Friday, April 01, 2011

HP's memristor is the greatest hoax in the history of electronics

The "memristor" is a type of circuit element in which there is a relationship between magnetic flux linkage and charge. This circuit element was proposed by a circuit theorist named Leon Chua in 1971 as a missing fourth fundamental passive circuit element after the resistor, capacitor, and inductor. In 2008 researchers at HPLabs announced they found this circuit element in the form of thin films of titanium dioxide in which oxygen vacancies acted as mobile ions.

When HP had first announced the memristor it seemed like an interesting development to me and I had a short paper at the 2008 Nano-Net conference in Boston proposing a way to use memristors for signal processing (link). I was later invited to speak at the 1st Memristor and Memristive System Symposium at UC Berkeley held in Dec. 2008 (link). This was the point where the memristor hoax started to become evident to me.

I was already aware of several companies such as Axon Technologies and Unity Semiconductor which had been working on thin films having memory resistance properties similar to the memristor for a new form of non-volatile memory called RRAM or ReRAM. At the time when HP's competitors were developing their thin film memory resistors the researchers at HP were developing molecular memory. However, they could not get their molecular memory to work well and I believe that they decided to switch to thin film memory resistors. In order to appear as an initiator rather than simply copying their competitors they used Chua's "missing" memristor as a publicity tool.

This may seem like a bold claim which is simply one opinion or point of view. However, there are some facts which may sway reasonable people toward this view. These facts become evident to anyone who actually spends some time trying to look beneath the shiny surface of HP's memristor claims. Last summer I was invited to speak at the 2010 IEEE International Symposium on Circuits and Systems and I pointed out some of the inconsistencies with HP's memristor (link). Below I review some of these defects in a form familiar to patent practitioners.

1) Novelty

HP's memristor fails to meet the requirement of novelty in view of Argall "Switching phenomena in titanium oxide thin films," Solid-State Electronics (1968) (link). Argall discloses that thin films of titanium dioxide which exhibit a zero-crossing hysteresis behavior identical to that discussed in HP's 2008 memristor paper. Argall's titanium dioxide thin films are taught to be switchable to three distinct resistance states.

In addition the importance of oxygen vacancies are discussed in patents issued to Sharp (US 6972238, filed May 21, 2003) and Samsung (US 7417271, priority Feb. 27, 2006).

2) Obviousness

Chua's original concept of the memristor is obvious in view of Bernard Widrow paper "An Adaptive ADALINE Neuron using chemical memistors," (1960) (link). This paper coins the term memistor (memory resistor) eleven years prior to Chua and exhibited the charge dependent conductance effect (Fig. 6) which Chua later used to define his memristor. The difference between Chua's memristor and Widrow's memistor is that Chua's device is 2-terminal and Widrow's device is 3-terminal. However, it was known in the 1960's that 3-terminal devices such as transistors could be converted to 2-terminal devices such as diodes by connecting two of the terminals together. It would have been obvious to a person of ordinary skill in the electronics art to connect two of the three terminals of Widrow's memistor together to form a 2-terminal memristor. The motivation for doing so would have been to facilitate miniaturization by reducing the number of required external electrical connections. 

3) Enablement

The argument made by HP's 2008 memristor paper was that the motion of oxygen vacancies or ions in thin films of titanium oxide could be modeled by similar drift-diffusion equations used to model electron motion. This leads to equations which can be modeled using the magnetic flux linkage versus charge  relationship of a memristor. However, this analysis is incorrect because it assumes that the ions can be considered to have zero mass and it neglects the repulsive forces between ions in a thin film as well as hysteresis effects from the built-in voltage of semiconductors. A more careful analysis illustrates that the behavior of ions in thin films is more properly treated in terms of a damped driven harmonic oscillation of the ions (link).

Conclusion

As explained above the memristor fails to meet requirements of novelty, non-obviousness, and enablement. One may argue that the researchers at HP were unaware of these issues and should not be blamed for their misunderstanding or spreading false information. However, there are additional facts which would indicate that the researchers at HP already knew that their device was not a memristor as described by Chua even as they published the 2008 memristor paper. In 2008 they conspired with Chua encouraging other researchers in RRAM (including myself) to call their device "memristors" even if they did not meet the original definition. This is an attempt by HP to gain control of a potential multi-billion dollar market for a new form of non-volatile memory which has been under development by HP's competitors. And it appears to be working.

What a disgrace.

Labels: , , , , ,