Organic memristor Devices for Logic Elements with Memory

Victor Erokhin, Gerard David Howard, Andrew AdamatzkyInt. J. Bifurc. Chaos 2012 · DOI

TL;DR

This paper presents the physical realization and characterization of organic memristor-based logic gates (AND, OR, NOT) that exhibit memory properties. These devices offer a promising pathway towards novel computing architectures that integrate logic and memory, mimicking biological systems and enabling "learning" capabilities.

What Problem Does This Paper Address?

Traditional computing architectures separate processing and memory, leading to energy-intensive data transfer between the CPU and memory units. Memristors, with their inherent memory capability, offer a potential solution by enabling "in-memory computing" where logic operations are performed directly where data is stored. While memristors have been explored for memory applications, their use in creating functional logic gates that also possess memory and learning capabilities, particularly using organic materials, is a key area of investigation. This paper addresses the challenge of physically realizing and demonstrating such organic memristor-based logic elements for fundamental Boolean operations.

Methods

The authors fabricated organic memristive elements using a heterojunction between a conducting polymer (polyaniline, PANI) and a solid electrolyte (polyethylene oxide doped with lithium salt, PEO). The conductivity of these devices is modulated by ionic charge transfer, leading to a history-dependent resistance. They then designed and implemented three basic logic gates:

  • Memorized OR (MOR): A single organic memristor is used, where input signals (voltages) drive the device towards a conducting state. The output current is analog and depends on the duration of the input signals, preserving the state after the input is removed.
  • Memorized AND (MAND): This requires additional components, including a summator (operational amplifier) and a voltage divider. Both inputs must be active to drive the memristor to a conducting state, demonstrating a form of associative learning where reinforcement requires the presence of both signals.
  • Memorized NOT (MNOT): This circuit uses resistors and a voltage source to create a logic NOT gate. The output signal is dependent on the memristor's resistance, which changes based on the input signal and its duration.

Experimental characterization involved voltage sweep and drain current measurements using Keithley instruments, automated via MATLAB scripts. They also simulated a one-bit full adder constructed from these memristive logic gates.

Datasets

Not discussed in the paper. The experiments rely on fabricated devices and simulated circuit behavior.

Key Insights and Ideas

  • Organic Memristors as Synapse Analogues: The core insight is that organic memristors can mimic synaptic behavior by exhibiting a conductivity state that depends on the history of applied signals, allowing for "learning" and memory integration within logic circuits.
  • Analog Output with Binary Interpretation: The memristive logic gates produce analog outputs that vary with input duration, offering a richer functionality than traditional binary logic. However, these analog outputs can be interpreted in a binary fashion using thresholding, providing flexibility.
  • "Learning" in Logic Gates: The MAND gate, in particular, demonstrates a form of associative learning where reinforcement of the circuit's state requires the simultaneous presence of both inputs, mirroring biological learning mechanisms.
  • Potential for In-Memory Computing: The successful implementation of these logic gates suggests a pathway towards computing architectures where logic and memory are co-located, potentially leading to more efficient and powerful systems.

Key Results

  • Physical fabrication and characterization of organic memristor-based MOR, MAND, and MNOT logic gates.
  • Demonstration of analog output signals that vary with input signal duration and are preserved after the input is removed.
  • Experimental results showing the MOR gate's output varying from 0 to 1 based on input activation.
  • Experimental data for the MAND gate showing a gradual increase in output current only when both inputs are activated, indicating associative learning.
  • Experimental results for the MNOT gate demonstrating its ability to perform the NOT operation with a preserved state.
  • Simulation of a one-bit full adder composed of these memristive logic gates, showing its functionality and response to different input patterns.

Limitations and Future Work

  • Response Time and Saturation: While the devices exhibit memory, the kinetics of conductivity change and saturation levels are influenced by the material properties and device design. The paper notes that delays in the simulated adder are attributed to the time required for nodes to reach the oxidation potential, especially in MAND gates.
  • Analog Nature and Thresholding: The analog nature of the output, while a strength, also necessitates careful consideration of thresholding for reliable binary operation in complex circuits. The paper mentions that MNOT gates don't reach 0V, requiring a sensible threshold.
  • Scalability and Complexity: The paper demonstrates basic logic gates and a simulated adder. Scaling these to more complex circuits and addressing potential issues like cross-talk between elements in larger arrays are areas for future work.
  • Return to Initial State: The paper suggests that periodic inverted potentials could be used to reset the memristors to their initial insulating state, making the circuits more similar to traditional logic elements. However, achieving precise and controlled return to initial states over time is an ongoing challenge.
  • Further Exploration of Learning: The "learning" aspect of these memristive circuits, particularly in more complex architectures, warrants further investigation. The authors suggest that simple modifications could lead to oscillator circuits, opening up possibilities for dynamically oscillating networks.

Summary updated

BibTeX
@article{erokhin2012organic,
  author = {Victor Erokhin and Gerard David Howard and Andrew Adamatzky},
  title = {{Organic memristor Devices for Logic Elements with Memory}},
  journal = {Int. J. Bifurc. Chaos},
  year = {2012},
  doi = {10.1142/s0218127412502835},
}
Figure