Programming memristor arrays with arbitrarily high precision for analog computing

Abstract

In-memory computing represents an effective method for modeling complex physical systems that are typically challenging for conventional computing architectures but has been hindered by issues such as reading noise and writing variability that restrict scalability, accuracy, and precision in high-performance computations. We propose and demonstrate a circuit architecture and programming protocol that converts the analog computing result to digital at the last step and enables low-precision analog devices to perform high-precision computing. We use a weighted sum of multiple devices to represent one number, in which subsequently programmed devices are used to compensate for preceding programming errors. With a memristor system-on-chip, we experimentally demonstrate high-precision solutions for multiple scientific computing tasks while maintaining a substantial power efficiency advantage over conventional digital approaches.

Reservoir computing with a random memristor crossbar array

“…Image recognitions using MCAs have been extensively demonstrated, in which the full-precision off-line trained weights [22,39] were transferred to the conductances of memristors with limited precision, leading to accuracy degradation. To overcome this problem, on-line training [40,41] and hardware-aware training [22,[42][43][44] have been proposed to compensate device non-ideality. By contrast, our system exploits device non-ideality and greatly simplifies the training procedure.…”

Two-dimensional material-based memristive devices for alternative computing

“…Unlike a conventional transistor that operates as a current switch and handles binary information, a memristive device functions as a current switch that remembers the voltage or current that has passed through a device. Memristive devices offer several advantages including non-volatility, low power consumption, and parallelism, allowing for applications such as analog computing [ 2 , 3 ], in-memory computing [ 4 , 5 ] and most importantly for the current technological climate, neuromorphic computing [ 6 , 7 ]. The underlying mechanism varies depending on the type of memristor, but it generally involves the movement of atomic defects or ions within the device structure, often leading towards a dielectric material being placed between two conductive plates.…”