Memistor revisited
Massimiliano Versace | September 13, 2009
The author of this post is Blaise L Mouttet.
The name Ted Hoff is familiar to many electrical and computer engineers for his work in developing the first microprocessor introduced by Intel. A lesser known contribution of Ted Hoff was the suggestion for a new type of circuit element called the memistor, an electrolytic memory element developed in 1960 which formed the basis for a neural circuit architecture called ADALINE (ADAptive LInear NEuron) developed by Stanford University professor Bernard Widrow.
According to a technical paper entitled “An Adaptive Adaline Neuron Using Chemical Memistors,” the memistors were formed based upon the electroplating of copper in a copper sulfate-sulfuric acid solution and operated similarly to a 3-terminal transistor except that the resistance between two of the terminals was controlled by the time-integral of the current as opposed to the instantaneous value as in ordinary transistors. Thus the element was denoted as a memory transistor or memistor. The memistors served as the variable gain elements (a1,a2,a3,…) in the ADALINE neural circuit below.
Bernard Widrow obtained a patent (US 3,222,654) for memory, logic, and pattern classification circuitry based on the memistors and even started a company called the Memistor Corporation which was briefly successful in commercializing memistors and ADALINE circuits as detailed in the book “Talking Nets:An Oral History of Neural Networks.” Unfortunately the timing of the memistor was off since solid state electronics was becoming dominant and memistors were electrochemical in nature and thus incompatible with integrated circuit processing techniques. In addition, ADALINE circuitry was based on analog circuit components incompatible with the rise of digital processing. In this respect it is ironic that Ted Hoff, who contributed the original suggestion for implementing the memistor, also developed the digital microprocessor with which the analog circuits based on memistors could not compete.
However, fast forwarding to the present time, there is a new chapter that seems to be beginning in the memistor saga. Memristors (memory resistors), which were originally theorized in the 1970’s, have recently been realized as solid state circuit components having some similar properties to the memistor. In addition, mixed signal integrated circuit designs combining both analog and digital components have been growing with the popularity of applications such as cell phones. The growth of video game technologies and robotic systems requiring at least some analog sensing or actuation functions will likely continue to increase the importance of analog and digital circuit integration. Thus, while the memistor and the ADALINE circuitry of the 1960’s were incompatible with the technological trend of the time, present technology may be catching up with the original promise of the memistor and neural circuits.
In the summer of 1958, Dr. Dudley A. Buck, a colleague of Dr. Widrow, was developing methods to make memistors in the laboratory at M.I.T. (Dr. Widrow was not aware of this.) Dr. Buck’s larger view was developing methods to manufacture integrated circuits with electron beams. He was particularly interested in manufacturing cryotron computers. He was also interested in the work of Frank Rosenblatt and the perceptron (a self-learning computer).
For the self organizing systems of the day, they were only interested in analogue memory devices. Transistors were still very expensive and just beginning to achieve some level of reliability, and the tube… well.. they were tubes.
Some of Dr. Buck’s earliest experiments for making memistors were with growing tin dendrites in an electrolyte. Soon after he considered anodizing one of two plates of aluminum, the amound of anodizing being the stored value.
Cuprous Sulfide was attractive in that copper could be plated out of Cu2S. The term memister was not in use, Dr. Buck called his device a “synapse”, using a copper sulfate / copper iodide device enclosed in a cylinder of quartz.