Session:

The need for increased computational efficiency for deep learning applications has led to interest in in-memory computing. One example, three-terminal synaptic transistors with ion or proton intercalation mechanisms [1,2], have been recently studied for their CMOS compatibility and relatively fast switching time. Given the young nature of this technology, there are fundamental unanswered questions about these devices, which creates a gap between the current available models and tools required to simulate and accelerate material and architecture design. Though ion-intercalation materials have been heavily studied in the context of batteries, the large applied voltages to trigger the resistance state switching and nanoscale structure of these films lead to uninvestigated effects on the material reaction and transport properties. Additionally, the impact of these devices’ design and formation on performance is not fully understood. A fundamental investigation of the relationship between material properties and device performance metrics is needed.

Here we present a new phenomenological phase-field model to describe the resistance switching of WO₃ 3-terminal protonic transistors [1] by extending multiphase polarization theory outline by Tian and Bazant [3]. Through this process, we determine relationships between the effective surface reaction and material transport parameters and device performance parameters, such as switching time and energy. We address the importance of the underlying model for the surface ion-intercalation reaction, comparing different the phenomenological Butler-Volmer and the coupled ion-electron transfer reaction model [4]. Additionally, we investigate the impact of phase separation on these devices, which, due to the large electric fields needed to switch between resistance states, is shown to occur even when the bulk concentration exists in a single-phase region. Finally, preliminary extensions to 3d device topologies are investigated. This model will be indispensable in predicting the impact of material properties and architecture on device performance, accelerating design of similar synaptic transistor technologies.

References
1. Onen, Murat, et al. "CMOS-compatible protonic programmable resistor based on phosphosilicate glass electrolyte for analog deep learning." Nano Letters 21.14 (2021): 6111-6116.
2. Cui, Jinsong, et al. "CMOS-compatible electrochemical synaptic transistor arrays for deep learning accelerators." Nature Electronics 6.4 (2023): 292-300.
3. Tian, Huanhuan, Ju Li, and Martin Z. Bazant. "Multiphase Polarization in Ion-Intercalation Nanofilms: General Theory Including Various Surface Effects and Memory Applications." Advanced Functional Materials 33.23 (2023): 2213621.
4. Bazant, Martin Z. "Unified quantum theory of electrochemical kinetics by coupled ion-electron transfer." Faraday Discussions (2023).

Co-Author(s): Martin Bazant (Massachusetts Institute of Technology)

Michael Li

Appearances