limited by energy sources. Advantages over biochemical brainlets are unlimited write capacity and near-zero (0.00000001) probability of failure. Due to their high performance, soldiers are required to have military-grade rare-earth brainlets installed in their neural implants.
 * 3) Carbon-based photonic (solid-state) brainlets. These brainlets operate at a fraction of light-speed (regulated through a software-configurable photonic clock), and they have fourth-write-per-read capability. They're a cheaper alternative for rare-earth brainlets. They're mostly present in cellphones, android brains, and commercial-grade virtual reality pods.

Neuromorphic Computing: The Beginnings
Since the 20th Earth century, neuromorphic electronic circuits were already in use for experimental research, using transistors and memristors, but overheating and power usage limited practical applications. An example was the Neurogrid developed by Stanford bioengineers in Terran year 2014. Unfortunately, the technology was not mature enough, and each neurogrid circuit board costed around 40,000 US dollars. In the following years, more efficient versions of the board were created using the most advanced manufacturing techniques available at the time, creating a million-neuron board for $400 a copy. But the limitations of 2D circuitry and the limited knowledge on biological neurons' learning processes made this applicable only for research.

In the mid 21st century, more research was done and new models showed how theoretical neurons were able to create memories. As more neuromorphic circuits were created, humanity reached a point where the human connectome could finally be implanted in a massive infrastructure, but the technology to scan the human brain hadn't been perfected yet.

In the end of the 21st century, the first human connectome was finally deciphered. By that time, technology was already advanced enough to be able to store a fully functional human brain in a 20-story building. The building was destroyed in a terrorist attack orchestrated by fundamentalist christians, who claimed that replicating a human brain was an act against God. Other projects continued working on the same research until they were cancelled due to the eruption of World War III.

Renaissance of Neural Computing
After World War III ended, the demand for bionic limbs spiked, and neural chips were already cheap enough to guarantee the cheap fabrication of bionic limbs. The discovery of efficient artifiical muscles allowed people to accept cybernetic implants, which resulted in the creation of companies like CyberMotion and Braincorp (which later became Brycorp). At the same time, Space Programs also advanced when it became obvious that living on Earth was not a viable option for the survival of mankind.

First biochemical brainlet
Concerned about the environmental disaster produced by WWIII, scientists tried to rely less on rare earth computing and opted for more renewable technologies. After having proven that neural computers were more efficient at data recognition than conventional ones, new research was focused on integrating traditional data with neural chips.

The first biochemical brainlet was manufactured in the mid 22nd century by a team of Scientists of the MIT Computer Science and Artificial Intelligence Laboratory. The research, sponsored by Braincorp, resulted in the creation of the first array of bioprinted neurons, that could be readable and writable via a tree of artificial synapses, created with carbon nanotubes and quantum dots.

This brainlet, consisting of one billion neurons, was created in very controlled conditions, and required a completely sterile environment. The brainlet access array was around 100 times larger than the original brainlet, and occupied an entire room, but the brainlet could be trained and the first visual experiments rendered satisfying results.

Later Biochemical brainlets
To prevent meddling with known biology, biochemists created their own synthetic neurons, that could only multiply and grow under certain conditions. This also allowed scientists to modify the neurons so that they could easily attach to the electrode. Once the cells are matured, a biochemical signal is sent so that their walls are fortified against external corrosion while maintaining their flexibility and plasticity.

The next few advances in brainlet design were just improvements over the originals, but the process followed an equivalent of "Moore's Law" for connectomics, until cell density was equiparable with a human brain.

Suspension of research
During this time, and with the creation of the first androids, the possibility of a singularity became palpable, and due to increased international pressure (caused mostly by protests and riots caused by anti-AI extremists), the Terran United Nations called for heavy regulation on Artificial Intelligence research. The creation of neural supercomputers was banned, and several research facilities were dismantled across the world.

Photonic brainlets
With the discovery of photonic technology and the invention of the photonic neuristor, manufacturing of solid state brainlets became practical and cost-effective. The only inconvenient of photonic brainlets is the requirement of using rare earth minerals. Due to the ban on mining for rare earth minerals on Earth, mining was outsourced to other planets in the solar system, which limited photonic technology on Earth. After the colonization of Tau Ceti in the first Midorian century, the colonies in Midoria had no problem with controlled mining, and the fact that Midoria's red moon Aka was also rich on rare earth minerals. Eventually, New China began exporting their own rare earth minerals, resulting in the creation of the gray pits. Carbon-based photonic circuits with lower write performance also exist, but have been dismissed by the highly competitive market in Midoria.

Second generation biochemical brainlets
A hybrid type of brainlets using carbon-based photonic receptors was created, resulting in the so-called "Gen-2" biochemical brainlets. These brainlets form their connections based on light pulses rather than electricity, and are at least as efficient for writing as carbon-based brainlets. Due to the requirement of rare earth materials for high efficiency photonic circuits, these new biochemical brainlets are several times less expensive than rare-earth brainlets, even after taking into account brainlet nurturing fluids and their distribution and recycling system. Most Information Technology companies in Midoria use biochemical brainlets. Brainlet exhaustion can also happen when connecting biochemical brainlets with solid-state brainlets due to update frequency disparity. This can result in glitches when using massive computational environments with mixed brainlet types.

VPacks
A full brainlet has a storage capacity of 64 Terabytes for neural connections and neuron states.For ease of use in computers and systems like the Grid, brainlets are kept in packages along with their photonic microprocessors and random access memory. These packages are called VPacks.

SPacks
VPacks of 16 or more brainlets are enough to hold an average human brain. These are called SPacks and are used for military and industrial purposes. Android connectomes require a total of 64 BLs of storage, and they're often stored in a customized arrangement of solid-state brainlets, with a shape resembling a human brain, for skull compatibility.

Brainlet Units (BLU)
VPack and SPack capacity is specified in Brainlet Units (BLUs). Commercially available VPacks have 1/4, 1/8, 1/16 and even 1/256 BLUs. To save space, Brainlet Units are also abbreviated as BL. It's common in chips to specify their fractional size in white tint, with the initials BL appended at the end.

Micro Units (MU)
Pronounced "mew", a Micro Unit is 1/1048576th of a Brainlet Unit.

Virtual Units
In systems like the Grid, virtual objects are stored in what is called a "Virtual Unit", abbreviated VU (pronounced View). Virtual Units do not have a specific size in bytes, they are instead variable in size, and are partitioned on-the-fly as objects are allocated in the grid.

Popular Culture
As an example of brainlets In popular culture, there is Brainy, the official mascot of Babylon Research.