The God Particle

The Higgs Boson is a theoretical elementary particle predicted to exist in the Standard Model of particle physics.  It is called the God particle because if it exists, it explains how elementary particles with no mass can be used to construct mass in matter.  Even though it has never been observed, it is predicted to exist, thus the God reference.

In the micro-electronics field, the "God particle" is the memristor.  The concept of the memristor is based on fundamental circuit variables much like the resistor, capacitor, and inductor.  The word "memristor" is short for "memory resistor". 

A memristor works like a resistor whose resistance level is determined by an amount of electrical charge that has passed through the memristor previously, and can maintain the resistance value when uncharged (power off).  The memristor is a resistor capable of remembering a previous charge, much like the way neural cells work in our brains.

HP researchers recently announced that they had created a nanoscale version of a memristor.  In "HP Discovers Potential "God Particle" the device is touted as a possible next step for creating androids and other types of synthetic intelligence. The thought being that memristors could potentially act as simulated neural cells.

First off, I'm nothing if not a skeptic.  I think the potential of the memristor is great.  I have always been a bit bewildered by our acceptance of a purely binary transistor processing short bit (32 & 64) pathways.  I've some great discussions with my technical colleagues about a 'trinary' transistor, capable of holding a positive, negative, or neutral charge.  And I've considered processors with 1024 or even 64k bit pathways, allowing us to process more information per CPU cycle.  But I'm not a particle physicist or an electrical engineer, so the discussions are usually more science fiction.

Lets consider the memristor as a synthetic neural cell for a second.  If you've ever done any development work on artificial neural nets you should have general understanding of the basic principle.  Neural nodes, or artificial cells, connect to other nodes.  Each connection has a value or a weight assigned to it.  In most neural network patterns these weights are values between 0 and 1.  When neural input is received, the nodes transmit their weight values to their connecting neural nodes.  The neural nodes add up the weights, and determine if the signal value, and its own weight will cause the node to fire, or send its own message through the neural net.

Between 0 and 1.  Transistors are binary.  They only compute zeros and ones, not values in between, so they don't serve very well as synthetic neurons.  And they only cycle 0's and 1's when they are powered, losing any stored values with loss of electrical charge.  With a memristor these constraints are removed.  Memristors can contain values (resistance level) between 0 and 1, and they can maintain that value even while not being charged.

So by all accounts it appears that the memristor has the potential to make for a great synthetic neuron.  The HP article goes on about facial recognition, intelligent robots, androids, etc. as articles like this are notorious for doing.  But lets add a touch of reality here.  Memristors are to the human brain what the wheel is to walking.  Memristors are inanimate and unevolved.  Our brains have already been subject to centuries of natural selection and genetic mutation.  The memristor will not magically fill this gap.  It may make for some great advances in storage capabilities though.