Dec 4, 2024
8:00pm - 10:00pm
Hynes, Level 1, Hall A
Hong Heon Lim1,Seung Chul Chae1
Seoul National University1
Hong Heon Lim1,Seung Chul Chae1
Seoul National University1
With the rapid development of artificial intelligence (AI) technology, the demand of energy-efficient non-volatile electronic devices is increasing significantly. The energy inefficiency of today’s computing system is largely attributed to the von Neumann bottleneck caused by the separation of logic and memory unit. As innovative alternative to overcome this disadvantage, neuromorphic computing system has been studied extensively, requiring artificial neurons and synapses that have multiple conductance levels. Ferroelectric materials are considered promising candidates since their multilevel states of spontaneous polarization have a potential to realize synaptic weight elements. Although numerous ferroelectric materials have been reported, ferroelectric hafnium oxide-based thin films have attracted attention due to their robust ferroelectric properties and good compatibility with complementary metal-oxide-semiconductor (CMOS) technology. However, researches on stoichiometric characteristics related to ferroelectric synaptic weights are still lacking.<br/>In this work, we investigated the ferroelectricity of hafnium zirconium oxide (HZO) thin films by varying the zirconium composition ratio. 1x nm polycrystalline Hf<sub>1-x</sub>Zr<sub>x</sub>O<sub>2</sub> (0.31 ≤ x ≤ 0.58) thin films were prepared by atomic layer deposition (ALD) on Si substrate. Besides we obtained 3-bit multilevel states of subloop hysteresis for each thin film. Hf<sub>0.59</sub>Zr<sub>0.41</sub>O<sub>2</sub> thin film showed the largest remnant polarization value of 24.8 μC/cm<sup>2</sup> and superior reliability of intermediate ferroelectric polarization states. In order to explain the reason for its reliability, we examined the critical volume for nucleation of the ferroelectric domain through the conventional electrical measurements and the phase-field simulation based on time-dependent Ginzburg-Landau model. Our results suggest the possibility to implement analog behavior of ferroelectric HZO thin films.