How it works
A BPU brings living neural networks, electronics, and software together into a single computing system.
Digital information is translated into signals neurons can understand. The neurons process and respond to those signals, and their activity is translated back into digital information.
- Information
- Interface
- Living neural network
- Neural response
- Software
Components
Four systems, one device.
-
The biological network
Living human neurons form interconnected biological neural networks that act as the computational core of the BPU.
-
The neural interface
A dense array of electrodes allows the system to communicate with the neurons, sending electrical signals into the network and recording its responses.
-
The life support system
The biological network is maintained in a carefully controlled environment that provides the temperature, nutrients, gases, and conditions neurons need to remain healthy and active.
-
The digital system
Electronics and software translate between digital information and neural activity in real time, making the biological network accessible to conventional computing systems.
Growing the network
Neurons are grown, not harvested.
The process begins with human stem cells, undifferentiated cells that carry the potential to become almost any cell type in the body. Exposed to a defined sequence of signalling molecules, they are guided down a single developmental path until they mature into neurons, which are then cultured onto the interface where they wire themselves into a connected network.
-
01
Stem cell
Undifferentiated, and in principle able to become almost any cell type in the body.
-
02
Direction
A defined sequence of signalling molecules commits the cell to a neural path.
-
03
Maturation
The cell becomes a neuron, growing the processes it needs to send and receive signals.
-
04
Network
Neurons connect to one another, forming the biological neural network that performs the computation.
One computation
From data to neural activity, and back.
- Encode
- Stimulate
- Compute
- Read
- Decode
-
01
Encode
Digital information is converted into patterns of electrical stimulation.
-
02
Stimulate
Electrodes deliver those patterns to selected parts of the biological neural network.
-
03
Compute
The neurons respond collectively, producing patterns of activity shaped by the network's existing connections and state.
-
04
Read
Electrodes record the resulting neural activity.
-
05
Decode
Software interprets that activity and converts it into a usable digital output.
Learning inside a BPU
The network changes as it learns.
A BPU does not simply pass information through a fixed biological circuit. During training, the neurons repeatedly receive inputs and feedback about their responses.
That feedback changes the biological network itself. Connections between neurons strengthen, weaken, and reorganize as the network learns the task. Over time, the same input can produce a better response because the physical network performing the computation has changed.
The processor and the model are intertwined.
Conventional computing
Separate layers. The hardware is unchanged by what it runs.
A BPU
One substrate. Learning happens as physical change within the biology doing the computing.
The training loop
Each cycle gives the network another opportunity to adapt.
Once trained, the network can use what it has learned to respond to new inputs, and continue adapting as conditions change.
Biology, accessible through software.
The BPU handles the complexity of maintaining and communicating with the biological network.
To the systems around it, the BPU becomes another computational resource: information goes in, computation happens, and information comes back out.