Lingli Liu1,Putu Andhita Dananjaya1,Wen Siang Lew1
Nanyang Technological University1
Lingli Liu1,Putu Andhita Dananjaya1,Wen Siang Lew1
Nanyang Technological University1
Biologically-inspired neuromorphic computing systems are attractive for next-generation computing technologies. This non-conventional computing approach offers processing parallelism, cognitive capability and boasts high energy efficiency. In recent years, the redox transistor has emerged as a potential candidate for artificial synaptic devices that can concurrently execute signal transmission and memory operations. In this work, a complementary metal-oxide semiconductor-compatible redox transistor with a highly linear weight update is presented. The synaptic weight represented by the channel conductance is modulated under an electrochemical reaction followed by electric field-driven ion migrations in and out of the channel. This electrochemical reaction involved in such device operation is known to correlate strongly with humidity. However, available studies which specifically address this humidity aspect are still limited. We investigate the humidity-sensitive pre-oxidation process of the pristine W channel required to initiate the switching operation, which can be attributed to the proton defects-enhanced electrochemical process. Under the bipolar pulse scheme, the gradual oxidation of W derived from the progressive activation of the proton transport channels is observed. Performances of the redox transistor under different moisture levels are presented, i.e., pre-oxidation and potentiation/depression. Excellent endurance performance with more than 256 k synaptic weight updates can be obtained. Furthermore, a handwritten digit recognition accuracy of more than 90% is achieved in a 4-layer neural network simulation. Overall, it is concluded that the as-presented redox transistor is a promising candidate for realizing hardware implementation of the artificial neural network.