The semiconductor industry has been exploring the use of 3D crosspoint memory for years.
I recently gave a talk at IEEE International Memory Workshop on 3D crosspoint memory arrays and Intermolecular also announced an industry-first related to this topic: the first 4-element Chalcogenide-based OTS for 3D memory arrays.
The results of the work we did demonstrate the breakthrough combination of germanium, arsenic, selenium and tellurium (GeAsSeTe) on OTS device for 3D vertical memory arrays that will enable a 3D vertical non-volatile memory (NVM) memory architecture for customers to design chips for high density, high-performance computing applications at affordable costs. This will allow the realization of new architectures, paving the way for neuromorphic computing, AI and other new semiconductor designs that are needed for faster and more affordable digital applications, from gaming to data centers.
Why OTS and ALD Chalcogenides?
The Ovonic Threshold Switch (OTS) is one type of selector element, which uses chalcogenides materials containing Group VI elements of sulfur, selenium, or tellurium. 3D crosspoint memory arrays require selectors to minimize leakage and crosstalk between memory cells. The OTS must be highly selective with a high enough on/off ratio to sustain high current in order to program the memory. Typically, physical vapor deposition (PVD) has been used to deposit the chalcogenide films. The PVD process limits film conformality and homogeneity on a large scale precluding the integration of tens of decks in a 3D crosspoint architecture. However, atomic layer deposition (ALD) allows for future 3D vertical integration with higher density and reduced cost, enabling the expansion of the 3D crosspoint technology.
How We Got There
Up until now, research institutes and universities developed binary or quaternary OTS with a germanium-selenium or germanium-tellurium only. The electrical performances in terms of leakage and endurance were relatively poor and not comparable to the actual standard for PVD OTS. Intermolecular, for the first time, has demonstrated a quaternary ALD OTS film with germanium, arsenic, selenium and tellurium with the electrical performances similar or better compared with PVD OTS.
We achieved this by developing an ALD reaction strategy for both elemental and binary components of the material. The composition analysis was carried out using the previously calibrated XRF. We were able to define the stoichiometry of the film as Germanium-23, Arsenic-38, Selenium-15 and Tellurium-24. The introduction of selenium in the composition was done just to increase the energy gap of the material in order to increase the resistivity, and the high content of arsenic was introduced to allow higher thermal stability and prevent the crystallization of the film at higher temperature. So, we started from a base that was a germanium-tellurium, and we added arsenic and selenium in order to improve the performance of our film. The conformality of the film was verified by depositing a 10 nm thick ALD GeAsSeTe film into a 20:1 high aspect ratio silicon trench structure. Using TEM, we were able to observe good conformality with similar film thickness from the top to the bottom of the trench.
Device Integration and OTS Device Performance
To characterize device performance, we started by measuring the leakage in DC regime at 1V. We were able to clearly see that the leakage increased at a constant rate as a function of bottom electrode area. This would indicate that our material is a scalable solution.
The first cycle (forming operation) showed a lower subthreshold leakage and higher Vth (threshold voltage) as expected. The device demonstrated repeatable IV characteristics after forming as well as good stability under field stress during long DC operation. When applying a triangular pulse, we were able to see that the Vth was around 2.1 V and switched off at around 1.4 V.
As widely studied in the past, chalcogenides tend to change electrical properties over time due to structural relaxation which is generally interpreted by thermally activated annihilation of disorder-induced defects. To verify this behavior, we applied two subsequent pulses at different time delays extracting the Vth median value of 10 repetitions. We were able to extract the Vth evolution over time and create a simple model to reproduce the data demonstrating a Vth drift lower than 40mV/dec.
After testing 20 devices we found that our OTS film was not prone to crystallization with slow quenching time. The threshold voltage during the endurance experiment was stable with a slight decrease only after 108 cycles. We also measured stable subthreshold conduction using longer pulses with the help of a pre-amplifier. Currents in the ON state at 3V and in the OFF state at VTH/2 showed stable selectivity higher than 3 orders of magnitude for at least 109 cycles.
In order to verify the robustness of the model, we calculated the full DC-IV characteristic adopting the same parameters and equation used in previous literature. We were able to see that the calculated IV was in agreement with the experimental data and the model could be used for future development of ALD OTS devices through fine-tuning film thickness and composition.
We have successfully proposed a model for GeAsSeTe OTS subthreshold conduction for potential optimization of ALD film stoichiometry and thickness. I expect even more applications of ALD chalcogenide-based selectors as a leading technology for multiple stacks integration of crosspoint memory arrays in the future. To learn more, you can view the video of my presentation here. I would like to thank IEEE for inviting me to talk on this exciting development.