Random Variation in Nanoscale HfZrO$_2$ Ferroelectric FinFETs: Physical Origin and Neuromorphic System Implications

This paper unveils the impact of random phase distribution fomented variations on the characteristics of hafnium zirconium oxide based ferroelectric FETs, and the implications towards the application in neuromorphic system. TiN/HZO/TiN/SiO$_2$ ferroelectric capacitors are fabricated on silicon substrate. Statistical distribution of coercive voltage from polarization-voltage measurements show a two-peak distribution, which is attributed to the existence of two distinct groups of samples , ferroelectric dominant and paraelectric dominant ones. Subsequently, ferroelectric FinFET devices with single and multiple fins have been fabricated on SOI wafer using a gate first process, with gate lengths down to 40nm. Ferroelectric hysteresis with enhanced static memory window up to 3.5V in 10nm HZO based ferroelectric FET is observed. Short-channel multiple-fin devices show particularly good ferroelectric switching characteristics and high yield. In addition, these devices show highly linear and symmetric multi-level programming characteristics, which makes them an excellent candidate as synaptic device. Modelling of device-to-device and cycle-to-cycle variation is performed based on measured data and applied to system level neural network simulations using the CIMulator software platform. Device-to-device variation is mostly compensated during neural network online training and has virtually no impact on inference accuracy. On the other hand, cycle-to-cycle threshold voltage variation up to 400mV can be tolerated for MNIST handwritten digits recognition. An online training accuracy of 96.34 percent can be achieved given the measured variability. We further demonstrate the optimization of inference-mode gate voltage considering ferroelectric FET based neural network.