Resistion: If it was down to me I would just repeat the step down pulse train and that would give the adavantage of the step down pulse effect and be the equivalent of repeating the maximum pulse. There is a contact email address on the VLSI paper who may be able to provide a more accurate answer. It is email@example.com
The Shanghai group is not being entirely straightforward about monitoring current during the SET down ramp. What happens when the ramp reaches 0 without success? Why not stay at the highest level all the way?