Neuromorphic computing of SiC based memristor

Abstract

The developing computing and information technology such as AI and IoT, challenges the ability of computing system, demands higher energy efficiency, faster processing and larger data storage. The modern computer system relies on the Bon Neumann system, where the computing and memory units are separated. Due to the development of computing system, the traditional Von Neumann architecture cannot full fill the demands, in other words, it faces the Von Neumann bottle neck. The Von Neumann bottle neck refers to the energy consumption and the limitations of separation between computing and memory units. To overcome this problem, inspired by the human brain and neuro system, the in-memory computing was proposed, where the computing and memory shares the same unit. With this method, there is not necessary to transport data between units, which accelerates the computing and reduces the energy consumption. To perform in-memory computing, it requires more than the tradition semiconductor device, which demands the device works like synapse such as various synapse plasticity. The memristor is the device that can perform the neuromorphic behaviours like synapse. The memristor has the advantage of adjustable resistance state, which are ideal for the synaptic weigh control as the neuromorphic cells. There are many materials can be applied for the memristor, amorphous silicon carbide (SiC) is one of outstanding competitor not only because it has been already widely used in industry as the Back-end-of Line (BEOL) material but also its stable physical and chemical characteristics. This PhD project aims to develop the SiC memristor for neuromorphic computing applications. Silicon-rich SiC based memristors were fabricated and characterised in terms of physical and electric features. In chapter 4, The underlying physical mechanism is discussed according to the switch characterisations. The thin film memristors demonstrate excellent binary resistive switching with compliance-free and self-rectifying characteristics. By investigating the neuromorphic switching performances, it is possible to mimic synaptic functions and perform synaptic behaviours by tuning the strength of applied stimulation signals. This neuromorphic behaviour was demonstrated to further emulate several virtual synaptic functions. The short-term memory behaviour and neuromorphic applications of SiC memristor is further investigated in chapter 5. The learning-forgetting behaviours was emulated by the SiC memristor devices according to the STP characterisations. An advanced neuromorphic computing application called reservoir computing (RC) system was established. Within this system, the digital number patterns were encoded into stimulation signals and the SiC memristors were applied as the physical reservoir units. This physical reservoir computing system has been experimentally demonstrated to perform the task of pattern recognition. After training, this RC system has achieved 100% accuracy in classifying number patterns from 0 to 9. The results shown here indicate that our SiC-based memristor devices are strong contenders for potential applications in artificial intelligence, particularly in temporal and sequential data processing. The pivotal role of electrolyte layer material composition in shaping the performance and functionality of these promising electronic devices was investigated by SiC fabrication recipe modifications. The relationship between various electrolyte layer composition and their impact on memristor behaviours were researched in chapter 6, seeking to the effects on neuromorphic switching of memristor. The mechanism governing neuromorphic behaviour within the SiC film of SiC-based memristor was investigated according to the switching behaviours. This study advances our fundamental understanding of memristor operation and helps develop customised and optimised electrolyte materials, which are essential for implementing next-generation memristor based computing paradigms.

More information

Identifiers

ORCID for Dongkai Guo: orcid.org/0009-0001-3901-312X

ORCID for Ruomeng Huang: orcid.org/0000-0003-1185-635X

ORCID for Kees De Groot: orcid.org/0000-0002-3850-7101

ORCID for Liudi Jiang: orcid.org/0000-0002-3400-825X

Catalogue record

Export record