Realizing Neuromorphic Computing
with Ferroelectric 2D Materials

Introduction

(Zidan et al.,  Nat Electron 1, 22–29 (2018))
(Jiang et al., Mat. Tod. Nano, 8, 100059 (2019))
(Ayadi et al., Physica E, 114, 113582 (2019))

As mankind generates exponentially more data each year, the power consumed to process such data is also increasing at an overwhelming rate. Some predict that if this trend continues, projected global power generation may not keep up with the power consumed for computing alone by the year 2040. On the other hand, the human brain processes an enormous amount of data all day, yet is extremely power-efficient that only requires a few thousand kilo-calories. 

In an effort to mimic the working principle of the human brain comes the concept of neuromorphic computing. The neuromorphic architecture requires device elements that are capable of both processing and storing data--just like human neurons do. The elemental devices, named memtransistors or memristors, allow for direct access to locally stored information without the need for long transmission through resistive lines, hence reducing power consumption and time delay.  

Memtransistors can be implemented in various ways, but the use of ferroelectric materials is a particularly appealing approach since it is compatible with existing transistor geometry. Ferroelectrics are a family of materials that exhibits spontaneous polarization upon an external electric field, which is then retained even after the field is removed. The remnant polarization generates a memory effect that could non-volatily modulate opto-electronic response of a device. Combining 2D semiconductors' excellent electronic transport properties and room temperature-stable excitonic emission with ferroelectrics will allow for power-efficient opto-electronic devices for neuromorphic applications. 

Realizing Electronic Neuromorphics

(Stay tuned for updates!)

Realizing Optical Neuromorphics

(Stay tuned for updates!)