The Cell Matrix

Introduction:
The Cell Matrix is a novel kind of microprocessor chip. Its key feature is that it can implement self-configuring, self-repairing electronic circuits. It is also capable of massively parallel computation. Both of these features are in common with the architecture employed biologically in the brain. Could the Cell Matrix be a candidate architecture for future artificial brains?

Massively parallel:
As its name suggests, the Cell Matrix consists solely of a large array of tiny cells. Each cell is capable of performing only simple logic computations. However, due to the massive number of cells per processor (millions, billions, or even trillions of cells), the processor as a whole is extremely powerful and capable of performing inherently parallel algorithms very very quickly. What's more, due to its simplicity, there is no limit to the size of the processor. Its parallelism is infinitely scalable.

Self-configurable:
The Cell Matrix is also highly re-configurable. Each cell can be re-programmed to perform some required combinatorial logic. Re-configurable processors (FPGA's) are nothing new. The major difference with the Cell Matrix however is that it is able to reconfigure itself. Each and every cell, as well as being able to perform simple logic calculations, is also able to both reprogram, and be reprogrammed, by neighbouring cells. Thus circuits within the matrix can dynamically grow and modify themselves, and all during runtime.

How does it work?
The implementation of the Cell Matrix could vary. For example the shape of each cell can be a flat hexagon, or a 3-Dimensional cube. The standard implementation, however, consists of identical square cells where each cell is in contact with each of its four neighbours. Each side of a cell is connected to the neighbouring cell via two input and two output lines (see diagram).

The 'D' lines are known as the data lines. During standard operation of the cell (data mode) the D output lines are calculated based on the state of the 'D' input lines. The calculation performed is specified in a 16x8 truth table held within the cell. Thus each cell can behave like the standard logic gates AND, OR, XOR etc.

The 'C' lines, known as the configure lines, are responsible for the programming of the cell's truth table (and hence the cell's logic properties). If any one of the C input lines is high, then that cell will be in configure mode. This means that it can now be configured by the neighbouring cell which is asserting the config line. The cell is programmed by shifting bits in its truth table. With each tick of the chip wide clock the state of the cell's relevant D input line is shifted into the "lowest" bit of its truth table whilst the "highest" bit is shifted out on the corresponding D output line. Thus, because the truth table is of dimension 16x8, it takes 128 clock ticks to completely reprogram a cell.

Current Status:
The Cell Matrix has so far only been implemented in hardware on a small scale, i.e. a matrix of 3x4 cells. However extensive experimentation has been performed in simulation. Simulated arrays of many thousands of cells have been used to implement the circuits including: cell replicators, a wire builder, a self-replicating space filler, and genetic algorithms.

NASA Space Probes:
In December 2000 Cell Matrix Corporation was awarded up to $70k in funding from NASA. This is a 6 month research contract to look into possible application of self-healing circuits to deep-space probes. In the harsh environment of deep-space extremes of temperature and radiation exposure can damage the delicate silicon circuitry of processors. The idea is that self-healing chips could be issued a single "rebuild" command. The processor would then test itself for faults, and if need be reconstruct its circuitry using the remaining good hardware.

Brain Building:
Initial discussions are now underway between Cell Matrix Corporation and the brain building laboratory in Utah, USA. The Utah brain project has already built an artificial brain which can simulate a million neurons (see: Utah-brain project). It is thought that a "second generation" artificial brain could be built using the cell matrix architecture. A custom ASIC chip could be used to implement a cell matrix consisting of a million cells. Assuming several hundred cells per neuron, this chip could simulate a network of a thousand neurons. A thousand of these Cell Matrix chips would then be required to build a million neuron artificial brain.

Nanofabrication:
The Cell Matrix architecture, being theoretically infinitely scalable, is set to significantly benefit from the emergence of nanofabrication and nano-scale electronic circuits. Nanofabrication is very hard. The reliable manufacture of specific physical circuits (physically heterogeneous systems) poses a major challenge. A possible way around this hurdle is to fabricate undifferentiated Cell Matrix hardware (a physically homogeneous system). This system is then differentiated into various specific digital circuits after its manufacture.

Cell Matrix Corporation:
The Cell Matrix architecture was dreamt up and patented by Nick Macias and Lisa Durbeck. In 1998 they formed the Cell Matrix Corporation which is based in Salt Lake City, Utah, USA.

References:
Cell Matrix corporation website: www.cellmatrix.com
The Cell Matrix design is protected by US Patent #5,886,537.