I couldnt access the full text, but that was my impression, too, based on the summary. It appears to work on some analog of hysteresis where the technical balance of energy is maintained but the time scale of restitution is long enough that power can be “siphoned off”. Again - since conservation of energy must be preserved and no matter is created or destroyed, this would serve to reduce the temperature of the graphene. There doesn’t appear to be a scale for their experimental work and whether they’re extracting pico amps or microamps across the (I guessing form the publicly available graphs) 0-0.4 volt potential.
It’s not clear if they’re looking at nominally uniform temperature material which has fluctuations in temperature due to the surroundings, or if they are inducing temperature gradients in the material intentionally to produce the signal. I’m an engineer, not a theoretical physicist, so anyone claiming to end-run the second law of thermodynamics is going to be treated with a bit of skepticism as to the practicality or scalability of this “cheat”.