Researchers from Oxford University and Xi’an Jiaotong University present innovative machine learning model to simulate phase change materials in advanced memory technologies

Understanding phase change materials and creating cutting-edge memory technologies can greatly benefit from the use of computer simulations. However, direct quantum mechanical simulations can only handle relatively simple models with hundreds or thousands of atoms at most. Recently, researchers from the University of Oxford and Xi’an Jiaotong University in China developed a machine learning model that could help with atomic-scale simulation of these materials, accurately recreating the conditions under which these devices operate.

The model presented in the Nature Electronics study conducted by the University of Oxford and Xi’an Jiaotong University can quickly generate high-fidelity simulations, providing users with a deeper understanding of the operation of PCM-based devices. To simulate a variety of germanium-antimony-tellurium compositions (typical phase change materials) in realistic device configurations, they propose a potential machine learning-based model that is trained using quantum mechanical data. The speed of our model enables atomistic simulations of numerous heat cycles and sensitive operations for neuro-inspired computing, particularly cumulative SET and iterative RESET. Our machine learning method directly describes technologically relevant processes in phase-change material memory devices, as demonstrated by a model of the device size (40 20 20 nm3) comprising nearly half a million atoms.

The researchers demonstrate that thanks to ML machine learning-based modeling, fully atomistic simulations of phase changes along the GST composition line are possible under real device geometries and conditions. Interatomic potentials are fitted within the GAP framework using ML for various stages and GST compositions, and then the resulting reference database is iteratively improved. Atomistic processes and mechanisms in PCMs on the ten-nanometer length scale are revealed through simulations of cumulative SET and iterative RESET processes under conditions relevant to real operation, such as non-isothermal heating. This method allows a crosspoint memory device to be modeled in a model with more than 500,000 atoms, thanks to its greater speed and precision.

The team created a new dataset with labeled quantum mechanical data to train their model. After building an initial version of the model, they gradually began feeding it with data. The model developed by this group of researchers has shown great promise in preliminary tests, allowing precise modeling of atoms in PCM through numerous heat cycles and while simulated devices perform delicate functions. This indicates the feasibility of using ML for the simulation of PCM-based devices at the atomic scale.

Using a machine learning (ML) model, we significantly improve the time and accuracy of PCM GST simulation, enabling truly atomistic simulations of memory devices with realistic device shapes and programming conditions. Since ML-based simulations scale linearly with the size of the model system, they can be easily extended to larger, more complicated device geometries and over longer time scales using increasingly powerful computing resources. We anticipate that our ML model will enable nucleation sampling and atomic-scale observation of grain boundary creation in large GST models in isothermal or temperature gradient environments, as well as simulating crystal melting and development. As a result, the nucleation barrier and critical nucleus size for GST can be determined using ML-based simulations coupled with state-of-the-art sampling approaches.

Interface effects on adjacent electrodes and dielectric layers are an important topic for device engineering that could be explored in future research. For example, it has been reported that enclosing the PCM cell with aluminum oxide walls can significantly reduce heat loss; However, the effect of these atomic-scale walls on the thermal vibrations at the interface and the phase transition capability of PCMs cannot be studied using finite element method simulations alone. This effect can be investigated using atomistic ML models with extended reference databases to provide predictions of minimum RESET energy, crystallization time for various device geometries, and microscopic failure mechanisms to improve architectural design. Our results demonstrate the potential value of ML-based simulations in creating PCM phases and PCM-based devices.

Review the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to join. our 32k+ ML SubReddit, 41k+ Facebook community, Discord channel, and Electronic newsletterwhere we share the latest news on ai research, interesting ai projects and more.

If you like our work, you’ll love our newsletter.

we are also in Telegram and WhatsApp.

<!– ai CONTENT END 2 –>