Local Activity, Neuristors and an Electronic Action Potential
R. Stanley Williams
HP Senior Fellow and Senior Vice President, Foundational Technologies
Hewlett-Packard Laboratories
We are working on a project at HP Labs to explore the use of “locally-active memristors” as the basis for extremely low-power transistorless computation. We have analyzed the thermally-induced phase transitions from a Mott insulator to a highly conducting state in a family of correlated-electron transition-metal oxides, such as VO2 and NbO2. The current-voltage characteristic of a simple cross-point device that has a thin film of such an oxide sandwiched between two metal electrodes displays a current-controlled or ‘S’-type negative differential resistance (NDR) caused by Joule self-heating if the ambient temperature is below the metal-insulator transition (MIT). We derived simple analytical equations for the behavior of these devices [1] that quantitatively reproduce their experimentally measured electrical characteristics, and observe that these devices are essentially electron valves that emulate the ion channels of axons. The resulting dynamical model was mathematically equivalent to the “memristive system” formulation of Leon Chua and Steve Kang [2]; we thus call these devices “Mott Memristors”. Moreover, these devices display the property of “local activity” [3]; because of the NDR, they are capable of injecting energy into a circuit (converting DC to AC electrical power) over a limited biasing range. We built and demonstrated Pearson-Anson oscillators with no inductors based on a parallel circuit of one Mott memristor and one capacitor, and were able to quantitatively reproduce the dynamical behavior of the circuit, including the subnanosecond and subpicoJoule memristor switching time and energy, using the numerical circuit simulation program SPICE. We then built a neuristor, a ‘thought-device’ proposed by Hewitt Crane [4] in 1960, using two Mott memristors and two capacitors. This neuristor has four state variables and electronically emulates the Hodgkin-Huxley model of the axon action potential of a neuron, which has been recently shown by Chua et al. [5] to be a circuit with two parallel ionic memristors, and we show experimental results that are quantitatively matched by SPICE simulations of the output signal bifurcation, signal gain and spiking behavior in our inorganic and electronic circuit [6] that closely replicates behavior observed in biological systems. We are exploring various spike-based computing paradigms [7,8]. Through SPICE simulations, we demonstrate that spiking neuristors are capable of Boolean logic and Turing complete computation by designing a one dimensional cellular automaton [9] based on ‘Rule 137’.
1. Pickett, M. D. and Williams, R. S. Sub-100 femtoJoule and sub-nanosecond thermally-driven threshold switching,” Nanotechnology 23, art. #215202 (2012).
2. Chua, L. & Kang, S. Memristive devices and systems. Proceedings of the IEEE 64, 209-223 (1976).
3. Itoh, M. and Chua, L. O. Memristor Oscillators, International Journal of Bifurcation and Chaos 18, 3183-3206 (2008).
4. Crane, H. D. The Neuristor. IRE Transactions on Electronic Computers EC-9, 370-371 (1960).
5. Chua, L., Sbitnev, V. & Kim, H. Hodgkin-Huxley axon is made of memristors. International Journal of Bifurcation and Chaos 22, 1-48 (2012).
6. Pickett, M. D., Medeiros-Ribeiro, G. and Williams, R. S. A scalable neuristor built with Mott memristors, Nature Materials 12, 114-117 (2013).
7. Chua, L. O. Memristor, Hodgkin-Huxley, and Edge of Chaos, Nanotechnology 24, 383001 (2013).
8. Sah, M. P., Kim, H. and Chua, L. O. Brains Are Made of Memristors, IEEE Circ. Syst. Mag. 14, 12-36 (2014).
9. Pickett, M. D. and Williams, R. S. Phase transitions enable computational universality in neuristor-based cellular automata, Nanotechnology 24, 384002 (2013).