Transparent-conductor discovery on a public dataset
A multi-objective search over (Al,Ga,In)₂O₃ compositions on the public NOMAD2018 dataset: in a 60-second bandgap run the quantum-inspired search scored 123.5 against 100.5 for NSGA-II and 76.2 for TPE, and reached NSGA-II’s 60-second score in about 10 seconds. A computational preprint delivered through our platform network.
Impact
The problem
Finding compositions that balance conductivity and transparency means searching a large composition space where each evaluation costs a simulation. Evolutionary and Bayesian searches are the standard, and they are slow to reach a good set.
Our approach
A factorization-machine model of the composition space was searched with a quantum-inspired annealer on classical hardware and compared against NSGA-II and TPE at equal wall-clock time on the same dataset.
The evidence
- FMA score 123.5 against NSGA-II 100.5 and TPE 76.2 in a 60-second bandgap run; NSGA-II’s 60-second score reached in about 10 s. Quantum-inspired annealing on classical hardware; public NOMAD2018 dataset; preprint.
- Not claimed: a quantum-hardware contribution, or that the gain transfers beyond this dataset.
Status: Platform network + Simulation. Delivered through our platform network; quantum-inspired, classical hardware; computational preprint; one dataset.