ABSTRACT
This article demonstrates indium gallium zinc oxide-based onetime programmable ferroelectric memory devices with multilevel coding and lifelong retention capability. The entire integration process was conducted in the back-end- of-line with a maximum process temperature of 350°C. The fabricated devices demonstrate data retention up to 104 seconds, which was used to estimate the retention property up to 108 seconds. We observed a marginal drop in channel current after 108 seconds, which makes them suitable for inference engine application. The compatibility with the back-end-of line process enables monolithic 3D integration of the devices with standard technology. We have evaluated the performance of this indium gallium zinc oxide-based onetime programmable ferroelectric thin film transistor for inference engine applications. The system-level simulation was performed to gauge the performance of the devices as synapses in multilevel perceptron-based neural networks. The synaptic devices could achieve 97% for inferenceonly applications with MNIST data. The accuracy degradation was also limited to 1.5% over 10 years without retraining. The proposed inference engine also showed superior energy efficiency and cell area of 95.33 TOPS/W (binary) and 16F2, respectively.
