This must be an accurate simulation of a whole human brain run on a non-neuronal substrate with at least real-time subject experience (1-1 or better emulation-time to clock-time).
This would probably be silicon-chip-based substrate, perhaps GPU/CPU or analogue neuromorphic chips.
Would need confirmation of high degree of fidelity of neural behavior to human neurons, and high degree of fidelity of neuronal population dynamics.
Does not need to emulation physical behavior below neuron level resolution (e.g. protein complexes like ion channels, or movement of ions).
Does not need to be a faithful replica of a specific human, just within behavioral distribution of neuronal dynamics observed in the general human population.
This requires improvements in measuring neuronal population dynamics in the human population (e.g. by improved brain-computer-interface observations).