Flash memory devices--that is, non-volatile computer storage media that can be electrically erased and reprogrammed--are vital for portable electronics, but the scaling down of metal-oxide-semiconductor (MOS) flash memory to sizes of below ten nanometres per data cell presents challenges. Molecules have been proposed to replace MOS flash memory, but they suffer from low electrical conductivity, high resistance, low device yield, and finite thermal stability, limiting their integration into current MOS technologies. Although great advances have been made in the pursuit of molecule-based flash memory, there are a number of significant barriers to the realization of devices using conventional MOS technologies. Here we show that core-shell polyoxometalate (POM) molecules can act as candidate storage nodes for MOS flash memory. Realistic, industry-standard device simulations validate our approach at the nanometre scale, where the device performance is determined mainly by the number of molecules in the storage media and not by their position. To exploit the nature of the core-shell POM clusters, we show, at both the molecular and device level, that embedding [(Se(IV)O3)2](4-) as an oxidizable dopant in the cluster core allows the oxidation of the molecule to a [Se(v)2O6](2-) moiety containing a {Se(V)-Se(V)} bond (where curly brackets indicate a moiety, not a molecule) and reveals a new 5+ oxidation state for selenium. This new oxidation state can be observed at the device level, resulting in a new type of memory, which we call 'write-once-erase'. Taken together, these results show that POMs have the potential to be used as a realistic nanoscale flash memory. Also, the configuration of the doped POM core may lead to new types of electrical behaviour. This work suggests a route to the practical integration of configurable molecules in MOS technologies as the lithographic scales approach the molecular limit.