High-Precision Tuning of State for Memristive Devices by Adaptable Variation-Tolerant Algorithm (1110.1393v1)

Abstract: Using memristive properties common for the titanium dioxide thin film devices, we designed a simple write algorithm to tune device conductance at a specific bias point to 1% relative accuracy (which is roughly equivalent to 7-bit precision) within its dynamic range even in the presence of large variations in switching behavior. The high precision state is nonvolatile and the results are likely to be sustained for nanoscale memristive devices because of the inherent filamentary nature of the resistive switching. The proposed functionality of memristive devices is especially attractive for analog computing with low precision data. As one representative example we demonstrate hybrid circuitry consisting of CMOS summing amplifier and two memristive devices to perform analog multiply and accumulate computation, which is a typical bottleneck operation in information processing.

• The paper presents a feedback algorithm that achieves 1% relative accuracy in tuning memristive device conductance, equivalent to around 7-bit precision.
• It employs sequences of variable amplitude write and read pulses to validate rapid (∼1 µs) switching and reliable state retention in Pt/TiO2/Pt devices.
• The study demonstrates a hybrid CMOS-memristor circuit for analog multiply-accumulate operations, highlighting its potential for low-power, high-density computing.

High-Precision Tuning of State for Memristive Devices Using an Adaptable Variation-Tolerant Algorithm

The paper presents a significant advancement in the utilization of memristive devices, particularly focusing on titanium dioxide thin-film devices, and proposes a methodology for achieving high-precision tuning of device conductance. The authors introduce a write algorithm capable of tuning memristive states to 1% relative accuracy, equating to approximately 7-bit precision, despite substantial variations in switching behavior.

Core Contributions

1. Algorithmic Precision: The researchers have developed an innovative feedback-based algorithm. This algorithm builds upon active feedback systems previously applied to phase change memories. Utilizing sequences of variable amplitude write and read pulses, it hones device resistance within 1% relative accuracy throughout the dynamic range. This precision is imperative for the reliable deployment of memristive devices in analog computing, particularly where low precision data is processed.
2. Empirical Validation: The paper reports experimental results achieved with Pt/TiO2/Pt devices, demonstrating rapid switching capabilities of the order of microseconds and significant retention of states. These are achieved through alternating large amplitude write pulses with smaller read pulses. The distinct switching dynamics - exponential for SET and power law for RESET - are corroborated by detailed empirical data.
3. Hybrid Circuit Implementation: By integrating memristive devices with CMOS technology, the paper demonstrates analog computation operations. A practical implementation includes a CMOS summing amplifier coupled with memristive devices to execute multiply and accumulate (MAC) operations, a common computational bottleneck. This hybrid approach effectively leverages the high-density and low-power characteristics of memristor-based systems.

Numerical and Experimental Insights

• Switching Time and Retention:

The experiments highlighted switching times as short as 1 µs for both SET and RESET processes, with retention times significantly exceeding 1 second within a specified voltage range. Such performance metrics are crucial for the practical application of these devices in real-time processing environments.

• Precision Achievements:

The algorithm successfully achieved high-precision tuning across the entire dynamic range of the devices, suggesting potential for even greater precision with further iterations.

Theoretical and Practical Implications

This research holds substantial implications for the development of analog computing systems, offering solutions for high-density, low-power computational architectures. The proposed algorithm not only aligns with the precise demands of analog signal processing but also promises enhanced efficiency for applications requiring under 8-bit precision computations, prevalent in fields like robotics and sensor networks.

Furthermore, the authors present a compelling case for the relevance of their findings to nanoscale devices. Prior studies have indicated reduced variations at the nanoscale, suggesting that this algorithm might achieve even greater precision and efficiency in smaller devices.

Future Directions

Looking forward, this paper sets the stage for exploring the integration of memristive devices within passive crossbar architectures. Addressing potential challenges such as semi-selection and leakage currents will be key to advancing this technology. Continued investigation into the physical limitations, such as ionic noise and the precision threshold, remains critical to understanding the ultimate capabilities and boundaries of memristive systems.

In conclusion, the research represents a valuable contribution to the field of memristive device applications, providing a robust framework for achieving high-precision state tuning. It lays foundational work for the broader implementation of memristive technology in mixed-signal processors and beyond.