AI-Driven Material Modeling Feature “Autopilot” Implemented in Advance/NanoLabo
We have pioneered the development of the “Autopilot” feature, which automatically generates models simply by inputting the desired atomic structure model in words. For example, the corresponding models are generated with the following commands:
1. Command: “110 surface of NMC electrode and solid-liquid interface of 1 mol/L LiBF4 in EC (7:3) solution”
| ⇒ | After creating a supercell of layered LiCoO2 and substituting Co with Mn and Ni to create a (110) slab, the system packs EC, DMC, Li+, and BF4- in the specified ratio. Note: The AI understands the meaning of “NMC electrode” before modeling. |
2. Command: “Fe/Ni/Cr/Mn=7/1/1/1 alloy”
| ⇒ | After creating a supercell of Fe, the system substitutes Fe with 10.0% Ni, 11.1% Cr, and 12.5% Mn. Note: The AI automatically calculates the element substitution rates. |
3. Command: “ZrO2/Rh(111) + 2 x NO@hollow”
| ⇒ | The system creates an interface model of ZrO2 and Rh(111) and adsorbs two nitric oxide molecules at the hollow sites on the surface. Note: The AI understands ambiguous expressions using chemical formulas. |
Voice command input is also supported. Additionally, a “semi-auto mode” is available, allowing users to set details as needed while the AI performs automatic modeling. The Autopilot feature will be available in the next version of Advance/NanoLabo, with no additional cost beyond the Advance/NanoLabo license fee. The AI is implemented based on GPT-3.5-turbo.
For more details, please watch the video.